UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Characterizing the granzyme-perforin pathway and its utility as a cell-to-cell delivery system for cellular… Woodsworth, Daniel 2017

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2017_september_woodsworth_daniel.pdf [ 17.57MB ]
Metadata
JSON: 24-1.0348382.json
JSON-LD: 24-1.0348382-ld.json
RDF/XML (Pretty): 24-1.0348382-rdf.xml
RDF/JSON: 24-1.0348382-rdf.json
Turtle: 24-1.0348382-turtle.txt
N-Triples: 24-1.0348382-rdf-ntriples.txt
Original Record: 24-1.0348382-source.json
Full Text
24-1.0348382-fulltext.txt
Citation
24-1.0348382.ris

Full Text

Characterizing the granzyme-perforin pathway and its utility as acell-to-cell delivery system for cellular therapeuticsbyDaniel WoodsworthBSc (Honours), The University of British Columbia, 2008A THESIS SUBMITTED IN PARTIAL FULFILLMENTOF THE REQUIREMENTS FOR THE DEGREE OFDoctor of PhilosophyinTHE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES(Genome Science and Technology)The University of British Columbia(Vancouver)June 2017c© Daniel Woodsworth, 2017AbstractAlongside small molecules and biologics, cell-based therapies are emerging as a third class of medicaltherapy. Additional sensors, actuators and control circuits would greatly expand the range of functionand application of cellular therapeutics. To this end, a cell-to-cell delivery module has been developedby investigating and re-engineering the granzyme-perforin pathway of cytotoxic lymphocytes. A com-putational biophysical model of this process was developed and implemented using a spatial stochasticsimulation algorithm, which indicated that hindered diffusion in the immune synapse is critical to en-sure reliable granzyme internalization and that large amounts of granzyme escape the synapse, butshould not have toxic effects due to rapid spatiotemporal dilution. Additionally, these results indicatedthat passive diffusion is sufficient for granzyme entry into the target cell, which motivated efforts to usegranzyme as a molecular chaperone to transfer exogenous payloads from effector to target cells. Usinga fluorescent protein payload, the subcellular localization of several granzyme B derived chaperoneswas characterized using fluorescence microscopy, and then their capacity to transfer the payload totarget cells was evaluated in co-culture experiments. The results indicated that the motifs in granzymeB that are required for lytic granule loading are only functional and contiguous in the folded protein.Additionally, these experiments demonstrated that full length granzyme B is a suitable chaperone fordelivering protein payloads to target cells via the granzyme-perforin pathway. Attempts were then madeto use this system to deliver potent orthogonal toxins to apoptosis and lymphocyte resistant tumor cells.A range of granzyme B toxin fusion proteins were constructed, all of which retained toxic activity tovarying degrees. To render target cells resistant to lymphocyte attack both small molecule and proteinbased inhibitors of apoptosis were tested in several cell lines, which delayed cell death, but did not stopit. Using effector target dose response curves, a moderate increase in target cell death was observedin cells targeted by lymphocytes expressing granzyme toxin fusion proteins, as compared to wild typelymphocytes, but the biological significance of this effect is uncertain. Approaches to improve thisgranzyme-perforin mediated delivery system and its therapeutic utility are discussed and explored.iiLay AbstractUsing biological cells as therapeutic devices has great potential. Cells are mobile in the human body,can be genetically programmed to take specific actions in response to environmental signals, and canbe modified to have additional therapeutic functions that improve upon the cells’ natural capabilities.Cytotoxic lymphocytes are components of the immune system that kill infected or malignant cells.These lymphocytes adhere to target cells and release two molecules, granzyme and perforin, into theregion between the two cells, with perforin facilitating granzyme’s entry into the target cell, where-upon granzyme kills the target cell. In this thesis I have taken preliminary steps towards adaptingthis pathway as a cell-to-cell molecular delivery system which could be incorporated into the cellulartherapeutic devices described above. Using both computational biophysical models and experimen-tal implementation, I have provided proof-of-principle that such an approach is feasible, although itstherapeutic utility remains to be demonstrated.iiiPrefaceThe overall project was designed and conducted in collaboration with my supervisor Dr. Robert Holt.Considerations for specific chapters are as follows.Sections of Chapter 1 and 5 were published as below. This paper was primarily written by me,with input and editing by Dr. Holt.Woodsworth, D. J., & Holt, R. A. (2017). Cell-Based Therapeutics: Making a FaustianPact with Biology. Trends in Molecular Medicine, 23(2), 104-115.Chapter 2 was published as below. The conceptual approach was developed by myself, andmy co-authors Valentin Dunsing and my committee member, Dr. Daniel Coombs. I developedthe detailed model, wrote the code for implementation, and generated all figures. The data wasanalyzed by myself and Dr. Coombs. I primarily wrote the manuscript, with input and editingby Dr. Coombs.Woodsworth, D. J., Dunsing, V., & Coombs, D. (2015). Design Parameters for Granzyme-Mediated Cytotoxic Lymphocyte Target-Cell Killing and Specificity. Biophysical Journal,109(3), 477-488.Chapter 3 has been submitted as a manuscript as below. The concept, approach and designwas conducted by myself and Dr. Holt. I performed all experimental work, with the followingexceptions. Lisa Dreolini conducted some of the molecular biology work: plasmid cloning,sequence verification and preparation, as well as western blotting. Libin Abraham aided me inthe preparation of cells for microscopy, and image acquisition. I conducted all data analysis withinput from Dr. Holt. I wrote the manuscript, with input and editing from Dr. Holt.Woodsworth, D. J., Dreolini, L., Abraham, L., & Holt, R.A. Targeted cell-to-cell deliveryof exogenous protein payloads via the granzyme-perforin pathway.Chapter 4 is unpublished. I performed all experimental work in Chapter 4, except that LisaDreolini conducted some of the molecular biology work: plasmid cloning, sequence verificationand preparation. The data was analyzed and interpreted by me, with input from Dr. Holt.ivTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiLay Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiList of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiiiAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Existing cell therapies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.1.1 Regenerative medicine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.1.2 Gene therapy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.1.3 Mesenchymal stem cell therapy . . . . . . . . . . . . . . . . . . . . . . . . . 51.1.4 Adoptive cell therapy in cancer . . . . . . . . . . . . . . . . . . . . . . . . . 61.1.5 Engineered cellular therapeutics . . . . . . . . . . . . . . . . . . . . . . . . . 91.2 Cytotoxic lymphocyte biology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121.2.1 Cytotoxic lymphocytes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121.2.2 The granzyme-perforin pathway . . . . . . . . . . . . . . . . . . . . . . . . . 131.3 Overview of a lymphocyte-based delivery system . . . . . . . . . . . . . . . . . . . . 141.3.1 The granzyme-perforin pathway as a delivery module . . . . . . . . . . . . . . 141.3.2 Comparison with existing systems . . . . . . . . . . . . . . . . . . . . . . . . 151.4 Thesis overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16v2 A computational biophysical model of the granzyme-perforin pathway . . . . . . . . . . 182.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192.2.1 Biophysical model: Geometry & molecular processes . . . . . . . . . . . . . . 192.2.2 Mathematical description: Spatial stochastic simulation algorithm . . . . . . . 212.2.3 Hindered diffusion in the immunological synapse . . . . . . . . . . . . . . . . 232.2.4 Perforin oligomer diffusion and aggregation in the target cell membrane . . . . 242.2.5 Rate of granzyme translocation through perforin pores . . . . . . . . . . . . . 252.2.6 Model parameterization and implementation . . . . . . . . . . . . . . . . . . 262.2.7 Validation of SSSA computational implementation . . . . . . . . . . . . . . . 262.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.3.1 Free diffusion in the synapse is incompatible with granzyme internalization . . 322.3.2 Pore formation is influenced by the amount of perforin released . . . . . . . . 342.3.3 The amount of granzyme internalized depends strongly on the rate of pore for-mation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352.3.4 Hindered diffusion critically influences pore formation and granzyme internal-ization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362.3.5 Dependence of pore formation and granzyme delivery on perforin insertion anddiffusion in the target cell membrane . . . . . . . . . . . . . . . . . . . . . . 382.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383 Targeted cell-to-cell delivery of exogenous protein payloads via the granzyme-perforinpathway . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443.2.1 Screening chaperone designs by assessing lytic granule colocalization usingconfocal microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453.2.2 Transfer of fusion proteins from effector to target cells . . . . . . . . . . . . . 463.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533.4 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573.4.1 Computational identification of N-linked glycosylation motifs . . . . . . . . . 573.4.2 Plasmids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573.4.3 Cell culture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573.4.4 Transfection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 583.4.5 Flow cytometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 583.4.6 Microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 583.4.7 Image analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 593.4.8 Cell labeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59vi3.4.9 Co-culture experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 603.4.10 Statistical analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 603.4.11 Western blotting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 603.4.12 Crystal structure analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 614 Efforts to use granzyme-perforin mediated delivery of orthogonal toxins to enhancecytotoxic lymphocyte killing of apoptosis resistant tumour cells . . . . . . . . . . . . . . 624.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 624.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 654.2.1 Development of orthogonal granzyme-toxin fusions for enhancing lymphocytemediated cytotoxicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 654.2.2 Characterizing the lymphocyte resistance of the breast cancer cell line MCF-7 . 694.2.3 Efforts towards generating a lymphocyte resistant cell line . . . . . . . . . . . 714.2.4 Effector dose response curves as a means of resolving small increases in YTtarget cell killing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 754.2.5 Validating the GZB-HTK and GZB-NFSA toxin fusion proteins in MCF-7s . . 764.2.6 Predicted and measured enhancements of YT-killing of MCF-7s using granzyme-toxin fusion proteins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 774.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 824.4 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 864.4.1 Plasmids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 864.4.2 Cell culture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 874.4.3 Transfection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 874.4.4 Flow cytometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 894.4.5 Metabolic activity assay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 894.4.6 Direct toxin expression experiments . . . . . . . . . . . . . . . . . . . . . . . 894.4.7 Cell labeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 904.4.8 Cytotoxic lymphocyte co-culture experiments . . . . . . . . . . . . . . . . . . 904.4.9 Isolation of YT Targeted cells using the FRET reporter . . . . . . . . . . . . . 904.4.10 MCF-7 cell characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . 914.4.11 Lymphocyte resistance experiments . . . . . . . . . . . . . . . . . . . . . . . 914.4.12 Enhancement of YT-Indy killing experiments . . . . . . . . . . . . . . . . . . 924.4.13 Drug reconstitution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 924.4.14 Statistical analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 925 Discussion, conclusions and future directions . . . . . . . . . . . . . . . . . . . . . . . . 945.1 The utility of the granzyme-perforin pathway as a delivery system . . . . . . . . . . . 955.2 Further efforts to demonstrate the toxin-mediated enhancement of lymphocyte killing . 985.2.1 Development of lymphocyte resistant target cells . . . . . . . . . . . . . . . . 98vii5.2.2 Improving fusion protein granule loading and delivery . . . . . . . . . . . . . 995.3 Future directions for granzyme-perforin delivery systems . . . . . . . . . . . . . . . . 1005.4 Broader insights into cellular therapeutics . . . . . . . . . . . . . . . . . . . . . . . . 1025.4.1 Context dependencies in cellular therapeutics . . . . . . . . . . . . . . . . . . 1035.4.2 A framework for cell engineering . . . . . . . . . . . . . . . . . . . . . . . . 1045.4.3 Grand challenges for cell engineering . . . . . . . . . . . . . . . . . . . . . . 1055.4.4 The future of cellular therapeutics . . . . . . . . . . . . . . . . . . . . . . . . 109Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110A Generating granzyme B mCherry fusion proteins expressed from genomic granzyme Blocus using CRISPR/Cas9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131A.1 Design of guide RNAs and donor template . . . . . . . . . . . . . . . . . . . . . . . . 134A.2 pDN_GZB_E5-MCH plasmid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137A.2.1 Plasmid map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137A.2.2 Plasmid sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138B Mass spectrometry based investigation of granzyme B mCherry fusion protein regionsof instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141C Generation of a mutant EEF2 gene which protects cells from DTA when the two areco-expressed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146D Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149D.1 Core MATLAB scripts implementing computational biophysical model (Chapter 2) . . . 149D.1.1 IMS_Ex.m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149D.1.2 buildQ.m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164D.1.3 calcDiffPropens.m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164D.1.4 calcInterxnPropens.m . . . . . . . . . . . . . . . . . . . . . . . . . . 165D.1.5 percDown.m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166D.1.6 Qsort.m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167D.1.7 swap.m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167D.1.8 updateQ.m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167D.2 Image filtration and colocalization analysis: MATLAB source code (Chapter 3) . . . . . 168E Plasmids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172E.1 Granzyme B derived chaperone-mCherry fusion protein plasmids (Chapter 3) . . . . . 172E.1.1 Base pdL plasmid map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172E.1.2 Base pdL plasmid sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . 173E.1.3 Coding sequence inserts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176viiiE.2 Granzyme B-toxin fusion protein plasmids and IAP-FRET plasmids (Chapter 4) . . . . 177E.2.1 Base pdL plasmid map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177E.2.2 Base pdL plasmid sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . 178E.2.3 Base pMND plasmid map . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182E.2.4 Base pMND plasmid sequence . . . . . . . . . . . . . . . . . . . . . . . . . . 183E.2.5 Source of component parts . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187E.2.6 Coding sequence inserts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187ixList of TablesTable 2.1 Parameters used in the immune synapse computational model. . . . . . . . . . . . . 27Table 4.1 Fitted EC50 values for enhanced lymphocyte killing . . . . . . . . . . . . . . . . . 81Table E.1 Source of component parts of pdL vectors . . . . . . . . . . . . . . . . . . . . . . 187xList of FiguresFigure 2.1 Model geometry and molecular interactions . . . . . . . . . . . . . . . . . . . . . 20Figure 2.2 Mass action kinetics are accurately captured by the SSSA algorithm . . . . . . . . 28Figure 2.3 Free diffusion described by the two-dimensional diffusion equation compares wellwith that simulated by the SSSA algorithm . . . . . . . . . . . . . . . . . . . . . 30Figure 2.4 Free diffusion with attenuation due to target cell membrane insertion is well de-scribed by the SSSA algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31Figure 2.5 Diffusion in the target cell membrane diffusion described by the two-dimensionaldiffusion equation compares well with that simulated by the SSSA algorithm . . . 32Figure 2.6 A two-dimensional reaction-diffusion system is well described by our SSSA . . . 33Figure 2.7 Time evolution of GZB and PFN in the immunological synapse without hindereddiffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34Figure 2.8 Effect of the amount of PFN released on pore formation probability and GZB in-ternalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35Figure 2.9 Dependence of GZB internalization on rate of pore formation . . . . . . . . . . . 36Figure 2.10 Importance of hindered diffusion in pore formation and GZB internalization . . . . 37Figure 2.11 Reduced diffusion of perforin in the target cell membrane reduces pore formationand GZB internalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39Figure 2.12 Impact of the rate of perforin insertion into the target cell membrane on pore for-mation and granzyme internalization . . . . . . . . . . . . . . . . . . . . . . . . . 40Figure 3.1 Design of payload delivery module chaperones . . . . . . . . . . . . . . . . . . . 45Figure 3.2 Subcellular distribution of candidate chaperone-mCherry fusion proteins . . . . . . 47Figure 3.3 Automated image noise filtering pipeline . . . . . . . . . . . . . . . . . . . . . . 48Figure 3.4 Quantitative assessment of candidate chaperone colocalization with lytic granules . 49Figure 3.5 Transfer of granzyme B mCherry fusion protein to target cells . . . . . . . . . . . 50Figure 3.6 Comparison of MCH payload transfer to target cells by the two granzyme B derivedchaperones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51Figure 3.7 Western blot confirmation of GZB-MCH fusion protein transfer to target cells . . . 52Figure 3.8 Spatial context of putative N-linked glycosylation sites in granzyme B . . . . . . . 54xiFigure 4.1 Granzyme toxin fusion protein design . . . . . . . . . . . . . . . . . . . . . . . . 65Figure 4.2 GZB-DTA and GZB-PEA testing in Hela cells . . . . . . . . . . . . . . . . . . . 66Figure 4.3 GZB-DTA testing in 293T cells . . . . . . . . . . . . . . . . . . . . . . . . . . . 67Figure 4.4 GZB-HTK testing in 293T cells . . . . . . . . . . . . . . . . . . . . . . . . . . . 68Figure 4.5 GZB-NFSA AND GZB-NFSB testing in 293T cells . . . . . . . . . . . . . . . . . 69Figure 4.6 Characterizing YT delivery of GZB to MCF-7s . . . . . . . . . . . . . . . . . . . 71Figure 4.7 Long term survival of YT-targeted MCF-7s . . . . . . . . . . . . . . . . . . . . . 72Figure 4.8 Small molecule inhibition of YT killing of 721 target cells . . . . . . . . . . . . . 73Figure 4.9 Long term viability of targets co-cultured with YTs in the presence of apoptosisinhibitors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74Figure 4.10 Long term viability of YT-targeted target cells expressing IAPs . . . . . . . . . . . 75Figure 4.11 YT effector:target dose response of MCF-7s . . . . . . . . . . . . . . . . . . . . . 76Figure 4.12 Testing GZB-HTK and GZB-NFSA toxin/prodrug systems in MCF-7s . . . . . . . 77Figure 4.13 Estimates of toxin/prodrug enhancement of YT killing . . . . . . . . . . . . . . . 80Figure 4.14 Investigating the enhancement of YT killing of MCF-7s by GZB-NFSA/CB1954 . 83Figure 4.15 Comparing estimates with experimental measurements of enhancement of YT killing 84Figure A.1 Molecular characterization of genomically expressed granzyme B mCherry fusionproteins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133Figure A.2 pDN_GZB_E5-MCH plasmid . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137Figure B.1 Western blot and gel of samples used for mass spectrometry experiments . . . . . 142Figure B.2 Mass spectrometry identified tryptic peptides from whole cell lysates . . . . . . . 143Figure B.3 Mass spectrometry identified non-tryptic peptides from whole cell lysates . . . . . 145Figure C.1 Evidence that mEEF2 co-expression with GZB-DTA restores protein synthesisfunction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148Figure E.1 Base pdL vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172Figure E.2 Base pdL vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177Figure E.3 Base pMND vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182xiiList of AbbreviationsAAV Adeno-associated virusACT Adoptive cell therapyALL Acute lymphoblastic leukemiaBID BH3 interacting-domain death agonistBMT Bone marrow transplantCAR Chimeric antigen receptorCFP Cerulean fluorescent proteinCFSE Carboxyfluorescein succinimidyl esterCL Cytotoxic lymphocyteCRISPR Clustered regularly spaced palindromic repeatsCTL Cytotoxic T-lymphocyteDAPI 4DCI DichloroisocoumarinDMSO Dimethyl sulfoxideDTA Diphtheria toxin A fragmentE:T Effector:targetER Endoplasmic reticulumERSS Endoplasmic reticulum signal sequenceESC Embryonic stem cellFACS Fluorescence activated cell sortingFITC Fluorescein isothiocyanateFRET Forster resonance energy transferFSC Forward scatterGCV GanciclovirGFP Green fluorescent proteinGILT Glycosylation independent lysosomal targetingGSL Glycine-serine linkerGZB Granzyme BxiiiGZBSM Granzyme B sort motifGZBSS Granzyme B signal sequenceHDR Homology directed repairHIV Human immunodeficiency virusHLA Human leukocyte antigenHSC Hematopoietic stem cellHSCT Hematopoietic stem cell transplantHTK Herpes simplex thymidine kinaseIAP Inhibitor of apoptosis proteinIFN InterferoniPSC Induced pluripotent stem cellIS Immunological synapseKAR Killer activating receptorKIR Killer inhibiting receptorLG Lytic granuleMAD Median absolute deviationMCH mCherryMFI Median fluorescent intensityMHC Major histocompatability complexMSC Mesenchymal stem cellMTOC Microtubule-organizing centerNFAT Nuclear factor of activated T-cellsNHEJ Non-homologous end joiningNK Natural killer (cell)NTR NitroreductaseODE Ordinary differential equationPB PrestoBluePBS Phosphate buffered salinePCC Pearson’s correlation coefficientPCR Polymerase chain reactionPDE Partial differential equationPEA Pseudomonas exotoxin A fragmentPFN PerforinPI Propidium iodideRFP Red fluorescent proteinscFV Single chain variable fragmentSCID Severe combined immunodeficiencySELEX Systematic evolution of ligands by exponential enrichmentxivSIN Self inactivatingSMAC Supramolecular activation complexSSC Side scatterSSSA Spatial stochastic simulation algorithmSTS StaurosporineSV SubvolumeTALEN Trancription activator-like effector nucleaseTCR T-cell receptorTGF Transforming growth factorTIL Tumour infiltrating lymphocyteTNF Tumour necrosis factorTOX ToxinVC Vehicle controlXIAP X-linked inhibitor of apoptosis proteinYFP Yellow fluorescent proteinZFN Zinc-finger nucleasexvAcknowledgmentsFirst and foremost, I would like to thank my supervisor Dr. Robert Holt. I walked into his office with alot of ambitious ideas about engineering biology, and literally no idea what that looked like. More thananything, I thank him for looking past that and taking me on as his student. I am very grateful for thelatitude he gave me to explore the many half formed ideas I had and have, and the patient and helpfulanswers to my near incessant questions and interruptions, especially near the beginning of my PhD.Rob gave me a unique combination of freedom and access to him that I feel extremely fortunate to havereceived. I also greatly respect the broad scope of his scientific ideas and interests, and his willingnessto embark upon ambitious endeavors. The project that he and I eventually built would not have beenpossible without both of these traits. Some of the most enjoyable moments of my time in graduateschool were sitting in his office, brainstorming or sketching out new ambitious ideas or applications—Iwish we had the time and resources to pursue all of them. As I leave that office, I feel that he has beeninstrumental in giving me an approach with which to attempt to realize some of those ideas. My onlyregret is that we never got out for a bike ride.I would also like to thank Dr. Daniel Coombs. In addition to sitting on my committee, we workedtogether to create the biophysical model of the immune synapse. I very much enjoyed this part of mytime in graduate school. Again, some of my best memories are of standing in front of his chalk boardtrying to fit a few more symbols between the forest of equations already there. While I am obviouslyno mathematician, Dan was good enough to overlook that and I am grateful for our collaboration—whatever mathematical skill I retain is largely thanks to him. It has been a long time since Math 215,but I still remember the important lesson that a cooked turkey should be above room temperature.I would like to thank the other members of my supervisory committee, Dr. Jonathan Choy, Dr.Phil Hieter, and Dr. Kelly McNagny for their guidance and suggestions, all of which were helpful, andmany of which led to important additions to my project. Their patience with my often tight timelinesis greatly appreciated. Dr. Eric Yung, a staff scientist in our lab has also helped me along much of myPhD. I appreciate his early efforts to construct some viral constructs, and teach me how to do so, aswell as his willingness to hassle suppliers (although I think he may have enjoyed that).Our lab has recently expanded, but there was a time when it was very small, and Scott Brown,Govinda Sharma and I have been together since then. Thanks to both of them for their advice, supportand time spent kicking ideas around. Most of all, thanks to them for laughing. Graduate school is axvilong road with many ups and downs, and they have both provided a combination of support and levity,as friends and colleagues. Kyla Cochrane and Chris May are new additions, but I have very muchenjoyed our time together, and will be sad to leave them. Along with Jerry Tien, that student room hasbecome a boisterous place, and while productivity may not be optimal, the crosswords, jokes, politicalconversations, movie quotes and occasional bits of science are totally worth it. Payal Sipahimalanidoes not sit in that student room, which is probably best for her sanity and productivity, which in turnbenefits the rest of us. Without her patience and efficiency, many of us would be lost, broke, or haveno access to our own data. Sadly this means I never got to see as much of her as I would have liked.Finally, I would like to especially thank Lisa Dreolini. She was immensely helpful in jumpstartingmy project at a critical time. This would be a very different thesis without her efforts. Beyond that,I will miss our coffee conversations—at times interesting, at times hilarious. Her incredibly cheerfuldemeanor, ready laugh, friendship and support have meant a lot to me.Five years is a long time, and I have many others to acknowledge: other and former members of theHolt lab, faculty in our building and others, and the many other students and staff that I have workedwith or chatted with over the years. I especially thank the various members of the other labs at the GSC,many of whom I will miss. I would also like to thank the many good teachers I have had throughoutmy life, who have each contributed in some way to this. I think they are under appreciated in the rolethey play in getting us to a jumping off point.Finally, I would like to thank my parents, Anne and Bob, and my sister Alexandra. Saying this isthe first time I ever needed them would belie the incredible upbringing, nurturing, growth, education,guidance and love they have given me. Saying this is the first time I ever realized how much I neededthem would be closer to the truth. I am very grateful and fortunate to have such a family, and I thankthem.xviiChapter 1IntroductionThe ability to deploy active therapeutic devices capable of engaging directly with the fundamentalcellular and molecular causes of disease will be transformative for medicine. Using biological cellsas the chassis for these devices has three fundamental advantages. First, cells maintain a homeostaticenvironment distinct from their surroundings, and integrate a wide variety of input signals to executecontext-dependent actions. Second, there is incredible phenotypic and functional diversity across thevarious cell-types. Finally, this diversity is encoded in the genome of each cell, making it possible toencode logic in these cells in the form of additional, modified or deleted nucleic acid sequence. Thus,each of the distinct cell types in the human body is a potential basic chassis and platform from whichto build a tailored cellular therapeutic.However, this ecosystem is not necessarily optimal or functionally complete. Additional cellularsensors, effectors, and layers of control logic will be required to truly realize the potential for cell basedtherapies. Rather than de novo construction, these components can be best obtained by successfullyleveraging the unique and varied functions of existing biological molecules or pathways. By altering,adding or removing elements of these processes, novel sensory and effector components can be gener-ated that fill existing gaps in a cell’s endogenous set of biological functions. These new components canthen be inserted into a cellular chassis to create a cellular therapeutic with functions that are a compositeof the original chassis and the new component derived from a repurposed pathway or molecule.This thesis represents steps towards this vision. I have attempted to repurpose the granzyme-perforin pathway of cytotoxic lymphocytes as a cell-to-cell delivery module for therapeutic proteins.The overarching goal is that by using this pathway to deliver therapeutics, this module could be insertedin cytotoxic lymphocytes that are targeted to diseased tissue via lymphocyte surface receptors to yielda targeted and specific cellular delivery device, capable of trafficking throughout the body to find thedesired site of disease, and deliver, in a cell-specific manner, a therapeutic to treat that disease.11.1 Existing cell therapiesThe history of cellular therapies is mainly rooted in the stem-cell research conducted over the last 60years. This led to first bone marrow transplants, followed by hematopoietic stem cell transplants fromother stem cell sources, for treatment of various hematological disorders and malignancies. The adventof immunosuppressive drugs allowed the therapeutic transplant of cells from other tissues, as well asbulk organ or tissue transplants. With the advent of recombinant DNA technology, the possibility ofmodifying cells ex vivo prior to administration emerged. These possibilities continue to expand withcheap and rapid DNA synthesis, improved methods for DNA delivery to a variety cell types, and,recently, rapid and facile methods for editing a cells genome with single base pair resolution. As aresult, incorporating existing biological function into new cellular therapeutic devices is an approachthat is gaining traction and application across a range of human disease.1.1.1 Regenerative medicineCurrent methods for treating diseases arising from dysfunctional or dying cells rely on: (i) systemicadministration of therapeutics with often-poor specificity, (ii) surgical resection, or (iii) bulk tissue ororgan transplantation. Rather than attempting to ’fix’ these dysfunctional cells, the approach of re-generative medicine is to simply replace them with fresh, functionally identical cells, new ’parts’ thatretain appropriate function. The first challenge is obtaining these replacement cells. In a few instancesthey may be harvested from allogeneic donors, as in the case of Type I diabetes, where infusion ofunmodified, donor-derived beta-islet cells is now a well-tested and often effective approach [1, 2], al-though achieving stable engraftment is still a substantial challenge [3]. More generally, supplementingtissue stem cell compartments with unmodified, donor-derived stem cells, including those of humanembryonic origin, continues to be explored as a promising therapeutic approach for a broad array ofpathologies, including neurological, hepatic, endocrine, and musculoskeletal disorders [4].The recently developed methods for genetically reprogramming terminally differentiated autolo-gous cells into pluripotent stem cells (iPSCs) [5], together with in vitro techniques for re-differentiatingthese iPSCs into a variety of mature, replacement cell types and tissues is bringing new possibilities toregenerative medicine and tissue engineering [6]. This technology is allowing generation of both stemcells and differentiated cell types from a potentially sick patient that may not have stem cells or differ-entiated cells available for harvesting, and for whom a suitable donor may not be available. Importantlythese iPSC-derived cell products are autologous and therefore more likely to be immunologically com-patible with the recipient. Furthermore, the ex vivo manipulations necessary for the derivation andre-differentiation of iPSCs provide new opportunities for additional genetic modification to enhancetheir therapeutic properties. For example, engineering replacement iPSC-derived beta-islet cells forimmuno-resistance may reduce their sensitivity to the autoimmune mechanisms that eliminated theirnatural predecessors, making them more therapeutically relevant than unmodified replacement cellswould be. Additional examples from the field of tissue engineering include generating a liver bud with2appropriate vascularization and three dimensional architecture [7], as well as recellularizing a decellur-ized heart scaffold with the aim of generating personalized whole organs for transplant [8]. Throughoutregenerative medicine, major challenges remain, including sources of universal allogeneic cells, stableengraftment, avoiding rejection, and maintaining the viability of the graft over the long term [9].1.1.2 Gene therapyGene therapy can be broadly defined as the delivery of genetic material to a patient’s cells with the goalof modifying the genetic makeup of the cells for a therapeutic benefit. With exceptions, it has largelyfocused on inserting a working copy of a gene into a cell population in which the gene is damagedor absent. Functionally, it can be divided into two main approaches: gene delivery in vivo or ex vivo.The former approach typically relies upon the use of viral vectors as delivery platforms (although otherapproaches have been used including naked DNA). The former is challenging since it requires nearperfect control of tissue specificity, either through viral tropism or tissues-specific promoters or otherregulatory elements. Furthermore, post-insertion quality control is impossible. Finally, physicallydelivering sufficient viral vector to the diseased tissue is often difficult. Therefore, it is not surprisingthat the field of in vivo gene therapy has progressed most in diseases which are single gene disordersthat occur in tissues where physical viral delivery is more anatomically accessible: notably the eye andliver [10].While the original, and likely still ultimate goal of gene therapy was and is in vivo delivery, in manycases the technical considerations listed above have necessitated ex vivo gene delivery into autologouspatient-derived cells, followed by transplantation of these cells back into the patient.Due to the accessibility of the starting cell population from either peripheral blood or bone marrow,and the relative experience and familiarity that clinicians had developed from bone marrow transplantprograms, gene therapy using hematopoietic stem cells (HSC) was the first, and is therefore the mostmature, form of cell-based gene therapy [11]. Initial focus was on single gene disorders, mainly primaryimmune deficiencies (severe combined immunodeficiency disorder and Wiskott-Aldrich syndrome), aswell as neurodegenerative storage disorders (adrenoleukodystrophy and metachromatic leukodystro-phy) [12]. Early successes were reported in the treatment of pediatric patients with X-linked severecombined immunodeficiency (SCID-X1) [13]. These patients suffer from a total lack of lymphocytes,with concomitant susceptibility to infection. The underlying cause of the disease is a deficiency forthe IL2RG gene, which codes for the gamma chain of several interleukin receptors that are necessaryfor lymphocyte development. Patients received CD34+ stem cells that had been transduced with aretro-viral vector carrying a functioning IL2RG gene. Initial reports were very promising, with thereappearance of lymphocytes in most children [13]. Unfortunately however, several patients devel-oped leukemia [14]. This was determined to be due to insertional oncogenesis via viral vector insertedlong-tandem repeat (LTR) activation of adjacent oncogenes. While these results significantly slowedthe progress of gene therapy, several factors have led to a recent increase in enthusiasm. The successof self-inactivating (SIN) lentiviral vectors is likely to significantly increase the safety profile of these3therapies by reducing the likelihood of clonal expansion of a transduced cell. Lentiviral vectors havenow been employed across a similar range of immunodeficiency diseases with good efficacy, and littleevidence for clonal expansion [12]. They have been further used in the treatment of metachromaticleukodystrophy [15] and β -thalassemia [16], although oligoclonal expansion was reported in the latter.Thus the current status of gene therapy is still in flux. Improvements in vector design have renewedenthusiasm for the field, but safety concerns remain. Gene-editing technologies, consisting of zinc-finger nucleases (ZFNs), TALENs and the CRISPR-Cas9 system have the potential to address someof these issues, by virtue of their ability to target DNA at a specific locus. This opens the possibilitygene-correction of the endogenous diseased gene, rather than gene replacement by addition.These gene editing systems all revolve around fusion proteins consisting of a DNA recognitiondomain joined to a nuclease domain. Unlike meganucleases or restriction endonucleases, the DNArecognition domains are programmable. ZFN technology was the first to be developed. The DNAbinding domain consists of zinc fingers, which are ubiquitous protein domains that bind to short nucleicacid sequences [17]. By modifying small numbers of residues in the alpha helix of Cys2His2 zincfingers, the triplet nucleotide sequence that the zinc finger binds to can be altered [18]. Chainingtogether multiple zinc fingers of varying triplet specificity results in a protein which binds to a defined,extended DNA sequence. This zinc finger domain is fused to the non-specific nuclease domain of theFokI enzyme.Transcription activator-like effector nucleases (TALENs) are similar to ZFNs, with the main dif-ference being that TALENs can bind arbitrary DNA sequence, and they are more modular. Each tran-scription activator-like effector (TALE) is a 36 amino acid motif that binds a single nucleotide, with thespecificity of the TALE defined by a dipeptide at the protein-DNA interaction site [19]. Thus TALEscan be predictably strung together to target arbitrary nucleotide sequences. Furthermore, chainingTALEs is a fairly reliable process, unlike ZFNs whose DNA binding function typically require proteinlevel optimization to resolve adjacent zinc finger interactions [19]. As with ZFNs, the TALE domain isfused to a non-specific FokI nuclease domain.The clustered regularly interspaced short palindromic repeats (CRISPR) cas9 system is the newestgene editing technology. The cas9 protein complexes with a guide RNA (gRNA, which is a chimeraof the two separate RNAs found in the prokaryotic endogenous system, consisting of a cas9 bindingmotif and a DNA bas-pairing motif), which imparts the DNA binding specificity to the cas9 protein byforming a DNA:RNA heteroduplex, after which cas9 cuts the adjacent DNA region [20]. Targeting is asimple function of Watson-Crick base-pairing. The other great advantage of the CRISPR/cas9 systemis that no protein engineering is required to target different loci: only new gRNAs are required. Thesefeatures allow for robust predictable design, rapid feedback loops, and multiplex or library scale genetargeting [21].Regardless of the gene editing system, after the fusion proteins are expressed in target cells viatransfection or transduction, the DNA recognition domain binds in a sequence specific manner to itsmatching target DNA sequence, and then the fused nuclease introduces a double stranded break in4an adjacent, and predictable, region of the DNA. (Extensions of this approach that use nickases tointroduce single stranded breaks are similar [22].) The double stranded break is then repaired by non-homologous end joining (NHEJ), and, if a repair template is present, homology directed repair (HDR)[23]. NHEJ is nonspecific and the end-processing of the two DNA ends results in nucleotide insertionsor deletions (indels), making this approach suitable for knocking out genes [24]. If a DNA template ispresent that has homology to the two cut ends, the HDR pathway is also active [25], which results inthe incorporation of the homology regions in the template, as well as any additional sequence betweenthese two regions. This makes possible the introduction of additional sequence, either simple nucleotidemodifications, or the insertions of whole genes [26].These capabilities may be exploited in several ways. First it could simply be used to insert a geneat a known, ’safe harbor’ location, for example the AAVS1 site, which would decrease the chances ofgene activation or insertional oncogenesis [27]. Zinc fingers have also been used to correct the IL2RGgene in HSCs derived from a SCID-X1 patient[28]. Finally, gene therapy need not be limited to theaddition or correction of genes, it can also be used in the context of removing genes for therapeuticpurposes. Gene-editing of CD4 T-cells to knock out the CCR5 receptor used by HIV to gain entry toCD4 cells has even progressed to clinical trials, which demonstrated safety and improved viral controlupon temporary antiretroviral withdrawal [29]. Genome editing still has outstanding questions thatneed to be answered prior to widespread therapeutic application, most importantly the frequency andimpact of off-target activity, which is a known issue across all genome editing platforms [30–33]. Inthe case of gene correction, editing efficiency is also quite low, which poses a major challenge for invivo gene-editing.More broadly, in both traditional viral gene therapy, as well as newer non-viral gene therapy andgenome editing strategies, additional challenges remain. These include: (i) accurate and efficient tar-geting and delivery of the genes to diseased cells; (ii) achieving modification or delivery to a sufficientfraction of cells for therapeutic effect; and (iii) avoiding iatrogenic effects, primarily insertional onco-genesis in the case of gene therapy and off-target editing in the case of gene-editing.1.1.3 Mesenchymal stem cell therapyMesenchymal stem cells (MSCs) are emerging as promising therapeutic candidates for many diseases,due to MSCs’ immunomodulatory effects, natural tropism for tumors and other sites of inflammation,multipotency, and relative ease of use. Both naturally, and following therapeutic implantation at sitesof disease, MSCs have been observed to participate in tissue repair and regeneration in damaged ordegenerative tissue, as well as induce a return to immuno-homeostasis in autoimmune diseases [34].Hundreds of clinical trials are currently underway, with particular focus on two main applications:regeneration of bone-related injured tissue, as well as promoting an anti-inflammatory response [35].In many cases their natural therapeutic properties have been augmented by genetic modificationsthat improve MSCs’ homing to sites of disease [36], as well as their therapeutic activity once at thesesites, including the heart [37], brain [38], and in tumors [39, 40] (which have now progressed to clinical5trials [41]).1.1.4 Adoptive cell therapy in cancerBased on epidemiological evidence such as an increased incidence of some types of cancer in immuno-suppressed transplant patients, the importance of the immune system in controlling and eradicatingtumours has been acknowledged for some time [42, 43]. The observation that tumour infiltrating lym-phocytes (TIL) could be expanded from surgically resected tumours, and that these TIL were reactiveagainst tumour cells in vitro spurred initial interest in the therapeutic utility of these cells [44]. Thisobservation was first exploited clinically by Rosenberg and colleagues: autologous lymphocytes fromsurgically resected tumour sections were expanded ex vivo by culture with high dose IL-2, and reinfusedinto melanoma patients [45]. As the approach has been refined, and in particular with the addition ofprior lymphodepleting chemotherapy, the outcomes have improved—especially in melanoma patients,with a recent clinical trial reporting complete remission in 22% of patients with metastatic melanoma[46].TIL therapy has several limitations. First, not all cancers have such a heavy mutational burden asmelanoma, which limits the number of tumour neo-epitopes against which TIL have the potential toreact, and thus limits the presence of TIL itself [47, 48]. Second, a suitable source of lymphocytes toexpand is required, which is not always possible, either due to the nature of the cancer, or because theanatomical location of the tumour makes it unresectable [49]. Furthermore even in cases where a resec-tion is available, only 30-40% of biopsies yield suitable T-cell populations [50]. Third, in many casesTIL are in a state of near total exhaustion and anergy [51]. This is due to a host of factors, including theimmunosuppressive tumour microenvironment (further discussed below), as well as exhaustion due torepeated antigenic stimulus [52, 53]. This makes them challenging to expand ex vivo and of dubiousutility and efficacy once administered back into the patient [50].To address these issues, and to make adoptive transfer of T-cells a viable therapy for a wider rangeof cancers, it would be desirable to use cytotoxic lymphocytes isolated from peripheral blood as thestarting source from which to expand T-cells. This approach would in theory allow for the generationof younger, less terminally differentiated and anergic T-cells, and would not be dependent on the avail-ability of TIL, the tumour resection feasibility and quality, or on the successful expansion of a smallstarting population of tumour reactive TIL to clinically useable numbers [54]. However, this approachintroduces a new challenge: in general, T-cells from peripheral blood are highly polyclonal, with onlya small fraction having specificity for malignant cells [55].A solution to this problem is to genetically modify the T-cells with an additional surface receptorspecific for the tumour. This was first attempted by Rosenberg and colleagues using a MART-1 specificT-cell receptor (TCR) in melanoma patients [56]. TCR targeting allows for administered T cells totarget intracellularly derived antigens presented on the tumour cell surface in the context of a peptide-MHC (major histocompatibility complex). The great advantage of this approach is that the potentialtarget antigens encompass, in theory, the entire peptidome of the cell, thus increasing the theoretical6likelihood of finding a tumor specific antigen. There are however, several disadvantages. First, aTCR with appropriate reactivity is required, and current methods for TCR screening and discoveryare relatively low-throughput [57]. Second, since TCRs recognize a peptide-MHC complex, a givenTCR is only suitable for the subset of patients who have the appropriate HLA-type. Third, a keymechanism of immune escape by tumours is MHC-downregulation, and in this case tumours may berelatively invisible to TCR-targeted T-cells [58]. Finally, there is the potential for generating a chimericTCR resulting from pairing between the original endogenous α-chain and the inserted β chain (orvice versa). This has the potential to create a TCR with unknown reactivity, possibly targeting self-antigens and resulting in graft-versus host effects. While this has been observed to cause lethal toxicityin mice [59], it has not been in humans [60]. Despite these challenges, TCR-targeted T-cells have beensuccessfully used in a recent clinical trial using an NY-ESO-1 specific TCR in sarcoma and melanomapatients. While the results were positive, the improvement in survival was not dramatic [61].The main alternative to TCR targeting is the use of chimeric antigen receptors (CARs). These aresynthetic receptors consisting of an extra-cellular single chain variable fragment (scFV) fused to anintracellular domain consisting of various T-cell signaling components [62]. The scFV is a fusion ofthe variable region of the heavy and light chains of antibody, which retains the specificity of its parentalmolecule [63], which is selected to react against a surface expressed tumour antigen. The intracellulardomains in the initial CAR designs consisted of a transmembrane and hinge region domain, a CD3ζdomain, with the latter serving to provide stimulatory intracellular signaling, similar that of CD3ζ in anendogenous TCR complex [64]. The coupling of epitope recognition with initiation of TCR signalingresults in the modified T-cells targeting any malignant cell expressing the cognate antigen for the scFV.Since they were first reported [65], the design of these receptors has been improved upon significantly,most notably by the addition of either the CD28 or 4-1BB costimulatory domains, which has improvedthe CAR-mediated stimulus of T-cell proliferation and persistence [66].Using CARs to target T-cells to tumours has several advantages. The targeting is MHC indepen-dent, meaning a given CAR is usable in all patients, it is unaffected by MHC downregulation, andtargeting non-peptide epitopes, such as post-translationally modified proteins, lipids or carbohydrates,is feasible [67]. Furthermore, methods for scFV development to target a given antigen are relativelymature [63]. However, this presupposes the existence of a surface expressed tumour-specific targetantigen for which a reactive antibody exists. Unfortunately, this excludes two major classes of tumourspecific antigens: cancer/testis antigens and tumour neoantigens [68], thus greatly constricting the setof potential target antigens. Since these proteins are largely intracellular, the only way in which anypart of them is found on the surface of a tumour cell ‘ is as small MHC-presented peptides. Effortsto develop antibodies recognizing peptide-MHC complexes were initially unsuccessful in that the an-tibodies were primarily reactive to the MHC itself (and thus potentially every patient cell), rather thanthe peptide-MHC complex [69]. Recently some success has been made in developing a peptide-MHCspecific antibody, although porting it to a CAR eliminated its epitope specificity, which had to be re-paired by lowering its affinity via rational mutation [70]. If these issues can be resolved, CAR targeting7of peptide-MHC would have the potential to greatly expand the range of cancers in which CAR T-celltherapy might be employed.Clinically, CAR T-cell therapy has received the most attention for its use in hematological ma-lignancies [71–73]. The CAR used in all cases targets the B-cell surface marker CD-19, resultingin B-cell aplasia and agammaglobulinemia which necessitates immunoglobulin transfusion [74]. Thespecifics of generating the CAR-T cellular product vary across the various academic centers that arecurrently investigating this therapy, but the general approach is as follows. A population of peripheralblood derived T-cells is selected for modification, which may be all T-cells, or skewed towards mem-ory, naive, CD4 or CD8 compartments. The relative makeup of the starting population of cells hasbeen shown to be of significant importance [75]. While the optimal distribution is by no means clear,recent work suggests that equal amounts of CD4 and CD8 T-cells from less differentiated subsets withgreater proliferative capacity, such as naive or central memory cells, yield the best in vivo anti-tumouractivity [75–78]. This starting population are then modified to express the CAR, typically using a retroor lentiviral vector, although other methods including transposons, and RNA transfection have beenused. Modified cells are then activated and expanded using either agonistic antibodies against CD3and CD28, or using irradiated feeder cell populations that express the cognate antigen for the CAR. Avariety of proliferative interleukins are also included in the expansion, usually at least IL-2, althoughmany others have been studied, most notably IL-7 and IL-15 [64]. Finally, the cell product is infusedinto the original (now lymphodepleted) patient from whom the cells were collected.Optimizing the CAR design, ex vivo expansion protocols, and patient lymphodepletion preparativeregimens has required substantial investment, but has paid off in the last five years, with CAR-T therapyin hematological malignancy achieving response rates that are truly spectacular, ranging from 70-90%complete remissions [76, 79–81]. With this level of efficacy, there is understandable enthusiasm forextending these results to other cancer types. Currently there are clinical trials underway using CARsthat target a variety of antigens and corresponding tumour sites, including prostate specific membraneantigen (PSMA, prostate cancer), mesothelin (pancreatic cancer and mesothelioma, among others),GD2 (a ganglioside, used in neuroblastoma) and the oncogene HER2 (glioblastoma and sarcoma amongothers) [67].In extending TCR- or CAR-targeted T-cell therapy to solid tumours, numerous additional obstacleswill be encountered. Tumour specific antigens will be required, and despite significant investment thereare still few validated targets about which the community is confident [82, 83]. Furthermore, on-targetoff-tumour toxicity, with outcomes as severe as patient fatality, remains an ongoing issue [84]. Perhapsmost importantly, the ability of CAR-T cells to overcome the substantial capacity for immune evasionand suppression exhibited by tumours is an outstanding question. Chemokine mismatch is a commonfinding in solid tumours, which results in ineffectual CAR-T trafficking in, and extravasation from, thecirculation to the tumour [85]. Many tumours are surrounded by an external stromal layer consistingof fibroblasts, myeloid cells and extracellular matrix, all of which can inhibit CAR-T penetration intothe tumour proper [86, 87]. The tumour microenvironment is characterized by hypoxia [88], depletion8of key metabolites such as glucose and amino acids [89–91], and regulatory cytokines such as TGF-β , all of which inhibit T-cell proliferation and can induce T-cell anergy or conversion to a regulatoryphenotype. These factors can be produced by malignant cells, but also by a variety of stromal cells suchas myeloid derived suppressor cells (MDSCs), Tregs, and innate immune cells such as neutrophils [92].In particular, Tregs suppress T-cell function and proliferation, as well induce T-cell apoptosis, through ahost of effector functions [93], and their selective depletion can improve anti-tumour activity of CAR-Tcells in mouse models [94, 95]. If a CAR-T cell survives this gauntlet and finds a tumor cell expressingthe cognate antigen for the CAR—and assuming this antigen has not been deleted or downregulated,another mechanism of tumor immune evasion [96, 97]—the CAR-T cell encounters a target tumour cellthat potentially expresses a variety of inhibitory, tolerogenic and pro-apoptotic ligands, most notablyPDL-1 [98], although many others are under investigation [99]. While these challenges can appearinsurmountable, there are active research programs that seek to address virtually all of these challenges[83], and with a variety of active clinical trials studying CAR-T therapy for solid tumours [54], thecoming years should provide insight and clarification as to the broad applicability of CAR-T therapy.More generally, and true of all cell-based therapies, significant hurdles are posed by the complex-ity associated with a therapy that is, currently, entirely personalized and manufactured separately foreach patient. This makes regulatory approval, delivery and deployment, and public payment for thesetherapies all outstanding issues to be resolved.Finally, several patients recently died in two waves in a CAR-T clinical trial run by one of theleaders in the field, Juno Therapeutics [54]. This illustrates the caution that will be required in movingforward with such complex, potent and only partially understood therapeutics.1.1.5 Engineered cellular therapeuticsThe therapies described in the previous sections are the first wave of cellular therapeutics. For the mostpart they focus on the addition or deletion of a single gene, potentially one that has been substantiallyengineered, as in the case of CARs. Moving forward, the full potential of cellular therapeutics willbegin to be realized as layers of molecular function and control logic are built on top of the cellularchassis, and this process is well underway.In the field of adoptive cell therapy, a variety of cell engineering is already being pursued. Con-ditionally active suicide switches are being included in adoptively transferred cells, to allow for erad-ication of the therapeutic product in case of adverse events [100–102]. Notably this is, in general,impossible for small molecules or biologics, unless a rapid inhibitor is available. In an inversion ofthis approach, a two-component CAR has been developed, that is only able to dimerize and activatedownstream signaling pathways in the presence of an inert small molecule ligand [103]. This wouldallow for infusion of CAR-T cells, and then subsequent activation or deactivation of the effector cells asneeded. This is a more nuanced approach than suicide switches as it might allow for dose titration andtemporary interruptions in the therapeutic action of the CAR-T cells as needed, without fully destroy-ing an expensive and personalized cellular product. To solve the problem of the immunosuppressive9tumour microenvironment, additional factors are being added to T-cells [104, 105]. Examples includesecretion of the pro-inflammatory cytokine IL-12, as well as receptors that respond to IL-4, which isabundant in the microenvironment, by providing IL-2/IL-15 proliferative stimuli to the CAR-T cells[67].Two recent combinations of gene editing and CAR-targeting have shown significant promise. Thegoal of the first was to generate CAR-targeted T-cells from an allogeneic donor. To do this, TALENswere used to knock out two genes in the donor cells: (i) TCR alpha, to eliminate potential graft-versushost reactivity; and (ii) the lymphocyte surface antigen CD52, which renders the cells resistant to thelymphodepleting anti-CD52 antibody alemtuzumab. This allowed for transient host lymphodepletion,thus avoiding graft rejection, while maintaining the viability of the incoming cellular product. Afterthese modifications, the donor cells were virally transduced with a CD19-CAR coupled to the hybridantigen RQR8 (a combination of CD20 and CD34), which renders the cells sensitive to rituximab, as asuicide switch for increased safety. This approach was tested in two pediatric patients with ALL whohad progressive disease after multiple rounds of therapy, and thus was so lymphopenic that autologousCAR-targeted T-cells could not be generated. As of January 2017, the patients are in molecular re-mission [106]. A similar approach has been pursued using CRISPR-Cas9 to knock out three genes:(i) TCR-alpha to eliminate alloreactivity; (ii) beta-2-microglobulin, to eliminate donor T-cell MHCexpression, and thus increase the persistence of allogeneic donor CAR-T cells in the recipient host;and (iii) PD-1, to decrease tumour microenvironment inhibition of CAR-T efficacy. In preclinical workthese modifications resulted in increased persistence and efficacy [107], and a clinical trial is under way.Finally, preliminary work has shown the feasibility of using iPSCs to generate CAR-targeted T-cells,which in theory would provide a limitless supply of a highly controlled cellular product [108]. Whilein their early stages, these efforts may be the first steps towards an off-the shelf, universal CAR-T celltherapy, which would greatly alleviate the cost and administrative burdens currently faced by CAR-Ttherapy.As discussed above, there are still relatively few cancer types for which a reliable, suitable CAR-Ttarget antigen exists [82].To solve this problem, combinations receptors are under development thattarget multiple antigens to implement simple Boolean logic gates [109]. While no single antigen maybe unique to a tumour, it is more likely that the combination of two or three may be. In a particularlyimpressive work, Lim and colleagues have generated a novel synthetic receptor derived from the Notchsignaling pathway. In its natural configuration, the extracellular domain of the Notch receptor binds toa delta ligand on another cell. This results in intracellular proteolytic cleavage of a transcription factorthat activates downstream signaling. Somewhat surprisingly, this system is fairly modular: by simplyretaining the core component of the Notch receptor that induces proteolysis, they were able to use avariety of novel external receptors (such as CARs, nanobodies and Myc tags), to trigger the release andactivation of a variety of downstream transcription factors [110, 111]. Finally, the CAR concept hasbeen inverted to create inhibitory CARs (iCARs), consisting of an scFV extracellular antigen recogni-tion domain coupled to a CTLA-4 or PD-1 derived inhibitory intracellular domain. These receptors are10coupled with a conventional CAR, to confine the specificity of the CAR-T cell to tumour cells. Thisapproach could be employed in the context of a tumour cell antigen target that is present on healthytissue as well. In this case the iCAR would be specific for a second antigen found on the healthy tissue.When both antigens are present (healthy tissue), the T-cell would not be activated, while if only thetumour antigen were present, the CAR-T cell would be activated [112]. Together, these initiatives havethe potential to enable combinatorial sensing of a variety of signals, and integration of these signalswith minimal cross-talk.Moving beyond the domain of adoptive cell therapy, several efforts are underway to constructrelatively complex control logic, albeit in quite simple model cell lines. In a pioneering work, a cellclassifier circuit was built to detect HaLa cells as a model cancer cell. A panel of 5 micro-RNAs wasidentified, whose abundance (high or low) differentiated between HeLa and healthy cells. Using acombination of the lac operator and miRNA binding sites, a classifier was built that was active only inthe presence of the correct combination of miRNAs, and inactive otherwise. When the output of thisclassifier was set to be the pro-apoptotic protein Bax, and the whole circuit was transfected into mixedcell populations, only the HeLa cells exhibited substantial cell death [113]. Another substantial bodyof work has been undertaken by Fussennegger and colleagues, who have engineered cellular factoriesthat have existing sensory and synthetic pathways joined to create novel cell-based therapeutics termedprosthetic gene circuits [114]. Preclinical work has seen investigators combine a uric acid sensor andurate oxidase (which degrades uric acid and is notably absent in humans) from two different bacterialspecies to create a prosthetic gene circuit that maintains uric acid homeostasis in experimental modelsof gout [115]. In another example, expression of an IL-22 receptor was placed under conditionalcontrol of a TNF-responsive promoter. IL-22 activity at the synthetic receptor resulted in production ofanti-inflammatory cytokines IL-4 and IL-10. In this way, a cellular therapeutic implanted in a mousemodel of psoriasis was able to suppress inflammation in a highly specific, targeted manner: only whenboth TNF and IL-22 were present was the circuit active [116]. Similar circuits have been designed andtested for diabetes [117], metabolic syndrome [118], thyroid disease [119]. These cellular therapeuticsall have several common themes. First, they all rely upon mining the diversity of biological functionto find useful parts (for example a light sensor, or an enzyme that degrades uric acid). Second, theyinclude rewiring transcriptional logic circuits to combine the inputs and outputs of these novel partsto achieve the desired response. Finally, they are almost uniformly expressed in simple model celllines that are encased in inert alginate gels and implanted in the body. These gels consist of a polymermatrix with pore sizes large enough to permit entry of crucial metabolites and exit of the cell-secretedtherapeutic molecule, but small enough to block the host immune cells, thus preventing host rejectionof the engineered cellular therapeutic [120]. This approach has several potential advantages over simplereplacement of the diseased cell type or a small molecule therapy. First, immunologically compatibledonor sources are often unavailable, and hESC or iPSC derived cells will not be available for all tissuesfor some time. Second, replacement may require implantation in a challenging physical location inorder for function. Finally, replacement is not a viable option in the cases of increased metabolite levels,11or production of toxic metabolites due to dysfunctional enzymatic activity. Recombining biologicalpathways in a suitable chassis offers solutions to these problems. The chassis may be selected tooperate in a more suitable physical niche, and be devoid of any pathogenic immunological markers. Thepathways for sensing metabolite levels may be optimized, using multiple, engineered surface receptors.Similarly, the pathways for metabolite regulation (production or elimination) may also be optimized,for example to avoid toxic metabolite production. These applications are clearly years from beingapplied clinically. They have only shown efficacy in mouse models over short periods of time. Thelong term safety profile, and immunological reactivity of implanted cell-lines is very much in question,even if they are encapsulated. However, what these circuits represent are the early stages of a set ofparts and approaches for combining sensors, effectors, and control logic into cellular therapeutics.Incorporating existing biological function into new cellular therapeutic devices is an approach thatis gaining traction and application across a range of human disease. I have attempted to add to this partset by developing a cell-to-cell therapeutic delivery module.1.2 Cytotoxic lymphocyte biologyThis thesis focuses on understanding and engineering the granzyme-perforin pathway, a key effectormechanism of cytotoxic lymphocytes. Before further discussing their application, I first provide somedetail on their basic biology.1.2.1 Cytotoxic lymphocytesCytotoxic T-lymphocytes are key elements of the adaptive immune response that are mainly respon-sible for the recognition and clearance of cells infected by intracellular pathogens, as well as tumourimmunosurveillance [121]. T-cell identification of target cells is a complex process that hinges uponTCR engagement of a cognate peptide presented by cell surface MHC [122]. This interaction activatesthe key cytotoxic effector mechanisms of T-cells: [121] (i) the granzyme-perforin pathway; (ii) surfaceexpressed death receptor ligands such as Fas (FasL); and (iii) cytokines, most importantly interferongamma (IFNγ).Unfortunately, primary T-cells are relatively difficult to manipulate genetically, with viral trans-duction often required, which is unsuitable for exploratory work involving iterations of design, testing,validation and optimization. Furthermore, maintaining primary lymphocytes in culture is more onerousthan maintaining immortalized cell lines, and, due to their limited proliferation lifespan, any modifica-tions made to primary cells will eventually be lost when the cells enter senescence. While immortalizedT-T hybridoma cell lines do exist, none retain a functional granzyme-perforin pathway.An attractive alternative to working with primary T-cells is to use natural killer (NK) cell lines as amodel system. NK cells, which are the analogue of T-cells in the innate immune system, kill targetedcells using the same effector mechanisms as T-cells [123], and several NK cell lines exist with an intactgranzyme-perforin secretion pathway, and intact target-specific cytotoxic function. NKs differ from12T-cells in that NK activation is a complex balance between activating and inhibitory receptor mediatedsignals, transduced by killer activating and killer inactivating receptors (KAR, KIR) respectively [124].Importantly, this means that the cytotoxic machinery of NK cells can be mobilized against a targetcell without the requirement of antigen receptor mediated activation, as would be the case in a T-cell.However, this antigen specificity can be imposed by expressing a CAR in NK cells [125], and indeedCAR-targeted NK cells have been used in clinical trials [126].1.2.2 The granzyme-perforin pathwayIn humans there are four granzymes: A,B,K and M, of which granzyme B (GZB) is the best charac-terized and most abundant [127]. GZB is a serine protease with a classical trypsin-like catalytic triadthat initiates apoptosis in targeted cells [128]. Synthesized primarily in cytotoxic lymphocytes as a 247amino acid precursor protein, GZB is directed to the endoplasmic reticulum by a signal peptide, whichis subsequently cleaved, yielding the zymogen form of GZB, which is still inactive due to a N-terminaldipeptide. This proenzyme is sorted through the Golgi network in a pathway that involves the additionof mannose-6-phosphate, as well as the chaperone molecule serglycin, both of which promote local-ization of GZB to lytic granules (LGs), a type of specialized secretory lysososome. Once in the LG,the dipeptide is cleaved by cathepsin C, and the active form of GZB is safely sequestered in the acidicLG and stored there awaiting cytotoxic lymphocyte activation [127]. Despite these structural insights,the exact motifs responsible for GZB trafficking from synthesis through to the target cell cytosol areunknown. The other major component of this pathway is perforin, a long, thin protein that forms poresin targeted cells, and is stored in LGs along with granzyme-serglycin aggregates [129].Granule synthesis occurs during cytotoxic lymphocyte development, and granules are prepositionedand ready for secretion upon target cell recognition [130]. Initial target cell interaction is mediated byintegrins on surface of cytotoxic lymphocytes, a prominent example being LFA-1 [131]. T-cell acti-vation results from antigen specific TCR recognition of a short peptide in the context of MHC [132].More complicated is NK cell activation, which is a function of the relative balance between a host ofinhibitory and activating receptors [124]. In both cell types, these surface receptor interactions result inthe formation of the immunological synapse, a tight apposition between effector and target cell, witha peripheral ring of adhesion molecules (pSMAC) and a central region of target recognition molecules(cSMAC) [133]. Surface receptor ligand interaction (e.g. TCR-pMHC in T cells) results in activationof canonical cytotoxic lymphocyte intracellular signaling pathways, with a phosphorylation cascadeconverging on the assembly of the LAT signaling complex, which activates the MAPK/ERK pathwayand initiates calcium influx into the cytotoxic lymphocyte via the PLC pathway [131, 134]. These path-ways initiate cytoskeletal remodeling, with the microtubule organizing center (MTOC) polarizing to theimmunological synapse, with lytic granules driven by dynein following the MTOC along microtubules[129]. Arriving at the synapse, surface molecules on the granules and cytoplasmic cell membrane fa-cilitate docking, followed by fusion of the granule and membrane lipid bilayers, which results in theexocytosis of the lytic granule contents (including granzymes and perforin) into the synapse [129].13Perforin and granzyme diffuse across to the target cell membrane, into which perforin inserts, andthen aggregates to form multimeric, transmembrane pores [135]. Historically it was thought that theseperforin pores were directly responsible for target cell death, but it is now believed that physiologicalconcentrations of perforin alone are not cytotoxic. Instead, the pores seem to be only briefly patentbefore membrane integrity is restored, with their main function being a conduit for passive diffusion ofGZB into the target cell [136, 137]. Once in the cytosol, it is GZB that initiates apoptosis by cleavageof BH3 interacting-domain death agonist (BID) and caspases 3,7 and 8, which in turn activate themitochondrial and caspase apoptosis pathways respectively [122, 135].In summary, the synergistic activities of granzyme and perforin represent a unique pathway fortransferring molecules from cytotoxic lymphocyte to target cell exclusively, as the immunologicalsynapse confines granzyme and perforin between the two cells, and moreover, significant numbersof perforin molecules are required to form the pores required for granzyme’s entry into the target cell.1.3 Overview of a lymphocyte-based delivery systemThe overall objective of this project is to to develop a cell-to-cell therapeutic delivery system, thatis built on a cytotoxic lymphocyte chassis, targeted by a CAR or TCR, and that uses the granzyme-perforin pathway to deliver a protein payload to a target cell. The specific focus of this thesis is toengineer the granzyme-perforin pathway as the delivery module of this system.1.3.1 The granzyme-perforin pathway as a delivery moduleIn order to construct such a cell-to-cell protein transfer system, my approach is to use granzyme B asa molecular chaperone to mark the therapeutic payload for transfer to the target cell via the granzyme-perforin pathway. This will be achieved by fusing the payload to granzyme B (or derivatives thereof),such that the payload fusion protein will be expressed and packaged into LGs in preparation for releaseupon target cell encounter. Transfer of a granzyme-fluorescent fusion protein has been demonstratedpreviously, although the data is either in primary mouse cells [138] or of questionable validity [139]. Indeveloping this system, several biophysical parameters must be considered for any potential therapeuticpayload. First and foremost, the fusion protein must transit perforin pores that have been measured viaelectron microscopy to have an average luminal diameter of 13-20 nm [137]. Combining this data withthe diameters of GZB and GFP (5 nm [137] and 3.5 nm [140] respectively), gives an approximate sizerestriction on potential payloads. Another significant constraint on the payload is that it must alwaysbe at the C-terminal end of the fusion protein, since GZB must be at the N-terminus, to ensure that thesignal- and pro-peptides are appropriately processed. Therefore, any payload with critical motifs at itsextreme N-terminus may have decreased or absent functionality at the C-terminus of a fusion protein.The stability of these fusion proteins in the harsh, acidic, proteolytic environment of lytic granuleswill also need to be assessed for each payload. Furthermore, for transit through the granzyme-perforinpathway, and functional activity once in the target cell, other factors such as fusion protein folding and14solubilization, external exposure of important signaling motifs, steric and electrostatic interaction, andcharge distribution will affect the success of a particular fusion protein.Thus on demand cell-to-cell protein transfer is enabled by a combination of the prepositioned lyticgranules, the immunological synapse, and the granzyme-perforin pathway itself. The potential utilityof this core function justify studying these systems in an attempt to repurpose them for various celltherapy applications.1.3.2 Comparison with existing systemsThis approach has a range of theoretical advantages over other related therapeutic modalities. Com-pared to biologics and small molecules, the sequestration of a therapeutic inside a delivery lymphocytemay well improve bioavailability, and enable the delivery of therapeutics that would be toxic if admin-istered systemically. The combination of receptor targeting, and the confinement of the therapeutic inthe immunological synapse may enable a level of specificity in the delivery of a therapeutic that is oth-erwise unattainable, except in the case of antibody based drugs. Antibody-conjugate therapeutics are amature technology, and would likely have the same level of specificity as this approach. Interestingly,the best estimates for the amount of payload that would be delivered are fairly similar for cellular andantibody mediated delivery: on the order of hundreds of molecules per target cell [141, 142]. However,it is important to note that the estimate for granzyme delivery is taken from a computational studyof the unmodified pathway, and as such the comparison should be interpreted with caution. A poten-tial advantage of cellular delivery of a payload is that its bioavailability might be far greater. In thecase of antibody-drug conjugates, the payload is exposed in transit to the target tissue, as opposed tointracellularly sequestered, which may greatly decrease the immunogenicity and clearance of the de-livered payload. Furthermore, the potential for cellular control logic (for example a suicide switch, ormolecular sensors) may allow for ongoing, post-administration control of the effects of the therapeutic.Use of a cellular delivery system might be a useful bridge therapy to delay the decay of damagedtissues, but ultimately regenerative medicine and stem cell therapies will clearly be superior for actualreplacement of damaged tissue. However, this assumes that an appropriate replacement cell populationor tissue is available for all damaged tissues, which is not currently the case. Furthermore, dependingon the location of the damaged tissue, and the properties of the incoming graft, implantation maybe challenging or impossible. Conversely a cellular delivery system that is motile might be usefulthroughout the body for delivery of a therapeutic that partially regenerates damaged tissue, althoughthis would require additional modification of the delivery lymphocyte (discussed below).Many of these advantages would apply equally to gene therapy and mesenchymal stem cell therapy:reliable and specific activity at disease sites are challenges that both of these fields have struggled with.Again, the broad tissue distribution and target cell specificity of a lymphocyte delivery system mightaddress both of these issues. In the case of lymphocyte delivery, presumably all delivered therapeuticpayloads would have a half-life in the target cell. Depending on the therapeutic application, redosingmight be required. In some cases this might be a disadvantage as compared to other, more permanent15types of therapy such as regenerative medicine, stem cells, iPSC, gene therapy, or MSC therapy. Alter-natively this might be an advantage: the lack of permanent modification is a substantial safety benefit,and offers greater flexibility.A significant challenge to using a cytotoxic lymphocyte chassis, and one which is not encounteredin viral or mesenchymal technologies, is that unmodified cytotoxic lymphocytes will kill any target cellto which they deliver a payload. If used in an application in which the intent of payload delivery isto eliminate the target cell, as would be the case in a cancer or infectious disease context, this wouldnot be a concern. However, for most other applications, such as delivery of a pro-survival or anti-inflammatory payload in the context of degenerative disorders, the delivery cell chassis would have torendered non-cytotoxic. This could be achieved through knockdown or knockout of the genes that codefor lymphocyte cytotoxic effector proteins, or by reconstituting the granzyme perforin pathway in aninert cellular chassis. This is discussed further in Chapter 5 as a future direction, but is not a focus ofthis thesis, which is rather to provide proof-of-principle of granzyme-perforin mediated delivery.1.4 Thesis overviewThis thesis has three data chapters, followed by a final chapter of conclusions and discussion. As all ofthe chapters have or will be published as stand-alone manuscripts, discussion relevant to each chapteris presented at the end of that chapter. The final chapter is mainly concerned with ways to improve thedelivery system, future directions for the project, and some broader insights and questions surroundingcellular therapeutics.Chapter 2 is a computational biophysical study of the immunological synapse and the behavior ofgranzyme and perforin within the synapse. Based on my computational results, I question some of thecore assumptions surrounding the mechanism of cytotoxic lymphocyte specificity and the immunolog-ical synapse, and suggest this specificity is the result of granzyme-perforin spatiotemporal dynamics,rather than immunological synapse geometry.Chapter 3 demonstrates proof-of-principle that the granzyme-perforin pathway can be used to de-liver a protein payload to a target cell population. I first designed a suite of granzyme B derived molec-ular chaperones, and fused them to mCherry as a model payload. I then screened these chaperones fortheir ability to load mCherry into lytic granules, using fluorescence microscopy. This generated twocandidates, which I tested further to see if they were transferred to target cells. Using a model naturalkiller cell line, I demonstrate transfer of a granzyme B mCherry fusion protein to target cells.Chapter 4 collects efforts to use this approach to deliver potent, orthogonal toxins to target cellsthat are resistant to lymphocyte cytotoxicity. I generate a variety of granzyme B toxin fusion proteins,and investigate their activity as fusion partners. I then attempt to generate lymphocyte resistant celllines, efforts which are for the most part unsuccessful. Using an effector cell dose response curve,I attempt to demonstrate that effector natural killer cells armed with granzyme-toxin fusion proteinsexhibit enhanced killing of target cells. I observe moderate effect sizes. I conclude this application16merits further investigation and optimization prior to any final judgment regarding its therapeutic utility.17Chapter 2A computational biophysical model of thegranzyme-perforin pathway2.1 IntroductionUpon cytotoxic lymphocyte (CL) recognition of a target cell via surface receptor interactions, the so-called immunological synapse (IS) is formed – a region of tight proximity between the CL and targetcell membranes in which two distinct killing pathways unfold. The first is the death receptor pathway,which is mainly thought to be important in the context of maintaining T-cell homeostasis and deletingautoreactive T-cells. Fas ligand expressed on the surface of the CL stimulates Fas receptors on the targetcell, leading to receptor aggregation and activation of the extrinsic apoptosis pathway. The second mainway in which CLs kill their targets is via exocytosis of lytic granules containing, among others, perforinand granzymes into the IS [129, 135, 143]. Here we confine our discussion to granzyme B (GZB), asit is the most important member of the granzyme family in inducing target cell death, possibly alongwith granzyme A, although this is controversial [144]. Perforin and GZB diffuse across the IS to thetarget cell membrane, where GZB achieves entry to the cytosol in a perforin-dependent manner. Onceinternalized, GZB, a serine protease, initiates apoptosis by cleavage of BH3 interacting-domain deathagonist (BID) and caspase-3.Exactly how perforin mediates GZB access to the target cell in the context of the IS has been thesubject of debate for over two decades, with two principle models having been investigated [143].The simpler model proposes that perforin creates pores in the target cell membrane, allowing GZB todiffuse into the cytosol of the target. The more complex theory suggested that perforin and GZB bindregions of the target cell membrane within the IS which are then rapidly endocytosed. Perforin poresform within the endosomes, allowing GZB to be released into the target cell. However, recent highresolution microscopic studies strongly support the simpler model whereby perforin monomers insertinto the target cell membrane, and then combine to form multimeric pores, through which GZB cansubsequently diffuse [136]. These pores have recently been observed and characterized using cryo-18electron microscopy [137]. It has also been demonstrated that perforin pores are rapidly repaired bythe target cell, leaving only a short window of time for GZB to enter the cytosol [136].Taken together, these studies raise interesting questions about the relative timescales for diffusion,pore formation and GZB delivery. Despite a large investment in experimental effort, we are awareof no existing theoretical consideration of this system that allows these questions to be resolved; pre-vious theoretical work in which we developed analytic solutions for the concentration of a diffusingspecies in the synapse volume based on partial differential equations is restricted to a single diffus-ing chemical [145]. Here, we consider nonlinear kinetics of perforin aggregation and small numbersof multiple diffusing molecules. To accurately capture both these aspects of the problem, we apply aspatial stochastic simulation algorithm (SSSA). This method, although relatively time-consuming com-putationally, allows us to gain insight into this nonlinear system, and to obtain probability distributionsof events in the model rather than just the mean behaviour, both of which are unobtainable with differ-ential equation methods. Using this approach, we develop and analyze a mathematical model of GZBdelivery via perforin pores. Our model allows us to show that perforin pore facilitated GZB entry intothe target cell can support rapid, targeted killing. However, reliable pore formation requires previouslyunconsidered constraints on the rate of diffusive transport within the IS, which we hypothesize is dueto molecular crowding in the synapse.2.2 MethodsWe seek to describe the dynamics of GZB and perforin (PFN) from their release from lytic granules,through their diffusion throughout the IS, to PFN pore formation and GZB internalization. We firstprovide a description of our biophysical model of this system, followed by its mathematical and com-putational implementation.2.2.1 Biophysical model: Geometry & molecular processesThe IS is an irregular narrow region between the CL and target cell that has a very high aspect ratio:the radius of the enclosed region is on the order of microns, while the distance between the two cellsis on the order of tens of nanometers [146]. Therefore we model the IS as a very flat, broad disc ofradius R = 3µm and height h = 20nm, as shown in Figure 2.1, with the CL membrane taken to be theupper surface of the disc, and the target membrane considered explicitly, immediately below the lowersurface of the disc. Since h ∼ 20− 40nm, and the diameter of GZB and PFN are ∼ 5nm and ∼ 8nmrespectively [137, 147], we allow that molecules may escape through the synapse edge.Given that exocytosis of lytic granules is temporally synchronized [129], and that the time-scalesof both exocytosis (on the order of milliseconds [148]) and diffusion across the synapse (calculatedusing the Stokes-Einstein relationship to be on the order of microseconds using the dimensions of thesynapse given above) are much faster than pore formation (observed to be on the order of seconds [136,149, 150]), we assume that GZB and PFN are instantaneously released from the CL membrane as an19R = 3 um Target Membrane Immunological Synapse Pore h ~20 nm CL TARGET Immunological Synapse Target Cell PFN Monomer Membrane Inserted PFN Monomer Membrane Inserted PFN Oligomer Mul$mer(Perforin(Pore( Granzyme(B(RLG = 500 nm Figure 2.1: Model geometry and molecular interactions. We consider the synapse (blue) as abroad flat disc, with the upper surface the CL membrane, and the lower surface the target cellmembrane (yellow). GZB (purple circles) and PFN (green cylinders) are released from a centrallytic granule (red). We discretize this space into a two dimensional mesh of sub-volumes (upperleft). The time evolution of the system is then governed by diffusive jumps between sub-volumes,and interactions between molecules within a sub-volume. These interactions encompass PFNmembrane insertion and oligomerization leading to pore formation, followed by GZB internaliza-tion through pores (lower right).initial bolus. The exact location of granule release, the so-called secretory domain, has been variouslyreported as both central [151] and in between the central and peripheral supramolecular activationcomplexes [152]. For simplicity, we assume GZB and PFN are released from a single lytic granule ofradius RLG = 500nm [151] at the centre of the synapse. This assumption is also maximally conservativewith respect to molecular escape from the synapse (see discussion below).Within this geometry, we model the spatiotemporal dynamics of GZB and PFN by considering dif-fusive transport, as well as chemical interaction (schematically depicted in Figure 2.1). Both moleculesdiffuse throughout the synapse, eventually either escaping at its lateral edge or interacting with thetarget cell membrane, as described below. Due to the extreme aspect ratio of the synapse, the timescalefor diffusive transport across the height of the synapse is short compared to all other relevant processes,and therefore we approximate diffusion in the synapse as two dimensional in the horizontal plane.PFN monomers insert into the target cell membrane with rate kins, which we assume is slower thanthe diffusion limited rate, due to the energy requirements of lipid membrane displacement for perforininsertion. Membrane inserted monomers can then diffuse across the membrane and potentially combineto form pores. Based on an analysis of electron micrographs that indicates that pores consist of a ring20of 18-20 PFN monomers spanning the target cell membrane [137], we modelled pores as 18mers. Wemodelled the path to pore formation as a multistep, multi-pathway oligomerization process, which con-sists of monomer dimerization as well as monomer and dimer aggregation to form trimers. Monomers,dimers and trimers then combine with each other to form higher order oligomers. Since membranediffusivity scales inversely with molecular size, higher order oligomers will be decreasingly mobile.They will also be sparsely distributed and therefore it is very unlikely that a higher order oligomerwould encounter another higher order oligomer before a low order oligomer. We used this observationto simplify our model of pore formation by neglecting any interaction between two oligomers greaterthan a trimer: oligomers can only grow in size by combining with a monomer, dimer or trimer. Weassume the rate of PFN oligomer aggregation is diffusion limited, and denote this rate ki, j. We neglectreverse reactions for both membrane insertion and oligomerization.Once an 18mer has formed, this becomes a pore through which GZB can diffuse, which occurswith rate kg. We assume that this process is diffusion limited, and neglect the reverse reaction.Symbolically we have the following reaction scheme:Pkins−−→ P1Pi+Pjki, j−→ Pi+ j i = 1,2,3 j = 1, . . . ,18− iG+P18kg−→ Gint +P18(2.1)where P and G represent synaptic PFN and GZB respectively, Pi denotes a membrane inserted PFNi−mer for i = 1, . . . ,18. Gint denotes internalized GZB, and kins, ki, j and kg denote the rates of PFNmembrane insertion, perforin oligomerization and GZB pore transit respectively. We assume that dur-ing the short timescale of the processes we are modelling, no GZB or PFN molecules are lost dueto other processes such as irreversible non-specific binding of these proteins in the synapse, loss ofPj molecules in the target cell membrane due to endocytosis, or loss of functional activity of eithermolecule due to irreversible inactivation in the IS.In summary, our model consists of a broad, flat, disc-shaped IS between CL and target cell. GZBand PFN are released as an instantaneous bolus from a single, centrally located lytic granule, where-upon they diffuse throughout the synapse, with the potential for any molecule to diffuse out of the ISthrough the lateral edges. Perforin inserts into the target cell membrane and oligomerizes to form pores,through which GZB can then diffuse.2.2.2 Mathematical description: Spatial stochastic simulation algorithmSince the numbers of certain molecular species (such as 18-mer pores) are very low, continuous modelsderived from mass action kinetics are inappropriate for describing our system, and stochastic methodsare instead necessary. Furthermore, due to the localized release of molecules, combined with the rela-tive sparsity of these molecules in the system, spatial effects are important, and homogenous stochastic21models are likely to be insufficient. In order to accurately model small numbers of molecules in spaceand time, we therefore applied a discrete spatial stochastic simulation algorithm developed by Elf andEhrenberg [153, 154], which is an extension of the spatially homogenous next reaction method ofGibson and Delbruck [155], itself a computationally more efficient version of the original Gillespiealgorithm [156].The Gillespie method considers a system of reactions Xi that occur with rate constants ri accordingto standard chemical kinetics, and asks two questions: (i) when does the next reaction occur? and (ii)which reaction occurs? By first calculating the probability that each reaction occurs, given by ai = riNi(with Ni denoting the number of instances of reaction Xi), Gillespie showed that the answer to thesequestions may be obtained by randomly sampling two probability distributions:tnext =1a0ln(1n1) (2.2)X = Xi if1a0i∑j=1ai ≤ n2 < 1a0i+1∑j=1ai (2.3)where n1 and n2 are random numbers between zero and one, and a0 = ∑iai. The time evolution of thesystem may then be obtained by first incrementing the time by tnext and updating the species numbersaccording to the stoichiometry of reaction Xi, then recalculating the probability distributions accordingto the updated species numbers, and finally resampling the new probability distributions.The essence of the SSSA (the spatially inhomogeneous extension of the Gillespie algorithm) is todiscretize the physical simulation space into sub-volumes (SVs) of length l, chosen to be small enoughthat the spatial distribution of species is approximately homogeneous within each sub-volume. Thisjustifies using mass action chemical rate constants to describe the molecular interactions within a subvolume. Diffusion is modelled as another ‘reaction’ in which a molecule jumps from one sub-volumeto an adjacent one with a rate constant of d = nD/l2, where n = 4 are the spatial degrees of freedomfor a diffusive jump, and D is the diffusivity. This mapping of diffusion to a reaction allows for theformalism of the Gillespie algorithm to be employed. The algorithm is then similar to that of theoriginal Gillespie algorithm. An event occurs in the SVi that has the lowest tnext , which is calculatedby sampling a probability distribution analogous to those of Equation 2.2, with the probabilities aireplaced by si = ai+di. Having chosen when and in which SV an event occurs, the event is a chemicalinteraction if n3 < a0/s0 (s0 = Σsi, and n3 is a random number distributed between zero and one)and diffusive otherwise. If the event is diffusive, which species diffuses is selected by weighting arandom distribution by the number of each species within the subvolume, and the direction of diffusionis randomly selected. If the event is a chemical reaction, then the identity of the chemical reactionis determined as in the Gillespie algorithm. Following sampling of the probability distributions, thetime and species number (in both the origin and destination SV if the event was diffusive) are updated,the probability distributions recalculated. The spatiotemporal evolution of the system is obtained byrepeatedly iterating the above algorithm.22It should be noted that there are several nuances to the version of the SSSA presented by Elf andEhrenberg that was implemented in this work that result in increased computational efficiency, and fordetails we refer the reader to their work.We constructed a discretized IS with two-dimensional sub-volumes of side length l as shown inFigure 2.1. There are three regions: the central lytic granule which contains PFN and GZB initially, therest of the synapse, and a region external to the synapse. This last region is present to allow for escapefrom the synapse and return from the exterior is prohibited. Finally, within each sub-volume, chemicalreactions occur according to the chemical reaction scheme in Equation 2.1.2.2.3 Hindered diffusion in the immunological synapseWe consider molecules in the synapse as roughly spherical particles diffusing freely in a bulk fluid,and so estimate their diffusivity from the Stokes-Einstein relation, Dfree = KbT/(6piηwr) where KbTis the thermal energy, ηw is the solvent viscosity and r is the radius of the molecule. Additionally, thediffusivity is modulated by a hindered diffusion parameter, α , which is motivated by the observationof very high electron density in the IS, which we hypothesize is due to densely packed extracellularadhesion and signalling molecules. This molecular crowding in the IS has two effects, both resultingin a decrease in effective diffusivity. First, non-specific binding of GZB and PFN to intra-synapticmolecules decreases the total time during which GZB or PFN are free to diffuse. Second, the spaceoccupied by intra-synaptic molecules is not available for GZB and PFN diffusion. The effects of thelatter are accounted for by multiplying the free diffusivity by the volume fraction of the synapse stillavailable for free diffusion: (1− f ), where f is the volume fraction of the synapse occupied by the intra-synaptic molecules. To derive an expression to model the effects of non-specific binding we considera molecule (GZB or PFN) diffusing in the synapse which is filled with other molecules (referred tohereafter as binders) filling the synapse at a number density ρ . We assume that the diffusing moleculebinds a binder with a rate ρkon and disassociates from it at a rate koff. As these binders are attachedto either the CL or the target cell, we assume that the diffusing molecule is immobile when bound toa binder. Thus the time the molecule spends free to diffuse or bound is proportional to the inverse ofthe associated binding and unbinding rates respectively, and the fraction of time (τ) that a diffusingmolecule spends free and unbound is, after minor algebraic manipulation,τ =koffρkon+ koff=kDρ+ kDkD =koffkon(2.4)where the second equality is obtained by dividing through by kon, and kD is the dissociation constantfor nonspecific protein-protein interactions, which can be experimentally measured. To calculate ρwe note that it is equal to the number of molecules (N) in the synapse divided by the volume of thesynapse (Vsyn). We can approximate the number of molecules in the synapse by taking it to be thetotal volume of the synapse occupied by molecules (which itself is fVsyn) divided by the volume of an23average molecule in the synapse (Vavg):ρ =NVsyn=fVsyn/VavgVsyn=fVavg. (2.5)To obtain an expression for α , which relates the effective diffusivity of molecules in the synapse (Deff)to the corresponding free (Stokes-Einstein) diffusivity (Dfree), we multiply the volume fraction of thesynapse still available for free diffusion (1− f ) by the fraction of time which molecules are able todiffuse freely (τ):Deff = αDfree α =kDρ+ kD(1− f ) ρ = fVavg. (2.6)2.2.4 Perforin oligomer diffusion and aggregation in the target cell membraneThe diffusivity of proteins in cell membranes is a problem that has received considerable experimentaland theoretical attention [157–159]. These efforts have demonstrated that the diffusivity varies withthe total protein density in the membrane, as well as the radius of the diffusing protein, both of whichare relevant for our consideration of a perforin oligomer of changing radius in the highly crowded IS.Recent experimental work has shown that at low densities, the diffusivity scales according to a Saffman-Delbruck [157] type relationship (ln[1/r], where r is the radius of the diffusing molecule), whereas inhighly crowded membranes, diffusivity scales as 1/r [159]. This latter result was also arrived at in-dependently by Gambin and colleagues [158], and is consistent with the Stokes-Einstein description.It is difficult to estimate whether the membranes at the immunological synapse are highly crowded ornot in the terms of these experimental papers. We performed simulations using both relationships andfound that pore formation and granzyme delivery were extremely improbable under the Stokes-Einsteindescription. Therefore, to be maximally conservative in terms of minimizing the requirement for hin-dered bulk diffusion in the synapse, we present simulations using the Saffman-Delbruck relationship todescribe the diffusivity of a perforin oligo in the membrane,D j =KbT4piηlξ(ln(ηlξ/ηwrPj)− γ)(2.7)where KbT is the thermal energy, ηw and ηl are the solvent and membrane viscosity respectively, ξis the thickness of the cell membrane, rPj is the radius of a perforin j-mer and γ is Euler’s constant.The radius of a j-mer is taken to be the average of its long and short dimensions: rPj = RPFN( j+1)/2,where RPFN is the radius of a perforin monomer. If two oligomers are in the same sub-volume, thenthey may combine to form a higher order oligomer, subject to the conditions in Equation 2.1. We takethe rate constant ki, j for this process to be the diffusion limited rate for a particle (in our case the smaller24oligomer Pi) finding a trap (the larger oligomer Pj) on the surface of a membrane [160]:ki, j =2pi(Di+D j)ln(b/s)where b =√l2piNPiand s = rPi + rPj . (2.8)Here l2 is surface area of the sub-volume, which is the space the NPi i-mers explore in finding the j-mertrap, and s is the reaction radius, which we have taken to be the sum of the radii of the two oligomers.This is the maximum possible reaction rate, and so we are considering the fastest possible rate of poreformation in this model.2.2.5 Rate of granzyme translocation through perforin poresIn order to derive a rate for GZB internalization through a pore, we consider two processes: (i) findingthe pore; and (ii) translocating through the pore. The first step can be described by considering a particlein a cuboid with height h (the height of the synapse) and cross-section l2 (the area of the sub-volumethe particle is in). If we assume some diffusion driven flux Φ of particles hitting the bottom of thecuboid (namely the target cell membrane) then approximately ΦAp/l2 particles per unit time will hit apore, where Ap is the cross-sectional area of a pore. If we then divide this by the number of particlesin the cuboid, we will have a rate for the particle finding the pore. To derive an expression for the fluxΦ, we first note that lateral diffusion occurs when the particle jumps to the next sub-volume, whilewe wish to calculate the rate at which the particle hits the target cell membrane in those cases wherethe particle remains in the sub-volume. Therefore, the problem can be described by a one dimensionaldiffusion equation at equilibrium, with reflecting boundary conditions at the top of the cuboid (since weassume no particles are reabsorbed by the target cell) and absorbing boundary conditions at the bottomof the cuboid (since we are calculating the flux of all particles hitting the membrane). Solving the onedimensional diffusion equation allows us to calculate the flux Φ at the target cell membrane, whichyields the rate kg = 3piR2pD/l2h2 for GZB hitting a pore, where Rp = 20nm is the pore radius [137].To derive a rate for GZB translocation, we note that electron micrographs of PFN multimer poresshow a relatively smooth and uniform pore, with no evidence of gates or obstruction to particle entry[137]. Therefore, we assume that pore transit can be modelled as a one-dimensional diffusive processes,with a rate that scales as 1/ξ 2 where ξ ∼ 10nm is the thickness of the membrane [161]. Comparingthis rate to that derived above for GZB finding a pore, and recalling that the the height of the synapseis h ∼ 20nm we see that GZB finding the pore will be the rate limiting step. Given that the GZBinternalization process in our model is identical in form to an enzymatic reaction (Equation 2.1), it isconceivable that GZB might saturate pores, in which case Michaelis-Menten kinetics would be moreappropriate than mass action kinetics. However, using the initial (maximal) concentration of GZB inour system, and a pore diameter of 20nm [137] we estimate the average number of GZB per pore to beon the order of 10−2, and so we find that our assumption of mass action kinetics for pore translocationwould be valid even for much higher GZB concentrations.252.2.6 Model parameterization and implementationTo calculate the number of PFN and GZB molecules released into the synapse we first estimated thetotal number of these molecules in a cytotoxic lymphocyte. For perforin this was reported to be anaverage of 500 PFN molecules for CTLs and 3500 for NK cells [162], and so we selected an arbitraryintermediate value of 1500. We could not find any estimates of the number of GZB molecules per cell,so we instead used RNA expression data showing GZB mRNA copy number is ten times that of PFN[163] to set the number of GZB molecules at 15000. Given that mRNA copy number often correlatespoorly with protein expression levels [164], that the experimental work quantifying perforin numberper cell has several technical limitations, and the uncertainty surrounding the number of moleculesreleased in a lymphocyte-target cell interaction, we use these values only as a starting point from whichto subsequently explore the effects of PFN and GZB concentration (see Results).To derive the hindered diffusion parameter α , we must estimate the volume of an average proteinspanning the synapse (Vavg), the volume occupancy of these proteins in the synapse ( f ), and the dis-sociation constant for non-specific binding interactions of GZB or PFN with such an average protein(kD). To strengthen our arguments about the importance of molecular crowding, we chose our param-eters so as to reasonably maximize α and thus minimize the effects of hindered diffusion for a givenparameter set. To estimate Vavg, we used an average molecular weight of 200kDa as representativeof abundant signalling and adhesion molecules found in the synapse (e.g. ICAM1, LFA1, CD45), andan average protein density of 1.35g/cm3 (valid for molecules larger than 20kDa) [165], to calculateVavg = 250nm3. To estimate f we note that electron micrographs of the synapse show it as much moreelectron dense than the average density of the cytoplasm. Therefore we reasoned that the maximum es-timates of the volume occupancy of the cytoplasm would be an appropriate lower bound on the volumeoccupancy of the synapse, choosing f = 0.4 [166]. Finally, again aiming to maximize α , we chosekD = 10−3 M, based on estimates of 10−6−10−3 M for non-specific protein-protein interactions [167].Finally, we used recent experimental data to estimate the rate of perforin insertion into the membraneas kins = 2s−1 [168].A summary of the parameters used in this study, with literature sources, is shown in Table 2.1. Themodel was implemented in MATLAB and then compiled as a standalone executable. The core sourcecode is reproduced in the Appendix (Section D.1). For each parameter set, 100 independent simulationswere conducted. Simulations terminated when no further diffusive or chemical events were possible.Data analysis, visualization and plotting was conducted using the R statistical analysis language.2.2.7 Validation of SSSA computational implementationThe dynamics of perforin oligomerization and GZB translocation are nonlinear and no analytic solutionexists that describes our model, which motivated our choice of the SSSA method to study this system.However, to ensure that our algorithm was correctly implemented, we sought to compare its predictionsto those cases for which analytic solutions are tractable: chemical reactions according to mass action,26Symbol Value Units Description ReferenceR 3 µm IS radius [146]h 20 nm IS height [146]ξ 10 nm Thickness of cell membrane [161]RLG 0.5 µm Lytic granule radius [151]RGZB 2.5 nm GZB radius [147]RPFN 4 nm Perforin radius [137]Rp 10 nm Perforin pore luminal radius [137]NPFN 1500 - Number of Perforin monomers released [162, 169]NGZB 15000 - Number of GZB molecules released [162, 163, 169]Vavg 250 nm3 Avg. volume of synaptic molecules [165]f 0.4 - Fractional synaptic occupancy [166]kD 10−3 M Diss. constant for non-specific binding [167]kins 2 s−1 Rate of perforin monomer insertion [168]T 310 K Human body temperature -ηw 6.53×10−10 kg µm−1s−1 Viscosity of water -ηl 100ηw kg µm−1s−1 Viscosity of cell membrane -l 0.5 µm Sub-volume side length -Table 2.1: Parameters used in the immune synapse computational model.and molecular diffusion due to Brownian motion.Chemical kineticsWe consider the simple case of a three species system given by A+B k−→C, which can be described bythe following set of ODEs:dAdt= −kABdBdt= −kABdCdt= kAB(2.9)To implement this limiting case in our SSSA model we set all diffusivities equal to zero, and initialized asingle subvolume with 2000 perforin pores and 3000 granzyme molecules, and then simulated the timeevolution of granzyme translocation through the perforin pores. Due to the nonlinearity of the aboveODE system, there is no closed form analytic solution, so we solved the system numerically usingode45 in Matlab, with the initial conditions as described above, and k = kg. The results are shownin Figure 2.2, and show good agreement, indicating that our algorithm correctly describes chemicalinteractions.27Figure 2.2: Mass action kinetics are accurately captured by the SSSA algorithm. A single sub-volume was initialized with 2000 perforin pores (red) and 3000 granzyme molecules (blue). Thissystem was simulated using our SSSA, with all diffusivities set to zero, and all chemical rate con-stants set to zero, save the rate of granzyme translocation through perforin pores, which was setto kg. Data from 100 independent simulations is plotted. Analytic solutions to this system werecalculated from the corresponding ODE system in Matlab, and are plotted as black lines. Thegreen data set is internalized granzyme.Free diffusion in the synapseWe consider an initial bolus of No molecules released into the synapse from the centrally located lyticgranule (of radius RLG), and aim to describe the diffusive spread of these molecules. To model thisprocess analytically, we sought solutions of the diffusion equation: ∂u/∂ t =D∇2u , where u= u(x,y, t)is the molecular concentration and D the diffusivity. In order to naturally parameterize the square subvolumes, we chose to adopt a square geometry in which solve the diffusion equation, using cartesiancoordinates and a square synapse of side length 2R, chosen to be the same value as the diameter ofthe circular synapse in the SSSA algorithm. We assume that the aspect ratio is extreme so that thetimescale of diffusion in the z-direction (across the synapse) is very short compared to that for xy plane,and spatial gradients in this direction may be neglected. Therefore we sought solutions to the diffusionequation in two dimensions over the domain S = {(x,y)|−R ≤ x ≤ R,−R ≤ y ≤ R}. Since we allowmolecules to escape the synapse in our simulations, we adopt absorbing boundary conditions withu(±R,±R, t) = 0. Finally the initial state is taken to be u(x,y,0) = uo = No/R2LG for |x,y| ≤ RLG, andzero elsewhere.Using separation of variables, it is a fairly simple exercise [170] to find the solution for this problemin terms of sine functions:u(x,y, t) =∞∑m=1∞∑n=1cm,n sin(αm(R+ x))sin(αn(R+ y))e−D(α2n+α2m)t28where αn = npi/2R. Using the initial conditions and the orthogonality of sines, it is straightforward toobtain the coefficients:cm,n = uo4sin(αmR)sin(αnR)sin(αmRLG)sin(αnRLG)R2αmαnTo simulate simple diffusive spreading in our SSSA model, we initialized the sub volumes containingthe central lytic granule with No = 2000 perforin molecules, spread evenly over the sub volumes. Weset all rate constants to zero, and simulated the time evolution of the system 100 times, and calculatedthe average concentration and standard deviation for each sub volume, at each time point. We took asubset of this data (y = 0.25) that corresponds to a cross section of sub volumes that run as close aspossible to the centre of the lytic granule and synapse. To confirm that this approach was representative,we repeated the subsetting for many different cross-sections, and in all cases the diffusive spread wasGaussian as expected. The concentration profiles from both our analytic and simulated data are plottedfor various times in Figure 2.3. We found that our simulations reproduce the analytic solution well, andthat the error introduced by the choice of sub volume size (initially l = 0.5µm) is reduced at smallersub volume sizes. Since the agreement was already quite good for l = 0.5µm, we felt that the largeincrease in computational cost was not worth the minor correction in accuracy and all simulations inthe main paper are presented with l = 0.5µm as the sub volume dimension.Free diffusion in the synapse with membrane insertionWe next considered a system as above, with the addition that perforin can insert into the target cellmembrane. For our analytic solution the diffusion equation becomes ∂u/∂ t =D∇2u−kinsu, where kinsis the membrane insertion rate of perforin as in the main text. The analytic solution has a very similarform to that of the previous section, with the exponential term becoming e−D(α2n+α2m)t−kinst . To simulatethis case we adopted the same approach as in the previous section, and set kins = 2s−1, rather thanzero. We then generated the analytic and simulated data in the same way, and the results are plotted inFigure 2.4. Again the analytic and simulated data compare well.Diffusion and reaction in the target cell membraneTo test our SSSA further, we considered the problem of perforin monomers diffusing and reacting inthe target cell membrane. Since our model does not allow membrane inserted perforin monomers toescape the synapse, we adopted reflecting (Neumann) rather than absorbing boundary conditions inderiving an analytic solution. In this case separation of variables yields a solution that is a series ofcosines rather than sines:29lllll llllllllll lllllllll lllllllll llllllllll lllll0100200300400−2 −1 0 1 2Distance from centre of synapse (µ m)Concentration (µ m−2 )Time (s)lllll0.0040.0080.0120.0160.02l llllllllllllllllllll llllllllll l l llllll llllllllllll ll lllllll lll l l l ll l ll ll llll l ll l l l l l l l l ll0100200300400500−3 −2 −1 0 1 2 3Distance from centre of synapse (µ m)Concentration (µ m−2 )Time (s)lllll0.0040.0080.0120.0160.02Figure 2.3: Free diffusion described by the two-dimensional diffusion equation compares wellwith that simulated by the SSSA algorithm. The subvolumes comprising the central lytic granulewere initialized with 2000 perforin molecules spread across the subvolumes. All chemical rateconstants were set to zero, resulting in free diffusion of perforin, and the system was simulated100 times. We then plotted the concentration profiles at various times for the plane y = 0.25, witheach dot representing the concentration in the subvolume with coordinates (x,0.25), where x isplotted on the x axis in the figure. The shaded region represents the standard deviation of theaverage concentration calculated from 100 independent simulations. The corresponding solutionsto the two-dimensional diffusion equation were calculated in Matlab up to the twentieth term inthe infinite sums, and are plotted as a solid line. Left panel: sub volumes of side l = 0.5µm. Rightpanel: sub volumes of side l = 0.25µm.u(x,y, t) = coo+∞∑n=1cn cos(αnx)e−Dα2n t +∞∑m=1cn cos(αmy)e−Dα2mt+∞∑n=1∞∑m=1cn,m cos(αnx)cos(αmy)e−D(α2m+α2n )t(2.10)where αn = npi/R. Using the initial conditions and the orthogonality of cosines, we obtain the followingexpressions for the coefficients:coo =uoR2LGR2, cn =2RLGuonpiRsin(αnRLG), cn,m =4uomnpi2sin(αnRLG)sin(αmRLG).The SSSA simulations were obtained in the same manner as in the initial case of free diffusion, ex-cept that the sub volumes were initialized with membrane inserted perforin monomers, rather than freeperforin. Perforin oligomer aggregation was suppressed by setting all chemical interaction rate con-stants to zero. The resulting simulations and analytic solutions are plotted in Figure 2.5. Again thecomparisons are reasonable and can be improved by decreasing the sub-volume size (not shown). Notethat since the membrane inserted perforin monomers cannot escape, the system reaches an equilibrium30lllll llllllllll lllllllll llllllll llllllll l ll0200400600800−2 −1 0 1 2Distance from centre of synapse (µ m)Concentration (µ m−2 )Time (s)lllll0.0020.0040.0060.0080.015Figure 2.4: Free diffusion with attenuation due to target cell membrane insertion is well describedby the SSSA algorithm. The data was generated and plotted in an identical manner to Figure 2.3save the additional exponential multiplicative term e−kinst in the analytic solution, and the perforinmembrane insertion rate kins was set at its value stated Table 2.1 (rather than zero as in the previousfigures).concentration given by the initial particle number (No) divided by the total area of the synapse. Sincethis area differs for the analytic (2R2) and SSSA (piR2) data, the equilibrium concentrations are slightlydifferent.Finally, we ensured that membrane reactions are correctly captured by our SSSA. We simulated thereaction A+B→C for membrane-bound diffusing molecules A and B which react to form an immobilecomplex C with bimolecular rate constant k. This system can be modelled using the following PDEsystem for the concentrations A(x,y, t),B(x,y, t) and C(x,y, t):∂A∂ t= DA∇2A− kAB∂B∂ t= DB∇2A− kAB∂C∂ t= kAB.We simulated this system using our SSSA and compared the results to those obtained from numerical31lllll llllllllll llllllllll llllll ll l l l l lll l l l l0200400600800−2 −1 0 1 2Distance from centre of synapse (µ m)Concentration (µ m−2 )Time (s)lllll0.10.20.40.81.5Figure 2.5: Diffusion in the target cell membrane diffusion described by the two-dimensional dif-fusion equation compares well with that simulated by the SSSA algorithm. The subvolumes com-prising the central lytic granule were initialized with 2000 membrane inserted perforin monomersspread across the subvolumes. All chemical rate constants were set to zero, resulting in perforinmonomers diffusing in the target cell membrane. The data was then generated and plotted in anidentical manner to previous figures.solution of the PDE system. We used a square domain of size 5µm×5µm. In the SSSA, the domain wasdivided into square sub-regions of side 0.5µm. We initialized 500 A molecules and 700 B molecules inthe central four sub-regions of the domain and simulated up to a fixed end-time with varying values ofk spanning two orders of magnitude. As shown in Figure Figure 2.6, the SSSA and PDE results agreewell.2.3 Results2.3.1 Free diffusion in the synapse is incompatible with granzyme internalizationWe initially built our model without including the hindered diffusion coefficient α . However, when weplotted the various molecular species against time, as in Figure 2.7, it became apparent that virtuallyall GZB, and a large majority of PFN, escaped the synapse before pore formation could occur. This32● ● ● ●●●●●●●●●●●●●●●●●●●●●● ● ● ●0100200300400−2 −1 0 1 2Distance from centre of synapse (μ m)Concentration (μ m−2)k (μ m2s−1)●●●●●●●●●●0.11.22.33.44.55.66.77.88.910Figure 2.6: A two-dimensional reaction-diffusion system is well described by our SSSA. TheSSSA was applied to the two-dimensional A+B→C system and concentrations of reaction prod-uct C were extracted for a line of sub-regions across the simulation domain, as in previous figures(coloured dots indicate averages and coloured regions± standard deviation over 100 simulations).The corresponding PDE system was solved using finite differences and the solution was then av-eraged over the regions corresponding to the same sub-regions of the SSSA to allow easy compar-ison (black dashed lines). SSSA and PDE results are shown for 10 different values of k spanningtwo orders of magnitude. We set DA = 2.9µm2s−1 and DB = 2.7µm2s−1continued to be the case when we increased the numbers of both GZB and PFN by one to two ordersof magnitude. Since lytic granule mediated cytotoxicity is known to be crucial for CL killing, thislack of internalization indicated that the aspect ratio of the synapse alone was insufficient to containmolecules in the IS. This motivated us to re-examine the physical environment of the IS. In contrast toa simple aqueous space, the IS contains a high density of signalling and adhesion molecules, giving ita characteristic electron dense appearance in electron micrographs. Incorporating a hindered diffusionparameter α (see Methods) to model the reduced diffusivity of GZB and PFN due to these condi-tions resulted in a marked decrease in the loss of both molecules, leading to increased pore formationprobability and GZB internalization.330.000.250.500.751.000.00 0.02 0.04 0.06 0.08 0.10Time (s)Fraction of total molecules per species in systemSpeciesSynaptic PerforinSynaptic Granzyme BEscaped PerforinEscaped Granzyme BPoresInternalized Granzyme BFigure 2.7: Time evolution of GZB and PFN in the immunological synapse without hindereddiffusion. Absolute numbers for each species were normalized to the initial total amount of thatmolecule in the system. These results show that in the absence of hindered diffusion there israpid and complete loss of GZB, and near complete loss of PFN, without pore formation or GZBinternalization. Here, NPFN = 1500.2.3.2 Pore formation is influenced by the amount of perforin releasedWhen we included a description of hindered diffusion in our model, the rate of GZB and PFN losswas dramatically attenuated, and occasionally pore formation did occur. However, the probability ofpore formation was still only 0.04, which we consider incompatible with physiological expectations.Therefore we explored the importance of the amount of perforin released into the synapse. Interest-ingly, increasing PFN number did increase pore formation probability (Figure 2.8a), with this lattervalue increasing from close to zero at low PFN numbers, to unity at high PFN numbers. Most no-tably, in control simulations without hindered diffusion (α = 1) at the maximum PFN value (6000Nin Figure 2.8a), virtually no pore formation occurs, as compared to consistent pore formation for thismaximum PFN value with hindered diffusion. This reinforces the argument that hindered diffusion iscritical for the system to function. These results also introduce the recurrent finding that appreciablequantities of GZB are internalized if pore formation occurs, but even in these cases the majority ofGZB still escapes the synapse. We return to these points in the discussion below.34a0.00.20.40.60.81.0600N 600 1200 1800 2400 3000 3600 4200 4800 5400 6000 6000NNumber of Perforin Molecules Released into SynapseProbability of Pore Formation Number of Pores Formed1 2 3 4 5 6 8b0.00.51.00.0 0.2 0.4 0.6 0.8Time (s)Probability of Pore FormationNumber of Perforin Molecules Released into Synapse600N 600 1200 1800 2400 3000 3600 4200 4800 5400 6000 6000Nl l l l llllllllc0.9000.9250.9500.9751.0001.025600N 600 1200 1800 2400 3000 3600 4200 4800 5400 6000 6000NNumber of Perforin Molecules Released into SynapseAverage Fraction of Escaped  Granzyme Molecules d02004006000.0 0.2 0.4 0.6 0.8Time (s)Average Amount of Granzyme B Internalized Number of Perforin Molecules Released into Synapse600N 600 1200 1800 2400 3000 3600 4200 4800 5400 6000 6000NFigure 2.8: Effect of the amount of PFN released on pore formation probability and GZB in-ternalization. (a,b) The probability of pore formation undergoes a transition from minimal poreformation to consistent pore formation over one order of magnitude of PFN number released intothe synapse volume, but even at the maximum PFN value we consider, hindered diffusion is re-quired for this effect. (c,d) Even when pore formation is certain, the majority of GZB still escapesthe synapse. Each error bar represents standard deviation over 100 simulations. The baseline hin-dered diffusion is α = 0.306, which is obtained from Equation 2.6 using values from Table 2.1 inthe methods. The N suffix indicates no hindered diffusion (α = 1).2.3.3 The amount of granzyme internalized depends strongly on the rate of poreformationSince the amount of PFN initially released had such a strong effect on the probability of pore formationand GZB internalization, we examined the potential correlation between the amount of GZB releasedand GZB internalized. Due to the uncertainty in the literature concerning the amount of GZB released,we present results over two orders of magnitude of NGZB. As can be seen from Figure 2.9a whilethe amount of GZB internalized does increase with GZB released, it is a very modest effect. Wehypothesized that the effect was so weak because GZB internalization is entirely dependent on poreformation: prior to pore formation, GZB is lost rapidly due to diffusive escape from the IS. Since35ll llll llllll01002003004005006006N 6 72 138 204 270 336 402 468 534 600 600NNumber of Granzyme Molecules Released into Synapse (x103)Average Number of Internalized Granzyme Molecules1e−051e−041e−031e−021e−011e+000.0 0.1 0.2 0.3 0.4 0.5Time to First Pore Formation (s)Fraction of Granzyme B Internalized (log scale)Number ofPores Formedlllll1234> 4Figure 2.9: Dependence of GZB internalization on rate of pore formation. (a) Despite an increaseof two orders of magnitude in the amount of GZB released, only a very modest increase in theaverage amount of GZB internalized is observed. The N suffix indicates no hindered diffusion(α = 1) resulting in no GZB internalization. Each error bar represents standard deviation over 100simulations. (b) Data from all simulations in this study is pooled, showing a tight dependency ofGZB internalization on the rate of PFN pore formation. The three different clusters correspond todifferent parameter regimes.the rate of loss is independent of total GZB number, increasing the total amount will not change thefraction of GZB preserved at the time of pore formation, but rather the absolute number. To confirmthis, we pooled data from all simulations in our study, across heterogeneous parameter sets, plotting thefraction of total GZB that is internalized against the time to first pore formation (Figure 2.9b). A distinctnegative correlation is observed, indicating that total GZB internalization is primarily a function of poreformation. Finally, we note the continued importance of hindered diffusion in maintaining appreciablelevels of GZB internalization (Figure 2.9a).2.3.4 Hindered diffusion critically influences pore formation and granzymeinternalizationEven when varying other parameters across several orders of magnitude, hindered diffusion provedcritical in all cases for pore formation and GZB internalization. Therefore, we sought to quantitativelyinvestigate its importance by varying α from 0.1 to 1, corresponding to marked hindered diffusion andfree diffusion, respectively. We chose to vary α rather than any of its constituent parameters fromEquation 2.6 to explore its effect in an unbiased manner. As expected, at high levels of hindereddiffusion, the loss of molecules is greatly attenuated, and pore formation occurs with a probabilitynearing unity, while when molecules diffuse freely, pore formation is rare (Figure 2.10a,b). Similarly,GZB internalization is significant at high levels of hindered diffusion and almost non-existent in the36a0.00.20.40.60.81.00.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1NHindered Diffusion Factor (α)Probability of Pore FormationNumber of Pores Formed1 2 3 4 5 6 8b0.010.101.000.0 0.5 1.0 1.5 2.0Time (s)Probability of Pore FormationHindered Diffusion Factor (α)0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Nll l l l l l l l lc0.850.900.951.000.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1NHindered Diffusion Factor (α)Average Fraction of Escaped  Granzyme Moleculesd1e−021e−011e+001e+011e+021e+031e+041e+050.0 0.5 1.0 1.5 2.0Time (s)Average Amount of Granzyme B Internalized Hindered Diffusion Factor (α)0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1NFigure 2.10: Importance of hindered diffusion in pore formation and GZB internalization. (a)Transition from consistent pore formation to minimal pore formation over decreasing levels ofhindered diffusion. The importance of α is highlighted by the variability of pore formation prob-ability (two orders of magnitude, (a,b)), and amount of GZB internalization (five orders of mag-nitude, (d)). While it is evident from (d) that appreciable GZB is internalized at high levels ofhindered diffusion, (c) shows that the majority of GZB still escapes the synapse in these cases.Each error bar represents standard deviation over 100 simulations. The N suffix indicates freediffusion.case of free diffusion (Figure 2.10c,d). Most notably, the variation in pore formation probability andGZB internalization extends over two and five orders of magnitude respectively, demonstrating theimportance of hindered diffusion in creating an IS environment that is conducive to GZB internalizationand effective target killing. We also note that even for α = 0.1, corresponding to high levels of hindereddiffusion, 95% of GZB escapes the synapse.372.3.5 Dependence of pore formation and granzyme delivery on perforin insertion anddiffusion in the target cell membraneTo investigate the effect of the diffusion coefficient for PFN oligomers in the target cell membrane, werepeated the analysis of the previous section, but with a tenfold-reduced membrane diffusivity. Theresults are shown in Figure 2.11. Comparing to Figure 2.10, we see approximately ten-fold lowerpore formation and GZB internalization, but the same general pattern of results: pore formation andgranzyme internalization depend heavily on a high level of hindered diffusion, and even in that casea large amount of GZB escapes. The main reason for this reduction in pore formation and granzymeinternalization is that the lower perforin oligomer diffusivity strongly effects the ability of large j-mersof PFN to find each other and form pores, but free GZB is still escaping from the synapse at the samerate as in the simulations shown in Figure 2.10.We also studied the effect of the PFN insertion rate into the membrane, kins. As one would expect,the likelihood of pore formation and GZB delivery increases with this parameter, but importantly,even at the highest value of kins we considered, successful delivery is strongly dependent on hindereddiffusion (Figure 2.12).2.4 DiscussionWe have used a mathematical model to investigate important features of the granzyme-perforin path-way, a two-component system used by cytotoxic cells of the immune system to kill infected and ma-lignant cells. Our key findings are as follows: (i) robust perforin pore formation and GZB delivery totarget cells requires rapid pore formation and molecular crowding that hinders diffusive transport in thesynaptic volume, thus slowing molecular escape; and (ii) even in regimes where we predict consistentformation of perforin pores and appreciable GZB internalization, the vast majority of GZB escapesfrom the synapse.Historically, the potency and specificity of cytotoxic lymphocyte killing has been understood in thecontext of two observational constraints: effective target cell lysis with bystander sparing [171]. Thishas been explained by a model in which the IS volume is effectively sealed by extremely close prox-imity between the cell membranes at its edges, which would physically prevent the escape of secretedmolecules and thus minimize bystander killing, while also ensuring adequate species concentrations inthe synapse for target cell killing. Our first models of the IS, consisting of a simple aqueous environ-ment, with no peripheral seal, clearly demonstrated that the high aspect ratio of the synapse in isolationwas insufficient for molecular confinement and therefore CL function. Rather than adding such a seal,for which no mechanism has been even proposed, let alone observed, we considered alternative mecha-nisms for molecular entrapment. We noted the electron density of the IS on electron micrographs, and,assuming that this was due to a high density of signalling and adhesion molecules, we hypothesizedthat these molecules effectively hinder diffusion in the synapse, resulting in sufficient confinement ofperforin and GZB in the IS for both bystander sparing and target cell killing.38a0.00.20.40.60.81.00.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1NHindered Diffusion Factor (α)Probability of Pore FormationNumber of Pores Formed1 2b0.010.101.000 1 2 3 4Time (s)Probability of Pore FormationHindered Diffusion Factor (α)0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Nll l l l l l l l lc0.960.981.001.020.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1NHindered Diffusion Factor (α)Average Fraction of Escaped  Granzyme Moleculesd1e−021e−011e+001e+011e+021e+031e+041e+050 1 2 3 4Time (s)Average Amount of Granzyme B Internalized Hindered Diffusion Factor (α)0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1NFigure 2.11: Reduced diffusion of perforin in the target cell membrane reduces pore formationand GZB internalization. In this figure, we have reduced the membrane diffusion constant for PFNby a factor of 10 compared to Figure 2.10. The main finding is that reduced membrane diffusionis detrimental to pore formation and GZB delivery. Each error bar represents standard deviationover 100 simulations. The N suffix indicates free diffusion.When we updated our model to reflect this hypothesis, we found consistent pore formation andGZB internalization under various parameter regimes, indicating hindered diffusion is sufficient fortarget cell killing. However, even with rapid pore formation, the majority of GZB still escapes thesynapse, along with a significant amount of perforin, raising the possibility of bystander killing due tothese escaped molecules. We believe that this is highly unlikely to be a significant issue, because of therequirement for very high and localized concentrations of perforin to effectively form pores. Based onour model, such high concentrations are only fleetingly present at the point of release of a lytic granule,and so will likely not occur at any distance from the IS, let alone at even more dilute concentrationsin the intracellular milieu. Even if a pore were to form, escaped GZB would be similarly diluted,making internalization of significant quantities of GZB highly unlikely. Additionally, cellular porerepair mechanisms act reasonably quickly [136], further reducing the potential for bystander killing inthe absence of a simultaneous high local concentration of GZB. This requirement of spatiotemporal39a0.000.250.500.751.000.2N 0.2 2.4 4.6 6.8 9 11.2 13.4 15.6 17.8 20 20NRate of Perforin Monomer Insertion into Membrane (s−1)Probability of Pore FormationNumber of Pores Formed1 2 3 4 5 6 7 8 9b0.00.51.00.0 0.1 0.2 0.3 0.4 0.5Time (s)Probability of Pore FormationRate of Perforin Monomer Insertion into Membrane (s−1)0.2N 0.2 2.4 4.6 6.8 9 11.2 13.4 15.6 17.8 20 20Nllllllllllllc0.800.850.900.951.000.2N 0.2 2.4 4.6 6.8 9 11.2 13.4 15.6 17.8 20 20NRate of Perforin Monomer Insertion into Membrane (s−1)Average Fraction of Escaped  Granzyme Moleculesd01000200030000.0 0.1 0.2 0.3 0.4 0.5Time (s)Average Amount of Granzyme B Internalized Rate of Perforin Monomer Insertion into Membrane (s−1)0.2N 0.2 2.4 4.6 6.8 9 11.2 13.4 15.6 17.8 20 20NFigure 2.12: Impact of the rate of perforin insertion into the target cell membrane on pore forma-tion and granzyme internalization. Pore formation and granzyme internalization increase dramat-ically with increasing membrane insertion rate. This is to be expected since a key determinant inthese processes is the ratio of perforin escaped to perforin inserted. By increasing the rate of in-sertion, more perforin is preserved to make pores. Importantly, as with all other parameters, evenat the highest rate studied, almost no pore formation or granzyme internalization occurs in theabsence of hindered diffusion. Each error bar represents standard deviation over 100 simulations.The N suffix indicates free diffusion.co-localization of high concentrations of perforin and GZB for pore formation and killing represents avery effective safety mechanism that avoids bystander killing, but nonetheless allows efficient targetedkilling. If we consider the requirement of both a GZB and perforin signal as binary in space and time,but the strength of the signal as analog, and therefore multiplicative, we see that this two componentsystem creates a filter which yields a signal that is very strong when co-localization occurs, but israpidly attenuated when the two species are spatially or temporally disparate. As opposed to the ‘sealed’IS model which conflates bystander sparing and target cell killing in a single mechanism (the seal),this hypothesis allows for the decoupling of pore formation, resulting from hindered diffusion, andbystander cell survival, resulting from the bimolecular filter.40Our hindered diffusion based model may also help explain the experimental observation that CTLscan kill multiple targets encountered simultaneously by polarizing lytic granules toward multiple targets[172]. Interestingly, a complete IS is not formed for every killed target. Under a sealed IS model, itis difficult to conceive how sufficiently high concentrations of GZB and perforin could be maintainedin such a ‘multi-target-leaky’ IS to achieve target cell killing. However, in our model, where highconcentrations are sustained by slowed diffusivity due to a crowded synapse, one could imagine thatwhile the synapse is not a tight, organized apposition, there are still significant amounts of adhesionand signalling molecules present. In this case, despite an incomplete appearance, the synapse wouldstill be crowded, with these molecules still slowing the escape of GZB and perforin, and thus enablingcontinued target cell lysis.From a biophysical standpoint, this model consists of four interacting processes, each with differenttimescales: (i) diffusion, whether hindered or free, in the synapse which influences the rate of molecularescape at the lateral edges of the synapse; (ii) the rate of perforin insertion into the target cell membrane;(iii) the rate of diffusion and aggregation of perforin oligomers in the target cell membrane whichinfluences the rate of pore formation; and (iv) the rate that granzyme finds perforin pores. The relativetimescales of these processes determine whether sufficient granzyme and perforin are retained in thesynapse, for a long enough time, to allow for pore formation and granzyme internalization. In reality,the first three are the rate limiting steps: once pore formation occurs, granzyme internalization is veryrapid.There are myriad effects that could influence these processes such as the volume occupancy of thesynapse, specific or non-specific interactions of the two species with each other or other molecules,spatial variations in the width or height of the synapse, active membrane processes at the target cellinvolving surface molecules or the cytoskeleton, and the possible presence of a heretofore unobservedphysical seal at the edges of the synapse, to name but a few. Given the sparse quantitative data re-garding these effects, rather than attempt to incorporate them into first-principles descriptions for thethree processes listed above, we have systematically investigated the influence these processes have onlymphocyte function. Our quantitative model allowed us to delineate the sensitivity of the granzyme-perforin pathway to these parameters by calculating the effect that varying a parameter has on theprobability of pore formation, the key determinant of cytotoxic lymphocyte killing in our model. Oneway to quantitate this is to use a metric of maximum difference in pore formation probability normal-ized to the fold change in the parameter value that was varied (∆Pmax), with the normalization allowingfor comparison between parameters with different units. Using this approach we investigated the de-gree of hindered diffusion (∆Pmax = 0.335, Figure 2.10), the rate of perforin insertion (∆Pmax = 0.162,Figure 2.12), and the diffusivity of perforin in the target cell membrane (∆Pmax = 0.069, Figure 2.11),and found that the most critical parameter of these three for pore formation and granzyme internaliza-tion is the degree of hindered diffusion in the IS. This can be most clearly seen by noting that whenhindered diffusion is replaced by free diffusion in our model, pore formation is dramatically attenuatedor eliminated, even at the extreme values of the other parameters we investigated.41While our model is appealing in its mechanistic simplicity, there are certainly others that are pos-sible, such as a peripheral ‘seal’, or transient, localized confinement of perforin in the target cell mem-brane [173–175] enhancing pore formation. We investigated this latter hypothesis (Figure 2.11) andfound that our model predicts significantly reduced target cell killing due to slowed PFN pore forma-tion. Importantly none of these models are mutually exclusive, and further computational and experi-mental work to delineate the relative importance of these mechanisms, as well as to further characterizethis important system, would be welcome. To investigate the ‘sealed’ IS model using our computationalimplementation, the most reasonable approach would be to use a spatially-dependent hindered diffu-sivity that is highest at the IS boundary as a model for the ‘seal’. Experimentally, there are two testablecharacteristics which would help distinguish between the two models. The first is the maximum sizeof molecule that can enter or exit the synapse, which could be tested by adding fluorescently taggedinert polymers of increasing size to the extracellular milieu of a CL-target cell conjugate, and usingsingle-molecule microscopy to determine the maximum size of molecule that enters the synapse. Com-paring this maximum size with the geometry of the synapse could provide insight into the nature ofa peripheral seal. Second, our model rests on the notion of hindered diffusion, which implies that thediffusivity of a molecule in the synapse should be well below free diffusion. To test this, fluorescentlytagged molecules might be observed within the synaptic region, and their diffusivity measured usingfluorescence recovery or single particle tracking. A diffusivity similar to that of free diffusion in anaqueous environment would argue against our model.In closing, we note that there is rapidly increasing excitement surrounding cancer immunotherapiesin general [176], and recognition of the central role that cytotoxic lymphocytes play in these modalities.As the granzyme-perforin pathway is crucial to the lytic capability of these cells, we believe that a betterunderstanding of the pathway may enable rational design of improved effector mechanisms for cellbased therapies that circumvent apoptosis resistant malignancies. In particular, our model’s predictionthat free diffusion of GZB is sufficient for its internalization was a crucial finding that enabled ourapproach to using GZB as a molecular chaperone for cell-to-cell delivery via the granzyme-perforinpathway.42Chapter 3Targeted cell-to-cell delivery of exogenousprotein payloads via thegranzyme-perforin pathway3.1 IntroductionWith their ability to sense and integrate a wide range of signals, actively move to specific tissue com-partments, and actuate context-dependent responses, engineered cell-based therapeutics are emergingas the next major class of medical intervention [177]. Chimeric antigen receptor T-cells (CAR-Ts)are highly effective in treating hematological malignancies [79–81], and many mesenchymal stem celltherapies [178] are at various stages of development for use in cardiac [36, 37], neurological [38],and malignant [39–41] disease. These advances are a result of recombining the diverse functional-ity of biological systems [179, 180] to generate new functional biological molecules and pathways[110, 111, 115, 116, 119]. Current cell-based therapeutics are limited by the small repertoire of avail-able modules and there is an unmet need for additional sensory and effector components for engineeredcell therapies.Cytotoxic lymphocytes have exceptional utility as cellular therapeutics because they are targetable,expandable, and amenable to genetic manipulation [181]. Cytotoxic lymphocytes possess a unique ef-fector mechanism, the granzyme-perforin pathway, one of the main ways in which they kill target cells[121]. The main components of this pathway are the serine protease granzymes, and the pore formingprotein perforin, both of which are stored in membrane bound secretory lysosomes, or lytic granules,in the cytosol of cytotoxic lymphocytes [182]. Upon target cell recognition, the cytotoxic lymphocyteforms a tight apposition with the target cell, forming the immunological synapse. Surface receptorsignaling results in the endocytic release of granzymes and perforin from the lytic granules into thesynapse between the two cells [129]. Perforin then inserts in the target cell membrane and oligomer-izes to form transient pores, through which granzyme diffuses into the target cell [136, 137, 139].43Finally, granzyme cleaves caspases and BID to initiate target cell apoptosis. Importantly, surround-ing bystander cells typically do not receive appreciable quantities of granzymes [142, 171, 183]. Insummary, surface receptor mediated target cell recognition results in specific, cytoplasm-to-cytoplasmintercellular transfer of granzymes to that same target cell.Granzyme B (GZB) is a well studied effector molecule that transits the granzyme-perforin pathway.Here I engineer GZB-derived chaperones and trace chaperone mediated trafficking of a functionalfluorescent protein payload through this pathway from effector to target cells. This constitutes a cell-to-cell transfer module that can be used in cellular therapeutics to deliver ectopic protein payloads totargeted tissues or cells.3.2 ResultsDesign of granzyme B derived molecular chaperonesGranzyme B is synthesized as a pre-pro-protein, with an 18 amino acid N-terminal ER signal peptide,followed by an inhibitory dipeptide, followed by the rest of the protein [182]. Upon initiation oftranslation, the ER signal peptide directs the nascent protein to the ER, where it is co-translationallyinserted into the ER. As the protein is synthesized in the ER, an N-glycan is added, which targetsthe protein to the Golgi network once synthesis is complete. In the Golgi, the N-glycan is furtherphosphorylated. This phosphosugar moiety on granzyme B then binds to the mannose-6-phosphatereceptor, which targets the protein to lytic granules, where it is sequestered until target cell recognition,resulting in granzyme B release into the immune synapse [129]. Importantly, recent work has shownthat following release into the immune synapse, the trafficking of GZB to the target cell membrane andentry into the target cell via perforin pores is likely a result of passive diffusion only [136, 137, 142].I used this information to guide my design of chaperones for granzyme-perforin mediated delivery.Since the steps in this process that are downstream of lytic granule exocytosis appear to be passive,I hypothesized that a chaperone that successfully directed a payload to be loaded into lytic granuleswould also be sufficient for payload delivery to the target cell. In designing such a chaperone, I adoptedtwo strategies: rational and empirical.For the rational design, I sought to develop a set of minimal granzyme B domains that would shuttlea protein payload to lytic granules. I took this set to be an ER localization domain, and a lytic granulelocalization domain. For the former, I used the GZB ER signal peptide (GZBSS). For the latter, Iused a 53 amino acid motif surrounding two computationally predicted N-linked glycosylation sites(GZBSM). The final rational design consisted of GZBSS at the N-terminus, followed by the modelpayload, followed by GZBSM (Figure 3.1). For a model payload, I selected crmCherry (hereaftermCherry or MCH), a derivative of the mCherry red fluorescent protein, that is known to be stable inthe acidic environment of lysosomes [184].The behaviour of chimeric proteins consisting of domains derived from multiple proteins that have44been rearranged is unpredictable. Therefore, I also selected full length granzyme B as an empiricalchaperone. My rationale for this choice was that if there were unknown domains within granzyme Bother than the region surrounding the N-linked glycosylation sites that were necessary for lytic granuleloading, or if the necessary domains are adjacent to the N-linked glycosylation sites only in the tertiarystructure of granzyme B, then they would be captured in the full length protein. To keep granzyme B inas native a form as possible, I fused the MCH payload to the C-terminus of GZB, with the two proteinsconnected by a flexible glycine serine linker (Figure 3.1).As controls, I also generated two additional constructs: MCH alone, and GZBSS followed by MCH(Figure 3.1).A) B)   matGZB GZBSS GSL crmCherry GZBSS crmCherry GZBSM GZBSS crmCherry crmCherry Figure 3.1: Design of payload delivery module chaperones. (a) Granzyme B as a model proteinthat transits the granzyme-perforin pathway. The full length coding sequence is shown in green,with the ER signal peptide in light green (GZBSS). The two putative N-linked glycosylation motifsare shown in blue, with the encompassing putative sort motif (GZBSM) in yellow. (b) Schematicof the constructs used in this study. mCherry (red), an RFP protein was used as a model payload,and a flexible glycine serine linker (GSL, purple) was used to join granzyme B to mCherry.3.2.1 Screening chaperone designs by assessing lytic granule colocalization usingconfocal microscopySince my hypothesis was that lytic granule loading of a payload would be sufficient for payload de-livery, I first investigated the subcellular localization of the chaperones using confocal microscopy. Iexpressed the candidates in the natural killer cell line YT-Indy (hereafter YT), which retains a functionalgranzyme-perforin pathway and has well characterized target cell lines [185]. Following enrichmentfor mCherry+ cells via cell sorting, I stained the cells for the lysosomal and lytic granule marker Lamp145and then acquired images of the cells using confocal microscopy. As expected, the Lamp1 distributionwas punctate in nature, but the MCH distribution was highly variable (Figure 3.2).Due to the range of phenotypes observed in the images, I sought to evaluate the degree of payload(MCH) colocalization with Lamp1 in an unbiased manner. To do this I developed a semi-automatedimage filtration and analysis pipeline. The algorithm is illustrated in Figure 3.3 and seeks to eliminateboth local background and bleed, as well as pixel noise. This is achieved using both local and globalimage information to filter each pixel. This filtering is critical to enable quantitation of colocalization,as it eliminates the background noise from the regions of the image in which there are no cells, as wellas regions that are adjacent to granules that have moderate signal intensity, both of which could give aspurious contribution to any quantitative metric of colocalization. The efficacy of this method can beobserved by examining the progression of the three columns from top to bottom of Figure 3.3: notethat pixel intensities of the punctate structures remain relatively intact, whereas the binarized images(showing the extent of the background signal) change from containing large homogeneous structuresto puncta that closely resemble those in the pixel intensity images.Using this approach, I quantified the colocalization between MCH and Lamp1 in these filtered im-ages using the Manders M1 coefficient and Pearsons Correlation Coefficient (PCC) (Figure 3.4). Bothmetrics indicated that MCH alone had a low degree of colocalization with lytic granules, which wouldbe expected as the lytic granules are small dense granules, and unfused MCH is distributed through-out the cytosol. GZBSS-MCH had high colocalization with Lamp1, with a perinuclear and membranedistribution, consistent with it entering the secretory pathway. GZB-MCH also had high Lamp1 colo-calization, but with punctate cytosolic distribution consistent with granule loading. GZBSS-MCH-GZBSM exhibits a partially punctate granular distribution (similar to GZB-MCH), but also a moderateintensity, diffuse cytosolic distribution (similar to unfused MCH). The mixed phenotype of GZBSS-MCH-GZBSM suggested this chaperone may not be effective at loading payloads into lytic granules.These imaging results suggested that of the two chaperone candidates, GZB had the most potential.3.2.2 Transfer of fusion proteins from effector to target cellsI next characterized the capacity of the two candidate chaperones (GZBSS-MCH and GZBSS-MCH-GZBSM) to facilitate transfer of the payload through the granzyme perforin pathway to target cells. Todo this I conducted a series of co-culture experiments, again using mCherry as a fluorescent reporterpayload that was easily traceable. Effectors expressing a variety of mCherry fusion proteins were co-cultured with fluorescently labeled target cells, and then analyzed for evidence of mCherry in the targetcell populations. I used the B-cell lymphoblastoid cell line 721.221 (hereafter 721) [186] as target cells,as they are a well known YT target.I started by testing GZB-MCH, GZBSS-MCH and MCH alone. GZB-MCH was the chaperone themicroscopy images had suggested was most likely to load payloads into lytic granules, while MCHalone was clearly not loaded into lytic granules, and thus a good control. GZBSS-MCH was includedto confirm that it was not being loaded into lytic granules and hence would not be transferred to target46GZBSS-MCH GZB-MCH GZBSS-MCH-GZBSM MCH Figure 3.2: Subcellular distribution of candidate chaperone-mCherry fusion proteins. YTs ex-pressing the candidate fusion proteins (labeled at left) were stained for the lytic granule markerLamp1 and imaged using confocal microscopy. Shown are merged red (mCherry) and green(Lamp1) channels for three representative cells for each sample.471D pixel intensity 2D pixel intensity 2D pixel binaryRawLocal backgroundBackground subtractedPixel noise thresholdThreshold appliedFigure 3.3: Automated image noise filtering pipeline. All panels are derived from a single channelof a single image. The first column consists of pixel intensity traces of a single horizontal line ofpixels through the corresponding whole two dimensional image, shown in the second column. Thethird column shows binarized versions of the middle column: all pixels with intensity greater than0 are set to 1. The first row is the raw image data. The second row is the local background of theimage. The third row is the background subtracted image, literally the second row subtracted fromthe first. The fourth row is identical to the third, except the first column plot has been enlargedto a small region (gray dashed lines) to better show the pixel noise (small fluctuations near zero).The horizontal red line is the threshold that will be applied to filter pixel noise. The bottom row isthe final processed image.48lllllllll lllllllllllllllllllllllllll ll llllllllllllllllllllllllllllllllllllll l0.00.20.40.6MCH GZBSS−MCH GZB−MCH GZBSS−MCH−GZBSMCandidate chaperoneColocalization scoreMetricllM1PCCFigure 3.4: Quantitative assessment of candidate chaperone colocalization with lytic granules.Two colocalization metrics were calculated: Manders M1 (quantifying the fraction of pixel inten-sity of mCherry positive pixels that also contains Lamp1 signal), and Pearson correlation coeffi-cient (PCC, quantifying the degree to which red and green pixel intensities are correlated). Thesewere calculated using both channels from each image. Each circle is the score of a single image,and each image contained between 2 and 5 cells. Overlaid are box and whisker plots.cells. YT cells were transfected with plasmids coding for these chaperones and then FACS enriched forRFP+ cells. The various effector cell types were separately co-cultured with 721 target cells that hadbeen labeled with a fluorescent dye, to distinguish between the effector and target cells. This mixed cellpopulation was then analyzed via flow cytometry (Figure 3.5). Target 721 cells that were co-culturedwith YTs expressing GZB-MCH show an increase in mCherry signal, as compared to 721s alone, 721sco-cultured with unmodified YTs, and 721s co-cultured with YTs expressing either MCH or GZBSS-MCH. Notably, this increase is most prominent in the dead cell fraction (DAPI+) of the target cells.Since the majority of these cells are dying due to YT attack, DAPI positivity can be viewed as a proxyfor YT targeting. Therefore, the increase in MCH signal in dead (DAPI+) targets co-cultured withYTs expressing GZB-MCH, but not MCH or GZBSS-MCH, suggests that GZB-MCH is transferred totarget cells specifically via chaperone mediated trafficking through the granzyme-perforin pathway.I then investigated if the rationally designed chaperone (GZBSS-MCH-GZBSM) would performsimilarly to GZB-MCH. I conducted the same type of experiment as above, comparing YTs expressingGZBSS-MCH, GZBSS-MCH-GZBSM and GZB-MCH. I selected GZBSS as the comparator so that allconstructs would have the same N-terminal ER signal peptide and potential for secretion. This would49mCherry Count DAPI mCherry FITC FSC DAPI Cells Targets Effectors A: target gating B: gated targets labeled by effector in co-culture YT:GZB-MCH YT:GZBSS-MCH YT:MCH YT ALONE Figure 3.5: Transfer of granzyme B mCherry fusion protein to target cells. YTs expressingvarious mCherry fusion proteins were co-cultured with CFSE labeled target 721 cells, and themixed population was analyzed via flow cytometry. (a) Gating strategy for isolating target cells.Debris was eliminated (top panel) and then FITC+ targets selected (bottom panel). Not shown isan intermediate hierarchical gating step in which doublets are excluded using forward and sidescatter width vs height gates. (b) Target 721 cell mCherry fluorescence. Each column is labeledby the effector that was present in the co-culture, but only target cells are plotted, using the gatingfrom (a). Each column contains the same data showing target cell populations from all co-cultures,but only a single target population is highlighted in blue, which corresponds to the effector partnerthat was present in the co-culture partner.allow us to differentiate between non-specific mCherry signal in target cells, and mCherry signal in tar-get cells resulting from chaperone mediated transfer. The data (Figure 3.6) is consistent with the initialco-culture experiments, and indicates that GZB transfers MCH to target cells, but GZBSS or GZBSSin combination with GZBSM does not. The fact that GZBSS-MCH-GZBSM did not transfer MCHto target cells is interesting. Given the mixed phenotype observed in YTs expressing this construct inthe microscopy images, I thought there was a possibility that it might also traffic to target cells. Thefact that it does not provides useful information concerning the nature and location of the granzyme Bmotifs responsible for its trafficking through the granzyme-perforin pathway, which I discuss below.Finally, to confirm at the protein level that GZB transfers the MCH payload to target cells, I re-peated the above experiments, with the additional, post co-culture, step of collecting live and dead(DAPI- and DAPI+ respectively) 721 target cells from each co-culture via FACS. Data from this sortis shown in Figure 3.7a, and it is consistent with the flow cytometry data from previous experiments.Whole cell lysates from the sorted target cell populations were then size-separated by gel electrophore-sis, and probed for mCherry via western blot (Figure 3.7b). A prominent 60 kDa band consistentwith GZB-MCH is observed in lysates of 721s co-cultured with YTs expressing GZB-MCH. Thereis also a background band of approximately 30 kDa consistent with unfused mCherry in the lysate of721s co-cultured with YTs expressing GZBSS-MCH. This is not unexpected as the ER signal peptidein GZBSS-MCH directs mCherry to the secretory pathway [187], resulting in extracellular mCherry,5002505007501000All Live DeadTarget cell populationTarget cell mCherry MFIEffector cell populationGZBSS−MCH GZBSS−MCH−GZBSM GZB−MCHp < 0.001 p < 0.001 p < 0.05 p < 0.05 p < 0.001 p < 0.001 n.s. n.s. n.s.Figure 3.6: Comparison of MCH payload transfer to target cells by the two granzyme B derivedchaperones. YTs expressing GZBSS-MCH (red), GZBSS-MCH-GZBSM (green) or GZB-MCH(blue) fusion proteins were co-cultured with CFSE labeled target 721 cells, and the mixed popu-lation was analyzed via flow cytometry. The same gating strategy from Figure 3.5 was used andonly target cells are plotted. Live and dead cells were selected using a DAPI vs MCH dot plot. Barplots show mean fluorescent intensity of the RFP channel (MCH MFI) of 721 target cells, witherror bars denoting the standard deviation of duplicate samples. p-values were calculated usingTukey’s HSD test applied to the results of a single factor ANOVA that was conducted for eachtarget cell population separately.some of which is likely taken up by the target 721s. However, if non-specific uptake were the mainmechanism of MCH transfer from effector to target cell for all samples, then I would not expect tosee any difference between 721s co-cultured with YTs expressing GZB-MCH compared to 721s co-cultured with YTs expressing GZBSS-MCH. Instead, of the DAPI+ 721 samples, only the sample fromtargets that were co-cultured with YTs expressing GZB-MCH has detectable amounts of mCherry pro-tein, and this band is detected at approximately 60 kDa, the expected size of GZB-MCH. Conversely,there is no detectable analogous 30 kDa band corresponding to MCH in the lysates from DAPI+ targets51mCherry YT:GZB-MCH None Targets Dead targets Live targets YT:GZBSS-MCH Effector in co-culture: 62	  49	  38	  28	  191	  LIVE	  DEAD	  LIVE	  DEAD	  LIVE	  DEAD	  None	  YT:	  GZBSS-­‐MCH	  YT:	  GZB-­‐MCH	  CFSE mCherry DAPI mCherry anti-vinculin anti-mCherry Co-culture Targets Effectors Targets Live Dead A B Figure 3.7: Western blot confirmation of GZB-MCH fusion protein transfer to target cells. (a)FACS sort data. YTs expressing either GZB-MCH or GZBSS-MCH were co-cultured with CFSElabeled target 721 cells, stained with DAPI and FACS sorted. Target cells were first selected(upper left panel), and then divided into live and dead (upper right), which were sorted separatelyand analyzed in (b). The bottom panel shows the mCherry fluorescence of targets (bottom left),live targets (bottom middle) and dead targets (bottom right), for 721s co-cultured alone (blue), withYTs expressing GZBSS-MCH (orange), or co-cultured with YTs expressing GZB-MCH (red). (b)Western blot of sorted target cell populations from (a). Equal cell-equivalent amounts of wholecell lysates of sorted target 721 populations were separated by gel electrophoresis, transferred toblots and probed for mCherry and vinculin (as a loading control). Expected protein sizes: MCH= 30 kDa; GZB-MCH = 60 kDa; vinculin = 130 kDa. Numbers displayed are sizes in kDa of theprotein ladder.co-cultured with YTs expressing GZBSS-MCH. That the putative GZB-MCH band in the DAPI+ 721sample co-cultured with YTs expressing GZB-MCH is even detectable is noteworthy given the actualamount of protein loaded is quite small, as demonstrated by the lack of a vinculin loading control band.This is despite equal cell numbers for all lanes being sorted and lysed, and is because the DAPI+ deadcells are apoptotic and rapidly degrading which results in a loss of protein.Taken together these results and analysis suggest that while there is some background, non-specific721 uptake of MCH from the co-culture media, YTs expressing GZB-MCH specifically transfer the52fusion protein to targeted 721s, while YTs expressing GZBSS-MCH and GZBSS-MCH-GZBSM donot specifically transfer MCH to targeted 721s. Thus GZB appears to be a suitable chaperone proteinfor delivery of protein payloads via the granzyme-perforin pathway.3.3 DiscussionCellular therapeutics that repurpose and recombine biological function in a cellular chassis are trans-forming medicine [177]. These efforts will rely heavily on the development of modules and systemsthat perform specific sensory, computational and effector functions [179, 180]. I report here efforts todevelop a cell-to-cell delivery module for cellular therapeutics, by repurposing the granzyme-perforinpathway of cytotoxic lymphocytes. The results support the use of granzyme B as a molecular chaperonefor inserting protein payloads into this pathway and facilitating payload delivery to target cells.I hypothesized that lytic granule loading of a payload would be sufficient for transfer to a targetcell, and that loading could be achieved by fusing a chaperone to the payload. I designed two candi-date chaperones derived from granzyme B, fused them to mCherry and investigated their subcellularlocalization in the natural killer cell line YT-Indy.All constructs containing an N-terminal ER signal peptide (GZBSS-MCH, GZB-MCH and GZBSS-MCH-GZBSM) exhibited a high degree of colocalization with Lamp1. This result is best understoodby considering the biological distribution of Lamp1 and the cellular compartments in which the colo-calization occurs. The primary route of newly synthesized Lamp1 follows the secretory pathway toexosomes at the cell membrane, and is then recaptured in early endosomes and eventually fuses withnascent lysosomes [187, 188]. In the case of GZBSS-MCH, the ER signal peptide would direct theprotein to the secretory pathway, so GZBSS-MCH is expected to be found co-localized with Lamp1 ina perinuclear distribution in the ER and Golgi and in punctate granules at the cell membrane, but notin cytoplasmic lytic granules. This is what I observe in cells expressing GZBSS-MCH, in contrast tothose expressing GZB-MCH, in which the observed colocalization is primarily in cytoplasmic puncta,consistent with lytic granules. These observations are supported by the co-culture experiments thatindicated that GZB-MCH was transferred from effector to target cell, but not GZBSS-MCH. This in-terpretation is predicated on the assumption that entry into the target cell via perforin pores is passive,which while historically controversial, is supported by most recent experimental and theoretical data[137, 142, 150, 182, 189].I postulated that combining an ER signal peptide with a putative N-linked glycosylation motifwould be sufficient for payload delivery, but the results from the co-culture experiments clearly refutedthis. That GZBSS-MCH-GZBSM did not transfer to target cells has several interesting implicationssurrounding the intracellular trafficking of granzyme B. The first is that the putative N-linked glyco-sylation sites I computationally identified, and their flanking amino acids, are insufficient for granuleloading. While I cannot rigorously exclude the possibility, I do not believe that these results are simplydue to faulty glycosylation of GZBSM, since this process occurs cotranslationally and only depends on53B) C) N71 N104 D) N71 E) N104 N71 M34 W36 H173 D176 A) Figure 3.8: Spatial context of putative N-linked glycosylation sites in granzyme B. (a) Schematicof primary amino acid structure of granzyme B. The coloring corresponds to the crystal structuresbelow. Numbers are amino acid residues. (b-c) Three dimensional crystal structure of granzymeB, colored as in (a), highlighting the potential importance of surface exposed residues that areimmediately adjacent (ADJ-N71, purple, (b); ADJ-N104, red, (c)) to the N-linked glycosylationsites (blue), but are not contained within the GZBSM (yellow). Note that these regions are quitefar from the GZBSM in primary amino acid space, as shown in (a), and in (b,c) by the labelingof representative amino acids in these regions. (d) Lysines have been colored in white, to showtheir inverted triangular pattern surrounding the N71 putative glycosylation site. (e) Locationof putative N-linked glycosylation sites (blue) throughout the protein. The residues have beencolored from red through white, to cyan, according to their position in the primary amino acidsequence.54local sequence [190]. This suggests then that GZBSM is not being phosphorylated, likely because thebinding domain for the GlcNAc-1-phosphotransferase, which adds a phosphate group to the mannoseof the N-linked glycan in the Golgi [187], is not faithfully recapitulated in GZBSS-MCH-GZBSM.This could either be due to a lack of actual amino acids that are present elsewhere in the full lengthGZB protein, or that the phosphotransferase binding domain is conformation dependent, as has beensuggested elsewhere in the literature [191], or both. Much work has been invested into characterizingthis domain, but its exact nature remains elusive. These results suggest that whatever the exact compo-sition, its constituent residues are likely distributed throughout the primary amino acid sequence, andhence were not captured in GZBSM, which is why it failed to facilitate transfer of the MCH payloadto target cells. This conclusion is consistent with the location of the asparagine residues within thecontext of GZBSM and the full length granzyme B protein. As shown in Figure 3.8, both of the N-linked glycosylation sites I computationally identified are located immediately adjacent to residues thatlie external to GZBSM. In particular, N104 is located at a junction in which residues on one side ofN104 are located within GZBSM, while those on the other side are located in the other half of the GZBprotein (Figure 3.8a,c). Also of note is that the other N-linked glycosylation site (N71) is surroundedby a triangular pattern of lysines (Figure 3.8d), a pattern which some experimental data suggests is thephosphotransferase binding site [192, 193].In summary, these results argue that granzyme B trafficking to lytic granules requires residues ordomains beyond those immediately flanking the putative N-linked glycosylation motifs. In particular,this data implies that the GlcNAc-1-phosphotransferase binding domain is not a contiguous amino acidsequence, but rather a conformation dependent motif composed of residues located throughout thelength granzyme B.This analysis implies that full length granzyme B is necessary for delivery of a payload to a targetcell. Both the flow cytometry and western blot data from the co-culture experiments demonstrated thatit is also sufficient: YTs expressing GZB-MCH transfer it to 721 target cells. Importantly, these samedata also indicated a background level of accumulation of mCherry signal in target cells co-culturedwith YT cells expressing the comparator constructs (GZBSS-MCH and MCH alone). While this mightinitially appear to undermine the utility of this system, I in fact believe the opposite: it demonstratesthe need for specific, cell-to-cell delivery, the activation of which is controlled by surface receptorinteractions. In the case of comparator effector populations expressing GZBSS-MCH or MCH, themCherry signal is the same in both live and dead target cells, indicating a non-specific effect. If theseeffector cells were used to deliver the payload, the specificity of delivery would be at best localized.However, in the case of the GZB-MCH expressing effector population, there is a significant increasein the RFP signal in dead target cells, indicating that YT-targeted 721s specifically received the mostGZB-MCH. Furthermore, my observations consistently have been that MCH is much brighter thanGZBSS-MCH, which is in turn brighter than GZB-MCH. If the transfer were non-specific and occurredat roughly equal rates for all mCherry fusion proteins, then I would expect to see 721s co-culturedwith YTs expressing MCH alone to display the greatest increase in RFP signal, followed by those55co-cultured with GZBSS-MCH and finally those with GZB-MCH. Instead, I see the opposite: with thegreatest increase in RFP signal in cells co-cultured with YTs expressing GZB-MCH, despite this fusionprotein having the dimmest fluorescent intensity. Together this data suggests there is a basal level ofbackground accumulation of the mCherry payload in all cases, but substantial, target specific transferof the payload in the case of YTs expressing GZB-MCH.The successful transfer of GZB-MCH highlights two unique and highly desirable features of thegranzyme-perforin pathway: modularity and prepositioning. The first is important in that all that isrequired to deliver a payload is to fuse it to the chaperone. In principle, no further modificationsare required, regardless of the payload. This modularity suggests that the system might be widelyapplicable as a means of cellular delivery, either in cytotoxic lymphocytes, or in the long term, in other,orthogonal, highly engineered cellular chassis. The second advantage is that, as opposed to producinga payload in response to target cell recognition using transcriptional control, a presynthesized payloadloaded into a lytic granule can be released on the timescale during which the immune synapse remainsintact, and hence cell-to-cell specificity is maintained.A critical consideration in using this system is the stability of future payloads in the harsh envi-ronment of the lytic granule, which is acidic and contains many proteases. Some desirable payloadsmay not be as stable in this environment as mCherry. This might limit the range of applications sucha system could be used for, although it is possible that the payload could be engineered to increase itsability to survive the lytic granule, for example by removing a protease cleavage site. The size of thepayload is also important, since the internal lumen of the perforin pore has been observed to be 10-20nm [137], which sets an upper limit on the payload size. However, the diameter of granzyme B is only5 nm [147], leaving an appreciable window for a variety of payloads.Any eventual application would also have to consider the native cytotoxic effector mechanismsof the lymphocyte chassis. Unmodified cytotoxic lymphocytes are appropriate vehicles to deliver pay-loads to target cells with the intent of killing them, as would be the case with tumour cells. However, forother applications—for example delivery of pro-survival factors in degenerative diseases, or deficientenzymes in metabolic diseases—the granzyme-perforin delivery functionality would have to be decou-pled from the delivery cell cytotoxicity. This may be possible either through attenuation or knockoutof the native effector mechanisms in a cytotoxic lymphocyte, or by reconstituting the pathway in anindependent, non-cytolytic cell chassis. The granzyme chaperone itself should be readily catalyticallyinactivated, as with other serine proteases [194].I have repurposed the granzyme-perforin pathway as a cell-to-cell delivery module for cellular ther-apeutics. By facilitating targeted transfer of arbitrary payloads with single-cell precision, this systemis an important addition to the part set of synthetic immunology.563.4 Methods3.4.1 Computational identification of N-linked glycosylation motifsThe granzyme B coding sequence was downloaded from NCBI RefSeq gene (accession NG_028340.1).I then used NetNGlyc 1.0 [195] to predict putative N-linked glycosylation sites, of which there weretwo, 33 residues apart. Since the NX(S/T) consensus sequence is necessary but insufficient for glyco-sylation, and the glycosylation occurs co-translationally [196], it follows that local sequence contextsurrounding the consensus site is critical. Therefore, I extracted a 53 amino acid domain from GZB,extending from 10 amino acids N-terminal of the first putative glycosylation site, to 10 amino acidsC-terminal of the second site. Intriguingly, this domain was also present in human granzyme H.3.4.2 PlasmidsA custom mammalian expression vector was used in this work. This pdL vector was constructed inhouse, based on a pcDNA3.1(+) (Thermo Fisher Scientific) backbone. Specifically, the mammalian andbacterial selectable markers and all origins of replication are derived from pcDNA3.1(+), correspondingto bases 1670 (CGATTTCGGCCTATTGGTTA...) to 5396 (...TAAACAAATAGGGGTTCCGC). Acustom expression cassette was cloned into this backbone. This cassette consisted of eukaryotic andprokaryotic promoters and ribosomal binding sequences, followed by the open reading frame, followedby eukaryotic and prokaryotic transcriptional termination sites. For the mammalian promoter I usedthe CAG promoter for its ability to drive high levels of expression in a variety of tissues. The sequencewas amplified from pEMS1157 [197]. This was followed by a hybrid T7 prokaryotic promoter, takenfrom pCMVTnT (Promega). This was followed by consensus Shine-Dalgarno and Kozak sequences.Following this is the open reading frame, which varies by plasmid. Following the end of the codingsequence, there is a BGH polyA sequence, and then a T7 terminator (with both sequences taken frompcDNA3.1(+)). Restriction enzyme cleavage sites flank all components to facilitate subcloning.The vector map and full plasmid sequence for the base pdL vector is in the Appendix, along withthe full coding sequence for all plasmids used. All plasmids were constructed through a combinationof PCR, synthesis and restriction/ligation cloning. All PCR amplicons and coding sequences weresequence verified.3.4.3 Cell cultureYT-Indy and 721.221 cells were a gift from Judy Lieberman (Harvard University). YT cells werecultured in RPMI 1640 media, supplemented with 20% heat inactivated fetal bovine serum, 1X Gluta-MAX, 1mM sodium pyruvate, 10 mM HEPES, 0.1 mM beta-mercaptoethanol. 721 cells were culturedin DMEM, supplemented with 10% heat inactivated fetal bovine serum and 1X GlutaMAX. All cellculture reagents were purchased from Thermo Fisher Scientific.573.4.4 TransfectionYT cells were electroporated using the Neon system (Thermo Fisher Scientific), using the 100 µl tip.6×106 cells were washed once in PBS, and resuspended in Buffer R along with 20 µg plasmid DNA,in a final volume of 110 µl. The extra volume ensures no bubbles are generated in aspirating the cellmixture into the electroporation tip. Critically the plasmid DNA must be of a concentration of at least1 µg/µl, and it must be prepared using an endotoxin free method. The quality of the plasmid prep greatlyinfluences the electroporation efficiency as well as the post-electroporation viability. The apparatuswas prepared as in the manufacturer’s manual, using the E2 electrolytic buffer. The electroporationconditions were 3×10 ms pulses at 1250 V. The electroporated cells were then immediately added to5 ml media spread across two wells of a 6 well plate.3.4.5 Flow cytometryCells were harvested and resuspended in PBS supplemented with 10% complete media and 1 µg/mlof DAPI (Sigma) as a viability stain. If cells were to be sorted, they were passed through a 35 µmnylon filter (BD Falcon). Cells were kept on ice and then analyzed on a BD Fortessa II, or sortedon either a BD Aria III or Fusion. For sorting, cells were sorted into complete media. In all flowcytometry experiments two initial gating steps were used. Debris was excluded by excluding cells atthe bottom left corner of a PI vs FSC-A (forward scatter area) gate. Doublets were excluded using ahierarchical gating scheme: all cells with a wider pulse width signal were excluded first in FSC-W vsFSC-H (forward scatter width vs height) and then SSC-W vs SSC-H (side scatter width vs height). Allflow cytometry data was analyzed in FlowJo.3.4.6 MicroscopyTransfected cells were first FACS sorted for moderate intensity RFP+ cells. 2.5×105 cells resuspendedin 50 µL complete media (RPMI-1640 supplemented with 10% fetal calf serum, 2 mM glutamine, 1 mMpyruvate, 2 µM 2-mercaptoethanol, 50 U/ml penicillin and 50 µg/ml streptomycin) were adhered to0.01% poly-L lysine (Sigma cat # P4707) coated, pre-cleaned 12 mm coverslips (#1.0, Fisherbrandcat # 12-0545-80) for 15 minutes at 37 ◦C. Cells were fixed with 2% paraformaldehyde (ElectronMicroscopy cat # 15710) for 15 minutes, washed with PBS, and then permeabilized with 0.1 % TritonX-100 (Sigma cat # T8787) for 1 minute. Samples were washed with PBS, and then blocked in 10%goat serum in PBS (blocking buffer) (Jackson Immunoresearch Labs cat # 005-000-121) for 1 hour.Subsequently, cells were stained with polyclonal mouse anti-Lamp1 primary antibody (Abcam cat #24170) at 1:250 dilution in blocking buffer for 1 hour. Samples were washed with PBS, and then stainedwith AlexaFluor 488 conjugated goat anti-mouse secondary antibody (Thermo Fisher Scientific cat #A-11008) at 1:1000 dilution in PBS for 45 minutes. After washing with PBS, coverslips were thenmounted on glass microscope slides using Prolong Diamond (Thermo Fisher Scientific cat # P36961)overnight. All steps were completed at room temperature unless otherwise noted.58The following day samples were imaged using a spinning disk confocal system (3i Intelligent Imag-ing Innovations) based on an inverted Zeiss Axiovert 200M microscope equipped with 100 NA 1.45Oil Plan Fluor objective and a QuantEM 512SC Photometrics camera. 10 images were acquired foreach sample, with each image containing 2-5 cells in the field of view. All exposure parameters werekept constant across all samples.3.4.7 Image analysisImage filtering was done using a custom script written in MATLAB. The green channel (Lamp1) wasfiltered as follows. The localized background of the image was calculated for each pixel as the medianintensity of a 25x25 pixel square centered on that pixel. This background pixel intensity was subtractedfrom the original pixel intensity. Any pixels with negative intensity after this step were set to zero. Thisstep aids in distinguishing small punctate structures from one another. Next, a pixel noise thresholdwas calculated as follows. First the median absolute deviation (MAD) of all nonzero pixels from theraw image was calculated. From this the standard deviation of the pixel intensity was approximated as1.4 times the MAD, which is a reasonable estimate of the pixel noise. Finally the noise threshold wastaken as 6 times this value (that is 6× 1.4×MAD). Any pixels in the background subtracted imagewhose intensity were below this value were set to zero. The red channel (mCherry) was filtered in thesame way, except localized background was not subtracted. The MATLAB script implementing thisalgorithm is in the Appendix.Colocalization analysis was also conducted in MATLAB. For paired red and green channel images,with pixel intensities Ri j and Gi j respectively, Pearson’s correlation coefficient was calculated asPCC =∑i∑j(Ri j−R)(Gi j−G)√∑i∑j(Ri j−R)2∑i∑j(Gi j−G)2(3.1)where G and R are the mean pixel intensities. The Manders M1 coefficient was calculated asM1 =∑i, jci jRi j∑i, jRi j, ci j =1, Gi j > 00, Gi j = 0 (3.2)These colocalization scores were calculated separately for each sample of each image, and thenplotted using RStudio.3.4.8 Cell labelingCells were fluorescently labeled with CFSE (eBioscience) following the manufacturer’s protocol, ex-cept that only 1 PBS wash prior to labeling was done and only 1 media wash after labeling was done.593.4.9 Co-culture experimentsYTs were transfected with mCherry fusion proteins and 48 hours later FACS sorted for viable RFP+cells. The following day 4×105 YT effector cells were combined with 1×105 CFSE labeled target721 cells at a 4:1 effector:target (E:T) ratio in a final volume of 500 µl YT media in 5 ml polystyreneround-bottom tubes (BD Falcon). The cell suspension was gently pelleted by spinning it at 200×gfor 15 seconds. The tubes were then incubated at 37 ◦C for 90 minutes, and then prepared for flowcytometry or FACS sorting as above.3.4.10 Statistical analysismCherry median fluorescent intensity was tabulated for each target cell population using flow cytome-try data from above. For each target cell population, a single factor analysis of variance was conductedto determine if the MCH MFI means were the same for all effector cell populations using the modelMFI ∼ EffectorPopulation. I then used these results as input for a Tukey’s HSD test of thedifference between sample means within each target cell population. Statistical tests were conductedin R, using the aov and TukeyHSD commands respectively.3.4.11 Western blotting3×104 cells were sorted into PBS in microcentrifuge tubes. Cells were kept on ice thereafter. Cellswere then pelleted, resuspended in 10 µL PBS and lysed directly by adding 10 µL 2X Laemmli samplebuffer. Samples were incubated at 95 ◦C for 10 minutes and then stored at −20 ◦C.For blotting, samples were boiled again at 95 ◦C for 10 minutes and then loaded onto pre-cast4-12% Bis-Tris polyacrylamide gels (Thermo Fisher Scientific). Proteins were size separated by gelelectrophoresis by running the gel at 150 V for 75 minutes. Proteins were transferred to a nitrocellulosemembrane using a standard wet transfer, at 300 mA for 2 hours.The blot was cut horizontally at 100 kDa, and then was blocked in TBS-T with 5 % skim milkpowder at room temperature for 1 hour, and then incubated with primary antibody in sealed pouchesat 4 ◦C overnight. The primary antibodies used were rabbit anti-mCherry (Biovision cat # 5993-100,lot 1A085993) and rabbit anti-vinculin (Abcam cat # EPR8185, lot GR82271-16), as a loading control.The dilutions were 1:500 (mCherry) with 5% skim milk powder, 1:10000 (vinculin) with 2% skimmilk powder, both in TBS-T. Blots were then washed with TBS-T, and incubated with horseradish-peroxidase conjugated goat anti-rabbit secondary antibody (Santa Cruz Biotechnology cat # sc-2004,lot H1015) for 1 hour. The dilution was 1:5000 in TBS-T, with 5% skim milk powder (anti-mCherry),and 2% skim milk powder (anti-Vinculin). Finally, the blots were washed with TBS-T, and then devel-oped using Bio-Rad Clarity Western ECL (enhanced chemiluminescence) Substrate reagent (Bio-Radcat # 170-5061), following the manufacturers protocol. Blots were imaged using a Bio-Rad ChemidocMP Imaging System, with exposure times ranging from 1 to 100 seconds.603.4.12 Crystal structure analysisTo visualize the location of the various motifs of GZB in the three dimensional protein, I downloadedthe granzyme B crystal structure from the Protein Data Bank (accession 1FQ3) and rendered the basecrystal structure and custom annotations using PyMOL (Schroedinger, LLC). Surface exposed residuesadjacent to the N-linked glycosylation sites were determined by first selecting all residues that werewithin 15 Å of the glycosylated residue. I then selected the subset of these residues that were surfaceexposed, using a custom PyMOL script written by Jason Vertrees.61Chapter 4Efforts to use granzyme-perforinmediated delivery of orthogonal toxins toenhance cytotoxic lymphocyte killing ofapoptosis resistant tumour cells4.1 IntroductionIt is well known that tumours frequently elude the immune response [198]. This is partially due totumor cell evasion of cytotoxic lymphocyte recognition, via mechanisms that include downregulationof antigen processing and surface MHC expression (for CTL targeting) [199] and upregulation of KIRreceptors (for NK attack) [200]. The advent of chimeric antigen receptor therapy has begun to ad-dress this aspect of the problem [201–203]. However, independent of evading recognition by lympho-cytes, tumour cells have often been shown to be resistant to lymphocyte cytotoxic effector mechanisms[64, 198, 204–206]. Disruption of apoptosis pathways mediated by death receptors (such as the Fassystem) has been found across a range of cancers, and has been implicated in carcinogenesis as wellas apoptosis resistance [207, 208]. Downregulation of the executioner caspases 3 and 7 is widelyobserved and correlates with poor survival [209–218]. Inhibitor of apoptosis proteins (IAPs) are con-sistently overexpressed in tumors, have been shown to initiate hematological malignancies in vivo, areresponsible for metastatic potential, have been found to cause resistance to adoptively transferred lym-phocytes, and are being actively pursued as small molecule targets, with these efforts having progressedto clinical trials [219–227]. Direct inhibition of granzyme by overexpressed serpins is well character-ized [228, 229]. Most importantly, overexpression of XIAP [230], survivin [231], and serpinb9 [232]confer apoptosis resistance to tumor cells, disrupt key nodes in apoptotic pathways and are directly andspecifically responsible for the resistance of these cells to lymphocyte mediated cytotoxicity, despite62effective targeting, both in vitro and in vivo. Thus, apoptosis resistance and specifically resistance tolymphocyte-induced apoptosis is a real, unsolved challenge in the field of cancer therapy.I have developed the granzyme-perforin pathway as a delivery module for cellular therapeutics(Chapter 3). Briefly, by fusing protein payloads to granzyme B and expressing these fusion proteinsin cytotoxic lymphocytes, granzyme B acts as a molecular chaperone that inserts the payload into thegranzyme-perforin pathway, resulting in the payload fusion proteins being loaded into lytic granulesin the cytosol. Upon target cell recognition, the cytotoxic lymphocytes release these fusion proteins,along with other granzymes and perforin, into the immunological synapse between the lymphocyte andtarget cell. Perforin forms a transient pore in the target cell membrane, through which the granzyme-payload fusion proteins diffuse. Here, I sought use this approach to deliver potent toxins to lymphocyteresistant tumour cells. My hypothesis was that by expressing granzyme B-toxin fusion proteins incytotoxic lymphocytes, these toxin fusion proteins would be transferred to targeted cells, resulting inenhanced killing of lymphocyte resistant target cells. This approach would have several advantages:(i) TCR- or CAR-mediated specific delivery of potent toxins to tumour cells, minimizing off-targettoxicity; (ii) sequestration of toxins inside the delivery lymphocyte would enable selection of toxinsthat could not be administered systemically, either due to toxicity or poor bioavailability.There is one significant caveat to selecting the problem of killing apoptosis and lymphocyte resis-tant tumour cells as an application in which to establish the utility of the granzyme-perforin mediateddelivery system. Unlike the delivery of pro-survival payloads with the intent of rescuing target cells,in the case of tumour cells, the intent of any payload delivery to a tumour cell is target cell death.This simplifies the experimental implementation of the latter application, since unmodified cytotoxiclymphocytes can be used as a cellular chassis, which would be inappropriate for pro-survival payloaddelivery as the lymphocytes would naturally deliver their own endogenous cytotoxic payload, killingthe target cell. In the case of target tumour cells, the purpose of the payload is to augment native lym-phocyte cytotoxicity. Unfortunately however, this also implies that for any enhancement of target cellkilling to be measurable, it is critical to generate a model system in which the target cells induce lym-phocyte reactivity and degranulation, but are completely resistant to the effects of that degranulation,which is a significant challenge in an in vitro system.With this potential concern in mind, I searched for toxins that might be capable of both killingapoptosis and lymphocyte resistant tumour cells, as well as amenable to cell-based granzyme-perforindelivery. These two requirements implied several criteria which guided my selection of toxins. Firstand foremost, the toxin ought to have an orthogonal mechanism of action, that is one that is at leastpartially independent of cytotoxic lymphocyte killing mechanisms, to maximize the toxin’s additiveeffect, and minimize the chance that a lymphocyte resistant tumour cell might also be resistant to thetoxin. Additional criteria for compatibility with granzyme-perforin mediated delivery dictated that thetoxin be: (i) genetically encodable (so that it may be fused to granzyme); (ii) relatively small (so that itmay translocate through perforin pores as a granzyme fusion); (iii) act in the cytosol of targeted cells.Based on these criteria I selected a suite of candidate toxins for study: the diphtheria toxin A fragment63(DTA), pseudomonas exotoxin A (PEA), herpes simplex virus thymidine kinase (HTK), and the E. colinitroreductases nfsA and nfsB (NFSA, NFSB).Diphtheria and pseudomonas toxins are bacterial exotoxins secreted by Corynebacterium diphthe-riae and Pseudomonas aeruginosa respectively. These toxins are composed of three major domains,each responsible for either membrane binding, membrane translocation, or ribosomal inhibition. Thelatter is the main mechanism of toxicity of these two proteins, and is a result of inhibition of elongationfactor 2 by ADP-ribosylation, thus preventing polypeptide elongation during protein synthesis [233].Both DTA and PEA have been used in recombinant immunotoxin fusion proteins in clinical trials,wherein the membrane binding domain is replaced by a single chain antibody or cytokine to target thetoxin to a specific cell population [234].Herpes simplex thymidine kinase and the E. coli nitroreductases both exert their toxic effect byactivating otherwise inert prodrugs [235]. HTK phosphorylates the prodrug ganciclovir, which onceactivated acts as a nucleoside analogue and thereby terminates DNA synthesis [236]. nfsA and nfsBmetabolize the prodrug CB1954 to potent DNA alkylating agents, resulting in DNA crosslinking andinterruption of DNA synthesis [237, 238]. These toxin/prodrug systems have two important features.First, the toxins themselves are not toxic to the delivery cell, obviating the need for additional chassismodifications to protect the host delivery lymphocyte from the toxin it carries. And second, prodrugsactivated in one target cell may kill adjacent target cells, a process known as the bystander effect. Thisis advantageous in that the toxicity of any single toxin armed lymphocyte can be greatly amplified, andby extension, near total killing of a target cell population can be achieved even in absence of near totaltumour cell targeting by delivery lymphocytes. This might be particularly relevant in the context of aimmunosuppressive tumour microenvironment, in which the lifetime of active cytotoxic lymphocytes(prior the microenvironment inducing lymphocyte anergy, conversion to a regulatory phenotype, orapoptosis) may be quite short. In this case successful lymphocyte targeting of only a fraction of tumourcells prior to succumbing to the tumour microenvironment might still result in a substantial anti-tumoureffect.Here I report my efforts to enhance lymphocyte cytotoxicity be delivery of these toxins via thegranzyme-perforin pathway. I generated granzyme-toxin fusion proteins, attempted to create lym-phocyte resistant target cells, and attempted to demonstrate enhanced target cell killing mediated bygranzyme-toxin delivery. My main findings are that granzyme-toxin fusion proteins have a variable,toxin dependent, activity, and that fully lymphocyte resistant target cells are likely required for com-pelling demonstration of the efficacy of this approach. However, using a dose response curve ex-perimental design, I was able to demonstrate significant enhancement of target killing by granzymedelivered toxins, pending repetition of the experiment.644.2 Results4.2.1 Development of orthogonal granzyme-toxin fusions for enhancing lymphocytemediated cytotoxicityFor developing my granzyme-toxin fusions, I used the same fusion protein design (shown in Figure 4.1)that was validated and successful previously, as discussed in detail in Chapter 3. I fused the toxins at theC-terminus of full length granzyme B, with the two components separated by a flexible glycine-serinelinker. I used full length granzyme B (including the pre and pro peptides at the N-terminus) so that thefusion protein would be appropriately processed and loaded into lytic granules. This has two importantimplications: (i) by virtue of its sequestration in lytic granules, the toxin domain of the fusion proteinought to have less, and perhaps no, detrimental effect on the delivery cell; and (ii) since the inhibitorydipeptide of GZB is only removed by cathepsin C once in lytic granules, if the fusion proteins weredirectly expressed in a non-lymphocyte cell type, the N-terminal inhibitory dipeptide ought to remainintact, keeping the granzyme B domain catalytically inactive.Using this design I generated a variety of granzyme B-toxin (GZB-TOX) fusion proteins. To fa-cilitate delivery cell tracking and cell sorting, I included a C-terminal GFP protein, separated fromthe granzyme-toxin fusions by a 2A ribosomal skipping sequence. This configuration results in a sin-gle mRNA transcript coding for GZB-TOX-2A-GFP in a single open reading frame. However, the2A sequence results in ribosomal skipping during translation, yielding two separate protein products,GZB-TOX and GFP [239].                     matGZB ERSS IN GSL TOXIN 2A GFP Figure 4.1: Granzyme toxin fusion protein design. The overarching design is, from N- to C-terminus: full length granzyme B (shades of blue) fused to a toxin (red) via a glycine serine linker(GSL, purple), followed finally by GFP (green). The polyprotein shown is synthesized as a singlepolypeptide. The 2A peptide (grey) induces ribosomal skipping, resulting in two mature proteins:GZB-TOX and GFP. Full length granzyme B was used, consisting of, from left to right (N- toC- terminal), the ER signal sequence (ERSS), the inhibitory dipeptide (IN) and finally maturegranzyme B (matGZB)Prior to testing the potential for cell mediated delivery of these toxin fusion proteins, it was firstimportant to determine if the toxins retained their activity as C-terminal fusion proteins. I did this bydirectly expressing these toxin fusions in a variety cells, and then assaying their viability. Importantly,as noted above the propeptide included in full length granzyme B ought remain uncleaved in thesenon-lymphocyte cell lines, keeping granzyme B inactive and non-toxic. As such any toxicity observedin these transfected cells would be expected to be a result of the toxin domain.65Bacterial inhibitors of protein synthesis: diphtheria and pseudomonas toxinsI first studied two bacterial toxins: the A fragment of diphtheria toxin (DTA) and pseudomonas exo-toxin A (PEA). I tested the toxicity of GZB-PEA and GZB-DTA fusion proteins in Hela cells, usingpropidium iodide (PI) staining as a measure of cell death (Figure 4.2). Hela cells were transfectedwith plasmids coding for GFP, GZB-DTA and GZB-PEA. Untransfected Hela cells were treated withthe apoptosis inducer staurosporine (STS) as a positive control. 48 hours after transfection, cells werestained with propidium iodide and analyzed via flow cytometry. I found that GZB-DTA was substan-tially more potent than GZB-PEA, and therefore selected GZB-DTA for further testing.●●●●●●●●●●●●0204060GFP GZB−PEA GZB−DTA STSPercent dead (PI+) cellsp < 0.005 Figure 4.2: GZB-DTA and GZB-PEA testing in Hela cells. Data shown is the dead cell percentage48 hours after transfection, as assessed by PI staining. Each circle is an independent replicate.To confirm that the observed toxicity was not due to granzyme, but rather DTA, I expressed GFP,GZB and GZB-DTA in 293T cells and assessed their viability (Figure 4.3). I also included untreatedcells as a negative control, and cells treated with the microtubule inhibitor colcemid as a positive con-trol. 48 hours after transfection, cell viability was assayed using a fluorescent metabolic activity assay(PrestoBlue). This reagent contains a cell permeable resazurin dye, which is not fluorescent. The reduc-ing environment of live cells reduce this dye to produce resorufin, which emits red fluorescence whenexcited by green light. Dead cells lose their reducing environment, and thus do not activate resazurin.Therefore, the fluorescent intensity of a sample of cells that is stained with PrestoBlue is proportional66to the number of live cells in the sample. Using this assay allowed indirect measurement of cell deaththroughout the experiment, rather than only at the time of measurement, since samples with apprecia-ble cell death will have fewer cells at the time of assay, resulting in a lower fluorescent signal. This isimportant as dead cells eventually disintegrate to the point that they are not measurable when using, forexample, viability dye stains such as DAPI or PI.Raw fluorescence values were obtained for all samples. I then subtracted the colcemid readingsfrom all other values, and corrected the values for transfection efficiency (quantified by flow cytometryto be 70 %). Finally I normalized the viability of each treatment condition to the viability of the un-treated cells. The results show that the viability of 293Ts transfected with GZB alone was comparableto GFP, while 293Ts transfected with GZB-DTA exhibited substantial cell death.●●●●●●0.00.51.0GFP GZB GZB−DTANormalized cell viabilityp < 0.05 p < 0.05 Figure 4.3: GZB-DTA testing in 293T cells. Shown is the viability of each transfected cellpopulation, normalized to untransfected controls. Each circle is a biological replicate, which as anaverage of technical triplicates.The herpes simplex virus thymidine kinase ganciclovir systemI next investigated the herpes simplex derived thymidine kinase ganciclovir (HTK/GCV) system, usinga similar approach to what was used in testing GZB-DTA. 293T cells were transfected with plasmidscoding for GZB or GZB-HTK and then treated with either ganciclovir (GCV) or vehicle control. I67also included cells treated with colcemid (COL) as a positive control. 5 days after transfection, thecells were imaged via brightfield microscopy, and then trypsinized, reseeded in plates, and assayedusing a fluorescent metabolic activity assay. I normalized each sample’s viability to the viability ofcells transfected with GZB and treated GCV. The results demonstrate both that the GZB-HTK fusionprotein can activate GCV, and illustrate the synergy of the HTK/GCV prodrug system: GCV or HTKin isolation is not toxic, while the combination is highly toxic Figure 4.4.GZB+GCVGZB-TK GZB-TK	+	GCVCOLCEMID●●●●● ●0.00.40.81.2GZB+GCV GZB−HTK GZB−HTK+GCV COLNormalized cell viabilityp < 0.0005 p < 0.0005 Figure 4.4: GZB-HTK testing in 293T cells. At left, optical microscopy images are shown ofsamples that were then used to produce the data plotted at right. Shown is the viability of eachtransfected cell population, normalized to the viability of cells transfected with GZB and treatedwith GCV. Each circle is a biological replicate, which as an average of technical triplicates. GCV= ganciclovir, COL = colcemid.The E. coli nitroreductase CB1954 systemUnlike the HTK/GCV system, the nitroreductases (NTR) nfsA nfsB are not well characterized. There-fore, rather than a simple binary comparison of drug compared to vehicle control, I chose to assessthe activity of the GZB-NTR fusion proteins using a dose response approach. 293Ts were transfectedwith plasmids coding for HTK (as a negative control, since HTK should not activate the NTR prodrugCB1954), NFSA, NFSB, GZB-NFSA, GZB-NFSB, and then treated with increasing levels of CB1954.5 days after transfection, the cells were reseeded in plates and assayed using a fluorescent metabolicactivity assay. The raw values were corrected for media background and then normalized to the val-ues measured for each transfected cell population (e.g. HTK, NFSA, etc) that were treated with 0 µMCB1954. The results, shown in Figure 4.5, have three key findings: (i) both nitroreductases exhibit aloss of activity as C-terminal granzyme fusions as compared to their unfused counterparts; (ii) GZB-NFSA retains sufficient activity to activate CB1954 to produce a toxic effect, while GZB-NFSB doesnot, and is indistinguishable from the negative control HTK population; (iii) 50 µM is tentatively anappropriate dose at which CB1954 alone is non-toxic, but becomes so in the presence of GZB-NFSA.Since I did not have biological replicates for this experiment, it is impossible to draw firm conclusions,68but the data was sufficient to select GZB-NFSA as a candidate worth pursuing further.0	  0.2	  0.4	  0.6	  0.8	  1	  1.2	  0.01	   0.1	   1	   10	   100	   1000	  Normalized	  cell	  viability	  CB1954	  dose	  (uM)	  HTK	   GZB-­‐NFSB	   GZB-­‐NFSA	   NFSB	   NFSA	  Figure 4.5: GZB-NFSA and GZB-NFSB testing in 293T cells. Shown is the viability of eachtransfected cell population as a function of the CB1954 dose. Each data point has been normalizedto the 0 µM viability for each transfected plasmid (e.g. HTK, NFSA, etc.). Data plotted is theaverage of technical triplicates. Error bars shown represent the standard deviation of the triplicates.Note that the CB1954 dose is plotted on a log scale, and to accommodate this the 0 µM data pointsare falsely plotted at 0.05 µM.Taken together the results in this section demonstrate that a variety of toxins retain their activityas C-terminal granzyme fusions, albeit to varying degrees, and that these GZB-TOX fusions kill targetcells when expressed directly in them. This encouraged us to develop a lymphocyte resistant modelcell-line in which to further test the granzyme-toxin fusions.4.2.2 Characterizing the lymphocyte resistance of the breast cancer cell line MCF-7As a model system for apoptosis resistance, I selected the breast cancer cell line MCF-7 as a target cell.MCF-7s lack caspase 3, one of two key executioner caspases in the apoptotic cascade [240]. Caspase3/7 downregulation in tumour cells is widely observed, correlates with poor survival, and thus MCF-7shave been frequently used to study the chemotherapeutic and apoptosis resistant phenotypes that canresult from caspase downregulation [209, 210]. I used the natural killer cell line YT-Indy (hereafterYT) as a model for cytotoxic lymphocytes, as it has an intact granzyme-perforin pathway [185].I am aware of only one report in the literature in which MCF-7s were used as a target cell forYT-Indy [241], which provides data only that obliquely suggests that YTs target MCF-7s. Therefore,I first set out to characterize MCF-7’s susceptibility to YT attack. I co-cultured CFSE labeled MCF-7swith YTs, and then stained the mixed cell population with propidium iodide (PI) and analyzed the cellsvia flow cytometry. After excluding debris and doublets and gating on CFSE+ target cells, I found thatthere was minimal change in the viability of MCF-7s cultured with YT’s as opposed to those cultured69alone (89.5% compared to 92.9%). This is in stark contrast to the 721.221 (hereafter 721) cell linewhich is known to be killed by YTs. Using an identical experimental design as was used for MCF-7s,YTs killing of 721s was measured to be roughly 50% (721 alone viability of 92.2% compared to 42.8%viability of 721s co-cultured with YTs).To exclude the possibility that this apparent lack of cell death was due to mechanisms of resis-tance involving the granzyme perforin pathway, I used a granzyme B FRET reporter (Sharma et al., inpreparation) to confirm that granzyme B does indeed enter MCF-7s. This reporter consists of CFP andYFP fluorescent proteins separated by the granzyme B consensus cleavage substrate. In the absenceof granzyme B there is a CFP to YFP FRET transfer signal. The presence of granzyme B results inthe cleavage of the substrate, separating the FRET partners, which results in an increase in the CFPsignal, and a decrease in the FRET transfer signal. I cultured target MCF-7 cells expressing this FRETreporter either alone, or in the presence of YT effector cells, and then analyzed the target MCF-7 cellsfor granzyme B status via flow cytometry (Figure 4.6). MCF-7s cultured with YTs exhibited a clearFRET-shifted population as compared to MCF-7s cultured in isolation, indicating that granzyme B istransferred by YTs to MCF-7s. The fraction of FRET-shifted MCF-7s (9%) in Figure 4.6 is at thelower end of the range I have measured, which can be as high as 30%. This percentages are sub-stantially lower than those observed for other target cell lines such as 721s, which commonly exhibitapproximately 60-70% death in similar YT co-culture conditions. The potential reasons for this differ-ence are many. An obvious possibility is that the complement of surface receptors on 721s produce amuch more robust YT response than do those on MCF-7s. MCF-7s are also very large and adherentcells, so it is possible that the YTs are less able to fully explore the physical environment of the co-culture. An important final note is that the relatively low percentage of YT-targeted MCF-7s (10-30%)is an important consideration when interpreting YT-MCF-7 co-culture experiments, as only these cellswill receive the granzyme-payload fusion proteins, which, depending on the application, puts an upperlimit on the potential observable effect size.The viability of the FRET+ cells (98.2%) was essentially the same as both the overall viabilityof the target MCF-7s co-cultured with YTs (97.1%) and the viability of MCF-7s cultured in isolation(98.1%). This suggested that MCF-7s might be resistant to YT attack. To confirm this, I investigatedthe long-term viability of YT-targeted MCF-7s. To do this I co-cultured YTs with MCF-7s expressingthe granzyme B FRET reporter, and then isolated viable, FRET shifted, granzyme B positive, MCF-7s,via cell sorting. As controls, I also sorted viable, non-shifted (FRET-) MCF-7s, and dead MCF-7s. Icultured these sorted cells for one week, and then assessed their viability using a commercial fluores-cent metabolic activity assay (PrestoBlue). After measuring raw fluorescent intensity, I subtracted thebackground signal measured from the dead cell wells from the other two populations, and then nor-malized the fluorescent values to the FRET- population. The resulting data shows that all MCF-7s thatwere FRET shifted ultimately died (Figure 4.7). Combining these results, my working model for YTkilling of MCF-7s is that the absence of caspase-3 significantly slows MCF-7 cell death, but that YTeffector mechanism are ultimately sufficient to initiate target cell death, and therefore MCF-7s are not70lymphocyte resistant.FRET+9.16FRET+0.082CFP CY-FRET MCF-7 YT-Indy + MCF-7 Figure 4.6: Characterizing YT delivery of GZB to MCF-7s using a FRET reporter. After exclud-ing debris and doublets, YFP+ target cells were selected, which are plotted here. The CFP to YFPFRET transfer signal (CY-FRET) signal is plotted on the y-axis, and CFP signal on the x-axis. Thetriangle gate is set on the MCF-7 alone population (left), and copied to the co-culture population(right). FRET-shifted cells are plotted in red.In summary I found that: (i) YTs target MCF-7s; (ii) granzyme B is transferred from YTs toMCF-7s; and (iii) MCF-7 cell death due to YT attack is slow, but occurs nonetheless, meaning thatunmodified MCF-7s are not suitable as a model lymphocyte resistant target cell line.4.2.3 Efforts towards generating a lymphocyte resistant cell lineBased on the results of the previous section it was evident that in order to observe enhanced lymphocytekilling due to orthogonal toxin delivery to MCF-7s, the cell line would have to rendered fully resistantto lymphocyte attack by further modifying the apoptosis pathway. I attempted to do this both by treatingtarget cells with small molecule inhibitors of apoptosis, as well as by overexpressing various genes thatinhibit apoptosis in target cells. I tested the effects of these modifications in two cell lines, MCF-7 aswell as the B-cell lymphoblastoid cell line 721 [186], as they are a well known YT target.Small molecule inhibition of apoptosisI first conducted experiments to see if the combination of the granzyme B inhibitor dichloroisocoumarin(DCI) [242] and the pancaspase inhibitor Q-VD-OPh would attenuate YT killing of 721 target cells.I co-cultured CFSE labeled target cells with YTs, and treated these co-cultures with either DMSO asa vehicle control, or the two inhibitors. I then stained the cells with propidium iodide and quantifiedtarget cell viability via flow cytometry. The results clearly show that the inhibitors essentially abrogate71●●●●0.000.250.500.751.00FRET− FRET+Normalized cell viabilityp < 0.001 Figure 4.7: Long term survival of YT-targeted MCF-7s. Normalized MCF-7 viability post YTco-culture and week long culture in isolation is shown. Each circle is a biological replicate, whichas an average of technical duplicates.YT killing of 721s (Figure 4.8).Since my experiments in Section 4.2.2 demonstrated that post co-culture viability is not necessarilyindicative of long term survival, it was necessary to determine if this was the case for DCI and Q-VD-OPh inhibition of YT killing. As target cells I used 721s or MCF-7s, both expressing the granzymeFRET reporter (Section 4.2.2). I co-cultured target cells with YTs in the presence of Q-VD-OPh. I wasunable to use the granzyme B inhibitor DCI since it has a fluorescence spectra that overlaps with CFP, acomponent of the FRET reporter which is required for isolating YT-targeted cells. After the co-culture,I stained the mixed population with propidium iodide (PI) and then isolated three target cell populationsvia cell sorting: PI-FRET-, PI-FRET+, and PI+. After one week in culture I assessed the viability ofthese cells using a commercial metabolic activity assay (PrestoBlue). I subtracted the readings fromthe PI+ cells from the other data, and then normalized the remaining values to the PI-FRET- valuesfor each cell type, with the results shown in (Figure 4.7). Similar to my results for MCF-7s withoutinhibitors, virtually all YT-targeted (FRET+) 721s and MCF-7s ultimately die.72●●●●●●0255075100DMSO INHIB YT + DMSO YT + INHIBNormalized cell viabilityp < 0.0005 p < 0.0001 p < 0.0001 Figure 4.8: Small molecule inhibition of YT killing of 721 target cells. After excluding debrisand doublets, CFSE+ 721 target cells were selected. Plotted is the percent of the 721 populationthat is dead (PI+). Each circle is a biological replicate.Overexpression of inhibitor of apoptosis proteinsI also investigated the potential of rendering target cells resistant to lymphocyte attack by overex-pressing inhibitor of apoptosis proteins (IAPs). Overexpression of IAP family members blocks theactivation of executioner caspases 3 and 7 by granzyme or other intrinsic activators, and IAP upreg-ulation is a natural apoptosis resistance mechanism of tumors [209, 222, 243]. I chose to focus ontwo IAP proteins survivin (SURV) and X-linked Inhibitor of Apoptosis Protein (XIAP), as these havebeen experimentally shown to render cells resistant to NK-mediated cytotoxicity, despite effective NKtargeting, engagement and degranulation [230, 231]. My hope was that the more biologically relevantmechanism, as well as the potential for ongoing production of the IAPs in the target cells might besufficient to render them resistant to lymphocyte attack.I constructed plasmids that polycistronically co-expressed an IAP protein (either XIAP or SURV)along with the FRET reporter. I used the same 2A peptide design as for the granzyme toxin fusionproteins. In this case, each polyprotein consisted of (from N- to C-terminus) IAP-2A-FRET. As in thetoxin case, the 2A peptide causes ribosomal skipping, resulting in separate IAP and FRET proteins. Itransfected MCF-7s with plasmids coding for XIAP-FRET, SURV-FRET, and FRET (as a control), andco-cultured these target cells with YTs. Following the co-culture I stained the mixed cell populationwith PI, and isolated PI- FRET+ target cells. As controls, I also collected live and dead (PI- and PI+)73-­‐0.2	  0	  0.2	  0.4	  0.6	  0.8	  1	  1.2	  1.4	  721	   MCF-­‐7	  Normalized	  cell	  viability	  PI-­‐FRET-­‐	  PI-­‐FRET+	  Figure 4.9: Long term viability of targets co-cultured with YTs in the presence of apoptosisinhibitors. Normalized target cell viability post YT co-culture and week long culture in isolationis shown. Data shown is the mean of technical duplicates, and the error bars are the standarddeviation.unmodified MCF-7s. After one week in culture I measured their viability using a fluorescent metabolicactivity assay, and normalized all viabilities to the values measured for the PI- MCF-7s. The results inFigure 4.10 indicate that the IAP proteins do not provide any increase in MCF-7 resistance to YT-attackas compared to MCF-7s expressing FRET alone.I was somewhat surprised by this result, as there are reports in the literature that overexpressionof IAP proteins in target cells can protect them in vitro from lymphocyte mediated cytotoxicity [220,230, 231]. However, there are several possible reasons for this discrepancy. My model system useshuman effector and target cells, while much of the data in the literature uses murine cells. Moregenerally, comparing different immortalized cell lines, even within species, is challenging, and none ofthe experiments in the literature used either MCF-7s or YT-Indys. Furthermore most of the experimentsin the literature only follow the viability for a matter of hours, and thus it is very possible that, as withour experiments, the IAPs are actually only slowing death rather not stopping it. In the experimentswhich did measure the viability of the cells over the long term, the protective effect sizes diminisheddramatically, and were only preserved at extreme effector to target ratios.In summary, my efforts at rendering target 721 and MCF-7 cells resistant to YT killing were ulti-mately unsuccessful. Despite in many cases slowing it, neither small molecule inhibition of caspases,nor IAP mediated inhibition, ultimately prevented YT-induced target cell death. While I did not havethe biological replicates in these experiments necessary to rigorously exclude the possibility that thesetreatments might render target cells fully resistant to YT killing, the effect sizes are sufficiently largeand reproducible across multiple experiments that I did not think it worthwhile to continue pursuingthis avenue of experiments.74-­‐0.2	  0	  0.2	  0.4	  0.6	  0.8	  1	  1.2	  MCF7:	  PI-­‐	   FRET	   SURV	   XIAP	   MCF7:	  PI+	  Normalized	  MCF7	  viability	  Figure 4.10: Long term viability of YT-targeted target cells expressing IAPs. Normalized targetcell viability post YT co-culture and week long culture in isolation is shown. Data shown is themean of technical duplicates, and the error bars are the standard deviation.4.2.4 Effector dose response curves as a means of resolving small increases in YTtarget cell killingBased on the results of the preceding three sections it was evident that the only way to enhance YTkilling of MCF-7s was to deliver a payload which would have toxic effects that extend beyond thecell to which the payload is delivered. This is because all YT-targeted MCF-7s ultimately die, so anyadditional payload toxicity in the targeted cell would not be observable. In light of this I chose toproceed using the toxin/prodrug systems and the bystander effect they provide. Observing this effectwould only be possible at effector target ratios at which there are significant populations of both YT-targeted and non-targeted MCF-7s, the former as a pool of toxin delivered and prodrug activating cells,and the latter as a pool of cells that are susceptible to the activated prodrug. To meet these requirementsI chose to proceed with my experiments by attempting to deliver granzyme-toxin fusion proteins in thecontext of an effector cell dose response experiment using MCF-7 target cells. Using a dose responsecurve has the added benefit of resolving smaller effect sizes than simple binary comparison. To confirmsuch an approach was feasible, I characterized the effector dose-response behavior of fluorescentlylabeled MCF-7s subjected to YT attack by co-culturing the two cell types at varying effector target(E:T) ratios. I then isolated the MCF-7s via cell sorting. The total MCF-7 population was isolated,rather than only YT-targeted MCF-7s, since the goal of this experiment was to determine appropriateconditions for future toxin-delivery experiments. Since YT-targeted MCF-7s die, these experimentswould require a pool of non-targeted MCF-7s to be present as targets for a toxin activated prodrugbystander effect. As controls I included MCF-7s that were not co-cultured with YTs, and dead MCF-7s. After one week in co-culture, I assessed their viability using a metabolic activity assay, with theresults show in (Figure 4.11). At higher E:T ratios there was near total MCF-7 death, while at lowerE:T ratios it was quite minimal. Based on my results above, this minimal cell death is likely due to thelow absolute number of YTs in the co-culture.75l l lllllll0.000.250.500.751.000.01:1 0.1:1 1:1 10:1Effector:target ratioNormalized MCF−7 cell deathFigure 4.11: YT effector:target dose response of MCF-7s. Target cell death post YT co-cultureand week long culture in isolation is shown. Fluorescent values were normalized to those fromMCF-7s not co-cultured with YTs, and then these values were subtracted from 1 to transform thedata from viability to cell death. Data points are the mean of biological duplicates, which are themean of technical triplicates. Error bars are the standard deviation of the biological duplicates.4.2.5 Validating the GZB-HTK and GZB-NFSA toxin fusion proteins in MCF-7sThe requirement for a bystander effect —namely delivered payloads having a toxic effect that extendsbeyond the cell to which the payload was delivered, necessary to show enhanced lymphocyte killingsince YT-targeted MCF-7s uniformly die—constrained the rest of my work to the two toxin/prodrugsystems I had evaluated: GZB-HTK/ganciclovir and GZB-NFSA/CB1954, as both ganciclovir andCB1954, once activated in a cell, can produce toxicity in adjacent cells. Prior to testing the poten-tial for cell mediated delivery of these toxin fusion proteins, it was first important to investigate ifthe granzyme-toxin fusion proteins were capable of killing caspase-3 deficient MCF-7s. I did thisby directly expressing these toxin fusion proteins in MCF-7s and assessing their viability. MCF-7sexpressing either GFP or GZB-HTK fusion proteins were treated with either ganciclovir or vehiclecontrol. After one week I assayed the cell viability using a metabolic activity assay (Figure 4.12a). Idid not have biological replicates for this pilot experiment, but the results suggest that neither ganci-clovir nor GZB-HTK treatment alone produces MCF-7 toxicity, while the GZB-HTK fusion protein incombination with ganciclovir is toxic to MCF-7s.I validated the GZB-NFSA fusion protein in a very similar manner, with the additional step oftesting multiple CB1954 doses, since this system is less commonly used, and so suitable doses werenot clearly available from the literature. I transfected MCF-7s with plasmids coding for either GFP76or GZB-NFSA fusion proteins and treated these cells with several concentrations of CB1954. Afterone week I assayed the cell viability using a metabolic activity assay (Figure 4.12b). Again, withoutbiological replicates, I cannot make any definitive conclusions, but these experiments suggest twoimportant findings to guide my next set of experiments. Unlike ganciclovir, CB1954 seems to have aninherent toxicity to MCF-7 cells, regardless of NFSA activation, at doses above 1 µM (19% and 59%at 10 µM and 50 µM respectively). However, GZB-NFSA appears to increase this toxicity (by 25% and57% at 10 µM and 50 µM respectively), with the greatest increase measured at 50 µM. Despite this Ichose to proceed with 10 µM as the working concentration of CB1954 for future experiments, since itoffered the best balance of low baseline toxicity but significant GZB-NFSA activated toxicity.0	  0.2	  0.4	  0.6	  0.8	  1	  1.2	  GFP	   GZB-­‐HTK	   GFP	   GZB-­‐HTK	  VC	   GCV	  Normalized	  MCF-­‐7	  viability	  0	  0.2	  0.4	  0.6	  0.8	  1	  1.2	  0	  uM	   1	  uM	   10	  uM	   50	  uM	  Normalized	  MCF-­‐7	  Viability	  CB1954	  concentra=on	  GFP	  GZB-­‐NFSA	  A B Figure 4.12: Testing GZB-HTK and GZB-NFSA toxin/prodrug systems in MCF-7s. (A) GZB-HTK. MCF-7s were transfected with plasmids coding for either GFP or the GZB-HTK fusionprotein. Five thousand transfected (GFP+) cells were then sorted into each well of a 96 well plate.The cells were then treated with either 5 µM ganciclovir (GCV), or vehicle control (VC). Cellviability was assessed using a metabolic activity assay. Raw fluorescent values were normalizedto the GFP + VC value. Error bars represent the standard deviation calculated from triplicatewells. (B) GZB-NFSA. MCF-7s were transfected with plasmids coding for either GFP (blue) orthe GZB-NFSA (red) fusion protein. Five thousand transfected (GFP+) cells were then sorted intoeach well of a 96 well plate. The cells were then treated with a range of CB1954 doses. Cellviability was assessed using a metabolic activity assay. Raw fluorescent values were normalizedto the GFP 0 µM value. Error bars represent the standard deviation calculated from triplicate wells.4.2.6 Predicted and measured enhancements of YT-killing of MCF-7s usinggranzyme-toxin fusion proteinsHaving validated the granzyme B toxin fusion proteins directly in MCF-7s, I attempted to use YTsto deliver these fusions to target MCF-7 cells. This would serve as proof of principle that cell baseddelivery via the granzyme perforin pathway could be used for practical applications such as toxin77delivery to apoptosis resistant tumour cells in the context of adoptive cell therapy.To do this, I designed an effector dose response experiment, in which I co-cultured various types ofgranzyme-toxin fusion expressing YT effector cells with target MCF-7 cells at a range of effector target(E:T) ratios. I selected this approach for several reasons. First, at high effector target ratios, most MCF-7 target cells are killed, and so the potential window for increased YT killing due toxin delivery is small.Similarly, the GZB-NFSA fusion only provides around a two-fold increase in CB1954 toxicity. Smalleffect sizes are much easier to resolve as a shifted curve, rather than a binary comparison. Furthermore,all YT-targeted MCF-7s do die eventually (Section 4.2.2). Therefore, any additional toxin-mediatedkilling would necessarily be via the bystander effect, in which toxins delivered to a YT-targeted MCF-7 activate prodrugs that then diffuse to and kill adjacent, non-YT-targeted MCF-7s. Thus the largesttoxin-mediated increase in YT killing should be expected at intermediate E:T ratios, where there is alarge pool of both targeted MCF-7s with toxins capable of activating prodrugs, as well as a large poolof non-targeted MCF-7s that have the potential to be killed by activated prodrugs. Finally, given thatmost solid tumours are large masses, and that the achieved E:T ratios in clinical adoptive cell therapyare reported to be well below unity [244], I feel this is a biologically realistic experimental design inwhich to test the utility of using granzyme-toxin fusions to enhance effector cell killing of tumour cells.I first used the results from previous sections to predict what enhancement in YT killing could bereasonably expected, and then conducted the dose response experiments to determine if any enhance-ment actually occurs.Estimated toxin/prodrug mediated enhancement of YT killingIt is challenging to provide an accurate prediction of the magnitude of the effect size that would beexpected, but an upper bound can be estimated by considering the results of Section 4.2.5 and Sec-tion 4.2.4. To derive a useful expression that can be parameterized from the data available from theseexperiments, I assume that there are three potential causes of MCF-7 cell death: YT killing, baselineprodrug toxicity, and toxin activated prodrug toxicity. I further assume that these effects are indepen-dent (i.e. there is no synergy between them), and that they act in an order of precedence determinedby their application to the target MCF-7s, that is first the YT killing, then the baseline drug toxicity,and then the activated prodrug toxicity. Finally I assume that a given effect is only exerted on what-ever fraction of cells were not already killed by effects earlier in the order of precedence. Under theseassumptions, the fractional death D of MCF-7s in the co-culture can be taken to be:D = γ+(1− γ)β +(1− γ)(1−β )τ (4.1)where γ is the fraction of cells killed by YT-Indys, β is the fraction of cells that die due to baselineprodrug toxicity, and τ is the fraction of cells that die due to toxin activated prodrug toxicity. Thesevalues are taken to represent the cell death that occurs when the effect is applied in isolation. Themultiplicative prefactors that appear in front of β and τ account for the attrition due to effects that are78earlier in the order of precedence: these later effects can only kill the fraction of cells not already killedby a previous effect. This assumption allows for estimating γ , β and τ from previous experiments inwhich the respective effects were applied in isolation: γ can be estimated from Figure 4.11, while βand τ can be estimated from Figure 4.12. Specifically:β = 1− VG,DVG,VCand τ = α ∗ (1− VT,DVT,VC) (4.2)where VG,D is the viability of MCF-7s transfected with GFP and treated with the prodrug (either GCVor CB1954), VG,VC is the viability of MCF-7s transfected with GFP and treated with vehicle control,VT,D is the viability of MCF-7s transfected with the granzyme B toxin fusion protein (either GZB-HTKor GZB-NFSA) and treated with the corresponding prodrug (GCV or CB1954 respectively), and VT,VCis the viability of MCF-7s transfected with the granzyme B toxin fusion protein and treated with thevehicle control. I have included α to model the efficiency with which MCF-7s that have received thegranzyme-toxin fusion activate the prodrug. This is almost certainly a function of the effector: to targetratio, likely initially increasing as increasing amounts of the prodrug activating toxin are delivered, andthen decreasing at high effector target ratios as the MCF-7s are so heavily attacked by the YTs that theydie more rapidly and so produce less active prodrug for a shorter period of time. However, since I haveno way of estimating the form of this dependence, in order to proceed I set α = 1. I have also ignoredany toxicity associated with granzyme B toxin fusion protein expression directly in MCF-7s (a smallamount of which is evident in Figure 4.12), which is justified since here I am estimating cell death in aco-culture experiment, in which the only MCF-7s that contain the toxin fusion protein will also receivea YT-hit from which they will already die, and thus are already accounted for by γ .Using the data from Figure 4.12 to estimate the viabilities V in Equation 4.2 gives β = 0.03 andτ = 0.78 for the GZB-HTK/GCV system, and β = 0.11 and τ = 0.25 for the GZB-NFSA/CB1954system (assuming 10 µM CB1954). This allows estimation of the death D from Equation 4.1 for anygiven γ , which is a function of the effector to target ratio, and can be estimated from Figure 4.11. InFigure 4.13 I have plotted these estimates for wild type effectors, as well as effectors delivering bothtoxin/prodrug systems. Notably, the death due to the GZB-HTK/GCV system is very high even forlow effector to target ratios. This is not due to baseline ganciclovir toxicity, which is actually very low(β = 0.03), and rather because of a very high toxin activated prodrug toxicity (τ = 0.78), and, critically,the assumption of perfect prodrug activation efficiency in the presence of any GZB-HTK (that is settingα = 1). Again it is likely that at low E:T ratios, this conversion would not be perfect (i.e. α < 1), andso the total cell death would be lower than is plotted. Nevertheless these plots demonstrate the potencyof the GZB-HTK/GCV system, as can also be seen in Figure 4.12.Finally to provide a summary metric for the estimates of maximum possible enhancement of YTkilling due to toxin delivery, I selected an intermediate E:T ratio of 0.5:1, for which γ = 0.31 in Fig-ure 4.11. I then calculated the absolute increase in cell killing ∆ = DA−DY , and the fractional foldincrease in cell killing f =DA/DY using Equation 4.1, where DA is the total cell death due to all effects,79l lllllll lllll ll llllll0.000.250.500.751.000.05:1 0.5:1 5:1Effector:target ratioNormalized MCF−7 cell deathYT effectorl l lGZB−HTK/GCV GZB−NFSA/CB1954 WTFigure 4.13: Estimates of toxin/prodrug enhancement of YT killing. Effector dose responsecurves for MCF-7s co-cultured with YTs. These were experimentally determined for wild typeYTs (black, from Figure 4.11), and calculated using Equation 4.1 for YTs expressing GZB-HTK(red), and GZB-NFSA (blue), assuming treatment of the MCF-7s with the corresponding prodrug.and DY = γ is the cell death due to YT killing only, i.e. Figure 4.11. These were ∆= 0.54 and f = 2.72for the GZB-HTK/GCV system and ∆= 0.24 and f = 1.78 for the GZB-NFSA/CB1954 system. Thesecalculations provide a very rough estimate of the upper bound on the increase in cell killing that I wouldexpect to see experimentally, and suggest that the GZB-HTK/CB1954 system might be very effective.Experimental measurement of enhancement of toxin/prodrug enhancement of YT killing ofMCF-7sBased on my granzyme-toxin validation work, I set out to characterize the potential for enhancement ofYT killing provided by both GZB-HTK and GZB-NFSA. However, when I transfected YTs with GZB-HTK, I observed massive genotoxicity in the YTs due to GZB-HTK expression. Since I have expresseda variety of granzyme B fusion proteins in YT cells and this was the first observation of genotoxicity, Iam fairly confident this is due to the HTK domain. Given that HTK has been expressed in lymphocytespreviously as a suicide gene in cell therapies [245], I was surprised by this result and repeated theexperiment several times, but the finding was consistent. It is possible that the difference between my80Sample EC50 σEC50 p-valueYTN-CB1954 0.30 0.037 0.0110YTN 0.52 0.03 0.0035YT-CB1954 1.12 0.015 0.0006YT 3.32 0.65 -Table 4.1: Fitted EC50 values for enhanced lymphocyte killing. Samples are listed in increasingEC50 (decreasing potency). σEC50 is the standard error of the estimate for EC50 resulting from thefit. The p-values are the Tukey HSD adjusted p-value comparing the sample to its next-nearestneighbor.observations and those in the literature are due to expression levels: those in the literature primarilyuse viral transduction, while I used electroporation, likely resulting in much higher expression levels.Since HTK catalyzes the synthesis of ADP from ATP, it is mechanistically conceivable that extremelyhigh levels of this foreign kinase (not subject to autoregulatory mechanisms) could sufficiently depleteATP levels to cause cell death, although there are no reports of this in the literature. Another possibilityis that the HTK domain could have dislodged the inhibitory dipeptide at the catalytic site of GZB,resulting in GZB-mediated toxicity.Fortunately this effect was not observed with GZB-NFSA expression, allowing me to proceed witheffector dose response experiments. YTs expressing GZB-NFSA, as well as unmodified YTs as acontrol, were co-cultured with fluorescently labeled MCF-7 target cells at a range of effector to targetratios. Target MCF-7 cells—consisting of YT-targeted and YT-naive cells—were then isolated fromthe mixed cell population via FACS, sorted into 96 well plates, and treated with CB1954, or DMSOas a vehicle control (giving all wells equivalent DMSO concentration). Finally, the viability of thesecells was measured using a metabolic activity assay. The results, shown in Figure 4.14, have three mainfindings. First, there is a clear E:T ratio dose response: higher YT-cell numbers result in higher levels ofMCF-7 cell death, for all conditions, as expected from my pilot experiments. Second, there is increasedMCF-7 cell death in samples that were either co-cultured with YTs expressing the GZB-NFSA toxinand then treated with vehicle control, or in samples that were co-cultured with unmodified YTs and thentreated with CB1954. Again, I expected this based on experiments in which I directly treated MCF-7swith this toxin/prodrug system, and found intermediate toxicity due to either component of the system.Third and most importantly, the highest level of MCF-7 cell death is in samples that were co-culturedwith YTs expressing the GZB-NFSA toxin, and then treated with the CB1954 toxin. Conversely, thelowest amount of MCF-7 cell death is in target cells co-cultured with unmodified YTs and not treatedwith CB1954.To confirm that these results were statistically significant, I fit the data from each co-culture condi-tion with a sigmoid logistic function of the form y = (1+ em(EC50−x))−1, where EC50 is the E:T ratio atwhich target cell death reaches half of its maximum value. The fit was done using a non-linear least-squares algorithm, and the results are shown in Table 4.1 and Figure 4.14. Using the EC50 estimates81and associated standard errors resulting from the fit, I conducted a single factor ANOVA test with thenull hypothesis that all EC50s resulted from the same distribution, which I rejected with p= 3.6×10−5.To determine which EC50s were different, I conducted a post-hoc Tukey’s HSD test, and found that allEC50s were different with p < 0.05 minimum (see Table 2.1).I also compared these results with the estimates of maximum enhancement of cell killing that Iderived above. I was surprised to find that the experimentally measured ∆= DA−DY was even greaterthan the theoretical predictions (Figure 4.15). It is possible that this is partly due to the higher levelof wild type YT killing in the data used for the theoretical predictions (from Section 4.2.4)—whichis likely due to minor variations in experimental conditions, as well as cell health, functionality andpassage number. However, this is unlikely to be the only cause of this effect, since the two different YTpopulations in this co-culture experiment were treated identically and concurrently. The other likelyfactor is that the assumptions made to calculate the estimate are wrong. In particular, this comparisonmight suggest that the three sources of cell death (YT-killing, baseline CB1954 toxicity and toxin-activated CB1954 toxicity) are not independent, and are rather this is an indication that they are actingsynergistically to enhance YT-killing. However, the simplistic nature of Equation 4.1 and its lackof validation, along with the fact that this discussion is based on comparing data from two separateexperiments, makes any firm conclusions inappropriate.Together, these results demonstrate a moderate enhancement in YT-killing of MCF-7s due to somecombination of GZB-NFSA expression, CB1954 treatment, and possibly synergistic activity betweenthe two, although the latter is uncertain. Furthermore, repetition of this experiment has been con-founded by technical issues—flow cytometry sorting errors and YT effectors with an exhausted, non-cytotoxic phenotype—and therefore I cannot make any final definitive conclusions at this time regard-ing the enhancement of YT-killing by the GZB-NFSA/CB1954 system. Efforts are ongoing in the labto remedy this.4.3 DiscussionCell based therapies are in the process of transforming medicine [177, 180]. In cancer therapy, adoptivecell therapy using tumour infiltrating lymphocytes and CAR targeted T-cells can be a curative treatmentin large fractions of patients with malignant melanoma [181] and B-cell malignancies [246], respec-tively. However, many patients still fail to respond to these treatments, for reasons which have notbeen fully elucidated [247], but are sure to include the immunosuppressive tumour microenvironment,antigen escape, and resistance to cytotoxic effector mechanisms.Here I have attempted to begin to address one aspect of the issue of apoptosis and lymphocyte re-sistance by using cytotoxic lymphocytes to deliver potent orthogonal toxins via the granzyme-perforinpathway. The approach I used to do this was using granzyme B as a molecular chaperone to insertthe toxins into the granzyme-perforin pathway, as I had previously shown this was possible using afluorescent protein (Chapter 3).82l l lllllllllllll lllllllllllllllll llll lllllll ll0.000.250.500.751.000.05:1 0.5:1 4:1Effector target ratioNormalized MCF−7 cell deathEffector cell population and drug treatmentl l l lYT YT−CB1954 YTN YTN−CB1954Figure 4.14: Investigating the enhancement of YT killing of MCF-7s by GZB-NFSA/CB1954.Fluorescently labeled target MCF-7s were co-cultured with YTs (YT) or YTs expressing GZB-NFSA (YTN). The target cell number was fixed, and the effector cell number adjusted for therange of E:T ratios. After the co-culture, the mixed cell population was sorted via FACS. Afterdebris and doublet gating, target MCF-7 cells were gated upon using two colour discrimination(one colour for each cell type). One thousand target MCF-7 target cells were sorted into each wellof a 96-well plate. These cells were then treated with either 10 µM CB1954 or vehicle control(0.1% DMSO). Cell viability was assessed one week later using a fluorescent metabolic activityassay. Raw fluorescent values were normalized to those derived from wells containing untreatedMCF-7s, and then these values were subtracted from 1 to transform the data from viability to celldeath. Each data point is a reading from a single well of a plate. Solid lines are the fitted logisticfunctions, with the shaded region denoting the 95% confidence bands for the fitted function. Thesolid squares with horizontal error bars are the fitted EC50 and associated error the estimate. YT= wild type YT; YTN = YT expressing GZB-NFSA; DMSO = vehicle control treatment afterco-culture; CB1954 = 10 µM CB1954 treatment after co-culture.I have generated a variety of granzyme-toxin fusion proteins, and shown that the toxins retain theiractivity as C-terminal fusions to granzyme B. However, my results also demonstrated that substantialattenuation of toxin potency can occur, as was the case with the nitroreductase fusion proteins GZB-NFSA and GZB-NFSB.MCF-7s are widely used as a model cell line both for breast cancer, as well as apoptosis resistance,due to their deficiency in caspase 3. I characterized their susceptibility to killing by the natural killercell line YT-Indy. I found that while MCF-7s are resistant to YT-killing over the duration of typical83l l lllllllllllllllll llll lll0.000.250.500.750.05:1 0.5:1 4:1Effector target ratioNormalized MCF−7 cell deathEffector cell population and drug treatmentl lYT YTN−CB1954Figure 4.15: Comparing estimates with experimental measurements of enhancement of YTkilling. Comparison of the experimental data (solid lines, circles and shaded bands, all as inFigure 4.14) and estimates from Equation 4.1 (dashed lines).co-culture and assay timescales (several hours to a day), ultimately YT-targeted MCF-7s do die (withina week). I encountered similar results in my attempt to render target cells resistant to YT-attack byinhibition of apoptosis. When using both small molecule inhibition of caspases, as well as overexpres-sion of IAP proteins, I found these modifications delayed target cell death due to YT attack, but did notactually prevent it. This recurring theme of delayed target cell death is relevant to the wider communitystudying both apoptosis resistance and methods to overcome it, in that long term evaluation of cellviability is critical.These results, and the failure of multiple approaches to render target cells fully lymphocyte resis-tant, raises two important points. First, it is natural to wonder why, despite reports in the literature(Section 4.1), inhibition of apoptosis was insufficient to render MCF-7s lymphocyte resistant. In thisregard it is important to note that I did not directly functionally verify that the apoptotic cascade wasinhibited, for example by monitoring caspase activity. However, both IAP overexpression and smallmolecule inhibition did delay cell death on the timescale of hours, so it is fairly clear that these treat-ments were exerting some pro-survival activity. Further, as discussed above, most of the reports ofin vitro IAP-mediated lymphocyte resistance only measured target cell viability after short timescales(at which point our data and that in the literature is congruent), and the few data points in the litera-ture from experiments that monitored the target cell viability over several days have small effect sizes84(Section 4.2.3). Beyond these specific reports, much of the evidence for IAP involvement in cancercomes from either mouse models or patient samples [219–227]. In these in vivo contexts, it is possiblethat IAP upregulation provides a proliferative effect rather than, or in addition to, resistance to immunemediated toxicity. Also, as it is the dominant therapeutic modality, much of the evidence for IAP medi-ated tumour resistance to apoptosis is in the context of chemotherapy [248–250], which is not directlycomparable to lymphocyte resistance.More broadly, the challenges I encountered in generating a fully lymphocyte resistant cell-linespeak to the complexity of the cytotoxic lymphocyte-tumour cell interaction, and the difficulty in mod-eling it in vitro. My results indicate that simple modification of one core pathway, no matter howcentral, is perhaps insufficient to prevent target cell death in a highly artificial co-culture environment,where the effector cells are free to attack over an extended period of time, with none of the metabolic,soluble or cell-mediated inhibitory factors encountered in a tumour. A more physiologically rele-vant model system might include multiple modifications to a variety of components of the apoptosispathway, perhaps with redundancies. These results may also suggest that in the context of cytotoxiclymphocyte therapy of tumour cells, apoptosis and lymphocyte resistance are secondary issues. Cer-tainly inhibition of cytotoxic lymphocyte trafficking to the site of the tumour, inhibition of lymphocyterecognition of tumour-cells, and inhibition of lymphocyte binding and engagement to tumour cells areall well characterized immunoresistance mechanisms [96]. Furthermore the immunosuppressive tu-mour microenvironment, containing regulatory T-cells and myeloid derived suppressor cells, as wellas tumour upregulation of PD-1 and IDO expression are key considerations [105, 251]. Perhaps thesemechanisms are the core issues that need to be addressed. More likely it is some combination of allthree factors: the tumour microenvironment, lymphocyte tumour cell recognition and engagement, andtarget cell susceptibility to lymphocyte toxicity. My results seem to suggest that simply focusing on thelatter in isolation, especially in an isolated co-culture system is perhaps insufficient.The limitations of my model system are likely part of the reason why the effect sizes I observed inmy final series of experiments are modest. When I attempted to use the GZB-NFSA/CB1954 systemto enhance YT-killing of MCF-7s, I was aiming at a small therapeutic window. Any YT-targetedMCF-7s I knew would ultimately die, and the therapeutic index of the GZB-NFSA/CB1954 system inMCF-7s is also very small. I conducted a series of dose response experiments and found that MCF-7s co-cultured with YTs expressing the toxin fusions and treated with CB1954 exhibited greater celldeath than those that were co-cultured with unmodified YTs and not treated with CB1954. However,the relative contribution of CB1954 treatment and GZB-NFSA delivery to target cells, and the potentialsynergy thereof, is unclear. This, along with the moderate magnitude of the enhancement of cell killing,I attribute at the very least to my lack of a YT-resistant cell line, as well as the small therapeutic indexof NFSA as a granzyme fusion. It is also possible that the presence of a host immune system mightincrease the efficacy of GZB-NFSA/CB1954, in that even moderate levels of bystander killing couldresult in exposure of new antigens, providing a degree of antigen spread, and thus restarting the hostimmune response.85Unfortunately, CB1954 has a dose dependent toxicity in MCF-7s even in the absence of NFSA,which results in a small therapeutic window. Several of the granzyme-toxin fusion proteins I developedhave significantly greater therapeutic windows, in particular GZB-HTK/ganciclovir and GZB-DTA.Viral transduction of GZB-HTK might reduce or eliminate the genotoxicity of this fusion protein uponexpression in YTs, which would make this system very attractive. Since GZB-DTA does not producea bystander effect, using it would require a target cell line that was fully lymphocyte resistant. Further-more, the delivery chassis would have to be protected from DTA toxicity. This is feasible, as a mutantform of EEF2—recall DTA inhibits protein synthesis by ribosylating EEF2 (Section 4.1)— has beenreported, the overexpression of which renders cells resistant to DTA toxicity [252]. I have shown thatco-expressing this mutant EEF2 along with GZB-DTA abrogates the toxicity of GZB-DTA (as shownin the Appendix). If these two issues were resolved, DTA would be another attractive payload to moveforward with.In conclusion, I have demonstrated that granzyme-toxin fusion proteins retain their activity to vary-ing degrees. Total, long-term resistance to lymphocyte attack in an artificial co-culture system is chal-lenging to achieve, and perhaps not as relevant as other elements of tumour immunoresistance. Finally,I have shown measureable enhancement in MCF-7 target cell killing by YTs expressing GZB-NFSA.The effect sizes are moderate, the synergy of the toxin/prodrug interaction unclear, and any firm con-clusion is pending repetition of the experiment. I am hopeful that with a more relevant target cell line,and potentially a toxin with improved activity as a granzyme fusion, this system may be shown to beworth pursuing for further development in the context of adoptive cell therapy of cancer.4.4 Methods4.4.1 PlasmidsTwo plasmid systems were used for this work. The MND plasmids are based on a lentiviral transfervector, and have been described elsewhere [253, 254]. They use the MND promoter to drive highlevels of expression in hematopoietic lineages. The pdL vector is a custom mammalian expressionplasmid I constructed in house, based on a pcDNA3.1(+) (Thermo Fisher Scientific) backbone. Specif-ically, the mammalian and bacterial selectable markers and all origins of replication are derived frompcDNA3.1(+), corresponding to bases 1670 (CGATTTCGGCCTATTGGTTA...) to 5396 (...TAAA-CAAATAGGGGTTCCGC). A custom expression cassette was cloned into this backbone. This cas-sette consisted of eukaryotic and prokaryotic promoters and ribosomal binding sequences, followed bythe open reading frame, followed by eukaryotic and prokaryotic transcriptional termination sites. Forthe mammalian promoter I used the CAG promoter for its ability to drive high levels of expressionin a variety of tissues. The sequence was amplified from pEMS1157 [197]. This was followed bya hybrid T7 prokaryotic promoter, taken from pCMVTnT (Promega). This was followed by consen-sus Shine-Dalgarno and Kozak sequences. Following this is the open reading frame, which varies by86plasmid. Following the end of the coding sequence, there is a BGH polyA sequence, and then a T7terminator (with both sequences taken from pcDNA3.1(+)). Restriction enzyme cleavage sites flank allcomponents to facilitate subcloning.A variety of coding sequences were inserted into these two base vectors. The coding sequencesfall into two general categories: toxin related and inhibition of apoptosis related. For the granzymeB toxin (GZB-TOX) fusions, the basic structure was always: full length granzyme B, followed by aglycine-serine linker, followed by the toxin, followed by a P2A ribosomal skipping peptide, followedby GFP. For plasmids expressing inhibitor of apoptosis genes the inserted coding sequence had thegeneral form of: inhibitor of apoptosis protein (either XIAP or Survivin), followed by the same P2Apeptide, followed by a granzyme B FRET reporter. This last component consists of CFP, followed bythe consensus granzyme B cleavage substrate, followed by YFP. It was constructed in house, and thedetails have been published ([255], Sharma et al., in preparation). In all cases restriction enzyme cutsites flank all components.For clarity and completeness, actual plasmid information is organized as follows. Plasmid mapsand full plasmid sequences for the base MND and pdL vectors are in the Appendix (Section E.2). Thefull coding sequence that was inserted into these base vectors is also in the Appendix for all plasmidsused. In this way the full plasmid sequence for every plasmid used in these experiments is captured.Finally, the source of each component of all coding sequences are listed in the Appendix (Table E.1).All plasmids were constructed through a combination of PCR, synthesis and restriction/ligationcloning. All PCR amplicons and coding sequences were sequence verified.4.4.2 Cell cultureHeLa cells were a gift from Jonathan Choy (Simon Fraser University). MCF-7 cells were a gift fromGregg Morin (Canada’s Michael Smith Genome Sciences Centre). 293T cells were a gift from Ki-eth Humphries (British Columbia Cancer Research Centre). YT-Indy and 721.221 cells were a giftfrom Judy Lieberman (Harvard University). YT cells were cultured in RPMI 1640 media, supple-mented with 20% heat inactivated fetal bovine serum, 1X GlutaMAX, 1 mM sodium pyruvate, 10 mMHEPES, 0.1 mM beta-mercaptoethanol. All other cells were cultured in DMEM, supplemented with10% heat inactivated fetal bovine serum and 1X GlutaMAX. All cell culture reagents were purchasedfrom Thermo Fisher Scientific.4.4.3 TransfectionHeLa cellsHeLa cells were transfected using Lipofectamine LTX. 2×105 cells were seeded in 2 mL in a well ofa 6 well plate, the day prior to transfection. The day of transfection, DNA was diluted in OptiMEM,followed by the addition of PLUS reagent, followed by a 5 minute room temperature incubation. LTX87reagent was added, followed by a 20 minute room temperature incubation, during which time thecell growth media was replaced with fresh media. Finally the transfection mixture was added to thecell media, rocking the plate to mix the added reagents. The quantities used were: 1.25 µg plasmidDNA; 1.25 µl PLUS reagent; 3.75 µl LTX reagent; 500 µl OptiMEM. (All reagents from Thermo FisherScientific).293T cells293T cells were transfected using TransIT.For GZB-DTA and GZB-HTK experiments 1/32 of a confluent 10 cm plate was seeded in 2 mlmedia in one well of a 6 well plate, the day prior to transfection. The day of transfection, DNA andTransIT were both diluted in OptiMEM, followed by a five minute room temperature incubation. Thetwo mixtures were combined, vortexed and incubated for 30 minutes at room temperature. Finally,the transfection mixture was added to the cell media, rocking the plate to mix the added reagents.The quantities used were: 1.25 µg plasmid DNA; 7.25 µl TransIT reagent; 62.5 µl OptiMEM. (TransITpurchased from MirusBio and OptiMEM from Thermo Fisher Scientific).For GZB-NFSA experiments 1.5×105 cells were seeded in 500 µL in a well of a 24 well plate, theday prior to transfection. The same protocol was used as above, with the following quantities: 0.5 µgplasmid DNA; 1.5 µl TransIT reagent; 25 µl OptiMEM.MCF-7 cellsMCF-7 cells were transfected using Lipofectamine 3000. 1/2 of a confluent 10 cm plate was seededin 10 mL in a 10 cm plate, the day prior to transfection. DNA and P3000 reagent were diluted inOptiMEM, as well as L3000 reagent separately. These mixtures were combined, incubated at roomtemperature for 10 minutes, and then added to the cell media, rocking the plate to mix the addedreagents. The quantities used were: 14 µg plasmid DNA; 27.5 µl P3000 reagent; 41 µl L3000 reagent;500 µl OptiMEM. (All reagents from Thermo Fisher Scientific).YT cellsYT cells were electroporated using the Neon system (Thermo Fisher Scientific), using the 100 µl tip.6×106 cells were washed once in PBS, and resuspended in Buffer R along with 20 µg plasmid DNA,in a final volume of 110 µl. The extra volume ensures no bubbles are generated in aspirating the cellmixture into the electroporation tip. Critically the plasmid DNA must be of a concentration of at least1 µg/µl, and it must be prepared using an endotoxin free method. The quality of the plasmid prep greatlyinfluences the electroporation efficiency as well as the post-electroporation viability. The apparatuswas prepared as in the manufacturer’s manual, using the E2 electrolytic buffer. The electroporationconditions were 3×10 ms pulses at 1250 V. The electroporated cells are then immediately added to5 ml media spread across two wells of a 6 well plate.884.4.4 Flow cytometryCells were harvested (via trypsinization if adherent), and resuspended in PBS supplemented with10% complete media and 1 µg/ml of either propidium iodide (PI) (Thermo Fisher Scientific) or DAPI(Sigma) as viability stains. If cells were to be sorted, they were passed through a 35 µm nylon filter (BDFalcon). Cells were kept on ice and then analyzed on a BD Fortessa II, or sorted on either a BD Aria IIIor Fusion. For sorting, cells were sorted into complete media. In all flow cytometry experiments twoinitial gating steps were used. Debris was excluded by excluding cells at the bottom left corner of a PIvs FSC-A (forward scatter area) gate. Doublets were excluded using a hierarchical gating scheme: allcells with a wider pulse width signal were excluded first in FSC-W vs FSC-H (forward scatter widthvs height) and then SSC-W vs SSC-H (side scatter width vs height).4.4.5 Metabolic activity assayCell viability was assessed using PrestoBlue (Thermo Fisher Scientific cat # A13261), which is aresazurin based fluorescent metabolic activity assay. It is provided at a 10X concentration. Cell viabilitywas measured at 1X PrestoBlue concentration in complete growth media in black walled, flat bottom,optically clear 96-well plates (BD Falcon cat # 353219). The fluorescence of each well was measuredusing a Tecan Safire2 plate reader. The data was acquired with the plate lid removed, with four readsper well, and a gain setting empirically determined by the instrument for each plate.4.4.6 Direct toxin expression experimentsGZB-HTK in 293T cells293T cells were transfected as above, and then the media was refreshed on all cells 24 hours aftertransfection. 5 µM ganciclovir was added at this time to appropriate wells, as was 70 ng/ml colcemid(Thermo Fisher Scientific cat # 15210-040) to control wells. 48 hours after transfection, cells wereharvested via trypsinization, with all cells retained from supernatant, PBS wash and trypsinized cells.Cells were reseeded 1/8 in fresh media in 6 well plates and cultured for another 72 hours, with ganci-clovir and colcemid supplemented where appropriate. 5 days after transfection, cells were harvested asabove, and 1/80 of this cell suspension was reseeded in 100 µl media. These plates were incubated at37 ◦C for 2 hours, and then 10 µl PrestoBlue was added to each well, followed by another incubation at37 ◦C for 2 hours, followed by acquisition as above.GZB-DTA in 293T cellsThe GZB-DTA experiments were conducted as above, except that no ganciclovir was added, and thecells were assayed 48 hours after transfection, at which point the cells were harvested and preparedexactly as at the 5 day timepoint for the GZB-HTK experiments.89GZB-NFSA AND GZB-NFSB in 293T cellsThe GZB-NFSA and GZB-NFSB experiments were conducted as for GZB-HTK, with the followingdifferences. 48 hours after transfection, CB1954 was added at varying doses, with the DMSO concen-tration kept constant at 0.1%. 5 days after transfection, cells were harvested and prepared as above,with 1/10 of the cell suspension reseeded into 96 well plates for the viability assay.GZB-NFSA and GZB-HTK in MCF-7 cellsMCF-7 cells were transfected as above, and 48 hours after transfection, GFP+ PI- cells were sorteddirectly into 100 µl media in black walled, flat bottom, optically clear 96-well plates (BD Falcon). Themedia in the plates was supplemented with CB1954 or DMSO as a vehicle control prior to the sort. TheDMSO concentration was kept constant at 0.1%. 72 hours after transfection the media was changedon all wells and replaced with fresh media. 9 days after transfection, 11 µl PrestoBlue and 9 µl media(20 µl total volume) was added to all wells. (This was done to pipette a larger volume to minimizepipetting error.) The plates were incubated for one hour at 37 ◦C and then read as above.GZB-HTK experiments were performed according to the same protocol, except that ganciclovirwas used and the media was not changed at the 72 hour timepoint.4.4.7 Cell labelingCells were fluorescently labeled with CFSE, Cell Proliferation Dye eFluor 450 (eF450), or Cell Prolif-eration Dye eFluor 670 (eF670) (all from eBioscience) following the manufacturer’s protocol, exceptthat only 1 PBS wash prior to labeling was done and only 1 media wash after labeling was done.4.4.8 Cytotoxic lymphocyte co-culture experimentsUnless otherwise noted, 4×105 YT effector cells were combined with 1×105 target cells (either MCF-7 or 721) at a 4:1 effector:target (E:T) ratio in a final volume of 500 µl YT media in 5 ml polystyreneround-bottom tubes (BD Falcon). The cell suspension was gently pelleted by spinning it at 200g for 15seconds. The tubes were then incubated at 37 ◦C for 4 hours, and then prepared for flow cytometry orFACS sorting as above.4.4.9 Isolation of YT Targeted cells using the FRET reporter48 hours after transfection target cells expressing the granzyme B CY-FRET reporter were FACS sorted.The gating strategy was to first select YFP+ cells, and then select cells from a tight diagonal band inthe CY-FRET (405 nm excitation; 525 nm emission) vs CFP gate. This latter step is to ensure atight, clear, homogeneous FRET signal. These cells were then used in co-culture experiments withYT effectors. After the co-culture the cells were either analyzed for granzyme B status (via the FRETshift, as discussed in the main text) on a flow cytometer, or sorted and cultured for long-term survivalexperiments. In the latter case, target cells were identified as YFP+ (to differentiate from effectors). A90triangular FRET+ gate was set immediately below the diagonal line exhibited by targets cultured in theabsence of effectors in the CY-FRET vs CFP gate (as shown in the main text). FRET+ cells are thosethat are shifted down into the triangular gate.4.4.10 MCF-7 cell characterizationYT killingMCF-7s expressing the FRET reporter were transfected, FACS sorted and co-cultured with YTs asabove. After co-culture 2×103 PI-FRET+ or PI-FRET- cells were sorted into 100 µl media in a well ofa 96 well plate. These cells were cultured for 7 days, and then cell viability assessed using PrestoBlueas above.YT E:T ratio dose responseMCF-7s were labeled with eF670 and then co-cultured with YTs as above. For all E:T ratios, 1×105targets were used, and YT cell numbers were adjusted according to the E:T ratio. After the co-culture1×103 viable targets (DAPI-eF670+) were sorted into 100 µl media in a well of a 96 well plate. Thesecells were cultured for 7 days, and then cell viability assessed using PrestoBlue as above.4.4.11 Lymphocyte resistance experimentsSmall molecule inhibitionTarget cells were CFSE labeled and then pretreated with DCI and Q-VD-OPh, both 10 µM, for 90minutes at 37 ◦C. The cells were then co-cultured with YT effectors, with the drug concentrations keptconstant in the co-cultures. The cells were then either immediately analyzed by quantifying PI stainingvia flow cytometry, or isolated via FACS sorting. In the latter case 1×103 PI-FRET+ or PI-FRET-cells were sorted into 100 µl media in a well of a 96 well plate. The media was presupplemented withthe appropriate drug or vehicle control. These cells were cultured for 7 days, and then cell viabilityassessed using PrestoBlue as above.Inhibitor of apoptosis gene overexpressionMCF-7s expressing the IAP-FRET constructs were transfected, FACS sorted and co-cultured with YTsabove. After the co-culture 1×103 PI-FRET+ or PI-FRET- cells were sorted into 100 µl media in awell of a 96 well plate. These cells were cultured for 7 days, and then cell viability assessed usingPrestoBlue as above.914.4.12 Enhancement of YT-Indy killing experimentsYTs were transfected as above, and enriched for PI-GFP+ cells via FACS sorting as above. TargetMCF-7s were labeled with eF450 as above. Both cell populations were cultured separately overnight.Effector and target cell populations were then co-cultured as above. For all E:T ratios, 1×105 targetswere used, and YT cell numbers were adjusted according to the E:T ratio. After the co-culture 1×103viable targets (DAPI-GFP-eF670+) were sorted into 100 µl media in a well of a 96 well plate. Themedia was presupplemented with either 10 µM CB1954 or 0.1% DMSO as a vehicle control. Thefollowing day, the media was changed to drug free media for all wells. These cells were cultured for afurther 6 days, and then cell viability assessed using PrestoBlue as above.4.4.13 Drug reconstitutionAll drugs were prepared fresh prior to experiments.GanciclovirGanciclovir was purchased from Invivogen (cat #sud-gcv). A 1000X stock solution was reconstitutedby first combining ganciclovir powder in PBS, followed by vortexing for one minute. 10 µl 12 M hy-drochloric acid was then added per 1 ml PBS. The solution was then vortexed until the solid is com-pletely dissolved. A vehicle control was also prepared in parallel.CB1954CB1954 was purchased from Sigma (cat # C2235). A 1000X stock solution was reconstituted in anhy-drous DMSO.Dichloroisocoumarin and Q-VD-OPhBoth reagents were purchased from Sigma (cat # D7910 and SMC0063 respectively), and reconstitutedat 10 mM in anhydrous DMSO.4.4.14 Statistical analysisAll analysis was done in R. For all experiments except those in Section 4.2.6, comparison of meanswas done using a single factor ANOVA (using the aov command in R with the functional form Mean∼ Treatment) to test the null hypothesis that all means were sampled from the same distribution,followed by a post hoc Tukey’s HSD test (TukeyHSD in R) to test the pairwise difference betweenall means in the set. The p-values annotated on all figures are the adjusted p-values resulting from theTukey’s HSD test.The following approach was used to analyze the dose response curves of Section 4.2.6. Mediabackground was first subtracted from raw fluorescent values, which were then normalized to readingsfrom wells with no effectors. This created a dataset in which the MCF-7 viability ranged from 092to 1. Subtracting these values from 1 converted the viabilities to cell death. The E:T ratios werethen log transformed, giving a dataset of normalized viability vs. log(E:T), for each of the 4 sampletypes. These sigmoid dose response curves were then fit with a logistic function of the form y(x) =(1+ em(EC50−x))−1, with y equal to the normalized MCF-7 cell death, and x equal to the log of theE:T ratio. The fit was done separately for each sample, using a nonlinear least squares method (thenls function in R). These fits produced estimates for EC50 for each sample and the standard errorof these estimates. I then used a random sampling algorithm (specifically the sim values from thepredictNLS function in the propagate package) to generate the 95% confidence bands for fitteddose response curves.To test if the predicted EC50s were significantly different I conducted a single factor ANOVA. Isimulated observations of each EC50 by sampling a normal distribution with a mean of the EC50 andstandard deviation of the standard error from the non-linear least squares fit— that is I sampled thesampling distribution of the mean. I sampled either N = 2 observations of each EC50 (the number oftechnical replicates for each dose response curve), or N = 12 (the degrees of freedom from the leastsquares fit). I then conducted a single factor ANOVA using these simulated observations of the EC50s(using the aov command in R with the functional form EC50 ∼ Sample). For both N = 2 andN = 12 observations, the null hypothesis that the EC50s were sampled from the same distribution wasrejected (p = 3.6×10−5 and p = 2×10−16 respectively). To be maximally conservative, I proceededwith the lower powered N = 2 set of simulated observations, and conducted a post hoc Tukey’s HSDtest (TukeyHSD in R) to test the pairwise difference between all EC50s. These were all significantlydifferent, with a minimum p-value of 0.011. It is these p-values that are reported in Table 4.1.Finally to plot the dose response data and fits, I back transformed the log(E:T) ratios, and the fittedEC50s, by taking their exponential. To convert the standard error of the fitted parameters I used the fol-lowing formula (derived using standard propagation of uncertaintity: σEC50 = |EC50log(EC50)σlogEC50 |93Chapter 5Discussion, conclusions and futuredirectionsCellular therapeutics are likely to be a key medical intervention in the coming years. The ability togenetically modify cells raises the possibility of harnessing their diverse array of molecular function. Torealize this potentially fully, efforts across a range of disciplines are underway to re-engineer biologicalsystems into functional modules. Combining several of these modules in a cellular chassis has thepossibility of offering incredibly useful cellular devices.In this thesis I sought to contribute to this effort by repurposing the granzyme-perforin pathway as adelivery module for use in cellular therapeutics. This would enable cell-to-cell delivery of a therapeuticpayload, from a prepositioned secretory lytic granule in a delivery lymphocyte, through perforin poresand into a target cell. When combined with other modules such as receptor targeting and prostheticnetworks of control logic this could be used to deliver a large range of therapeutics throughout thebody.Towards this goal, in Chapter 2 I developed a computational biophysical model of the immuno-logical synapse between a cytotoxic lymphocyte and its target. These results had one very importantimplication for this project: namely that simple diffusion is a perfectly plausible mechanism by whichgranzyme transits perforin pores. Receptor mediated endocytosis or granzyme-perforin interactionsare not required, as have been suggested periodically over the last several decades. This is a crucialinsight as it suggests that therapeutics that are present in the immunological synapse have the potentialto transit perforin pores and enter the target cell. These results suggested that the problem of cell-to-celldelivery might be reducible to the problem of developing a method for releasing therapeutics into thesynapse upon target cell recognition.Two other biologically important insights were suggested by my model. First, the various adhesionand signaling molecules in the synapse create a crowded environment that is critical for lymphocytekilling of target cells, as without this crowding virtually all granzyme and perforin escape the synapse.Second, even in the presence of this crowding, substantial amounts of granzyme and perforin still94escape the synapse. I propose that the mechanism by which target cell specificity is maintained is thatthe requirement for high local concentrations of perforin (for pore formation) and granzyme (for cellentry) in the same place and time acts as a strong bimolecular filter. In the synapse, these conditionsare met and granzyme enters the target cell; in the case of escaped granzyme or perforin, they bothdilute rapidly and so those conditions are not met and thus granzyme does not enter bystander cells.These findings challenge the concept of a tight seal at the edges of the immunological synapse, whichhistorically has been accepted as a given [256, 257], even though experimental evidence is mixed andconflicting [152, 183].I then attempted to implement granzyme-perforin delivery experimentally in Chapter 3. To dothis I focused on granzyme B as a molecular chaperone for inserting and trafficking a payload intoand through the pathway. I designed a set of granzyme B derived tags and fused them to mCherry as amodel payload. I screened these tags by assessing if they colocalized with lytic granules, using confocalmicroscopy. These experiments had interesting biological implications, in that they suggested that thedomains of granzyme B that are responsible for directing it from the endoplasmic reticulum to the lyticgranules are not contiguous in amino acid sequence space, and potentially are not localized to a singlemotif in the tertiary structure either. They also indicated that two candidate granzyme chaperones werepromising. I moved on to test the capacity of these chaperones to fully transfer the payload from aneffector to a target cell and found that full length granzyme fusions were indeed capable of this. Iconfirmed these results at the protein level.Having demonstrated that granzyme-perforin mediated delivery was at least possible in principle,I attempted to use it to enhance lymphocyte killing via delivery of additional toxins to tumour cells inChapter 4. I generated a variety of granzyme toxin fusions and found that all retained their activity,albeit with varying attenuation in potency. Efforts to develop truly lymphocyte resistant cells wereultimately unsuccessful. In light of these results, I attempted to still demonstrate enhanced killingusing an effector dose response curve, with the aim of achieving a modest, but measurable increased,effect. Using an E. coli derived toxin/prodrug system, I did show that effector cells expressing the toxinfusion induced greater cell death than wild type effector cells. However the effect sizes were modest,and the relative contribution of the prodrug itself and the toxin fusions remains to be fully determined.In the remainder of this chapter, I discuss some of the issues and outstanding problems encounteredin this work, as well as ways these might be addressed. I close with some broader thoughts aboutengineering cellular therapeutics.5.1 The utility of the granzyme-perforin pathway as a delivery systemThe balance of evidence for the broad scale utility of granzyme-perforin mediated delivery presented inthis thesis is equivocal. While I have confirmed that granzyme mCherry fusion proteins are transferredto target cells, it is only in a single effector and target cell line. Certainly the efficiency of this systemis not perfect: in my proof-of-principle experiments I observed substantial background transfer of95unfused mCherry. However, there is always background in any biological system. In particular in thiscase the experiment was conducted in an artificial co-culture system in which there is ample time forspontaneous absorption of the mCherry protein from the media, which in turn might be present due tospontaneous effector cell lysis. When I attempted to use this system with a toxin payload, the resultingenhancement in cell killing was moderate, but measurable. I feel that the modest signal-to-noise ratioachieved in this work is not unreasonable: many mature technologies leveraging a new concept startout with poor efficiency. Subsequent iterative engineering can greatly improve the efficacy, robustness,broad scale applicability and so on.A major unsolved problem in using the granzyme-perforin pathway as a delivery system is the re-quirement for the fusion proteins to be loaded into, and spend significant sequestered in, lytic granules.These granules are acidic, and contain a variety of proteases. Thus for every new protein payload, thereis a chance that it will be substantially degraded, or at least decoupled from the granzyme chaperone. Ihave observed varying amounts of fusion protein breakdown for all fusion proteins tested in YT-Indys.In early experiments I observed considerable breakdown in another natural killer cell line (NK-92MI).This context dependency, which is unknowable for each protein until it is empirically tested, is a majordrawback of this approach.To address this problem, I have considered alternative methods for secretion into the immunologicalsynapse. This approach is supported by the computational results presented here that indicate simplediffusion through perforin pores can transfer of a molecular species from the immunological synapseinto the target cell. If this were true, and it has not yet been demonstrated experimentally, then theproblem of developing a lymphocyte based delivery system becomes reduced to developing a secretorysystem that is activated by lymphocyte recognition of a target cell. In principle this could be achievedby placing the gene coding for the desired payload under control of a TCR response promoter, forexample the Nuclear Factor of Activated T-cells (NFAT) promoter. If the gene contained a secretorysignal sequence at its 5’ end (for example the commonly used IL-2 or IgG sequences) TCR or CARrecognition of a target cell would initiate synthesis and then secretion of the payload. The problem withthis approach is that by the time the protein was synthesized, the lymphocyte would have disengagedfrom the target cell. Thus, the secretion would not occur into the synapse: the cell-to-cell nature ofthe system would be lost. The result of this secretion would depend on the context. If the lymphocytewas in a target rich environment, it is possible that secretion would occur as the lymphocyte formed asynapse with another target, thus providing perforin pores the payload could access the target cell. Inthe case of a lymphocyte in a tumour this perhaps would be tolerable, since the secreted therapeuticwould still target cells in a tumour. In other applications where cell-specific targeting is required, thiscould cause unacceptable adverse events.This illustrates the main advantage of using the granzyme-perforin pathway: granules and theircontents are already synthesized and prepositioned, ready for release. Therefore, to build a system thatsecretes a therapeutic into the synapse on the appropriate time scales, the therapeutic must be preposi-tioned as well. The mechanisms by which TCR signaling initiates lytic granule release are somewhat96understood, mainly involving MTOC polarization to the synapse followed by lytic granule polarizationalong microtubules. At the synapse the LGs first dock to the membrane, prime by undocking but re-maining tethered, fuse to the membrane and finally exocytose their contents [129]. While it might betheoretically possible to insert a presynthesized payload into this chain of events at a late stage (andthereby avoid lytic granule degradation), there is certainly no obvious, tractable approach to interfacewith this complex machinery.More broadly, a general consideration of the interaction between the payload and the delivery cellis important. As discussed above, lytic granule sequestration can be detrimental to the fusion proteinintegrity and activity. But the payload could easily be harmful to the cellular chassis as well. Anobvious case is that of delivery of toxins. In my work here, I avoided this difficulty by using toxinprodrug systems, in which the delivery lymphocytes would not be affected until the prodrug were ad-ministered. Importantly, prodrug dosing would eliminate the delivery lymphocytes, which dependingon the application might or might not be appropriate. But in the general case, such two componentsystems are unlikely to consistently be available. There are two ways in which this issue could beaddressed or at least mitigated. The first is via sequestration: either in lytic granules, or another secre-tory vesicle. This could well be sufficient to prevent any major activity of the payload in the deliverycell. However, it reintroduces the problem of payload degradation. The second approach would beto engineer the delivery lymphocyte with additional components that protect it from the payload. Forexample, in the case of diphtheria toxin (which exerts its toxic effect by inhibiting elongation factor 2),overexpression of a mutant form of elongation factor 2 that is resistant to diphtheria toxin inhibitioncan protect cells from concurrent diphtheria toxin expression [258]. Again, this approach is certain tonot be generally applicable, as a gene encodable factor that protects the delivery cell will frequentlybe unavailable. Furthermore, as increasing numbers of modifications are made to a cellular chassis,there is a significant metabolic load placed on the cell, and the potential for unexpected interactions be-tween the engineered components increases with their number. This combinatorial increase in potentialinteractions highlights the need for orthogonal parts as cellular therapeutics increase in complexity.Finally, a discussion of the cellular chassis itself is required. Exploiting the granzyme perforinpathway requires the use of a cytotoxic lymphocyte, as they are the cells that contain the pathway.While it is one of the main effector mechanisms of cytotoxic lymphocytes, it is not the only one.Death receptors and cytokines are other prominent examples. In order for this system to be generallyapplicable and deliver a range of payloads, and not simply toxins, these mechanisms would have to beeliminated. This is in principle achievable via either RNAi attenuation, or gene knockout. Two factorsmake this a significant challenge. The first is elucidating exactly which genes must be silenced in orderto abrogate lymphocyte cytotoxicity. Regardless of their exact nature, it is likely to be more than five,possibly more than ten. This is certainly a substantial undertaking, but given the multiplex capabilitiesof the CRISPR/Cas9 system, it is not impossible. However, when considering the feasibility of thisapproach, it is important to also consider that in a clinical application, whatever modifications aremade to the cellular chassis are likely to be made in primary cells, at least in the near term. (While the97use of immortalized cell lines has been reported clinically [259], their widespread use seems unlikelyfrom a regulatory standpoint due to the risks of oncogenesis or otherwise uncontrolled proliferation.In the long term, standardized starting populations of heavily modified precursor cells that are thencustomized to a final personalized cellular product are conceivable, but this is a distant goal.) Germlinemodification of primary cells is a substantially greater challenge than for cell-lines, and indeed as-of-yetunproven: methods for simultaneously isolating, in a non-destructive manner, a population of cells thatuniformly have a bi-allelic knockout of five to ten genes do not currently exist. Doing so and recoveringmeaningful, clinically useable numbers of cells would likely be harder still. A potential, speculativesolution to this problem would be to reconstitute the granzyme-perforin pathway in a non-cytotoxiccellular chassis. While this sounds ambitious in the extreme, an analogous feat has been achieved forthe initial steps of T-cell signaling in 293T cells [260].It is clear that there are many large impediments to using the granzyme-perforin pathway andcytotoxic lymphocytes as a general delivery system. However, the core functionality that it provides,precise, cell-to-cell delivery, is sufficiently attractive, and my results so far sufficiently encouraging,that it is worth further pursuing this goal.5.2 Further efforts to demonstrate the toxin-mediated enhancement oflymphocyte killingDue to the limitations discussed above surrounding the cytotoxic nature of the delivery chassis, I viewdelivery of toxins to cancer cells in the context of adoptive cell therapy as the most likely potentialnear term application. This eliminates the need for chassis attenuation, and my results suggested thatthis approach may be feasible, but were not definitive. There are several avenues that could be pursuedto increase the magnitude of the toxin-mediated enhancement of lymphocyte killing of target cells, inorder to make such a determination.5.2.1 Development of lymphocyte resistant target cellsFirst and foremost, a target cell population that is near or fully resistant YT-Indy would greatly increasethe potential observable enhancement of lymphocyte killing via toxin delivery. In the experimentspresented in this thesis, the maximum theoretically attainable enhancement in killing was quite small,and was dependent on a bystander effect. Based on initial experiments it was known that any directlytargeted cell would die, thus greatly increasing the baseline level of cell death, regardless of toxintransfer. If a target cell population were available in which YT-targeted cells survived, this woulddecrease that baseline, thereby increasing the maximum enhancement in killing that might be observed.Furthermore, it would expand the suite of potential toxins that could be employed, since the choice oftoxins would no longer be confined to those that can produce a bystander effect.I attempted to render cells apoptosis and lymphocyte resistant by inhibiting apoptotic cascades. Amore direct approach would be to directly eliminate key mediators of that cascade. Using RNAi or98CRISPR/Cas9 to knock out the key executioner caspases 3 and 7 would be worth pursuing, especiallyin MCF-7s which are already caspase 3 deficient [240]. Additional inhibitor of apoptosis genes couldalso be tested, for example the BCL family [220], as well as the granzyme inhibitor PI-9 [232]. Finally,these approaches could be combined combinatorially. While this would be a substantial effort, it isalso probably the most likely to succeed, as each component (caspase knockout or attenuation, caspaseinhibition, and granzyme inhibition) would work in parallel, absorbing whatever leakage or overflowdeath signals that overcome the inhibition from the other components.In the pursuit of fully lymphocyte cell lines, two points are worth considering. First, it is possiblethat such a hypothetical resistant cell population might also be resistant to additional toxins. However,this would be testable by direct expression of these toxins in the target cells. Second, if such a modelsystem cannot be generated, it is worth considering that it may not be biologically relevant. Perhapstrue lymphocyte resistance is not a characteristic of tumour cells. While there is a body of literatureon the subject (Section 4.1), there is a far greater focus on the mechanisms by which tumours escapelymphocyte recognition and prevent lymphocyte adhesion to tumour cells [199]. Perhaps this is thegreater issue. The immunosuppressive environment is also unquestionably an important factor in thisregard [261], one which is not modeled in my current experimental design. Therefore, perhaps it is un-realistic to expect a target cell to survive an extended period of repeated attack in a completely isolated,artificial co-culture environment. Perhaps a more realistic model might be found in a tumour mousemodel which might more fully capture the complexities of an actual tumour, including the tumour mi-croenvironment. While this approach might allow for the observation of much more compelling effectsizes, it is hard to justify moving into an animal model system without convincing evidence of efficacyex vivo, even if that evidence is harder to obtain. In closing, it is important to emphasize that this isnot to say that lymphocyte delivery of toxins to tumour cells, if achieved, would not be therapeuticallyrelevant. Rather, demonstrating its efficacy might require a more representative model system.5.2.2 Improving fusion protein granule loading and deliveryIncreasing enhancement of effector killing could also be achieved by addressing the other side of theproblem: increasing the amount of toxins that are delivered to target cells. In this regard there are twostrategies that would be worth immediately pursuing. The first is generating effector cells that expresstoxin fusion proteins from the endogenous granzyme locus. This could be done by using a CRISPR/-Cas9 system to introduce a double stranded break at the 3’ end of the granzyme B coding sequence, andsupplying a suitable repair template that inserts a C-terminal toxin as a fusion protein. As is describedin the Appendix, I have already done this with an mCherry fusion protein as a proof-of-principle, soI am confident that this step is feasible. A fusion protein that is expressed from the endogenous lo-cus might increase the amount of fusion protein that is loaded into lytic granules, as compared to theamount loaded in cells where the fusion protein is being expressed ectopically. The main reason forthis hypothesis is the possibility of a competition effect: namely that granzyme produced from its ownlocus is loaded more efficiently into lytic granules. This could due to expression that is coordinated99with cell division or degranulation (when granules need to be populated or repopulated respectively).Or it could be that granzyme fusion proteins load less efficiently than does unfused granzyme. If thiswere the case, then expressing endogenous fusion proteins would essentially eliminate the competitionposed by the unfused granzyme. In this case a biallelic modification of the granzyme locus would benecessary. Finally, in the case of ectopic overexpression, it is almost certain that granule loading is sat-urated. Excess fusion proteins are likely secreted, which in vivo could result in unwanted toxin activityat off-target sites. Expression from the endogenous granzyme promoter would likely greatly decreasethis background.A second approach worth considering would be the development of a minimal chaperone or tagthat was still sufficient for delivery via perforin pores, but had improved stability and persistence in theharsh environment of lytic granules. Having observed appreciable fusion protein degradation acrossseveral fusion proteins and lymphocyte cell types, I initiated a series of mass spectrometry experimentsto investigate if there were any regions in the fusion proteins that were more susceptible to breakdown.The data is shown in the Appendix, and indicates that the region surrounding and including the glycine-serine linker could be modified to minimize fusion protein degradation. In terms of developing asuperior chaperone, efforts to map the critical domains of granzyme B would be informative, as woulda similar investigation of other granzymes and lytic granule constituents. Finally, synthetic tags suchas the glycosylation independent lysosomal tag (GILT) domain could also be investigated [262].5.3 Future directions for granzyme-perforin delivery systemsThe concept of granzyme-perforin mediated delivery is still in its infancy, and its practical utility re-mains to be robustly demonstrated. This makes any in depth discussion of its possible extensions,additions and applications of increased complexity premature. However, a certain amount of thoughtand attention has been invested in preparation for such an eventuality, and a brief summary is presentedhere.Addition of receptor mediated targetingFor simplicity, natural killer cell lines with known target cell lines were used here as the most basicmodel system. As the approach is developed a return to primary T-cells would likely be appropriate. Inboth primary and immortalized cells, and in both T-cells and NK cells, targeting could be achieved viathe addition of either a synthetic TCR, or a CAR, as discussed in the Introduction.Addition of suicide switchesEspecially given a recent series of fatalities in a CAR-T trial, the ability to rapidly and safely eliminatethe administered cell product in the event of unforeseen toxicity will be a necessity moving forward.There are already clinically validated methods for doing this, such as the iCasp system [263]. Anotherpromising approach is a dual CD20-CD34 marker which, when expressed on the surface of adminis-100tered CAR-T cells, allows for selective depletion of the administered cells using rituximab, a clinicallyapproved anti-CD20 antibody [102]. However, in this latter case, whether the depletion is sufficientlyrapid for clinical use, and its general clinical utility remain to be demonstrated in clinical trials.Extension to non-protein payloadsAll designs and approaches that have been presented here are limited to protein or peptide payloadsthat can be fused to a molecular chaperone that inserts the payload into the granzyme perforin pathway.However, it is possible to envision ways to extend this approach to nucleic acids and small molecules.In the case of small molecules, DNA aptamers against the protein chaperone (for example granzymeB), as well as the small molecule itself, could be generated using SELEX [264–266]. These couldbe joined using a DNA linker sequence, to create a bi-specific aptamer [267, 268]. In this way itwould be in theory possible to couple the small molecule to the chaperone. In order to deliver largeDNA constructs, a similar strategy could be pursued, using the bacterial plasmid segregation machinery[269]. The protein chaperone could be fused to a centromere binding protein (for example ParR). TheDNA payload would then be cloned into a plasmid, which would contain a centromere-like DNA siteto which the matching centromere binding protein would bind (for ParR this would be parC). Againthis would couple the DNA payload to the protein chaperone. The first and most obvious unknown iswhether these couplings would survive trafficking through the granzyme-perforin pathway, especiallysequestration in the lytic granule. While these ideas are highly speculative, the delivery of wholenucleic acid circuits in particular is especially appealing, and perhaps eventually even entire syntheticgenomes [270, 271].Neutralization of the delivery chassisFor any application beyond the delivery of a payload to a target cell population for which death is thetherapeutic objective, the cytotoxicity of the delivery lymphocyte chassis would need to be eliminated.Otherwise, whatever therapeutic or diagnostic effects that were achieved by payload delivery would becountered by the lymphocyte cytotoxicity. Here gene attenuation or knock out would both be relevant.Both an RNAi approach as well as CRISPR/Cas9 approaches, either actual knockout using Cas9, orknockdown using dCas9 fused to a repressor [272] could be considered. There would be two mainchallenges in this undertaking. First, it would be necessary to delineate exactly which genes wouldneed to be targeted in order to eliminate lymphocyte cytotoxicity, and to what degree in the case ofgene knockdown. A minimal set of genes would likely be granzymes A and B, Fas ligand and TNF-α , but others could well be required. Second, simultaneous, biallelic knockout of multiple genes isa challenging task. Furthermore, additional modifications of the cellular chassis are required for thisapplication, such as the insertion of the payload. Finally, the efficiency of the entire process must besufficiently high so as to recover a cell population of a viability, size and purity that is clinically usableIn this regard, attenuation is perhaps more feasible, since there is no requirement for isolation of cellswith inactive target genes, but rather simply cells that have been successfully modified to express the at-101tenuating constructs (either RNAi based or Cas9 based). Using an immortalized cell line as the cellularchassis would make this undertaking slightly more feasible, since iterative modification steps would bepossible. However, clinical use of immortalized cell lines would raise substantial regulatory concerns,although there is precedent in clinical trials [273]. Thus, it is likely that chassis neutralization will notbe feasible without substantial improvements in genome editing methodology, which fortunately arenot unlikely.A tool for screening CARs for on-target off-tissue activityA significant outstanding problem in the development pipeline of CARs is that of off-target activity.While in vitro cross-reactivity testing using ELISPOT is helpful, it cannot capture all eventualities[274, 275]. The payload delivery system presented here might be used as a unique screening tool forCAR off-target activity in animal models. I would envision generating delivery lymphocytes expressinga CAR under development, and a Cre recombinase payload. These cells would then be adoptivelytransferred to a mouse that had a transgenic lacZ Cre recombinase reporter germline modification [276].Any tissue targeted by the CAR-T cells would also receive the Cre recombinase payload, triggeringexpression of the lacZ reporter. Following necropsy, standard X-gal staining would provide valuableinformation about which tissues were targeted, and these tissues could be further investigated. Animportant caveat to this approach is that the time for lacZ expression would have to be shorter thanthe time for target cell death, which is perhaps unlikely—making this another application for which aneutralized delivery cell chassis would be very useful. In this event, other payloads could be used whichdo not rely on target cell transcription for reporter activity. Epitope tags might be suitable payloads inthis case, as tissue sections could be analyzed using standard immunohistochemistry techniques, toascertain the same information. Using such a payload would not address the related issue of balancingthe timing of necropsy such that the CAR-T cells are adequately distributed throughout the mouse,while ensuring that targeted cells have not progressed sufficiently in their apoptosis that they cannotbe identified histologically. Finally, even were these challenges are overcome, it is important to notethat the antigen profile of a mouse is clearly not entirely representative of that of a human. As variousforms of humanized mouse models are developed, this discrepancy may decrease [277].5.4 Broader insights into cellular therapeuticsUnlike modern computing or aerospace, which rely on parts built from the ground up in a step-by-step and hierarchical fashion, biology is not yet an engineering discipline. Natural development ofmolecules, pathways, and cells has not been guided by the principles of modularity, insulation, androbustness. Rather, biological systems evolved in an entirely opportunistic manner to occupy a nichethat has a specific parameter set. Therefore, these systems can be highly unstable when perturbed orplaced in an environment with a different parameter space. These intricate evolutionary origins havesignificant consequences for the systems we attempt to engineer. It is important to recognize that102progress will be impeded by the complexity of molecular and cell biology, which imposes a heavyburden of noise, unpredictability, and context dependency on the enterprise of engineering cellulartherapeutics [278]. Successful examples are typically the result of trial and error, and contain manyaccessions to the messy realities of building functional biological systems. In these cases, countlessdesign choices have been made, which are rational and empirical, explicit and implicit, deliberate andaccidental. While the delivery system presented here cannot be classified as a successful cell-basedtherapeutic, several useful insights have been gleaned while attempting to make it so.5.4.1 Context dependencies in cellular therapeuticsBiological molecules and pathways are highly evolved and networked [279], very sensitive to pertur-bation [280], and modifying them often results in cell death or unexpected failure modes [281]. Thefunctions of individual molecules or pathways may be overlapping, integrated, redundant, and degen-erate [282]. Rather than an insulated collection of pathways with specified interactions, the cell canexhibit characteristics of a single large network of multiplexed, interacting parts. This network mayexhibit behavior that is probabilistic rather than deterministic, and in some cases may lack damping orinput filtering to determine output responsiveness. Locally, the physical and biochemical niche, alongwith stromal cells and the extracellular matrix all provide critical stimuli and mechanical cues thatinfluence cellular, tissue, and organ level differentiation and development.This connectivity, and sensitivity injects a substantial amount of noise and unpredictability intoengineered biological systems [283–285], and gives rise to what I would call context dependency.That is, the behavior, function, and stability of engineered components are entirely dependent on themolecular and cellular milieu in which they operate, and these characteristics may vary dramatically ifthis context is altered. Similarly, the connectivity and function of cellular pathways and the viabilityof the cellular chassis itself may also be highly sensitive to the introduction of a new component.While it may simply reflect our incomplete knowledge, the complexity inherent in these biologicalsystems is such that their behavior can verge upon chaotic [286]. Predicting the effects of simple, smallmodifications within or between systems is in many cases impossible.This has practical implications in the laboratory. Minor alterations to reagents or protocols canresult in outsized effects, completely unexpected catastrophic failures, or worst of all, spurious results.Assays to test, debug, and validate cellular therapeutics are imperfect: the results are often ambigu-ous, and can frequently be misleading such that following insertion into the cell chassis, engineeredcomponents can display partial or incomplete failure modes that can be easy to miss. These issuesare compounded by the sensitivity of the cellular chassis, and its tendency to initiate apoptosis uponperturbation. This is problematic, since the time window from the first signs of cellular dysfunctionto cell death can be very short indeed, which makes troubleshooting and debugging engineered cellsextremely challenging.Thus, when we attempt to repurpose specialized molecules, interconnected pathways and sensitivecells, we should not be too surprised when they reveal themselves to be brittle and subject to failure.103After all, we are attempting to tweak highly evolved systems. However, evolution does not logicallyimply optimization, a crucial distinction that offers a way forward, and a window in which to work.5.4.2 A framework for cell engineeringWhere does this leave us? How can we succeed in this difficult environment, and manage the issuesI have discussed above, while realizing the potential of repurposing biological systems? My solutionhas been to adopt empiricism and pragmatism as guiding principles. A set of rules has emerged fromthis approach, which my supervisor and I have proposed as a framework for building robust cell-basedtherapeutics in the complex, networked, and sensitive mammalian cell chassis [180].Parsimony and simplicityThe design phase should always be guided by these two core principles. Unlike other engineeringdisciplines, in which efforts to consider and address edge cases may yield more robust function, in thisfield, these extra layers of design are more likely to have unforeseen negative consequences rather thanimprove system stability. Avoid the common desire to ’overbuild’ systems by adding excessive featuresfor downstream and long-term functionality that are, in reality, far more likely to result in unintended,and potentially very damaging and confounding, consequences.ReusePre-existing, validated designs and components should be used wherever possible. Unless novel orimproved function is required, if a component has been shown to work sufficiently well for a givenpurpose, it should be used in place of any alternative, untested design, even if the latter is in theorysuperior. This concept should be extended all the way down to the level of nucleotide sequence, andthe local sequence context should be maintained if possible.AdaptationIf a molecule, pathway or cell with new or improved function is required, two options should be con-sidered: (i) refinement and engineering of the existing part using either rational design or evolutionarymethods, or (ii) sourcing a new component from other biological systems by testing those that havesimilar function, an approach that might be termed panning. Crucially, in either approach, the new andimproved component should be considered as untested. Its behavior should not be inferred from thatof the related component.Step-wise testingAlways test, at the molecular level, each step in the production, maturation, and function of an engi-neered molecule, pathway or cell. While certainly useful during pilot work or as screening tools, proxyreporter constructs or assays should not be relied upon to definitively confirm the function and behavior104of engineered pathways. Similarly, investigators should make efforts to work in the intended cellularchassis, and as much as possible avoid reliance on model cell lines that are easy to work with, but lessphysiologically relevant. This should continue throughout the path to clinical use, with each modi-fication tested in relevant animal models, and increasingly, in organ-on-a chip testbeds [287], whichhopefully will continue to improve our ability to avoid adverse events that were missed in pre-clinicalscreening [275].UniquenessEach combination of cellular chassis, genetic construct, experimental protocol, hardware apparatus,and so on, contains sufficient intricacies and permutations so as to render it unique. Extreme cautionshould be applied when mapping information from one experimental context to another, or betweentwo types of biological molecule, pathway or cell. This concept extends to the level of the patientreceiving the cellular therapeutic. The complex interplay between the immunogenicity of the cellulartherapeutic and the patient’s immune system has mostly necessitated cellular therapies that are eitherautologous or at least HLA matched. Even if these immune constraints are able to be overcome viainnovation, it is likely that the complexity and immune network of each patient will necessitate a morepersonalized approach when administering cellular therapeutics, compared to historical small moleculetherapy.Empirical designThe interconnected web of biological context dependency remains opaque, especially at a molecular,mechanistic level. This makes predictive design of meaningful systems challenging. It is verydifficult to predict how modifications to a given pathway component will affect other components inthe pathway, the pathway itself, other pathways in the network, and the cell as a whole. Thus, in mostcases theoretical predictions and designs are mainly useful at high level and preliminary stages ofdesign, and require careful validation once implemented in the laboratory.5.4.3 Grand challenges for cell engineeringThe complexity and diversity of biological function is a double edged sword. On the one hand itprovides a tantalizing set of parts from which to build cellular devices. On the other, its vagariesproduce context dependencies that necessitate the empirical and pragmatic approach outlined in theprevious section. This is not to minimize such an approach: the progress that has been made in thisproject has resulted from following such principles. I would venture the same is true for most othercellular therapies. Nor is it to suggest that biological knowledge is of secondary importance to a bruteforce, black box, empirical approach. Quite the opposite in fact: the intricacy of biological systemsmakes a deep and wide knowledge essential from design through development. Currently however,often this the knowledge is context dependent, making generalization difficult. This slows the progress105of developing specific cellular devices, since each effort is largely a new enterprise, rather than buildingon existing devices or technology. While this could be interpreted as being dismissive of cellularengineering, it is not meant to be, and indeed what has been achieved is remarkable. Rather it is simplya reflection of the early stage at which we find ourselves in developing the field. I suggest here twogrand challenges that if realized would greatly accelerate its maturation.Orthogonal systemsDevelopment of biological modules or systems that operate independently of their surrounding envi-ronment. This would reduce or eliminate context dependencies of individual parts which would greatlyincrease the robustness and predictability of the systems which are built upon those parts, thus increas-ing the success rate of designs. Initially such requirements would likely necessitate a highly artificialand fragile system, which would be confined to use as a testbed, for prototyping or debugging. Eventhis would be of great benefit: consider a wind tunnel or a breadboard. In the long run it is possible thatthese systems could be sufficiently strengthened for practical use.Orthogonal systems could be developed by following two main strategies: reconstitution, or denovo construction. Both are already underway. The most straightforward approach to the problem is toremove it: remove the cell. Cell free systems consisting of cellular extract have been available for sometime, and are mature enough that commercial kits are available. However, more recently this approachis being used for rapid prototyping, and has been used to develop novel RNA circuitry [288] whichcould be used for RNA-based control logic. Immunological insight motivated the reconstitution of aT-cell receptor in a 293T model system, but it also provides inspiration for further reconstitution of afunctional component in a non-native cellular chassis, which might allow for a greatly expanded rangeof function [260].More ambitious still are the efforts underway to expand, adapt and create novel translational ma-chinery. Synthetic ribosomes that recognize quadruplet codons, artificial amino acids and tRNAs, andE. coli strains that have redundant codons collapsed onto a single codon per amino acid have all beenreported [289–291]. This opens the possibility of entirely parallel protein machineries operating ina cellular chassis: the endogenous machinery performing basic homeostatic functions, while the syn-thetic translational system produces the added functional components. These two systems could operateindependently, free of cross-talk any the associated problems for both the viability host cell chassis andthe performance of the added components. Even more far reaching is the complete chemical synthesisof the Mycoplasma mycoides genome [292], as well as the delineation of its minimal gene set for via-bility [271]. Intriguingly, approximately one third of this essential gene set could not be annotated witha known function, indicating the work yet to be done to realize a truly orthogonal minimal cell.Obviously true orthogonality is likely impossible: there will always be context dependencies. Butwhatever insulation from these dependencies can be achieved will almost certainly bring substantial in-sight and biological knowledge, accelerate development cycles, and improve the robustness of cellularproducts. Perhaps most importantly, it would greatly improve the predictability of engineered systems,106perhaps releasing their design from the current empirically approach.Predictive designDevelopment of a method for predicting the behavior of biological systems based on a minimal setof parameters. The application of quantitative, predictive models to the design and development ofproducts is critical across all engineering disciplines. It allows for a streamlined, cheap, and rapiddesign process. It also massively increases the confidence that a design validated by the predictivetheory is likely to work once constructed, albeit usually with modification.In the fields of engineering, these models and theories are for the most part based on mathematicalformalism built on top of physical and chemical principles. An extension of this approach to cellularand molecular biology would be transformative. Why has this not been achieved? Is it possible?At first glance, it seems as though it should be straightforward. No one would question that ul-timately biological systems are governed by physics and chemistry. Incredibly refined, mature andaccurate theories exist for all of the molecular constituents of these systems. What is more, the lastfew decades have also seen the development of incredible computing power with which to implementthese models. In my view the problem comes down to two fundamental challenges: complexity andparameterization.For any molecular system of any size, we as a community have the appropriate mathematical andphysical machinery with which to model it. Using the formalism of molecular dynamics, and theunderlying electrodynamics and quantum chemistry where needed, we understand in principle all of theinteractions governing the molecular processes in biology. The system of equations could be writtendown, and in theory integrated forward in time. The problem is that we cannot even begin to applythe appropriate theories at the length and time scales that are necessary to usefully model biologicalsystems. There would be too many constituent parts. Even initializing the simulation in a meaningfulway would be near impossible. If such a feat were achieved, despite our vast computational resources,they would be massively insufficient for the task. This is the complexity problem. Solving it seemsunlikely in the near term. Even a transformative advance in computational power (and accompanyingalgorithms for efficiently tracking and storing the particle trajectories, energy states and so on) wouldstill leave the problem of initialization. It is unclear how this could be accomplished either manually orin an automated fashion.As a result of these challenges, mathematical models of biology typically make use of supra-molecular methods. That is, for the most part, forces and interactions are modeled as homogenousbulk effects that are the same for all instances of a object. Even stochastic methods, such as thoseemployed in Chapter 2 in which finite particles are tracked, still rely on descriptions of the forces andeffects at supramolecular scales. This sufficiently reduces the complexity of these models such thatthey generally are tractable in terms of both initialization and computational requirements.The cost of these simplifications is that these bulk effects are described by numerical parameters,and very often the models are incredibly sensitive to these parameters. Moreover, it is difficult to107appropriately assign values to these parameters that reflect the underlying biological process. This iswidely acknowledged, and is usually addressed by exploring the parameter space in an unbiased man-ner, thus producing predictions for a range of parameter values selected to correspond to the range ofpossible biological extremes. This is a valid approach, but it presupposes the existence of an actualbiological data set that the model can be compared to, so as to select the appropriate parameter rangebefore making further predictions. These sets in general are rare. Thus there is a gap between the pre-dictive models and the experimental biological implementation. This is the parameterization problem.That is, the amount of experimental work required to calibrate a predictive model by appropriately pa-rameterizing it is sufficient to often undermine the model’s predictive utility as an engineering tool. Arelated problem is that in many cases, the necessary experimental tools simply do not exist to generatea biological data set that correlates well with biophysical model parameters. Here it is important tonote that the effort required to generate such a data set may be well worth it in the context of scientificdiscovery.It is in resolving this parameterization problem that I think progress may be made. Improvementsin microscopy technology may provide a wealth of experimental, time resolved data which may greatlyincrease the accuracy of empirical parameters for biophysical models, and provide a broader rangeof validation data sets. Another fruitful approach may be a systematic effort to characterize a hostof biophysical parameters of certain model systems, for example a HeLa cell. This could providequantitatively accurate models that could be tentatively extended to other systems, and incrementallyadjusted as needed. Whatever approaches are used, the parameterization problem seems tractable, ifthere is a desire to pursue it.Finally, I leave open the possibility for a new, non-reductionist theory of biology that is not basedin mathematical physics or chemistry. This is not to say that such principles do not apply to biology,for they do. Rather, it seems at least possible that an alternative theory could be developed in parallel.It might be one better suited to the volume of constituent parts in a biological system, the multitudeof different time and length scales that are relevant, the high number of different classes of objects,and the staggering complexity and heterogeneity of those classes. In general, scientific theory is sodeeply rooted in some variation of reductionist mathematical physico-chemical approaches that it ishard to even conceive of what such a theory might look like. With no justification, I conjecture thatan information centric approach might be a fruitful lamppost under which to look. It has an associatedformalism, but is free from obvious physical or chemical constraints. However speculative this mayseem, it seems equally rigid to not at least admit the possibility that an alternative, rigorous theory ofbiology might exist.Of course, this but one vision of a way forward for cellular therapeutics. Perhaps both of thosechallenges are impossible. Perhaps they are unnecessary. New approaches may emerge, in which highthroughput screening methods will render both biological knowledge and predictive design unneces-sary. Or the payoff of cellular therapeutics will be sufficiently high that bespoke devices remain thenorm. Or the whole field will remain a boutique cottage industry. I think the last outcome unlikely.108The power of encoding logic in cells equipped with a variety of biological functions will be impactful,even if the eventual realization is currently unclear.5.4.4 The future of cellular therapeuticsAs the cell is the basic unit of life, it seems evident that the ideal therapeutic modality would directlyengage, cell-to-cell, with a diseased cellular target. Combined with every cell’s capacity for molecu-larly encoded logic, this one-to-one interaction would allow for real-time therapeutic decision-makingand target-cell discrimination at the site of active disease. Target cell surface molecules as well as en-vironmental variables such as acidity or oxygen content could all be considered in selecting if, when,and which target cells to treat. The ability of a cell to sequester a therapeutic molecule intracellu-larly while actively trafficking to disease sites for molecular delivery would maximize the therapeuticmolecule’s biological activity, and minimize its off-target activity. Finally, disorders not amenable tosmall molecule or biologic therapy might respond to cell-to-cell therapy: engagement of multiple sur-face receptors or elimination via phagocytosis are both possible. The age of cell-based therapeutics hasarrived, and their impact will continue to expand across medicine [177]. The array of functional parts,pathways, and cellular chassis already available within our cell repertoire represents an incredible re-source for building these devices, one that will not be matched by ground up synthesis for a very longtime. Instead, modifying and re-purposing biological systems will yield novel functional componentsand new waves of cell-based therapeutics.109Bibliography[1] Barton, F. B. et al. Improvement in Outcomes of Clinical Islet Transplantation: 1999-2010.Diabetes Care 35, 1436–1445 (2012). DOI 10.2337/dc12-0063. → pages[2] Farney, A. C., Sutherland, D. E. R. & Opara, E. C. Evolution of Islet Transplantation for theLast 30 Years. Pancreas 45, 8–20 (2016). DOI 10.1097/MPA.0000000000000391. → pages[3] Bruni, A., Gala-Lopez, B., Pepper, A. R., Abualhassan, N. S. & Shapiro, A. J. Islet celltransplantation for the treatment of type 1 diabetes: recent advances and future challenges.Diabetes, metabolic syndrome and obesity : targets and therapy 7, 211–23 (2014). DOI10.2147/DMSO.S50789. → pages[4] Kimbrel, E. A. & Lanza, R. Current status of pluripotent stem cells: moving the first therapiesto the clinic. Nature Reviews Drug Discovery 14, 681–692 (2015). DOI 10.1038/nrd4738. →pages[5] Takahashi, K. & Yamanaka, S. Induction of pluripotent stem cells from mouse embryonic andadult fibroblast cultures by defined factors. Cell 126, 663–76 (2006). DOI10.1016/j.cell.2006.07.024. → pages[6] Wu, S. M. & Hochedlinger, K. Harnessing the potential of induced pluripotent stem cells forregenerative medicine. Nature Cell Biology 13, 497–505 (2011). DOI 10.1038/ncb0511-497.→ pages[7] Takebe, T. et al. Vascularized and functional human liver from an iPSC-derived organ budtransplant. Nature 499, 481–484 (2013). DOI 10.1038/nature12271. → pages[8] Lu, T.-Y. et al. Repopulation of decellularized mouse heart with human induced pluripotentstem cell-derived cardiovascular progenitor cells. Nature communications 4, 2307 (2013). DOI10.1038/ncomms3307. → pages[9] Tsukamoto, A. et al. Challenging Regeneration to Transform Medicine. STEM CELLSTranslational Medicine 5, 1–7 (2016). DOI 10.5966/sctm.2015-0180. → pages[10] Naldini, L. Gene therapy returns to centre stage. Nature 526, 351–360 (2015). DOI10.1038/nature15818. → pages[11] Little, M.-T. & Storb, R. History of haematopoietic stem-cell transplantation. Nature ReviewsCancer 2, 231–238 (2002). DOI 10.1038/nrc748. → pages[12] Kaufmann, K. B., Büning, H., Galy, A., Schambach, A. & Grez, M. Gene therapy on the move.EMBO Molecular Medicine 5, 1642–1661 (2013). DOI 10.1002/emmm.201202287. → pages110[13] Hacein-Bey-Abina, S. et al. Sustained Correction of X-Linked Severe CombinedImmunodeficiency by ex Vivo Gene Therapy. New England Journal of Medicine 346,1185–1193 (2002). DOI 10.1056/NEJMoa012616. → pages[14] Hacein-Bey-Abina, S. et al. Insertional oncogenesis in 4 patients after retrovirus-mediated genetherapy of SCID-X1. The Journal of Clinical Investigation 118 (2008). DOI10.1172/JCI35700.3132. → pages[15] Biffi, A. et al. Lentiviral Hematopoietic Stem Cell Gene Therapy Benefits MetachromaticLeukodystrophy. Science 341 (2013). → pages[16] Cavazzana-Calvo, M. et al. Transfusion independence and HMGA2 activation after genetherapy of human β -thalassaemia. Nature 467, 318–322 (2010). DOI 10.1038/nature09328. →pages[17] Laity, J. H., Lee, B. M. & Wright, P. E. Zinc finger proteins: new insights into structural andfunctional diversity. Current opinion in structural biology 11, 39–46 (2001). → pages[18] Urnov, F. D., Rebar, E. J., Holmes, M. C., Zhang, H. S. & Gregory, P. D. Genome editing withengineered zinc finger nucleases. Nature Reviews Genetics 11, 636–646 (2010). DOI10.1038/nrg2842. → pages[19] Joung, J. K. & Sander, J. D. TALENs: a widely applicable technology for targeted genomeediting. Nature reviews. Molecular cell biology 14, 49–55 (2013). DOI 10.1038/nrm3486. →pages[20] Sander, J. D. & Joung, J. K. CRISPR-Cas systems for editing, regulating and targetinggenomes. Nature biotechnology 32, 347–55 (2014). DOI 10.1038/nbt.2842. → pages[21] Hsu, P. D., Lander, E. S. & Zhang, F. Development and Applications of CRISPR-Cas9 forGenome Engineering. (2014). NIHMS150003. → pages[22] Tsai, S. Q. & Joung, J. K. Defining and improving the genome-wide specificities ofCRISPRâA˘S¸Cas9 nucleases. Nature Reviews Genetics 17, 300–312 (2016). DOI10.1038/nrg.2016.28. → pages[23] Chapman, M., Warren, E. H. & Wu, C. J. Applications of next-generation sequencing to bloodand marrow transplantation. Biology of blood and marrow transplantation : journal of theAmerican Society for Blood and Marrow Transplantation 18, S151–60 (2012). DOI10.1016/j.bbmt.2011.11.011. → pages[24] Lieber, M. R. The mechanism of double-strand DNA break repair by the nonhomologous DNAend-joining pathway. Annual review of biochemistry 79, 181–211 (2010). DOI10.1146/annurev.biochem.052308.093131. → pages[25] Liang, F., Han, M., Romanienko, P. J. & Jasin, M. Homology-directed repair is a majordouble-strand break repair pathway in mammalian cells. Proceedings of the National Academyof Sciences of the United States of America 95, 5172–7 (1998). → pages[26] Cox, D. B. T., Platt, R. J. & Zhang, F. Therapeutic genome editing: prospects and challenges.Nature medicine 21, 121–31 (2015). DOI 10.1038/nm.3793. → pages111[27] Lombardo, A. et al. Site-specific integration and tailoring of cassette design for sustainablegene transfer. Nature Methods 8, 861–869 (2011). DOI 10.1038/nmeth.1674. → pages[28] Genovese, P. et al. Targeted genome editing in human repopulating haematopoietic stem cells.Nature 510, 235–240 (2014). DOI 10.1038/nature13420. → pages[29] Tebas, P. et al. Gene Editing of <i>CCR5</i> in Autologous CD4 T Cells of Persons Infectedwith HIV. New England Journal of Medicine 370, 901–910 (2014). DOI10.1056/NEJMoa1300662. → pages[30] Fellmann, C., Gowen, B. G., Lin, P.-C., Doudna, J. A. & Corn, J. E. Cornerstones ofCRISPRâA˘S¸Cas in drug discovery and therapy. Nature Reviews Drug Discovery 16, 89–100(2016). DOI 10.1038/nrd.2016.238. → pages[31] Tsai, S. Q. et al. GUIDE-seq enables genome-wide profiling of off-target cleavage byCRISPR-Cas nucleases. Nature Biotechnology 33, 187–197 (2014). DOI 10.1038/nbt.3117. →pages[32] Yang, L. et al. Targeted and genome-wide sequencing reveal single nucleotide variationsimpacting specificity of Cas9 in human stem cells. Nature Communications 5, 5507 (2014).DOI 10.1038/ncomms6507. → pages[33] Cho, S. W. et al. Analysis of off-target effects of CRISPR/Cas-derived RNA-guidedendonucleases and nickases. Genome research 24, 132–41 (2014). DOI10.1101/gr.162339.113. → pages[34] Wei, X. et al. Mesenchymal stem cells: a new trend for cell therapy. Acta PharmacologicaSinica 34, 747–754 (2013). DOI 10.1038/aps.2013.50. → pages[35] Squillaro, T., Peluso, G. & Galderisi, U. Clinical Trials With Mesenchymal Stem Cells: AnUpdate. Cell Transplantation 25, 829–848 (2016). DOI 10.3727/096368915X689622. → pages[36] Cheng, Z. et al. Targeted Migration of Mesenchymal Stem Cells Modified With CXCR4 Geneto Infarcted Myocardium Improves Cardiac Performance. Molecular Therapy 16, 571–579(2008). DOI 10.1038/sj.mt.6300374. → pages[37] Suresh, S. C. et al. Thioredoxin-1 (Trx1) engineered mesenchymal stem cell therapy increasedpro-angiogenic factors, reduced fibrosis and improved heart function in the infarcted ratmyocardium. International Journal of Cardiology 201, 517–528 (2015). DOI10.1016/j.ijcard.2015.08.117. → pages[38] van Velthoven, C. T., Braccioli, L., Willemen, H. L., Kavelaars, A. & Heijnen, C. J. TherapeuticPotential of Genetically Modified Mesenchymal Stem Cells After Neonatal Hypoxic-IschemicBrain Damage. Molecular Therapy 22, 645–654 (2014). DOI 10.1038/mt.2013.260. → pages[39] Niess, H. et al. Selective Targeting of Genetically Engineered Mesenchymal Stem Cells toTumor Stroma Microenvironments Using Tissue-Specific Suicide Gene Expression SuppressesGrowth of Hepatocellular Carcinoma. Annals of Surgery 254, 767–775 (2011). DOI10.1097/SLA.0b013e3182368c4f. → pages112[40] Sage, E. K. et al. Systemic but not topical TRAIL-expressing mesenchymal stem cells reducetumour growth in malignant mesothelioma. Thorax 69, 638–647 (2014). DOI10.1136/thoraxjnl-2013-204110. → pages[41] Sage, E. K., Thakrar, R. M. & Janes, S. M. Genetically modified mesenchymal stromal cells incancer therapy. Cytotherapy 18, 1435–1445 (2016). DOI 10.1016/j.jcyt.2016.09.003. → pages[42] Corthay, A. Does the immune system naturally protect against cancer? Frontiers inimmunology 5, 197 (2014). DOI 10.3389/fimmu.2014.00197. → pages[43] Cramer, D. W. & Finn, O. J. Epidemiologic perspective on immune-surveillance in cancer.Current opinion in immunology 23, 265–71 (2011). DOI 10.1016/j.coi.2011.01.002. → pages[44] Muul, L. M., Spiess, P. J., Director, E. P. & Rosenberg, S. A. Identification of specific cytolyticimmune responses against autologous tumor in humans bearing malignant melanoma. Journalof immunology (Baltimore, Md. : 1950) 138, 989–95 (1987). → pages[45] Rosenberg, S. A. et al. Use of Tumor-Infiltrating Lymphocytes and Interleukin-2 in theImmunotherapy of Patients with Metastatic Melanoma. New England Journal of Medicine 319,1676–1680 (1988). DOI 10.1056/NEJM198812223192527. → pages[46] Rosenberg, S. A. et al. Durable Complete Responses in Heavily Pretreated Patients withMetastatic Melanoma Using T-Cell Transfer Immunotherapy. Clinical Cancer Research 17,4550–4557 (2011). DOI 10.1158/1078-0432.CCR-11-0116. → pages[47] Brown, S. D. et al. Neo-antigens predicted by tumor genome meta-analysis correlate withincreased patient survival. Genome research 24, 743–50 (2014). DOI 10.1101/gr.165985.113.→ pages[48] Rooney, M. S., Shukla, S. A., Wu, C. J., Getz, G. & Hacohen, N. Molecular and geneticproperties of tumors associated with local immune cytolytic activity. Cell 160, 48–61 (2015).DOI 10.1016/j.cell.2014.12.033. → pages[49] June, C. H. Principles of adoptive T cell cancer therapy. The Journal of Clinical Investigation117 (2007). DOI 10.1172/JCI31446.1204. → pages[50] Dudley, M. E., Wunderlich, J. R., Shelton, T. E., Even, J. & Rosenberg, S. A. Generation oftumor-infiltrating lymphocyte cultures for use in adoptive transfer therapy for melanomapatients. Journal of immunotherapy 26, 332–42 (2003). → pages[51] Jiang, Y., Li, Y. & Zhu, B. T-cell exhaustion in the tumor microenvironment. Cell Death andDisease 6, e1792 (2015). DOI 10.1038/cddis.2015.162. → pages[52] Staveley-O’Carroll, K. et al. Induction of antigen-specific T cell anergy: An early event in thecourse of tumor progression. Proceedings of the National Academy of Sciences of the UnitedStates of America 95, 1178–83 (1998). → pages[53] Ahmadzadeh, M. et al. Tumor antigen-specific CD8 T cells infiltrating the tumor express highlevels of PD-1 and are functionally impaired. Blood 114, 1537–44 (2009). DOI10.1182/blood-2008-12-195792. → pages113[54] Johnson, L. A. & June, C. H. Driving gene-engineered T cell immunotherapy of cancer. CellResearch 27, 38–58 (2017). DOI 10.1038/cr.2016.154. → pages[55] Gros, A. et al. Prospective identification of neoantigen-specific lymphocytes in the peripheralblood of melanoma patients. Nature Medicine 22, 433–438 (2016). DOI 10.1038/nm.4051. →pages[56] Morgan, R. A. et al. Cancer regression in patients after transfer of genetically engineeredlymphocytes. Science (New York, N.Y.) 314, 126–9 (2006). DOI 10.1126/science.1129003. →pages[57] Sharma, G. & Holt, R. A. T-cell epitope discovery technologies. Human immunology 75,514–9 (2014). DOI 10.1016/j.humimm.2014.03.003. → pages[58] Hicklin, D. J., Marincola, F. M. & Ferrone, S. HLA class I antigen downregulation in humancancers: T-cell immunotherapy revives an old story. Molecular medicine today 5, 178–86(1999). → pages[59] Bendle, G. M. et al. Lethal graft-versus-host disease in mouse models of T cell receptor genetherapy. Nature medicine 16, 565–70, 1p following 570 (2010). DOI 10.1038/nm.2128. →pages[60] Rosenberg, S. A. Of Mice, Not Men: No Evidence for Graft-versus-Host Disease in HumansReceiving T-Cell ReceptorâA˘S¸Transduced Autologous T Cells. Molecular Therapy 18,1744–1745 (2010). DOI 10.1038/mt.2010.195. → pages[61] Robbins, P. F. et al. A Pilot Trial Using Lymphocytes Genetically Engineered with anNY-ESO-1-Reactive T-cell Receptor: Long-term Follow-up and Correlates with Response.Clinical Cancer Research 21, 1019–1027 (2015). DOI 10.1158/1078-0432.CCR-14-2708. →pages[62] Barrett, D. M., Singh, N., Porter, D. L., Grupp, S. a. & June, C. H. Chimeric antigen receptortherapy for cancer. Annual review of medicine 65, 333–47 (2014). DOI10.1146/annurev-med-060512-150254. → pages[63] Ahmad, Z. A. et al. scFv Antibody: Principles and Clinical Application. Clinical andDevelopmental Immunology 2012, 1–15 (2012). DOI 10.1155/2012/980250. → pages[64] Dotti, G., Gottschalk, S., Savoldo, B. & Brenner, M. K. Design and development of therapiesusing chimeric antigen receptor-expressing T cells. Immunological reviews 257, 107–26(2014). DOI 10.1111/imr.12131. → pages[65] Gross, G., Waks, T. & Eshhar, Z. Expression of immunoglobulin-T-cell receptor chimericmolecules as functional receptors with antibody-type specificity. Proceedings of the NationalAcademy of Sciences of the United States of America 86, 10024–8 (1989). → pages[66] Sadelain, M., Brentjens, R. & Rivière, I. The Basic Principles of Chimeric Antigen ReceptorDesign. Cancer Discovery 3, 388–398 (2013). DOI 10.1158/2159-8290.CD-12-0548. → pages[67] Jackson, H. J., Rafiq, S. & Brentjens, R. J. Driving CAR T-cells forward. Nature ReviewsClinical Oncology 13, 370–383 (2016). DOI 10.1038/nrclinonc.2016.36. → pages114[68] Vigneron, N. & Nathalie. Human Tumor Antigens and Cancer Immunotherapy. BioMedResearch International 2015, 1–17 (2015). DOI 10.1155/2015/948501. → pages[69] Hulsmeyer, M. et al. A Major Histocompatibility Complex{middle dot}Peptide-restrictedAntibody and T Cell Receptor Molecules Recognize Their Target by Distinct Binding Modes:CRYSTAL STRUCTURE OF HUMAN LEUKOCYTE ANTIGEN (HLA)-A1{middledot}MAGE-A1 IN COMPLEX WITH FAB-HYB3. Journal of Biological Chemistry 280,2972–2980 (2005). DOI 10.1074/jbc.M411323200. → pages[70] Maus, M. V. et al. An MHC-restricted antibody-based chimeric antigen receptor requiresTCR-like affinity to maintain antigen specificity. Molecular Therapy - Oncolytics 3, 16023(2016). DOI 10.1038/mto.2016.23. → pages[71] Dai, H., Wang, Y., Lu, X. & Han, W. Chimeric Antigen Receptors Modified T-Cells for CancerTherapy. Journal of the National Cancer Institute 108, djv439 (2016). DOI10.1093/jnci/djv439. → pages[72] Maude, S. L., Teachey, D. T., Porter, D. L. & Grupp, S. A. CD19-targeted chimeric antigenreceptor T-cell therapy for acute lymphoblastic leukemia. Blood 125, 4017–4023 (2015). DOI10.1182/blood-2014-12-580068. → pages[73] Zhang, T. et al. Efficiency of CD19 chimeric antigen receptor-modified T cells for treatment ofB cell malignancies in phase I clinical trials: a meta-analysis. Oncotarget 6, 33961–71 (2015).DOI 10.18632/oncotarget.5582. → pages[74] Maude, S. L., Teachey, D. T., Porter, D. L. & Grupp, S. A. CD19-targeted chimeric antigenreceptor T-cell therapy for acute lymphoblastic leukemia. Blood 125 (2015). → pages[75] Riddell, S. R. et al. Adoptive therapy with chimeric antigen receptor-modified T cells ofdefined subset composition. Cancer journal (Sudbury, Mass.) 20, 141–4 (2014). DOI10.1097/PPO.0000000000000036. → pages[76] Turtle, C. J. et al. CD19 CARâA˘S¸T cells of defined CD4+:CD8+ composition in adult B cellALL patients. Journal of Clinical Investigation 126, 2123–2138 (2016). DOI10.1172/JCI85309. → pages[77] Sommermeyer, D. et al. Chimeric antigen receptor-modified T cells derived from definedCD8+ and CD4+ subsets confer superior antitumor reactivity in vivo. Leukemia 30, 492–500(2016). DOI 10.1038/leu.2015.247. → pages[78] Gattinoni, L. et al. A human memory T cell subset with stem cellâA˘S¸like properties. NatureMedicine 17, 1290–1297 (2011). DOI 10.1038/nm.2446. → pages[79] Maude, S. L. et al. Chimeric Antigen Receptor T Cells for Sustained Remissions in Leukemia.New England Journal of Medicine 371, 1507–1517 (2014). DOI 10.1056/NEJMoa1407222. →pages[80] Davila, M. L. et al. Efficacy and toxicity management of 19-28z CAR T cell therapy in B cellacute lymphoblastic leukemia. Science translational medicine 6, 224ra25 (2014). DOI10.1126/scitranslmed.3008226. → pages115[81] Lee, D. W. et al. T cells expressing CD19 chimeric antigen receptors for acute lymphoblasticleukaemia in children and young adults: a phase 1 dose-escalation trial. The Lancet 385,517–528 (2015). DOI 10.1016/S0140-6736(14)61403-3. → pages[82] Hinrichs, C. S. & Restifo, N. P. Reassessing target antigens for adoptive T-cell therapy. NatureBiotechnology 31, 999–1008 (2013). DOI 10.1038/nbt.2725. → pages[83] Newick, K., O’Brien, S., Moon, E. & Albelda, S. M. CAR T Cell Therapy for Solid Tumors.Annual review of medicine 1–14 (2016). DOI 10.1146/annurev-med-062315-120245. → pages[84] Bonifant, C. L., Jackson, H. J., Brentjens, R. J. & Curran, K. J. Toxicity and management inCAR T-cell therapy. Molecular Therapy - Oncolytics 3, 16011 (2016). DOI10.1038/mto.2016.11. → pages[85] Harlin, H. et al. Chemokine Expression in Melanoma Metastases Associated with CD8+ T-CellRecruitment. Cancer Research 69, 3077–3085 (2009). DOI10.1158/0008-5472.CAN-08-2281. → pages[86] Wang, L.-C. S. et al. Targeting Fibroblast Activation Protein in Tumor Stroma with ChimericAntigen Receptor T Cells Can Inhibit Tumor Growth and Augment Host Immunity withoutSevere Toxicity. Cancer Immunology Research 2, 154–166 (2014). DOI10.1158/2326-6066.CIR-13-0027. → pages[87] Caruana, I. et al. Heparanase promotes tumor infiltration and antitumor activity ofCAR-redirected T lymphocytes. Nature Medicine 21, 524–529 (2015). DOI 10.1038/nm.3833.→ pages[88] Hatfield, S. M. et al. Immunological mechanisms of the antitumor effects of supplementaloxygenation. Science Translational Medicine 7, 277ra30–277ra30 (2015). DOI10.1126/scitranslmed.aaa1260. → pages[89] Fischer, K. et al. Inhibitory effect of tumor cell-derived lactic acid on human T cells. Blood109, 3812–3819 (2007). DOI 10.1182/blood-2006-07-035972. → pages[90] Jacobs, S. R. et al. Glucose uptake is limiting in T cell activation and requires CD28-mediatedAkt-dependent and independent pathways. Journal of immunology (Baltimore, Md. : 1950)180, 4476–86 (2008). → pages[91] Ninomiya, S. et al. Tumor indoleamine 2,3-dioxygenase (IDO) inhibits CD19-CAR T cells andis downregulated by lymphodepleting drugs. Blood 125, 3905–3916 (2015). DOI10.1182/blood-2015-01-621474. → pages[92] Yong, C. S. M. et al. CAR T cell therapy of solid tumors. Immunology and Cell Biology 1–8(2016). DOI 10.1038/icb.2016.128. → pages[93] Vignali, D. A. A., Collison, L. W. & Workman, C. J. How regulatory T cells work. Naturereviews. Immunology 8, 523–32 (2008). DOI 10.1038/nri2343. → pages[94] Zhou, Q. et al. Program death-1 signaling and regulatory T cells collaborate to resist thefunction of adoptively transferred cytotoxic T lymphocytes in advanced acute myeloidleukemia. Blood 116, 2484–2493 (2010). DOI 10.1182/blood-2010-03-275446. → pages116[95] Liu, Y. et al. Inhibition of p300 impairs Foxp3+ T regulatory cell function and promotesantitumor immunity. Nature Medicine 19, 1173–1177 (2013). DOI 10.1038/nm.3286. → pages[96] Muenst, S. et al. The immune system and cancer evasion strategies: therapeutic concepts.Journal of Internal Medicine 279, 541–562 (2016). DOI 10.1111/joim.12470. → pages[97] Whiteside, T. L. Immune suppression in cancer: Effects on immune cells, mechanisms andfuture therapeutic intervention. Seminars in Cancer Biology 16, 3–15 (2006). DOI10.1016/j.semcancer.2005.07.008. → pages[98] Moon, E. K. et al. Multifactorial T-cell Hypofunction That Is Reversible Can Limit the Efficacyof Chimeric Antigen Receptor-Transduced Human T cells in Solid Tumors. Clinical CancerResearch 20, 4262–4273 (2014). DOI 10.1158/1078-0432.CCR-13-2627. → pages[99] Pardoll, D. M. The blockade of immune checkpoints in cancer immunotherapy. NatureReviews Cancer 12, 252–264 (2012). DOI 10.1038/nrc3239. → pages[100] Ciceri, F. et al. Infusion of suicide-gene-engineered donor lymphocytes after familyhaploidentical haemopoietic stem-cell transplantation for leukaemia (the TK007 trial): anon-randomised phase I-II study. The Lancet. Oncology 10, 489–500 (2009). DOI10.1016/S1470-2045(09)70074-9. → pages[101] Di Stasi, A. et al. Inducible apoptosis as a safety switch for adoptive cell therapy. The NewEngland journal of medicine 365, 1673–83 (2011). DOI 10.1056/NEJMoa1106152. → pages[102] Philip, B. et al. A highly compact epitope-based marker/suicide gene for easier and safer T-celltherapy. Blood 124 (2014). → pages[103] Wu, C.-Y., Roybal, K. T., Puchner, E. M., Onuffer, J. & Lim, W. A. Remote control oftherapeutic T cells through a small molecule-gated chimeric receptor. Science (New York, N.Y.)350, aab4077 (2015). DOI 10.1126/science.aab4077. → pages[104] Khalil, D. N., Smith, E. L., Brentjens, R. J. & Wolchok, J. D. The future of cancer treatment:immunomodulation, CARs and combination immunotherapy. Nature Reviews ClinicalOncology 13, 273–290 (2016). DOI 10.1038/nrclinonc.2016.25. → pages[105] Newick, K., Moon, E. & Albelda, S. M. Chimeric antigen receptor T-cell therapy for solidtumors. Molecular Therapy âA˘Tˇ Oncolytics 3, 16006 (2016). DOI 10.1038/mto.2016.6. →pages[106] Qasim, W. et al. Molecular remission of infant B-ALL after infusion of universal TALENgene-edited CAR T cells. Science Translational Medicine 9, eaaj2013 (2017). DOI10.1126/scitranslmed.aaj2013. → pages[107] Ren, J. et al. Multiplex genome editing to generate universal CAR T cells resistant to PD1inhibition. Clinical Cancer Research (2016). → pages[108] Themeli, M. et al. Generation of tumor-targeted human T lymphocytes from inducedpluripotent stem cells for cancer therapy. Nature Biotechnology (2013). DOI 10.1038/nbt.2678.→ pages117[109] Kloss, C. C., Condomines, M., Cartellieri, M., Bachmann, M. & Sadelain, M. Combinatorialantigen recognition with balanced signaling promotes selective tumor eradication byengineered T cells. Nature biotechnology 31, 71–5 (2013). DOI 10.1038/nbt.2459. → pages[110] Roybal, K. T. et al. Precision Tumor Recognition by T Cells With CombinatorialAntigen-Sensing Circuits. Cell 164, 770–779 (2016). DOI 10.1016/j.cell.2016.01.011. →pages[111] Morsut, L. et al. Engineering Customized Cell Sensing and Response Behaviors UsingSynthetic Notch Receptors. Cell 164, 780–791 (2016). DOI 10.1016/j.cell.2016.01.012. →pages[112] Fedorov, V. D., Themeli, M. & Sadelain, M. PD-1- and CTLA-4-based inhibitory chimericantigen receptors (iCARs) divert off-target immunotherapy responses. Science translationalmedicine 5, 215ra172 (2013). DOI 10.1126/scitranslmed.3006597. → pages[113] Xie, Z., Wroblewska, L., Prochazka, L., Weiss, R. & Benenson, Y. Multi-input RNAi-basedlogic circuit for identification of specific cancer cells. Science (New York, N.Y.) 333, 1307–11(2011). DOI 10.1126/science.1205527. → pages[114] Bojar, D. & Fussenegger, M. The best of both worlds: reaping the benefits from mammalianand bacterial therapeutic circuits. Current Opinion in Chemical Biology 34, 11–19 (2016). DOI10.1016/j.cbpa.2016.05.012. → pages[115] Kemmer, C. et al. Self-sufficient control of urate homeostasis in mice by a synthetic circuit.Nature Biotechnology 28, 355–60 (2010). DOI 10.1038/nbt.1617. → pages[116] Schukur, L., Geering, B., Charpin-El Hamri, G. & Fussenegger, M. Implantable syntheticcytokine converter cells with AND-gate logic treat experimental psoriasis. ScienceTranslational Medicine 7, 318ra201–318ra201 (2015). DOI 10.1126/scitranslmed.aac4964. →pages[117] Ye, H., Daoud-El Baba, M., Peng, R.-W. & Fussenegger, M. A synthetic optogenetictranscription device enhances blood-glucose homeostasis in mice. Science (New York, N.Y.)332, 1565–8 (2011). DOI 10.1126/science.1203535. → pages[118] Ye, H. et al. Pharmaceutically controlled designer circuit for the treatment of the metabolicsyndrome. Proceedings of the National Academy of Sciences of the United States of America110, 141–6 (2013). DOI 10.1073/pnas.1216801110. → pages[119] Saxena, P., Charpin-El Hamri, G., Folcher, M., Zulewski, H. & Fussenegger, M. Synthetic genenetwork restoring endogenous pituitary-thyroid feedback control in experimental Graves’disease. Proceedings of the National Academy of Sciences of the United States of America 113,1244–9 (2016). DOI 10.1073/pnas.1514383113. → pages[120] Orive, G. et al. Cell encapsulation: Promise and progress. Nature Medicine 9, 104–107 (2003).DOI 10.1038/nm0103-104. → pages[121] Barry, M. & Bleackley, R. C. Cytotoxic T lymphocytes: all roads lead to death. Nature reviews.Immunology 2, 401–9 (2002). DOI 10.1038/nri819. → pages118[122] Murphy, K., Travers, P. & Walport, M. Janeway’s Immunobiology (Garland Science, NewYork, 2007), 7th edn. → pages[123] Caligiuri, M. A. Human natural killer cells. Blood 112, 461–9 (2008). DOI10.1182/blood-2007-09-077438. → pages[124] Lanier, L. L. Up on the tightrope: natural killer cell activation and inhibition. Natureimmunology 9, 495–502 (2008). DOI 10.1038/ni1581. → pages[125] Kruschinski, A. et al. Engineering antigen-specific primary human NK cells against HER-2positive carcinomas. Proceedings of the National Academy of Sciences of the United States ofAmerica 105, 17481–6 (2008). DOI 10.1073/pnas.0804788105. → pages[126] Vivier, E., Ugolini, S., Blaise, D., Chabannon, C. & Brossay, L. Targeting natural killer cellsand natural killer T cells in cancer. Nature reviews. Immunology 12, 239–52 (2012). DOI10.1038/nri3174. → pages[127] Chowdhury, D. & Lieberman, J. Death by a thousand cuts: granzyme pathways of programmedcell death. Annual review of immunology 26, 389–420 (2008). DOI10.1146/annurev.immunol.26.021607.090404. → pages[128] Estébanez-Perpiña, E. et al. Crystal structure of the caspase activator human granzyme B, aproteinase highly specific for an Asp-P1 residue. Biological chemistry 381, 1203–14 (2000).DOI 10.1515/BC.2000.148. → pages[129] de Saint Basile, G., Ménasché, G. & Fischer, A. Molecular mechanisms of biogenesis andexocytosis of cytotoxic granules. Nature reviews. Immunology 10, 568–79 (2010). DOI10.1038/nri2803. → pages[130] Vivier, E., Nunès, J. A. & Vély, F. Natural Killer Cell Signaling Pathways. Science 306,1517–1519 (2004). DOI 10.1126/science.1103478. → pages[131] Krzewski, K. & Coligan, J. E. Human NK cell lytic granules and regulation of their exocytosis.Frontiers in Immunology 3, 335 (2012). DOI 10.3389/fimmu.2012.00335. → pages[132] Rudolph, M. G., Stanfield, R. L. & Wilson, I. a. How TCRs bind MHCs, peptides, andcoreceptors. Annual review of immunology 24, 419–66 (2006). DOI10.1146/annurev.immunol.23.021704.115658. → pages[133] Dustin, M. L. & Long, E. O. Cytotoxic immunological synapses. Immunological reviews 235,24–34 (2010). → pages[134] Brownlie, R. J. & Zamoyska, R. T cell receptor signalling networks: branched, diversified andbounded. Nature Reviews Immunology 13, 257–269 (2013). DOI 10.1038/nri3403. → pages[135] Trapani, J. A. & Smyth, M. J. Functional significance of the perforin/granzyme cell deathpathway. Nature reviews. Immunology 2, 735–47 (2002). DOI 10.1038/nri911. → pages[136] Lopez, J. A. et al. Perforin forms transient pores on the target cell plasma membrane tofacilitate rapid access of granzymes during killer cell attack. Blood 2659–2668 (2013). DOI10.1182/blood-2012-07-446146. → pages119[137] Law, R. H. P. et al. The structural basis for membrane binding and pore formation bylymphocyte perforin. Nature 468, 447–51 (2010). DOI 10.1038/nature09518. → pages[138] Mouchacca, P., Schmitt-Verhulst, A.-M. & Boyer, C. Visualization of cytolytic T celldifferentiation and granule exocytosis with T cells from mice expressing active fluorescentgranzyme B. PloS one 8, e67239 (2013). DOI 10.1371/journal.pone.0067239. → pages[139] Thiery, J. et al. Perforin pores in the endosomal membrane trigger the release of endocytosedgranzyme B into the cytosol of target cells. Nature immunology 12, 770–7 (2011). DOI10.1038/ni.2050. → pages[140] Ormö, M. et al. Crystal structure of the Aequorea victoria green fluorescent protein. Science273, 1392–5 (1996). → pages[141] Mason-Osann, E., Hollevoet, K., Niederfellner, G., Pastan, I. & Sun, X. Quantification ofrecombinant immunotoxin delivery to solid tumors allows for direct comparison of in vivo andin vitro results. Scientific Reports 5, 10832 (2015). DOI 10.1038/srep10832. → pages[142] Woodsworth, D. J., Dunsing, V. & Coombs, D. Design Parameters for Granzyme-MediatedCytotoxic Lymphocyte Target-Cell Killing and Specificit. Biophysical journal 109, 477–488(2015). → pages[143] Voskoboinik, I., Smyth, M. J. & Trapani, J. A. Perforin-mediated target-cell death and immunehomeostasis. Nat. Rev. Immunol. 6, 940–52 (2006). DOI 10.1038/nri1983. → pages[144] Lieberman, J. Granzyme A activates another way to die. Immunological reviews 235, 93–104(2010). DOI 10.1111/j.0105-2896.2010.00902.x. → pages[145] Coombs, D. & Goldstein, B. Effects of the geometry of the immunological synapse on thedelivery of effector molecules. Biophysical journal 87, 2215–20 (2004). DOI10.1529/biophysj.104.045674. → pages[146] Grakoui, A. et al. The immunological synapse: A molecular machine controlling T cellactivation. Science 285, 221–227 (1999). → pages[147] Waugh, S. M., Harris, J. L., Fletterick, R. & Craik, C. S. The structure of the pro-apoptoticprotease granzyme B reveals the molecular determinants of its specificity. Nature structuralbiology 7, 762–765 (2000). → pages[148] Burgoyne, R. D. & Morgan, A. Secretory granule exocytosis. Physiol Rev 83, 581–632 (2003).DOI 10.1152/physrev.00031.2002. → pages[149] Ishiura, S. et al. Calcium is essential for both the membrane binding and lytic activity ofpore-forming protein (perforin) from cytotoxic T-lymphocyte. Molecular immunology 27,803–7 (1990). → pages[150] Lopez, J. a. et al. Rapid and unidirectional perforin pore delivery at the cytotoxic immunesynapse. Journal of immunology 191, 2328–34 (2013). DOI 10.4049/jimmunol.1301205. →pages120[151] Brown, A. C. N. et al. Remodelling of cortical actin where lytic granules dock at natural killercell immune synapses revealed by super-resolution microscopy. PLoS biology 9, e1001152(2011). DOI 10.1371/journal.pbio.1001152. → pages[152] Stinchcombe, J. C., Bossi, G., Booth, S. & Griffiths, G. M. The immunological synapse of CTLcontains a secretory domain and membrane bridges. Immunity 15, 751–61 (2001). → pages[153] Elf, J., Don, A. & Ehrenberg, M. Mesoscopic reaction-diffusion in intracellular signaling.SPIE: Fluctuations and noise in Biological, Biophysical and Biomedical Systems 5110,114–125 (2003). → pages[154] Elf, J. & Ehrenberg, M. Spontaneous separation of bi-stable biochemical systems into spatialdomains of opposite phases. Systems biology. 1, 230–236 (2004). DOI 10.1049/sb. → pages[155] Gibson, M. a. & Bruck, J. Efficient Exact Stochastic Simulation of Chemical Systems withMany Species and Many Channels. The Journal of Physical Chemistry A 104, 1876–1889(2000). DOI 10.1021/jp993732q. → pages[156] Gillespie, D. T. A general method for numerically simulating the stochastic time evolution ofcoupled chemical reactions. Journal of Computational Physics 22, 403–434 (1976). DOI10.1016/0021-9991(76)90041-3. → pages[157] Saffman, P. G. & Delbruck, M. Brownian motion in biological membranes. Proceedings of theNational Academy of Sciences 72, 3111–3113 (1975). DOI 10.1073/pnas.72.8.3111. → pages[158] Gambin, Y. et al. Lateral mobility of proteins in liquid membranes revisited. Proceedings of theNational Academy of Sciences of the United States of America 103, 2098–102 (2006). DOI10.1073/pnas.0511026103. → pages[159] Ramadurai, S. et al. Lateral diffusion of membrane proteins. Journal of the AmericanChemical Society 131, 12650–6 (2009). DOI 10.1021/ja902853g. → pages[160] Lauffenberger, D. A. & Linderman, J. J. Receptors. Models for binding, trafficking andsignaling (Oxford University Press, 1993). → pages[161] Jaspard, F., Nadi, M. & Rouane, A. Dielectric properties of blood: an investigation ofhaematocrit dependence. Physiol. Meas. 24, 137–147 (2003). → pages[162] Maher, K. J., Klimas, N. G., Hurwitz, B., Fletcher, M. A. & Schiff, R. QuantitativeFluorescence Measures for Determination of Intracellular Perforin Content QuantitativeFluorescence Measures for Determination of Intracellular Perforin Content. Clinical andVaccine Immunology 9, 1248–1252 (2002). DOI 10.1128/CDLI.9.6.1248. → pages[163] Kelso, A. et al. The genes for perforin , granzymes A-C and IFN-g are differentially expressedin single CD8 + T cells during primary activation. International Immunology 14 (2002). →pages[164] Maier, T., Güell, M. & Serrano, L. Correlation of mRNA and protein in complex biologicalsamples. FEBS letters 583, 3966–73 (2009). DOI 10.1016/j.febslet.2009.10.036. → pages121[165] Fischer, H., Polikarpov, I. & Craievich, A. F. Average protein density is amolecular-weight-dependent function 2825–2828 (2004). DOI 10.1110/ps.04688204.studies.→ pages[166] Ellis, R. J. & Minton, A. P. Cell biology: join the crowd. Nature 425, 27–8 (2003). DOI10.1038/425027a. → pages[167] Ozbabacan, S. E. A., Engin, H. B., Gursoy, A. & Keskin, O. Transient protein-proteininteractions. Protein engineering, design & selection : PEDS 24, 635–48 (2011). DOI10.1093/protein/gzr025. → pages[168] Tang, J. et al. Using two fluorescent probes to dissect the binding, insertion, and dimerizationkinetics of a model membrane peptide. Journal of the American Chemical Society 131, 3816–7(2009). DOI 10.1021/ja809007f. → pages[169] Pipkin, M. E. & Lieberman, J. Delivering the kiss of death: progress on understanding howperforin works. Current opinion in immunology 19, 301–8 (2007). DOI10.1016/j.coi.2007.04.011. → pages[170] Haberman, R. Elementary Applied Partial Differential Equations (Prentice Hall, Upper SaddleRiver, NJ 07458, 1998), 3rd edn. → pages[171] Stinchcombe, J. C., Majorovits, E., Bossi, G., Fuller, S. & Griffiths, G. M. Centrosomepolarization delivers secretory granules to the immunological synapse. Nature 443, 462–465(2006). DOI 10.1038/nature05071. → pages[172] Wiedemann, A., Depoil, D., Faroudi, M. & Valitutti, S. Cytotoxic T lymphocytes kill multipletargets simultaneously via spatiotemporal uncoupling of lytic and stimulatory synapses. Proc.Natl. Acad. Sci. U S A 103, 10985–90 (2006). DOI 10.1073/pnas.0600651103. → pages[173] Dietrich, C., Yang, B., Fujiwara, T., Kusumi, A. & Jacobson, K. Relationship of lipid rafts totransient confinement zones detected by single particle tracking. Biophys J 82, 274–84 (2002).DOI 10.1016/S0006-3495(02)75393-9. → pages[174] Kusumi, A., Ike, H., Nakada, C., Murase, K. & Fujiwara, T. Single-molecule tracking ofmembrane molecules: plasma membrane compartmentalization and dynamic assembly ofraft-philic signaling molecules. Semin Immunol 17, 3–21 (2005). DOI10.1016/j.smim.2004.09.004. → pages[175] Lillemeier, B. F. & Klammt, C. How membrane structures control T cell signaling. Frontiers inImmunology 3 (2012). DOI 10.3389/fimmu.2012.00291. → pages[176] Couzin-Frankel, J. Cancer Immunotherapy. Science 342, 1432–1433 (2013). → pages[177] Fischbach, M. A., Bluestone, J. A. & Lim, W. A. Cell-Based Therapeutics: The Next Pillar ofMedicine. Science translational medicine 5, 179ps7 (2013). → pages[178] D’souza, N. et al. Mesenchymal stem/stromal cells as a delivery platform in cell and genetherapies. BMC Medicine 13, 186 (2015). DOI 10.1186/s12916-015-0426-0. → pages[179] Lim, W. A. Designing customized cell signalling circuits. Nature Reviews Molecular CellBiology 11, 393–403 (2010). DOI 10.1038/nrm2904. → pages122[180] Woodsworth, D. J. & Holt, R. A. Cell-Based Therapeutics: Making a Faustian Pact withBiology. Trends in Molecular Medicine 23, 104–115 (2017). DOI10.1016/j.molmed.2016.12.004. → pages[181] Hinrichs, C. S. & Rosenberg, S. A. Exploiting the curative potential of adoptive T-cell therapyfor cancer. Immunological reviews 257, 56–71 (2014). DOI 10.1111/imr.12132. → pages[182] Voskoboinik, I., Whisstock, J. C. & Trapani, J. A. Perforin and granzymes: function,dysfunction and human pathology. (2015). → pages[183] Cartwright, A. N. R., Griggs, J. & Davis, D. M. The immune synapse clears and excludesmolecules above a size threshold. Nature Communications 5, 5479 (2014). DOI10.1038/ncomms6479. → pages[184] Huang, L., Pike, D., Sleat, D. E., Nanda, V. & Lobel, P. Potential pitfalls and solutions for useof fluorescent fusion proteins to study the lysosome. (2014). → pages[185] Montel, A. H., Morse, P. A. & Brahmi, Z. Upregulation of B7 molecules by the Epstein-Barrvirus enhances susceptibility to lysis by a human NK-like cell line. Cellular immunology 160,104–14 (1995). → pages[186] Shimizu, Y. & DeMars, R. Production of human cells expressing individual transferredHLA-A,-B,C genes using an HLA-A,-B,-C null human cell line. The Journal of Immunology142, 3320–3328 (1989). → pages[187] Braulke, T. & Bonifacino, J. S. Sorting of lysosomal proteins. Biochimica et Biophysica Acta(BBA) - Molecular Cell Research 1793, 605–614 (2009). DOI 10.1016/j.bbamcr.2008.10.016.→ pages[188] Saftig, P. & Klumperman, J. Lysosome biogenesis and lysosomal membrane proteins:trafficking meets function. Nature Reviews Molecular Cell Biology 10, 623–635 (2009). DOI10.1038/nrm2745. → pages[189] Lopez, J. a. et al. Perforin forms transient pores on the target cell plasma membrane to facilitaterapid access of granzymes during killer cell attack. Blood (2013). DOI10.1182/blood-2012-07-446146. → pages[190] Breitling, J. & Aebi, M. N-linked protein glycosylation in the endoplasmic reticulum. ColdSpring Harbor perspectives in biology 5, a013359 (2013). DOI 10.1101/cshperspect.a013359.→ pages[191] van Meel, E. et al. Multiple Domains of GlcNAc-1-phosphotransferase Mediate Recognition ofLysosomal Enzymes. The Journal of biological chemistry 291, 8295–307 (2016). DOI10.1074/jbc.M116.714568. → pages[192] Warner, J. B., Thalhauser, C., Tao, K. & Sahagian, G. G. Role of N-linked oligosaccharideflexibility in mannose phosphorylation of lysosomal enzyme cathepsin L. The Journal ofbiological chemistry 277, 41897–905 (2002). DOI 10.1074/jbc.M203097200. → pages[193] Qian, Y. et al. Functions of the alpha, beta, and gamma subunits of UDP-GlcNAc:lysosomalenzyme N-acetylglucosamine-1-phosphotransferase. The Journal of biological chemistry 285,3360–70 (2010). DOI 10.1074/jbc.M109.068650. → pages123[194] Corey, D. R. & Craik, C. S. An investigation into the minimum requirements for peptidehydrolysis by mutation of the catalytic triad of trypsin. Journal of the American ChemicalSociety 114, 1784–1790 (1992). DOI 10.1021/ja00031a037. → pages[195] Gupta, R. & Brunak, S. Prediction of glycosylation across the human proteome and thecorrelation to protein function. Pacific Symposium on Biocomputing. Pacific Symposium onBiocomputing 310–22 (2002). → pages[196] Blázquez, M. & Shennan, K. I. J. Basic mechanisms of secretion : sorting into the regulatedsecretory pathway. Biochemistry and Cell Biology 191, 181–191 (2000). → pages[197] Portales-Casamar, E. et al. A regulatory toolbox of MiniPromoters to drive selective expressionin the brain. Proceedings of the National Academy of Sciences of the United States of America107, 16589–94 (2010). DOI 10.1073/pnas.1009158107. → pages[198] Igney, F. H. & Krammer, P. H. Immune escape of tumors: apoptosis resistance and tumorcounterattack. Journal of leukocyte biology 71, 907–20 (2002). → pages[199] Vinay, D. S. et al. Immune evasion in cancer: Mechanistic basis and therapeutic strategies.Seminars in Cancer Biology 35, S185–S198 (2015). DOI 10.1016/j.semcancer.2015.03.004. →pages[200] Marcus, A. et al. Recognition of tumors by the innate immune system and natural killer cells.Advances in immunology 122, 91–128 (2014). DOI 10.1016/B978-0-12-800267-4.00003-1. →pages[201] Brentjens, R. J. et al. CD19-Targeted T Cells Rapidly Induce Molecular Remissions in Adultswith Chemotherapy-Refractory Acute Lymphoblastic Leukemia. Science TranslationalMedicine 5, 177ra38–177ra38 (2013). DOI 10.1126/scitranslmed.3005930. → pages[202] Porter, D. L., Levine, B. L., Kalos, M., Bagg, A. & June, C. H. Chimeric antigenreceptor-modified T cells in chronic lymphoid leukemia. The New England journal of medicine365, 725–33 (2011). DOI 10.1056/NEJMoa1103849. → pages[203] Fulda, S. Tumor resistance to apoptosis. International journal of cancer 124, 511–5 (2009).DOI 10.1002/ijc.24064. → pages[204] Jensen, M. C. & Riddell, S. R. Design and implementation of adoptive therapy with chimericantigen receptor-modified T cells. Immunological reviews 257, 127–44 (2014). DOI10.1111/imr.12139. → pages[205] Bonavida, B. Tumor Cell Resistance to Apoptosis by Infiltrating Cytotoxic Lymphocytes. InYefenof, E. (ed.) Innate and Adaptive Immunity in the Tumor Microenvironment, 121–137(Springer, 2008). → pages[206] Töpfer, K. et al. Tumor evasion from T cell surveillance. Journal of biomedicine &biotechnology 2011, 918471 (2011). DOI 10.1155/2011/918471. → pages[207] Wong, R. S. Y. Apoptosis in cancer: from pathogenesis to treatment. Journal of experimental& clinical cancer research : CR 30, 87 (2011). DOI 10.1186/1756-9966-30-87. → pages124[208] Debatin, K.-M. Apoptosis pathways in cancer and cancer therapy. Cancer immunology,immunotherapy : CII 53, 153–9 (2004). DOI 10.1007/s00262-003-0474-8. → pages[209] Lin, Y.-f. et al. Targeting the XIAP / caspase-7 complex selectively kills caspase-3 âA˘S¸deficient malignancies. The Journal of Clinical Investigation 123, 3861–3875 (2013). DOI10.1172/JCI67951DS1. → pages[210] Devarajan, E. et al. Down-regulation of caspase 3 in breast cancer: a possible mechanism forchemoresistance. Oncogene 21, 8843–51 (2002). DOI 10.1038/sj.onc.1206044. → pages[211] de Heer, P. et al. Caspase-3 activity predicts local recurrence in rectal cancer. Clinical CancerResearch 13, 5810–5815 (2007). DOI 10.1158/1078-0432.CCR-07-0343. → pages[212] Huang, H. et al. Expression and prognostic significance of osteopontin and caspase-3 inhepatocellular carcinoma patients after curative resection. Cancer science 101, 1314–9 (2010).DOI 10.1111/j.1349-7006.2010.01524.x. → pages[213] Iolascon, A. et al. Caspase 3 and 8 deficiency in human neuroblastoma. Cancer genetics andcytogenetics 146, 41–7 (2003). → pages[214] Lakhani, S. A. et al. Caspases 3 and 7: key mediators of mitochondrial events of apoptosis.Science (New York, N.Y.) 311, 847–51 (2006). DOI 10.1126/science.1115035. → pages[215] de Oca, J. et al. Caspase-3 activity, response to chemotherapy and clinical outcome in patientswith colon cancer. International Journal of Colorectal Disease 23, 21–27 (2008). DOI10.1007/s00384-007-0362-3. → pages[216] Oudejans, J. J. et al. Absence of caspase 3 activation in neoplastic cells of nasopharyngealcarcinoma biopsies predicts rapid fatal outcome. Modern pathology : an official journal of theUnited States and Canadian Academy of Pathology, Inc 18, 877–85 (2005). DOI10.1038/modpathol.3800398. → pages[217] Sun, Y. et al. Differential caspase-3 expression in noncancerous, premalignant and cancertissues of stomach and its clinical implication. Cancer detection and prevention 30, 168–73(2006). DOI 10.1016/j.cdp.2006.02.004. → pages[218] Winter, R. N., Kramer, A., Borkowski, A. & Kyprianou, N. Loss of caspase-1 and caspase-3protein expression in human prostate cancer. Cancer research 61, 1227–32 (2001). → pages[219] Small, S., Keerthivasan, G., Huang, Z., Gurbuxani, S. & Crispino, J. D. Overexpression ofsurvivin initiates hematologic malignancies in vivo. Leukemia 24, 1920–6 (2010). DOI10.1038/leu.2010.198. → pages[220] Huber, C. et al. Inhibitors of apoptosis confer resistance to tumour suppression by adoptivelytransplanted cytotoxic T-lymphocytes in vitro and in vivo. Cell death and differentiation 12,317–25 (2005). DOI 10.1038/sj.cdd.4401563. → pages[221] Berezovskaya, O. et al. Increased Expression of Apoptosis Inhibitor Protein XIAP Contributesto Anoikis Resistance of Circulating Human Prostate Cancer Metastasis Precursor CellsIncreased Expression of Apoptosis Inhibitor Protein XIAP Contributes to Anoikis Resistance ofCircul. Cancer Research 2378–2386 (2005). → pages125[222] Fulda, S. & Vucic, D. Targeting IAP proteins for therapeutic intervention in cancer. Naturereviews. Drug discovery 11, 109–24 (2012). DOI 10.1038/nrd3627. → pages[223] Coumar, M. S., Tsai, F.-Y., Kanwar, J. R., Sarvagalla, S. & Cheung, C. H. A. Treat cancers bytargeting survivin: just a dream or future reality? Cancer treatment reviews 39, 802–11 (2013).DOI 10.1016/j.ctrv.2013.02.002. → pages[224] de Almagro, M. C. & Vucic, D. The inhibitor of apoptosis (IAP) proteins are critical regulatorsof signaling pathways and targets for anti-cancer therapy. Experimental oncology 34, 200–11(2012). → pages[225] LaCasse, E. C. Pulling the plug on a cancer cell by eliminating XIAP with AEG35156. Cancerletters 332, 215–24 (2013). DOI 10.1016/j.canlet.2012.06.015. → pages[226] Obexer, P. & Ausserlechner, M. J. X-linked inhibitor of apoptosis protein - a critical deathresistance regulator and therapeutic target for personalized cancer therapy. Frontiers inoncology 4, 197 (2014). DOI 10.3389/fonc.2014.00197. → pages[227] Altieri, D. C. Survivin , cancer networks and pathway-directed drug discovery. Nature reviewsCancer 8, 551–560 (2008). → pages[228] Bots, M. et al. SPI-CI and SPI-6 cooperate in the protection from effector cell âA˘S¸ mediatedcytotoxicity. Blood 105, 1153–1161 (2005). DOI 10.1182/blood-2004-03-0791.Supported. →pages[229] Bots, M., VAN Bostelen, L., Rademaker, M. T., Offringa, R. & Medema, J. P. Serpins preventgranzyme-induced death in a species-specific manner. Immunology and cell biology 84, 79–86(2006). DOI 10.1111/j.1440-1711.2005.01417.x. → pages[230] Ravi, R. et al. Resistance of cancers to immunologic cytotoxicity and adoptive immunotherapyvia X-linked inhibitor of apoptosis protein expression and coexisting defects in mitochondrialdeath signaling. Cancer research 66, 1730–9 (2006). DOI 10.1158/0008-5472.CAN-05-3377.→ pages[231] Végran, F. et al. Survivin-3B potentiates immune escape in cancer but also inhibits the toxicityof cancer chemotherapy. Cancer research 73, 5391–401 (2013). DOI10.1158/0008-5472.CAN-13-0036. → pages[232] Medema, J. P. et al. Blockade of the granzyme B/perforin pathway through overexpression ofthe serine protease inhibitor PI-9/SPI-6 constitutes a mechanism for immune escape by tumors.Proceedings of the National Academy of Sciences of the United States of America 98, 11515–20(2001). DOI 10.1073/pnas.201398198. → pages[233] Stirpe, F. Ribosome-inactivating proteins. Toxicon: official journal of the International Societyon Toxinology 44, 371–83 (2004). DOI 10.1016/j.toxicon.2004.05.004. → pages[234] Pastan, I., Hassan, R., Fitzgerald, D. J. & Kreitman, R. J. Immunotoxin therapy of cancer.Nature reviews. Cancer 6, 559–65 (2006). DOI 10.1038/nrc1891. → pages[235] Zhang, J., Kale, V. & Chen, M. Gene-directed enzyme prodrug therapy. The AAPS journal 17,102–10 (2015). DOI 10.1208/s12248-014-9675-7. → pages126[236] Fillat, C., Carrió, M., Cascante, A. & Sangro, B. Suicide gene therapy mediated by the HerpesSimplex virus thymidine kinase gene/Ganciclovir system: fifteen years of application. Currentgene therapy 3, 13–26 (2003). → pages[237] Searle, P. F. et al. NITROREDUCTASE: A PRODRUG-ACTIVATING ENZYME FORCANCER GENE THERAPY. Clinical and Experimental Pharmacology and Physiology 31,811–816 (2004). DOI 10.1111/j.1440-1681.2004.04085.x. → pages[238] Vass, S. O., Jarrom, D., Wilson, W. R., Hyde, E. I. & Searle, P. F. E. coli NfsA: an alternativenitroreductase for prodrug activation gene therapy in combination with CB1954. BritishJournal of Cancer 100, 1903–1911 (2009). DOI 10.1038/sj.bjc.6605094. → pages[239] Kim, J. H. et al. High Cleavage Efficiency of a 2A Peptide Derived from Porcine Teschovirus-1in Human Cell Lines, Zebrafish and Mice. PLoS ONE 6, e18556 (2011). DOI10.1371/journal.pone.0018556. → pages[240] Soule, H. D., Vazguez, J., Long, A., Albert, S. & Brennan, M. A human cell line from a pleuraleffusion derived from a breast carcinoma. Journal of the National Cancer Institute 51, 1409–16(1973). → pages[241] Baginska, J. et al. Granzyme B degradation by autophagy decreases tumor cell susceptibility tonatural killer-mediated lysis under hypoxia. Proceedings of the National Academy of Sciencesof the United States of America 110, 17450–5 (2013). DOI 10.1073/pnas.1304790110. →pages[242] Mahrus, S. & Craik, C. S. Selective chemical functional probes of granzymes A and B revealgranzyme B is a major effector of natural killer cell-mediated lysis of target cells. Chemistry &biology 12, 567–77 (2005). DOI 10.1016/j.chembiol.2005.03.006. → pages[243] Deveraux, Q. L. & Reed, J. C. IAP family proteins–suppressors of apoptosis. Genes &development 13, 239–52 (1999). → pages[244] Kalos, M. et al. T cells with chimeric antigen receptors have potent antitumor effects and canestablish memory in patients with advanced leukemia. Science translational medicine 3, 95ra73(2011). DOI 10.1126/scitranslmed.3002842. → pages[245] Greco, R. et al. Improving the safety of cell therapy with the TK-suicide gene. Frontiers inpharmacology 6, 95 (2015). DOI 10.3389/fphar.2015.00095. → pages[246] Fesnak, A. D., June, C. H. & Levine, B. L. Engineered T cells: the promise and challenges ofcancer immunotherapy. (2016). → pages[247] Peng, W. et al. Loss of PTEN Promotes Resistance to T CellâA˘S¸Mediated Immunotherapy.Cancer Discovery 6 (2016). → pages[248] Amantana, A., London, C. A., Iversen, P. L. & Devi, G. R. X-linked inhibitor of apoptosisprotein inhibition induces apoptosis and enhances chemotherapy sensitivity in human prostatecancer cells. Molecular cancer therapeutics 3, 699–707 (2004). → pages[249] Flanagan, L. et al. High levels of X-linked Inhibitor-of-Apoptosis Protein (XIAP) are indicativeof radio chemotherapy resistance in rectal cancer. Radiation Oncology 10, 131 (2015). DOI10.1186/s13014-015-0437-1. → pages127[250] Garg, H., Suri, P., Gupta, J. C., Talwar, G. P. & Dubey, S. Survivin: a unique target for tumortherapy. Cancer Cell International 16, 49 (2016). DOI 10.1186/s12935-016-0326-1. → pages[251] Turley, S. J., Cremasco, V. & Astarita, J. L. Immunological hallmarks of stromal cells in thetumour microenvironment. Nature Reviews Immunology 15, 669–682 (2015). DOI10.1038/nri3902. → pages[252] Wang, C. et al. High throughput sequencing reveals a complex pattern of dynamicinterrelationships among human T cell subsets. Proceedings of the National Academy ofSciences of the United States of America 107, 1518–1523 (2010). DOI10.1073/pnas.0913939107. → pages[253] Li, M. et al. Optimal promoter usage for lentiviral vector-mediated transduction of culturedcentral nervous system cells. Journal of Neuroscience Methods 189, 56–64 (2010). DOI10.1016/j.jneumeth.2010.03.019. → pages[254] Astrakhan, A. et al. Ubiquitous high-level gene expression in hematopoietic lineages provideseffective lentiviral gene therapy of murine Wiskott-Aldrich syndrome. Blood 119, 4395–4407(2012). DOI 10.1182/blood-2011-03-340711. → pages[255] Choi, P. J. & Mitchison, T. J. Imaging burst kinetics and spatial coordination during serialkilling by single natural killer cells. Proceedings of the National Academy of Sciences of theUnited States of America 110, 6488–93 (2013). DOI 10.1073/pnas.1221312110. → pages[256] Lieberman, J. The ABCs of granule-mediated cytotoxicity: new weapons in the arsenal. Naturereviews. Immunology 3, 361–70 (2003). DOI 10.1038/nri1083. → pages[257] Griffiths, G. M., Tsun, A. & Stinchcombe, J. C. The immunological synapse: a focal point forendocytosis and exocytosis. The Journal of Cell Biology 189, 399–406 (2010). DOI10.1083/jcb.201002027. → pages[258] Wang, Z. et al. Development of a nonintegrating Rev-dependent lentiviral vector carryingdiphtheria toxin A chain and human TRAF6 to target HIV reservoirs. Gene therapy 17,1063–76 (2010). DOI 10.1038/gt.2010.53. → pages[259] Klingemann, H., Boissel, L. & Toneguzzo, F. Natural Killer Cells for Immunotherapy -Advantages of the NK-92 Cell Line over Blood NK Cells. Frontiers in immunology 7, 91(2016). DOI 10.3389/fimmu.2016.00091. → pages[260] James, J. R. & Vale, R. D. Biophysical mechanism of T-cell receptor triggering in areconstituted system. Nature 487, 64–9 (2012). DOI 10.1038/nature11220. → pages[261] Rabinovich, G. A., Gabrilovich, D. & Sotomayor, E. M. Immunosuppressive Strategies that areMediated by Tumor Cells. Annual Review of Immunology 25, 267–296 (2007). DOI10.1146/annurev.immunol.25.022106.141609. → pages[262] Lebowitz, J. H. et al. Glycosylation-independent targeting enhances enzyme delivery tolysosomes and decreases storage in mucopolysaccharidosis type VII mice (2003). → pages[263] Di Stasi, A. et al. Inducible Apoptosis as a Safety Switch for Adoptive Cell Therapy. NewEngland Journal of Medicine 365, 1673–1683 (2011). DOI 10.1056/NEJMoa1106152. →pages128[264] Ellington, A. D. & Szostak, J. W. In vitro selection of RNA molecules that bind specificligands. Nature 346, 818–822 (1990). DOI 10.1038/346818a0. → pages[265] Zhu, Q., Liu, G. & Kai, M. DNA Aptamers in the Diagnosis and Treatment of HumanDiseases. Molecules 20, 20979–20997 (2015). DOI 10.3390/molecules201219739. → pages[266] Kimoto, M., Yamashige, R., Matsunaga, K.-i., Yokoyama, S. & Hirao, I. Generation ofhigh-affinity DNA aptamers using an expanded genetic alphabet. Nature Biotechnology 31,453–457 (2013). DOI 10.1038/nbt.2556. → pages[267] Merkle, T., Holder, I. T. & Hartig, J. S. The dual aptamer approach: rational design of ahigh-affinity FAD aptamer. Org. Biomol. Chem. 14, 447–450 (2016). DOI10.1039/C5OB02026C. → pages[268] Zhou, J. & Rossi, J. Aptamers as targeted therapeutics: current potential and challenges.Nature Reviews Drug Discovery (2016). DOI 10.1038/nrd.2016.199. → pages[269] Schumacher, M. A. Bacterial plasmid partition machinery: a minimalist approach to survival.Current opinion in structural biology 22, 72–9 (2012). DOI 10.1016/j.sbi.2011.11.001. →pages[270] Annaluru, N. et al. Total Synthesis of a Functional Designer Eukaryotic Chromosome. Science344, 55–58 (2014). DOI 10.1126/science.1249252. → pages[271] Hutchison, C. A. et al. Design and synthesis of a minimal bacterial genome. Science 351,aad6253–aad6253 (2016). DOI 10.1126/science.aad6253. → pages[272] Qi, L. S. et al. Repurposing CRISPR as an RNA-guided platform for sequence-specific controlof gene expression. Cell 152, 1173–83 (2013). DOI 10.1016/j.cell.2013.02.022. → pages[273] Suck, G. et al. NK-92: an âA˘Ÿoff-the-shelf therapeutic’ for adoptive natural killer cell-basedcancer immunotherapy. Cancer Immunology, Immunotherapy 65, 485–492 (2016). DOI10.1007/s00262-015-1761-x. → pages[274] Linette, G. P. et al. Cardiovascular toxicity and titin cross-reactivity of affinity-enhanced T cellsin myeloma and melanoma. Blood 122, 863–71 (2013). DOI 10.1182/blood-2013-03-490565.→ pages[275] Cameron, B. J. et al. Identification of a Titin-Derived HLA-A1-Presented Peptide as aCross-Reactive Target for Engineered MAGE A3-Directed T Cells. Science TranslationalMedicine 5, 197ra103–197ra103 (2013). DOI 10.1126/scitranslmed.3006034. → pages[276] Soriano, P. Generalized lacZ expression with the ROSA26 Cre reporter strain. Nature genetics21, 70–1 (1999). DOI 10.1038/5007. → pages[277] Ito, R., Takahashi, T., Katano, I. & Ito, M. Current advances in humanized mouse models.Cellular & molecular immunology 9, 208–14 (2012). DOI 10.1038/cmi.2012.2. → pages[278] Kwok, R. Five hard truths for synthetic biology. Nature 463, 288–90 (2010). DOI10.1038/463288a. → pages129[279] Barabási, A.-L. & Oltvai, Z. N. Network biology: understanding the cell’s functionalorganization. Nature Reviews Genetics 5, 101–113 (2004). DOI 10.1038/nrg1272. → pages[280] Hornung, G. & Barkai, N. Noise Propagation and Signaling Sensitivity in Biological Networks:A Role for Positive Feedback. PLoS Computational Biology 4, e8 (2008). DOI10.1371/journal.pcbi.0040008. → pages[281] Elmore, S. Apoptosis: a review of programmed cell death. Toxicologic pathology 35, 495–516(2007). DOI 10.1080/01926230701320337. → pages[282] Edelman, G. M. & Gally, J. A. Degeneracy and complexity in biological systems. Proceedingsof the National Academy of Sciences 98, 13763–13768 (2001). DOI 10.1073/pnas.231499798.→ pages[283] Del Vecchio, D. Modularity, context-dependence, and insulation in engineered biologicalcircuits. Trends in biotechnology 33, 111–9 (2015). DOI 10.1016/j.tibtech.2014.11.009. →pages[284] Tawfik, D. S. Messy biology and the origins of evolutionary innovations. Nature ChemicalBiology 6, 692–696 (2010). DOI 10.1038/nchembio.441. → pages[285] Tsimring, L. S. Noise in biology. Reports on progress in physics. Physical Society (GreatBritain) 77, 026601 (2014). DOI 10.1088/0034-4885/77/2/026601. → pages[286] Uversky, V. N. Dancing Protein Clouds: The Strange Biology and Chaotic Physics ofIntrinsically Disordered Proteins. Journal of Biological Chemistry 291, 6681–6688 (2016).DOI 10.1074/jbc.R115.685859. → pages[287] Esch, E. W., Bahinski, A. & Huh, D. Organs-on-chips at the frontiers of drug discovery. NatureReviews Drug Discovery 14, 248–260 (2015). DOI 10.1038/nrd4539. → pages[288] Takahashi, M. K. et al. Rapidly Characterizing the Fast Dynamics of RNA Genetic Circuitrywith Cell-Free TranscriptionâL´ŠTranslation (TX-TL) Systems. ACS Synthetic Biology 4,503–515 (2015). DOI 10.1021/sb400206c. → pages[289] Chin, J. W. Expanding and Reprogramming the Genetic Code of Cells and Animals. AnnualReview of Biochemistry 83, 379–408 (2014). DOI 10.1146/annurev-biochem-060713-035737.→ pages[290] Lajoie, M. J. et al. Genomically Recoded Organisms Expand Biological Functions. Science342 (2013). → pages[291] Ostrov, N. et al. Design, synthesis, and testing toward a 57-codon genome. Science 353 (2016).→ pages[292] Gibson, D. G., Smith, H. O., Iii, C. A. H., Venter, J. C. & Merryman, C. chemical synthesis ofthe mouse mitochondrial genome 7 (2010). DOI 10.1038/nMeth.1515. → pages130Appendix AGenerating granzyme B mCherry fusionproteins expressed from genomicgranzyme B locus using CRISPR/Cas9To try to improve granzyme B payload fusion protein loading into lytic granules, and hence potentiallytransfer to target cells and fusion protein stability, I generated YTs that express granzyme B mCherryfusion proteins from the genomic granzyme B locus. (See Section 5.2.2 for further discussion andmotivation.) Specifically I used the CRISPR/Cas9 system to insert a sequence consisting of a glycineserine linker followed by the crmCherry coding sequence immediately preceding the granzyme B stopcodon.I first designed two guide RNAs that had predicted cut sites within ten base pairs of the stop codon(see below for details of gRNA sequence and design). I cloned these gRNAs into pX330 (a plasmidexpressing spCas9 and gRNA from CBh and U6 promoters respectively, addgene cat #42230). I alsocloned a small fragment of the genomic granzyme B sequence immediately surrounding the stop codon(putative cut sites) into a plasmid that consists of GFP followed by RFP out of frame. The smallfragment was inserted between the two fluorescent proteins. Thus if the gRNA cuts in that region, theNHEJ repair will result in around 30% in frame RFP. I tested the two guide RNAs using this plasmid(data not shown) and found that gRNA-1 (with sequence CATGAAACGCTACTAACTAC) had a muchhigher cutting efficiency.I then designed a donor template, consisting of an insert of a glycine serine linker followed bycrmCherry, and left and right homology arms for regions immediately 5’ and 3’ of the stop codon in thegenomic granzyme B locus (see below for details of template design, homology arm PCR from genomicDNA, and template assembly and cloning). The homology arms were 1 kb, and were amplified fromgenomic DNA extracted from the cell line NK-92MI. This donor was initially assembled in a TOPOvector (using tdTomato as a payload), and then subcloned into an MND vector. For the results presentedhere, the donor was further subcloned into a modified pdL vector that had the CAG promoter deleted131(see Appendix E for details on the pdL vector) and the tdTomato payload was replaced by an mCherrypayload. This final donor plasmid was named pDN_GZB_E5-MCH (see Section A.2 for plasmid mapand sequence).I electroporated the pX330 (gRNA-1) and pDN_GZB_E5-MCH plasmids into YT-Indys using theNeon electroporation system (Thermo Fisher Scientific), using mainly the manufacturers recommen-dations. For details see the Methods sections of Chapters 3 and 4. Here I note any relevant details tothis particular experiment.• YTs were passaged 1/2 24 hours prior.• Plasmids and DNA amounts. The negative controls were pX330 or pDN_GZB_E5-MCH alone,20 ug. The positive control was 10 ug of pdL_MCH an mCherry expression plasmid. Finally,the experimental condition was both pX330 and pDN_GZB_E5-MCH at 1:1 molar ratio, 20 ugDNA.• 100 ul tip using buffer R, final volume 110 ul2 weeks after electroporation I FACS sorted cells for RFP signal, and then lysed 2×105 cells inLaemmli sample buffer. These lysates were size separated by gel electrophoresis transferred to blotsand probed for mCherry and granzyme B (see Methods sections of Chapter 3 for details on samplepreparation, Western blotting and antibodies).I also extracted genomic DNA from 8×105 cells using DNAzol (Thermo Fisher Scientific) fol-lowing the manufacturer’s instructions. I used this genomic DNA as template for a PCR in whichthe forward primer (ACAGCTGCTCACTGTTGGGG) annealed 5’ of the 5’ end of the left homologyarm (that is in the granzyme B gene, but not in the donor template) and the reverse primer (TTG-TACAGCTCGTCCATGCC) annealed in the mCherry insert (that is only in the donor template andnowhere in the granzyme B gene), so therefore the desired amplicon (2.8 kb) should only amplify fromgenomic DNA with the insert at the correct granzyme B locus. The PCR was conducted using Taq PCRsupermix (Thermo Fisher Scientific) with 0.25 ul gDNA as template, 1 ul each of 10 uM primers in a50 ul reaction. The cycling conditions were 57C annealing temperature, 3 minute extension time, for30 cycles.The results of the FACS sort, Western blot, and PCR screen are shown in Figure A.1 and demon-strate locus specific insertion of a granzyme B mCherry fusion protein expressed from the nativegranzyme B promoter. The fusion protein is functional (FACS and microscopy) and stable (I monitoredexpression for 6 weeks via FACS and microscopy). Important given the purpose of these experiments,is that the observed degradation of these fusion proteins is similar to what is observed when the fusionproteins are transiently overexpressed (data not shown).I repeated these experiments exactly as above, using a different natural killer cell line (NK-92MI).NK92s had previously been used extensively in this project and had been shown to degrade granzymepayload fusion proteins to a very large extent. I was interested to see if expression from the endogenous132locus decreased this degradation. The experimental protocols were identical to those for YTs. The re-sults are shown in Figure A.1. Again, the degradation is substantial and similar to what was previouslyobserved for fusions transiently overexpressed (data not shown).This data demonstrates that functional fusion proteins can be produced from the genomic granzymeB locus, but that the degradation of the fusion proteins is not substantially altered.YT-Indy NK-92MI62493828191YT YT:GZB-MCJHanti-vinculinanti-MCH anti-GZBYT YT:GZB-MCJH mCherryDAPIFigure A.1: Molecular characterization of genomically expressed granzyme B mCherry fusionproteins. CRISPR-cas9 generation of YT-Indy cells expressing granzyme-B-mCherry fusion pro-teins from the endogenous granzyme B locus. Left panel. A) FACS data showing that cellstransfected with both cas9, gRNA and donor plasmids had a clear RFP+ population (right), whilethose transfected with cas9 and gRNA alone did not (left). B) Agarose gel of PCR reactions us-ing genomic DNA from transfected YT-Indys as template, a forward primer that anneals at the5âA˘Z´ end of the granzyme B coding sequence, and a reverse primer that anneals at the 3âA˘Z´ endof mCherry. Amplicons consistent with mCherry insertion at the endogenous granzyme B locusare observed in samples from cells that were transfected with cas9, gRNA and donor plasmid(far right, DN + C9 + gR), but not in any other sample. C) Western blot of whole cell lysatesfrom transfected YT-Indys. Blot was probed with anti-mCherry antibody. Bands consistent withgranzyme B-mCherry fusion proteins are observed in samples from cells that were transfectedwith cas9, gRNA and donor plasmid (far right, DN + C9 + gR), but not in any other sample.Right panel As for left panel, but using NK-92MI cells. Abbreviations: DN: donor plasmid; C9:cas9; gR: guide RNA; NTC: no template control; TC: transfection control.133TEMPLATE	DESIGN	Design	of	donor	vector	for	cas9	mediated	fusion	of	payload	to	genomic	GZMB	locus.	Need	homology	arms	surrounding	GZMB	stop	codon,	flanking	desired	insert.	Want	this	whole	construct	to	be	embedded	inside	LTRs	of	viral	vector	so	that	it	may	be	packaged	into	virus.		Donor	Template:	LHA—GS_LINKER—PAYLOAD—2A—GFP—STOP—RHA		LHA:	1	kb	NK-92MI	genomic	sequence	immediately	upstream	of	GZMB	STOP	codon	RHA:	1	kb	NK-92MI	genomic	sequence	immediately	downstream	of	GZMB	STOP	codon	PAYLOAD:	In	assembly	process	will	be	tdTomato				Assembly	strategy:	1. PCR	amplify	homology	arms	with	tailed	primers.	Primer	tails	include	homology	overlaps	to	adjacent	components	for	Gibson	assembly,	as	well	as	NotI	sites	to	liberate	homology	arms	from	sequencing	vector.	2. TOPO	clone	amplicons	and	sequence	verify.	3. Liberate	INSERT	(GS_LINKER—PAYLOAD—2A—GFP—STOP)	from	pMND:GZB-TDT_GFP	(dHL_0355)	via	restriction	digest		(NotI/XhoI).	4. Open	pMND	(vHL_0021)	by	cutting	out	promoter	and	GFP	reporter	(ClaI/SalI)	immediately	inside	the	LTRs)	to	function	as	BACKBONE.	5. Liberate	RHA,	LHA	from	TOPO	vectors	by	NotI	restriction	digest.	6. 4	part	Gibson	Assembly:	BACKBONE—LHA—INSERT—RHA—BACKBONE	Homology	for	Gibson	assembly	provided	by	homology	arm	amplicons.	**NOTE:	This	assembly	strategy	requires	3’à5’	exonuclease	activity	of	the	polymerase	used.	The	standard	NEB	Gibson	kit	uses	a	polymerase	that	has	fairly	nonrobust	activity.	Thus	it	will	be	better	to	use	either	their	NEB	HiFi	builder	kit,	or	follow	original	Gibson	protocol	directly.		Primer	design:	5’—NOTI—GIBSON_OVERLAP—RE_SITE—ANNEAL—3’		1. Gibson	overlap	design:	Tm>48C;	length>20bp	(since	doing	more	than	3	part	assembly)	Choose	to	do	25	bp	overlaps,	all	of	which	had	Tm	high	enough.	For	restriction	digest	liberated	components,	there	remains	a	few	bases	of	the	old	RE	sequence.	To	completely	eliminate	old	RE	and	replace	with	new	one	(see	below)	move	Gibson	overlap	further	into	vector.	This	will	result	in	small	3’	mismatches,	which	requires	3’à5’	exonuclease	activity.	,A.1 Design of guide RNAs and donor template134BACKBONE—LHA:	GATTAGTGAACGGATCTCGACGGTA	LHA—INSERT:	TGGAGGCGGGGGTTCTGGCGGGGGT	(RC:	ACCCCCGCCAGAACCCCCGCCTCCA)	INSERT—RHA:	CGGCATGGACGAGCTGTACAAGTAA	RHA—BACKBONE:	TGATCAAATTCGAGCTCGGTACCTT	(RC:	AAGGTACCGAGCTCGAATTTGATCA)		2. RE	site	selection:	LHA.f:	Use	MauBI	to	replace	ClaI,	since	8	cutter	more	unique.	CGCGCGCG	LHA.r:	Use	PacI	to	replace	NotI,	which	is	not	unique	in	MND	vector	(NotI	cuts	in	backbone).	TTAATTAA	RHA.f:	Use	FseI	to	replace	XhoI,	since	8	cutter	more	unique.	GGCCGGCC	RHA.r:	Use	I-SceI	to	replace	SalI.	Choose	homing	endonuclease	so	guaranteed	is	always	unique	so	that	vector	can	be	linearized	for	transfection.	TAGGGATAACAGGGTAAT.	(RC:	ATTACCCTGTTATCCCTA)		3. Annealing	region	design:	Standard	4	requirements:	GC	content	roughly	50%,	length	~	20	bp,	Tm	~50-60C,	GC	clamp.	LHA.f:	CTACCTAGCAACAAGGCCCAG	LHA.r:	GTAGCGTTTCATGGTTTTCTTTATCC	RHA.f:	CTACAGGAAGCAAACTAAGCCC	RHA.r:	TTTGAACTCAAAGGGCTGATGTAGC				Sequencing	primers:	Designed	one	forward	and	one	reverse	in	the	middle	of	each	homology	arm.		135GUIDE	DESIGN	Took	70	bp	on	either	side	of	GZMB	stop	codon,	input	into	Zhang	lab	guide	design	tool:	http://crispr.mit.edu/.			Picked	top	two	guides;	all	others	have	cut	sites	>10bp	away	from	stop	codon	(e.g.	homology	junctions).		Ordered	top	and	bottom	oligos	for	annealing	with	overhangs	for	ligation	into	Zhang	lab	plasmids.					136A.2 pDN_GZB_E5-MCH plasmidA.2.1 Plasmid mapFigure A.2: pDN_GZB_E5-MCH plasmid. The annotations correspond to those in the genbankfile.137A.2.2 Plasmid sequenceLOCUS pDN_GZB_E5−MCH 4954 bp DNA c i r c u l a r UNA 23−MAR−2017DEFINITION Concatenat ion o f 2 sequences .ACCESSION urn . l o c a l . . .1490302855544.11VERSION urn . l o c a l . . .1490302855544.11KEYWORDS .SOURCEORGANISM.FEATURES Locat ion / Qua l i f i e r ssource 23..1022/ l abe l ="LHA"CDS j o i n ( <23. .238 ,882. . >1022)/ gene="GZMB"/ note =" Derived by automated computat iona l ana l ys i s usinggene p r ed i c t i o n method : BestRefSeq . "/ codon_star t=1/ product ="granzyme B precursor "/ p r o t e i n_ i d ="NP_004122 . 2 "/ db_xref ="GI :221625528"/ db_xref ="CCDS:CCDS9633. 1 "/ db_xref ="GeneID :3002"/ db_xref ="HGNC:HGNC:4709"/ db_xref ="MIM:123910"/ t r a n s l a t i o n ="MQPILLLLAFLLLPRADAGEIIGGHEAKPHSRPYMAYLMIWDQKSLKRCGGFLIRDDFVLTAAHCWGSSINVTLGAHNIKEQEPTQQFIPVKRPIPHPAYNPKNFSNDIMLLQLERKAKRTRAVQPLRLPSNKAQVKPGQTCSVAGWGQTAPLGKHSHTLQEVKMTVQEDRKCESDLRHYYDSTIELCVGDPEIKKTSFKGDSGGPLVCNKVAQGIVSYGRNNGMPPRACTKVSSFVHWIKKTMKRY"/ l abe l ="GZMB CDS"misc_feature 1032..1091/ note ="Geneious type : p o l y l i n k e r "/ l a be l ="GS_Linker_Rand "CDS 1101..1775/ l abe l ="mCherry "t e rm ina to r 1782..1787/ l abe l ="STOP"source 1794..2793/ l abe l ="RHA"misc_feature complement (3170. .3840)/ note ="Geneious type : Or ig in o f Rep l i ca t i on "/ l a be l =" pUC_ori "misc_feature complement (3985. .4845)/ note ="Geneious type : Marker "/ l a be l ="AmpR"ORIGIN1 ccccgtacgc gagataatcg a tc tacc tag caacaaggcc caggtgaagc cagggcagac61 atgcagtg tg gccggctggg ggcagacggc ccccctggga aaacactcac acacactaca121 agaggtgaag atgacagtgc aggaagatcg aaagtgcgaa t c t g a c t t a c gcca t t a t t a181 cgacagtacc a t t g a g t t g t gcgtggggga cccagagat t aaaaagactt c c t t t a agg t241 aagactatgc acctgcctgg a t t g g c t c t t gggagaaaga tg t t t gggga ata tc tgaga301 cctggagact caagtagtgg gggac tcc t t cacccactag ac t g t ga t a t t t c t c t c t g g361 aaagagaaga ggggactaga ctgagctggg gagaaattag ggcctc tgca aact taccag138421 gaggct ta tg gtggatggtg c t t c t t t g g a aggatgaat t tgcaacactc cacccactcc481 aggtcacaga ta t taggaaa ctg tgcccac tgggggtgca g taa t t a t aa ccagg tg tg t541 ct tcagaggc tggtacccaa cg t gg t t aa t gggctggtcc tcca tgg tgg acatcagccc601 t cc t t gccca c t t c t g gg t c ct taaacagc caacggtccc acatacc tcc gatctcagga661 tctgggggac atgacggagg ctggcccctg ggatgaggtg aagcagtaac aatgtccagg721 gccagagct t ggcagctggg ggccaccagc ggcctgccct gccc t c tgg t c tcccacatg781 taggc tg tgc aag t tggcc t t t t c t aaaag ggggcttgag atggaagaga gggcaggacc841 cggaggagca tcagc tcag t c c t t c cac t c t c t a t t c a ca gggggactct ggaggccctc901 t t g t g t g t a a caaggtggcc cagggcattg t c t c c t a t g g acgaaacaat ggcatgcctc961 cacgagcctg caccaaagtc t c aagc t t t g tacac tgga t aaagaaaacc atgaaacgct1021 acgcggccgc tggaggcggg gg t tc tggcg ggggtggatc agggggtgga ggctccggtg1081 gaggcgggtc gggcgcgccc atcatcaagg ag t tca tgcg c t t caagg tg cacatggagg1141 gctccgtgaa cggccacgag t tcgaga tcg agggcgaggg cgagggccgc ccctacgagg1201 gcacccagac cgccaagctg aaggtgacca agggtggccc cc t g ccc t t c gcctgggaca1261 t c c t g t c c cc t c ag t t c a t g tacggctcca aggcctacgt gaagcacccc gccgacatcc1321 ccgac tac t t gaagctgtcc t tccccgagg gc t tcaag tg ggagcgcgtg a tgaac t t cg1381 aggacggcgg cgtggtgacc gtgacccagg ac t c c t c c c t gcaggacggc gag t t c a t c t1441 acaaggtgaa gctgcgcggc accaact tcc cctccgacgg ccccgtaatg cagaagaaga1501 ccatgggctg ggaggcctcc tccgagcgga tgtaccccga ggacggcgcc ctgaagggcg1561 agatcaagca gaggctgaag ctgaaggacg gcggccacta cgacgctgag gtcaagacca1621 cctacaaggc caagaagccc gtgcagctgc ccggcgccta caacgtcaac atcaagt tgg1681 acatcacctc ccacaacgag gactacacca tcgtggaaca gtacgaacgc gccgagggcc1741 gccactccac cggcggcatg gacgagctgt acaaggaatt c taa tagc tc gagctacagg1801 aagcaaacta agcccccgct gtaatgaaac a c c t t c t c t g gagccaagtc caga t t taca1861 ctgggagagg tgccagcaac tgaataaata c c t c t t a g c t gagtggaaaa g c t g g t t t c t1921 t g t t t a t t c a t t gaccc t ca t t c t caggca ccacatc tgc gctatgcagg ccaatgacac1981 a a t t t t g c t g t t t t c t g c t t t c t c c t c t c c cc tcacccc t tgccacc tcc ccaaaccccc2041 acatgaagct gatactcagc t c c t t c c t a t ccacaccagt t t c t ccaggg cc t gccc t t c2101 tgccaaggct gaagctgagc accatcagga gacaacatgg accac t t t gg tcc tggggc t2161 t tgggtaaac t t c t t a c c t c c t t c t c c a g t g t taca tgac agagaaaaaa gggataatac2221 catgggacct aac t cc t ca t cccccactgg ggc t cc t ca t t c t c ccc t gg g c t t a g t t t c2281 t c t a ccc t c c tc tgagc tca aggctcagct cgtcc tccag c c t c t t g g c t g cccc t t c t c2341 t t c a t c c c t g c t gag t g t t c tcagaatcca ccaac t c t t g t c c t c t c cag accacactga2401 t c t ga t c t gg cccc tccc tc a ta t c taccc acctaagata cccagagacc ca t g t gg t t c2461 cataagggcc t tgccac tga gacgccagcc ca t c t ca tgc cctggcagag aggggcctca2521 gaaaaaccag gcc tg tg tgg caaccaggta agacccatgg aggacaaggc tggcacggtc2581 t c t c t c caac cc t t ggc t c c a t c t c t c c cc taggtagggc cagctcaacc cc tccca tcc2641 agcccagtgt cc tccca tac ac tcaagg t t cactgcccac ctgggcagtc agcaggctga2701 gcccc t t t aa a c c t g t t c c t c t t g g t c a c t gc tggcc t c t aggctaagat t ccc tgc tag2761 ccacctgggc tacatcagcc c t t t g a g t t c aaagtcgacc tc tagc taga gc t tggcg ta2821 a tca tgg tca t a g c t g t t t c c tg tg tgaaa t t g t t a t c c g c t cacaa t t c cacacaacat2881 acgagccgga agcataaagt gtaaagcctg gggtgcctaa tgagtgagc t aac tcaca t t2941 aa t t g cg t t g cgctcactgc ccgc t t t c ca gtcgggaaac c tg t cg tgcc agc tgca t ta3001 atgaatcggc caacgcgcgg ggagaggcgg t t t g c g t a t t gggcgc tc t t c cgc t t c c t c3061 gctcactgac t cgc tgcgc t cgg tcg t t cg gctgcggcga gcggtatcag ctcactcaaa3121 ggcggtaata cgg t ta t cca cagaatcagg ggataacgca ggaaagaaca tgtgagcaaa3181 aggccagcaa aaggccagga accgtaaaaa ggccgcgt tg c t g g c g t t t t t cca taggc t3241 ccgcccccct gacgagcatc acaaaaatcg acgctcaagt cagaggtggc gaaacccgac3301 aggactataa agataccagg cg t t t c c ccc tggaagctcc c t cg tgcgc t c t c c t g t t c c3361 gaccctgccg c t taccgga t acctg tccgc c t t t c t c c c t tcgggaagcg t g g cg c t t t c3421 tca tagc tca cgc tg tagg t a t c t c ag t t c gg tg taggtc g t t cgc t cca agctgggctg3481 tgtgcacgaa ccccccg t tc agcccgaccg c tgcgcc t t a t ccgg taac t a t c g t c t t g a3541 gtccaacccg gtaagacacg ac t t a t cgcc actggcagca gccactggta acaggattag3601 cagagcgagg ta tg taggcg gtgctacaga g t t c t t gaag tgg tggcc ta actacggcta1393661 cactagaaga acag t a t t t g g ta tc tgcgc tc tgc tgaag ccag t t acc t tcggaaaaag3721 agt tgg tagc t c t t g a t c c g gcaaacaaac caccgctggt a g c g g t t t t t t t g t t t g c a a3781 gcagcagatt acgcgcagaa aaaaaggatc tcaagaagat c c t t t g a t c t t t t c t a cggg3841 g tc tgacgc t cagtggaacg aaaactcacg t t aaggga t t t t gg t ca t ga gat ta tcaaa3901 aagga tc t t c acctagatcc t t t t a a a t t a aaaatgaagt t t t aaa t caa t c t aaag t a t3961 ata tgagtaa ac t t g g t c t g acagt tacca a t g c t t a a t c agtgaggcac c ta t c t cagc4021 ga t c t g t c t a t t t c g t t c a t cca tag t tgc c tgac tcccc g tcg tg taga taac tacga t4081 acgggagggc t t a c ca t c t g gccccagtgc tgcaatga ta ccgcgagacc cacgctcacc4141 ggctccagat t ta tcagcaa taaaccagcc agccggaagg gccgagcgca gaagtggtcc4201 t g caac t t t a tccgcc tcca t c c ag t c t a t t a a t t g t t g c cgggaagcta gagtaagtag4261 t t c g c cag t t aa t ag t t t g c gcaacg t tg t t g c ca t t g c t acaggcatcg tgg tg tcacg4321 c t c g t c g t t t gg t a t ggc t t ca t t cagc t c cggt tcccaa cgatcaaggc gagt taca tg4381 atcccccatg t tg tgcaaaa aagcggttag c t c c t t c g g t cc tccgatcg t tg tcagaag4441 taag t tggcc gcag t g t t a t cac tca tgg t tatggcagca c t g ca t aa t t c t c t t a c t g t4501 catgccatcc g taagatgc t t t t c t g t g a c tgg tgagtac tcaaccaagt ca t t c tgaga4561 a tag tg t a t g cggcgaccga g t t g c t c t t g cccggcgtca atacgggata ataccgcgcc4621 acatagcaga ac t t taaaag t g c t c a t c a t tggaaaacgt t c t t cggggc gaaaactctc4681 aaggatc t ta ccgc tg t tga ga tccag t t c gatgtaaccc actcgtgcac ccaactgatc4741 t t c a g ca t c t t t t a c t t t c a ccagcg t t t c tgggtgagca aaaacaggaa ggcaaaatgc4801 cgcaaaaaag ggaataaggg cgacacggaa a tg t t gaa ta c t c a t a c t c t t c c t t t t t c a4861 a t a t t a t t g a agca t t t a t c aggg t t a t t g t c t ca tgagc ggatacata t t t g a a t g t a t4921 t tagaaaaat aaacaaatag gggt tccgcg agct/ /140Appendix BMass spectrometry based investigation ofgranzyme B mCherry fusion proteinregions of instabilityTo try to decrease the amount of degradation of the granzyme B payload fusion proteins—and hencepotentially increase the efficiency of payload transfer to target cells, and expand the range of possibledelivery cell chassis— I conducted a mass spectrometry experiment to investigate if any regions inthe granzyme B mCherry fusion protein were more susceptible to breakdown. (See Section 5.2.2 forfurther discussion and motivation.) Specifically I expressed the fusion proteins in YT-Indy cells, ranthe whole cell lysate through a shotgun mass spectrometry pipeline, and then realigned the peptides tothe fusion proteins sequence, looking for regions of instability.I electroporated the pdL plasmids coding for mCherry (MCH) and granzyme B fused to MCH viaa glycine serine linker (GZB-MCH) into YTs using the Neon electroporation system (Thermo FisherScientific), using mainly the manufacturers recommendations. For details see the Methods sections ofChapters 3 and 4. The plasmids were pdL_MCH and pdL_GZB-MCH respectively (see Appendix Efor plasmid sequences and maps) and 20 ug of DNA was used. 48 hours after transfection I FACSsorted viable RFP+ cells. I then lysed these samples in Laemmli sample buffer, and size separate8×105 cell equivalents via gel ectrophoresis, transferred the samples to blots and probed for mCherry(see Methods sections of Chapter 3 for details on sample preparation, Western blotting and antibodies).Using an identical aliquot of the cell lysate sample, and loading equivalent cell amounts, I ran a secondgel using the same running conditions and stained this gel with a non-specific Coomassie blue stain toidentify all proteins in the gel.141anti-mCherry blot Coomassie stain62493828A B A BFigure B.1: Western blot and gel of samples used for mass spectrometry experiments. The im-ages come from two separate gels, loaded with identical aliquots of the same sample. A = YTsexpressing MCH; B = YTs expressing GZB-MCH. Handwritten numbers in middle are bands thatwere excised (approximate location hand drawn drawn on Coomassie stain) for mass spectrometryinvestigation. The letters A and B, and numbers 1-4 correspond to those in fig.... Numbers at farleft are sizes in kDa of molecular marker.The blot and gel images are shown in Figure B.1. For each sample, I identified four bands ofinterest, labeled 1-4, and consisting of (in order):1. Full length fusion protein2. Prominent breakdown product of full length fusion protein3. Full length mCherry and/or fusion protein cleaved or decoupled approximate at the linker (sincethe sizes of granzyme B and crmCherry are both approximately 30 kDa)4. Prominent breakdown product of mCherryThe Coomassie stained gel was submitted to the GSC Proteomics core facility. The bands wereexcised, the proteins in-gel digested with trypsin, and run on an Orbitrap Fusion mass spectrometer.142The resulting mass-charge ratio spectra were compared to the theoretical peptide spectra of GZB-MCHusing SeqQuest, generating a list of peptides. These peptides were then aligned to the GZB-MCHamino acid sequence using a custom script written in R. The data is split into two figures Figure B.2(for tryptic peptides) and Figure B.3 (non-tryptic peptides).12341234YT:MCHYT:GZB-MCHFigure B.2: Mass spectrometry identified tryptic peptides from whole cell lysates. Numbers attop and red schematic at bottom are reference GZB-MCH sequence . Each rectangle is individualpeptide, mapped to its location in the reference.143Together this data suggests two tentative findings. First, there appears to be some semi-conservedpattern of N-terminal degradation of mCherry, as can be seen from comparing the coverage going fromA3 to A4 in Figure B.2 (recall A = YT:MCH, and 3 = 30kda expected MCH size, 4 = 20kda breakdownregion). This results in a substantial loss of coverage in the N-terminal region of Cherry, not just aloss of depth (as I would expect if the proteins in this region were just random breakdown products).This conclusion is further supported by the data in Figure B.3, which shows a significant density ofnon-tryptic peptides near the N-terminus of the protein. Notably this degradation seems somewhatdependent on N-terminal exposure, as these peptides are far more common in the unfused samples(which have a free N-terminus) than in the fusion protein samples.Second, it appears that there is substantial ’breakage’ in the peri-linker region of GZB-MCH. Myargument here is the same: going from B1 to B3 (i.e. from the region I would expect full length fusionprotein to the region where GZB or MCH would run independently) in Figure B.2, you see a substantialloss of coverage, mainly focused around the linker (with an almost complete loss of linker peptides).My assessment of this pattern is that it is more of an ’explosion’ than a clean break. That is, if theprotein just snapped in two, then I would expect both halves to migrate at 30 kda, and thus retain all ofthe peptides found in the full length fusion protein (losing at most one unique peptide, if the break pointwere in the middle of that peptide, but then I would potentially expect a non-tryptic peptide). However,if the middle of the protein is cleaved or somehow otherwise degraded in a localized but not amino acidrestricted fashion, then all of the small fragments would migrate off the bottom of the gel, and thusgenerating the loss of peptides we see. One possible mechanism for is that if there are have two coreglobular proteins with hanging fragments of unstructured polypeptides, these are degraded away (byexopeptidases, mechanical shear etc). So I think a region of fragility, with frequent breakage, followedby further degradation from the break point to the core peptides, could explain these observations.Based on this data I think it would be worth considering re-engineering the fusion protein. Sincethe domains necessary for granzyme B to act as a suitable chaperone are not fully delineated, it wouldbe inadvisable to modify it. However, a different linker could be tested. This would be a judiciousfirst modification to make, based on the model of peri-linker instability described above. One possiblealternative is a shorter glycine serine linker that has been used in granzyme B tdTomato fusion proteinsin a mouse model system [138] (with nucleotide and amino acid sequences GGCGGGTCTGGCG-GTGGGGGATCGGCCAACGGATCC and GGSGGGGSANGS respectively).The crmCherry protein used here is a variant of mCherry with the 12 N-terminal amino acids ofmCherry deleted [184]. This was done in an effort to improve the stability of mCherry in lysosomes,after substantial N-terminal degradation was observed. It is possible that further elimination of N-terminal residues could improve the stability even more. This could be guided by the constraint ofavoiding disrupting the beta-barrel or fluorophore containing alpha helix of the mCherry protein.Finally, this pipeline provides a template for investigating the characteristics of fusion proteinbreakdown. This could be employed in attempts to extend this system to other cytotoxic lymphocytechassis or when developing new payloads for delivery.14434123YT:MCHYT:GZB-MCHFigure B.3: Mass spectrometry identified non-tryptic peptides from whole cell lysates. Numbersat top and red schematic at bottom are GZB-MCH sequence. Each rectangle is individual peptide,mapped to its location in the reference.145Appendix CGeneration of a mutant EEF2 gene whichprotects cells from DTA when the two areco-expressedThe diphtheria toxin A fragment (DTA) was shown to be potent as a C-terminal granzyme fusion(GZB-DTA) and would be an attractive payload to deliver to tumour cells Section 4.2.1. However,for lymphocyte delivery, it is likely that some mechanism would be required to protect the deliverylymphocyte from GZB-DTA autotoxicity. Fortunately, a mutant form of EEF2—recall DTA inhibitsprotein synthesis by ribosylating EEF2 (Section 4.1)— has been reported, the overexpression of whichrenders cells resistant to DTA toxicity [252]. I set out to confirm this result and test if co-expression ofthis mEEF2 with GZB-DTA would protect cells from toxicity.I first constructed a mEEF2 clone as follows. The wild type gene was obtained from the MGCcollection (BC126259). The required mutation is G717R (GGA to CGA). To introduce this muta-tion I conducted a two step PCR reaction the first to insert the mutation, which would generate twooverlapping fragments, and then a second fusion PCR with only external primers.The annealing regions of the 4 primers were:1. EEF2_G717R_IN.r: GATCTGGCCCCCTCGGCGGTGGATG2. EEF2_G717R_IN.f: CATCCACCGCCGAGGGGGCCAGATC3. EEF2_Ex.r: CAATTTGTCCAGGAAGTTGTCCAG4. EEF2_Ex.f: ATGGTGAACTTCACGGTAGACCAGThe first PCR reaction used 0.5 ul Accuprime PFX (Thermo Fisher Scientific), 5 ul 10X reactionmix, 1.5 ul each primer (1 with 4 and 2 with 3, two separate reactions) and 30 ng EEF2 cDNA in a finalvolume of 50 ul. The cycling parameters were 55C annealing temperature and a 2m30s extension time,146for 30 cycles. The amplicons were size separated by gel electrophoresis and 2 and 0.4 kb bands wereexcised and column purified. The second PCR was conducted using 25 ng of the larger fragment and5 ng of the smaller fragment (this to ensure equimolar amounts of each template). Only the externalprimers were used (3 and 4), otherwise the PCR reaction was as above. The cycling conditions were64C annealing temperature, a 3m10s extension time for 27 cycles. The amplicons were size separatedby gel electrophoresis and a 2.4 kb band was excised and column purified. This amplicon was ligatedinto a TOPO vector and sequence verified.I then constructed three separate plasmids coding for: (i) GFP; (ii) GZB-DTA-2A-GFP; (iii) mEEF2-2A-GZB-DTA-2A-GFP, where 2A denotes a 2A peptide which results in ribosomal skipping. Thusthese coding sequences are expressed as a single transcript, but result 1, 2 and 3 separate proteins re-spectively (see Section 4.2.1 and reference [239] for details). These coding sequences were expressedfrom the MND promoter in a MND lentiviral transfer plasmid (see Section 4.4 for details and referenceson the MND vector, and Section E.2 for plasmid maps and sequences).I transfected these three plasmids into 293T cells using TransIT (Mirus Bio). (See Section 4.4for details on 293T cells and transfection). 48 hours after transfection I imaged the cells using afluorescence microscope and then ran them on a flow cytometer. The results are shown in Figure C.1,and show that: (i) GZB-DTA nearly completely abrogates GFP expression and (ii) mEEF2 co-expressedwith GZB-DTA simultaneously restores GFP expression, albeit at lower a lower expression level, andin a lower fraction of cells.This data suggests that DTA could be delivered to target cells as a granzyme B fusion protein, ifmEEF2 were co-expressed along with the fusion protein, which could be done using 2A peptides aswas done here. Importantly, since the mature mEEF2 and GZB-DTA proteins are separated, mEEF2would not be loaded into lytic granules nor delivered to target cells, thus both preserving the integrityof the delivery lymphocyte, and avoiding counteracting the toxic effect of the delivered GZB-DTA inthe target cells.147GFP GZB-DTA-2A-GFP mEEF2-2A-GzB-DTA-2A-GFPPIGFPFigure C.1: Evidence that mEEF2 co-expression with GZB-DTA restores protein synthesis func-tion. Top and bottom panels are us the same samples: cells were imaged and then analyzed viaflow cytometry.148Appendix DCodeD.1 Core MATLAB scripts implementing computational biophysicalmodel (Chapter 2)In this section I include a minimal working example of the stochastic simulation algorithm implementa-tion. This consists of several MATLAB source code files, and a parameter file. For the work presentedin the main text, this core set of code was compiled into a standalone executable (using MATLAB’sbuilt in capacity for this) on Westgrid. The result is a binary executable and and execution shell script.A custom python wrapper script was used to initialize the simulations across the parameter ranges, andsubmit them to the scheduler. Finally a small bash cript was used to concatenate the various outputfiles, and a MATLAB script to summarize this output. Since these are ’computational’ steps, whichwould not change the actual data, and are fairly specific to Westgrid, they are not included.D.1.1 IMS_Ex.m%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Date :2015/07/12%%Wr i t ten by : Danie l Woodsworth ( dan ie l . woodsworth@gmail . com)%%%Summary : Simulates t ime evo lu t i on o f granzyme and pe r f o r i n molecules%w i t h i n immunological synapse , from molecule ’ s re lease from l y t i c granule%through synapse d i f f u s i o n , p e r f o r i n i n se r t i o n , pore fo rmat ion and granzyme%i n t e r n a l i z a t i o n .%%Uni ts : ( unless otherwise noted ) microns , seconds .%%Methods : Uses s p a t i a l s t ochas t i c s imu la t i on a lgo r i t hm ( adapted%from El f , 2004) . See E l f , Systems Biology , 2004 f o r f u r t h e r d e t a i l s%regard ing t h i s method .%149%Inpu t : s i ng l e parameter , an i n t ege r ( s imu la t i on number ) o f c lass char .%%Requires :%1. Helper s c r i p t s , w i t h i n same d i r ec t o r y , inc luded wi th t h i s f i l e .%2. Parameter f i l e named ’ params . t x t ’ , i n same d i r ec t o r y , inc luded .%%Output :%1. summary . t x t : Most human readable , has s imu la t i on endpoint data .%2. simVarSummary . csv : Has more endpoint data , csv format f o r data%ana lys i s .%3. xx_timeCourse . csv : ( xx i s number provided as inpu t ) Has t ime evo lu t i on%data o f species numbers , csv format ted .%%%For f u r t h e r d e t a i l s see paper : Woodsworth DJ et a l . , B iophys ica l Journal ,%2015.%%Free to use , please prov ide a t t r i b u t i o n .%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%func t i on IMS_Ex ( sim_num)t i c %S ta r t c lock to ca l cu l a t e t o t a l CPU time%−−−−−−−−−−−−−−−−−−−−−−−−−−Parameter Input−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−%f i d = fopen ( ’ params . t x t ’ , ’ r ’ ) ;%Parameter f i l e must have format : name of parameter=value%Read parameter f i l e i n t o c e l l a r ray : f i r s t column i s parameter names%second i s parameter values . A l l s t r i n g s a t t h i s stageC = tex tscan ( f i d , ’%s %s ’ , ’ d e l im i t e r ’ , ’ = ’ ) ;%Convert to hash tab le , w i th names as keys , and values as valuesparams = con ta ine rs .Map(C{1 } ,C{ 2 } ) ;%I n i t i a l i z e l o c a l va r i ab l es from hash tab le%GeometryR_syn = str2num ( params ( ’ Radius_of_synapse ’ ) ) ;h = str2num ( params ( ’ Height_of_synapse ’ ) ) ;r_LG = str2num ( params ( ’ Rad ius_o f_ ly t i c_granu le ’ ) ) ;l = str2num ( params ( ’ Mesh_cel l_size ’ ) ) ;%Molecule Sizesr_GzB = str2num ( params ( ’ Radius_of_granzyme_B ’ ) ) ;r_Pfn_XA = str2num ( params ( ’ Cross−sec t i ona l _pe r f o r i n_ rad i us ’ ) ) ;%Pore valuesn_pore = str2num ( params ( ’ Number_of_perforin_monomers_per_pore ’ ) ) ;r_pore = str2num ( params ( ’ Lumina l_rad ius_of_per fo r in_pore ’ ) ) ;%B iophys ica l Constants150e taL i p i d = str2num ( params ( ’ Viscosi ty_of_cel l_membrane_ ( l i p i d ) ’ ) ) ;etaH20 = str2num ( params ( ’ V iscos i ty_o f_water ’ ) ) ;Kb = str2num ( params ( ’ Boltzmann_constant ’ ) ) ;T37 = str2num ( params ( ’ Temperature_in_Kelvin ’ ) ) ;K_pfn_ins = str2num ( params ( ’ Rate_of_per for in_ inser t ion_ in to_membrane ’ ) ) ;%Hindered d i f f u s i o n parameter , between 0 and 1.%This a l lows f o r sys temat ic exp l o ra t i on o f space%To use b iophys i ca l c a l c u l a t i o n o f t h i s parameter , se t i t to −1hdc = str2num ( params ( ’ H inde red_d i f f us ion_coe f fec ien t ’ ) ) ;%I n t r a s ynap t i c parametersV_avg_mol = str2num ( params ( ’ Average_volume_of_molecule_in_synapse ’ ) ) ;f rac_occ = str2num ( params ( ’ Fract ion_of_synapse_occupied_by_molecules ’ ) ) ;Kd = str2num ( params ( ’ Average_dissoc ia t ion_constant_for_non−spec i f i c_b ind i ng ’ ) ) ;%Molecule numbersN_PFN = str2num ( params ( ’ Number_of_perfor in_released_into_synapse ’ ) ) ;N_GZB = str2num ( params ( ’ Number_of_granzyme_B_released_into_synapse ’ ) ) ;%Global Simulat ion_parametersT = str2num ( params ( ’ Maximum_simulation_time ’ ) ) ;ou t I n c r = str2num ( params ( ’ T ime_ in te rva l_ fo r_da ta_outpu t ’ ) ) ;pVal = str2num ( params ( ’ Current_Varied_Parameter_Value ’ ) ) ;%−−−−−−−−−−−−−−−−−−−−−−−Shu f f l e random number generator−−−−−−−−−−−−−−−−−−%%Ensure unique random number stream f o r s imu la t i on .rng ( ’ shu f f l e ’ ) ;sd = in t32 ( rand∗1e8 + str2num (sim_num) ) ;rng ( sd ) ;%−−−−−−−−−−−−−−−−−−−−−−−−−−Spa t i a l D i s c r e t i z a t i o n−−−−−−−−−−−−−−−−−−−−−−−−−%%Disc re t i z e c i r c u l a r synapse . I n sc r i be c i r c l e i ns i de square .%Def ine o r i g i n i n lower l e f t o f square . Mesh l o ca t i o n%def ined by cent re coord ina tes o f mesh .%Each c e l l indexed by r a s t e r i n g from bottom l e f t to top r i g h t%Mesh c e l l s i ze must be chosen to be l a r ge r than pe r f o r i n pore diametern_ce l l s_per_s ide = c e i l (2∗R_syn / l ) ; % number o f mesh c e l l s to go across squaren_ce l l s = n_ce l l s_per_s ide ^2 ; % number o f c e l l s i n system%−−−−−−−−−−−−−−−−−−−−−−−−Calcu la te De r i va t i ve Constants−−−−−−−−−−−−−−−−−−−%r_Pfn = 2∗r_Pfn_XA ; %volume averaged pfn rad ius151%rad ius o f PFN o l i go i s average of long and shor t dimensions o f chainr = @( i ) r_Pfn_XA∗( i +1) / 2 ;%D i f f u s i v i t i e s f o r s o l u t i o n p a r t i c l e s from E ins te i n Smoulchowski r e l a t i o nD_GzB = Kb∗T37 / (6∗ p i∗etaH20∗r_GzB ) ;D_Pfn = Kb∗T37 / (6∗ p i∗etaH20∗ r_Pfn ) ;%D i f f u s i v i t i e s f o r membrane inse r t ed pe r f o r i n o l i g o i s ca l cu la ted using%re su l t s from Saffman−Delbruckgamma = 0.5772; %Eulers constanth = 0 .01 ; % Thickness o f plasma membraneD_Pfn_jmer = @( j ) ( ( Kb∗T37 ) / (4∗ p i∗e taL i p i d∗h ) ) ∗( log ( e t aL i p i d∗h / ( etaH20∗ r ( j ) ) )−gamma) ;%−−−−−−−−−−−−−Hindered d i f f u s i o n & macromolecular crowding−−−−−−−−−−−−−−−−%%Assume tha t molecules d i f f u s i n g through synapse slowed down by both%occupied space of synapse due to o ther synap t i c molcules ( e . g . ICAM/LFA ,%CD45, TCR:MHC) and non−s p e c i f i c b ind ing wi th these molcules%So Def f = (Kd / (R+Kd) )∗(1− f ) Dfree , where Kd i s d i s so c i a t i o n constant , R i s%dens i t y o f nonspec i f i c binders , and f i s volume f r a c t i o n occupied by a l l%molecules .%R = f / ( Avg . Molecule i n Syanpse )%Switch : set hdc to −1 to use b iophys i ca l c a l c u l a t i o n ; otherwise can use%lumped hdc parameter to exp lore e f f e c t s o f vary ing hindered d i f f u s i o ni f ( hdc < 0)%l a s t term to conver t to Molar (NOTE t h i s term accounts f o r nm vs um ! ! )R_binder = ( f rac_occ / V_avg_mol ) ∗ (6 /10 ) ;%E f f e c t i v e d i f f u s i v i t i e sD_GzB_eff = (Kd / ( R_binder+Kd) )∗(1− f rac_occ )∗D_GzB;D_Pfn_eff = (Kd / ( R_binder+Kd) )∗(1− f rac_occ )∗D_Pfn ;e lseD_GzB_eff = hdc∗D_GzB;D_Pfn_eff = hdc∗D_Pfn ;end%−−−−−−−−−−−−−−−−−−−−−−−−−−System Size−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−%%Number o f species i n system : i nse r t ed pe r f o r i n jmers 1−18, synap t i c%pe r f o r i n and granzymeN_species = n_pore +2;%State mat r i x N, each row i s a subvolume%Order o f species i n s ta te mat r i x N: P1 , . . . , P18 , Psyn ,GsynN = zeros ( n_ce l l s , N_species ) ;152%−−−−−−−−−−−−−−−−−−−−−−−−−−−Set Reactions−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−%%%We have f o l l ow i ng reac t i ons :%% K i j%Pe r f o r i n o l i gome r i za t i on : Pj+Pi −−> Pi+ j , j =1 ,2 ,3; i = j , . . . , 18 − j%% K1%Bulk pe r f o r i n i n s e r t i o n i n t o membrane : Pb −−> P1% Kg%Granzyme t r an s l o ca t i o n through pe r f o r i n pore : G + P18 −−> Gi + P18%%Impor tant to note only a l low 1 ,2 ,3mers j o i n to form higher o l igos , >4mer%cannot j o i n w i th each other , on ly 1 ,2 ,3mer .%Number o f i n t e r a c t i o n s :%monomer to jmer (17)%dimer to jmer (15)%t r ime r to jmer (13)%pe r f r o i n i n s e r t i o n and granzyme t r ans l o ca t i o n (2 )n_rxns = 17+15+13+2;%Sto ich iomet ry mat r i x S .%Each row corresponds to a a given reac t i on i n the system , row order must%match column order o f A / a i ( see below ) .%Each column corresponds to a species i n the system , column order must%match t ha t o f N above .%S i j i s the change in number o f molecules o f the j t h species t ha t occurs%when the j t h reac t i on occurs .%Const ruct S i n 4 par ts , one each f o r each 1 ,2 ,3mer reac t ions , f i n a l l y f o r%bulk species .S = [ ] ;%cyc le through 1 ,2 ,3mersf o r j =1:3%Can only have reac t i ons up to n_pore (18) − j . E . g . i f have 3mer , can%only have 3mer j o i n i n g a 15mer to form an 18mer , whereas wi th 1mer can%have 1mer j o i n i n g 17mer to form 18mer , so more reac t i ons . This i s the%i n t e r p r e t a t i o n o f the n_pore−j below : number o f reac t i ons f o r a jmer .subS=zeros ( n_pore−j , N_species ) ;f o r i = j : ( n_pore−j )%Except ion f o r two i d e n t i c a l o l i gos combining to form mul t imer%In t h i s case lose 2 of one species , manually set app rop r i a t e l yi f ( i == j )subS ( i , j ) = −2;subS ( i , i + j ) = 1 ;153cont inue ;end%For a l l o ther o l i g o i n t e r a c t i o n s , Lose imer and jmer , gain i +jmersubS ( i , j ) = −1;subS ( i , i ) = −1;subS ( i , i + j ) = 1 ;end%Add t h i s subsect ion o f mat r i x to t o t a l s t o i ch iome t r y mat r i x%Since index ing s t a r t s a t j , w i l l have empty rows of subS from 1 to j−1%So only append rows from j onwardsS = [S ; subS ( j : end , : ) ] ;end%Last pa r t f o r pfn i n s e r t i o n and granzyme t r ans l o ca t i o nsubS = zeros (2 , N_species ) ;%Pfn i n s e r t i o nsubS (1 , n_pore+1) = −1;subS (1 ,1 ) = 1 ;%Granzyme t r an s l o ca t i o nsubS (2 , n_pore+2) = −1;%appendS = [S ; subS ] ;%−−−−−−−−−−−−−−−−−−−−−−−Set ra te constants−−−−−−−−−−−−−−−−−−−−−−−−−−−−%Per f o r i n o l i g o fo rmat ion ( K i j )%Rate constant f o r c rea t i on o f p e r f o r i n mul t imer from two pe r f o r i n o l i gos%Use r e s u l t from Lauf fenburger & Linderman f o r p a r t i c l e f i n d i n g a t rap%K = 2piD∗Kon / ( 2 piD+Kon∗ l n ( b / s ) ) , Kon i s the chemical ra te%b = sq r t (SA / ( p i∗Pj ) ) , Pj i s the number o f PFN o l i go jmers f i n d i n g the%la r ge r PFN imer t rap , D i s sum of d i f f u s i v i t i e s%s i s reac t i on radius , take t h i s to be sum rad ius o f PFN imer and jmer%Take d i f f u s i o n l im i t e d case−−> K= 2piD / l n ( b / s )%This k has un i t s o f L ^2 / s , so need to d i v i de by SA of mesh c e l l l ^2%So k i = 2piD / ( ( l ^2)∗ l ns ( b / s ) ) . Note t ha t s i s a f unc t i on o f i , and b i s%a func t i on o f P1%Since these values change over time , need to de f ine them i n t e r n a l l y to%s imu la t i on .b = @( Pj ) sq r t ( l ^ 2 / ( p i∗Pj ) ) ;s = @( i , j ) r ( i ) + r ( j ) ;K = @( i , j , Pj ) 2∗p i ∗( D_Pfn_jmer ( i ) + D_Pfn_jmer ( j ) ) / ( ( l ^2)∗ log ( b ( Pj ) / s ( i , j ) ) ) ;%Rate o f granzyme f i n d i n g pore%Assume granzyme in cuboid o f he igh t h , whose p r o j e c t i o n onto membrane%sur face i s a mesh c e l l154%Then ra te o f d i f f u s i v e movement across cuboid i s (3D / h^2) , so ley along z%ax is . This i s i n compet i t i on w i th d i f f u s i v e jumps out o f cuboid .%Assume tha t t h i s ra te i s modulated by A_p / A_tot , e . g . the f r a c t i o n o f the%pore s ize compared to the mesh c e l l s i ze%So ra te i s (3 D A_p ) / ( L^2 h^2)Kg = 3∗D_GzB_eff∗( p i∗ r_pore ^2) / ( ( l ^2) ∗(h^2) ) ;%−−−−−−−−−−−−−−−−−−− I n i t i l i a z i a t i o n −−−−−−−−−−−−−−−−−−−−−−−−−−−−−−%%F i r s t d i s c r e t i z e synapse i n t o square ce l l s , and ca l cu l a t e which c e l l s%correspond to cen t r a l reg ion where LGs are re leased . Track i f c e l l i s%ex te rna l , i n t e r na l , or i n t e r n a l to LG.%Raster from bottom l e f t to top r i g h t . For each c e l l check i f i t s cent re i s%w i t h i n rad ius o f synapse , and w i t h i n rad ius o f l y t i c granule .%For l y t i c granule , assign ind i ces to vec to r so t ha t molecules can be%i n i t i a l l y d i s t r i b u t e d there%BinMat mat r i x below def ines i f mesh c e l l i s i n s i de (1 ) or ou ts ide (0 )%synapse , or i n s i de LG (2 )BinMat = zeros ( n_ce l l s_per_s ide ) ;%Boolean vector , determines i f c e l l i s i n t e r n a l or ex te rna l , according to%c e l l indexI s I n t = zeros ( n_ce l l s , 1 ) ;%Ce l l index . S ta r t s from bottom l e f t , increases L−>R, Bot−>TopCi = 1 ;%For given subvolume , g ive x , y coord ina tesSVcoords = zeros ( n_ce l l s , 2 ) ;LGInd = [ ] ; %vec to r w i th i nd i ces o f a l l c e l l s w i t h i n LG rad ius%Raster over d i s c r e t i z ed spacef o r i x = 1 : n_ce l l s_per_s idef o r i y = 1 : n_ce l l s_per_s ide%spa t i a l coords o f cu r ren t c e l l ( def ined c e n t r a l l y )%Note −1 i s ’ one ’ , o thers are lowercase Lx = ( ix −1)∗ l + l / 2 ;y = ( iy −1)∗ l + l / 2 ;SVcoords ( Ci , : ) = [ x y ] ;%Check i f d isplacement from cent re syanpse less than rad ius o f%synapse minus mesh c e l l length , to ensure a t l eas t 1 c e l l i s%ex te rna l a l l the way roundi f ( ( x−R_syn ) ^2 + ( y−R_syn ) ^2 <= (R_syn−l ) ^2)BinMat ( ix , i y ) = 1 ;I s I n t ( Ci ) = 1 ;end155%Check i f d isplacement from cent re synapse less than LG rad iusi f ( ( x−R_syn ) ^2 + ( y−R_syn ) ^2 <= r_LG^2)BinMat ( ix , i y ) = 2 ;LGInd = [ LGInd , Ci ] ;endCi = Ci +1;endend%Now generate i n i t i a l d i s t r i b u t i o n s . Only need to update those c e l l s t ha t%have species i n them i n i t i a l l y .%Required matr ices%Rate mat r i x R. Row i s a subvolume , columns are sum ( w i t h i n t ha t subvolume )%of i n t e r a c t i o n p ropens i t i es , d i f f u s i o n p ropens i t i e s and a l l p ropens i t i e s .R = zeros ( n_ce l l s , 3 ) ;%Propens i ty mat r i x A . Each row corresponds to a subvolume , each column i s a%i n t e r a c t i o n ( e . g . a ’ chemical ’ r eac t i on )A = zeros ( n_ce l l s , n_rxns ) ;%D i f f u s i on mat r i x D. Each row corresponds to subvolume , each column i s%d i f f u s i v e propens i t y f o r a species to jump out o f subvolume . Column order%matches s ta te mat r i x ND = zeros ( n_ce l l s , N_species ) ;%I n i t i a l i z a t i o n l i s t f o r next event t ime f o r each c e l l used to bu i l d b inary%t ree%Since most mesh c e l l s have no species i n them , t h e i r next event t ime i s%i n f i n i t e%F i r s t row i s index o f mesh ce l l , second i s next event t imeI n i t L i s t = [ [ 1 : n_ce l l s ] ’ i n f ( n_ce l l s , 1 ) ] ;Qarray = [ 1 : n_ce l l s ] ’ ;%I n i t i a l i z e c e l l s t ha t have nonzero species numbers%1. D i s t r i b u t e species%2. Ca lcu la te d i f f u s i v e and reac t i on p ropens i t i es , update matr ices%3. Ca lcu la te next event t ime f o r these c e l l s%A l l o ther c e l l s have 0 species ; hence no chemical or d i f f u s i o n events%Since d i v i d i n g t o t a l number o f molecules across l y t i c granules , round to%nearest i n t ege r ; cannot have decimal molecules .N_PFN_per_LG_cell = round (N_PFN/ leng th ( LGInd ) ) ;N_GZB_per_LG_cell = round (N_GZB/ leng th ( LGInd ) ) ;%Adjus t t o t a l number o f molecules i n system accord ing ly f o r i n t e r n a l%cons is tency .N_PFN = N_PFN_per_LG_cell∗ l eng th ( LGInd ) ;156N_GZB = N_GZB_per_LG_cell∗ l eng th ( LGInd ) ;f o r i = LGInd%Note : a l l c a l c u l a t i o n s i n t h i s loop are f o r a given c e l l%Add pe r f o r i n and granzyme to s ta te mat r i xN( i , n_pore+1) = N_PFN_per_LG_cell ;N( i , n_pore+2) = N_GZB_per_LG_cell ;%Calcu la te i n t e r a c t i o n p ropens i t i e s f o r cu r ren t c e l la i = ca lc In t rxnPropens (N( i , : ) , n_pore ,K, K_pfn_ins ,Kg) ;%I n se r t t h i s nonzero row i n t o propens i t y mat r i xA( i , : ) = a i ;%now reca l cu l a t e a0a0 = sum( a i ) ;%Calcu la te d i f f u s i o n p ropens i t i e s f o r cu r ren t c e l ld i = ca lcD i f fP ropens (N( i , : ) , n_pore , D_Pfn_jmer , D_Pfn_eff , D_GzB_eff , l ) ;%I n se r t t h i s nonzero row i n t o propens i t y mat r i xD( i , : ) = d i ;%reca l cu l a t e d0d0 = sum( d i ) ;%update t o t a l p ropens is tys0 = a0 + d0 ;%Update ra te mat r i x f o r t h i s SVR( i , : ) = [ a0 d0 s0 ] ;%Calcu la te next event t ime f o r t h i s ce l l , which w i l l now be less than%i n f i n i t e s ince have molecules i n c e l ltau = −log ( rand ) / s0 ;%Put t h i s i n t o the i n i t i a l i z a t i o n l i s t . Note t ha t the cons t ruc to r f o r%the b inary t ree accepts an a r b i r t r a r y unordered l i s tI n i t L i s t ( i , 2 ) = tau ;end%Bu i ld event Q and Qarray[Q, Qarray ] = bui ldQ ( I n i t L i s t , Qarray ) ;%−−−−−−−−−−−−−−−−−−−−−−−−Simulat ion−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−%%Current s imu la t i on t imet = 0 ;157%Counting va r i ab l esN_GzB_Int = 0 ; %I n t e r n a l i z e d GzBLost = [0 0 ] ; %Lost PFN and GzBT_pore_form = [ ] ; %Time when each pore formsNum_pore_formed = 0; %Tota l number o f pores formedmaxmer = 0; %Maximum PFN ol igomer (18 i f pore forms )stuckCount = 0 ; %Boolean : 1 i f system i s stuck , 0 otherwise .%outMat i s mat r i x t ha t t r acks number o f species i n system over t ime%I n i t i a l i z e w i th i n i t i a l valuesoutMat = [ t ,N_PFN,N_GZB, Lost ( 1 ) , Lost ( 2 ) ,Num_pore_formed , ( Num_pore_formed > 0) ,N_GzB_Int , pVal ] ;%tOut i s next t ime at which to record s ta te o f system , occurs a t i n t e r v a l s%of ou t I n c rtOut = ou t I n c r ;wh i le ( t < T)%F i r s t check i f system i s ’ stuck ’ , e . g . no 1 ,2 ,3mers , no pores , no%f ree so l u t i o n molecules . In t h i s case noth ing can happen in system ,%but h igher order PFN jmers w i l l d i f f u s e fo reve r . To avoid t h i s useless%s imu la t ion , te rmina te .monomer = sum(N( : , 1 ) ) ;dimer = sum(N( : , 2 ) ) ;t r ime r = sum(N( : , 3 ) ) ;num_free = sum(N( : , n_pore+1) ) + sum(N( : , n_pore+2) ) ;i f ( num_free == 0 && monomer == 0 && dimer == 0 && t r ime r == 0)stuckCount = 1 ;break ;end%Otherwise , proceed according to SSSA%Next reac t i on w i l l be i n subvolume at top o f event t r eesv = Q(1 ,1 ) ;%W i l l occur a t t ime t from subvolume at top o f event t r eet = Q(1 ,2 ) ;%State vec to r f o r t ha t subvolumeNsv = N( sv , : ) ;%To ta l i n t e r a c t i o n p ropens i t i e s . Ca lcu la ted as above .a0 = R( sv , 1 ) ;%Tota l p ropens i t ys0 = R( sv , 3 ) ;%Choose i f event i s reac t i on or d i f f u s i o ni f ( rand < a0 / s0 )%React ion158%Propens i ty vec to r f o r t ha t subvolume ( i t was ca l cu la ted when an event%l a s t occured t ha t i n f l uenced the subvolume ) . Or , i n the case of%ICs , i t was ca l cu la ted dur ing i n i t i a l i z a t i o n .a_sv = A( sv , : ) ;num_pore_prior = Nsv ( n_pore ) ;%Determine reac t i on t ha t occursrxn_ ind = f i n d (cumsum( a_sv ) > rand∗a0 , 1 , ’ f i r s t ’ ) ;%Track i n t e r n a l i z e d GzBi f ( rxn_ ind == n_rxns )N_GzB_Int = N_GzB_Int + 1 ;end%Update species numbers , and update s ta te mat r i xNsv = Nsv + S( rxn_ind , : ) ;N( sv , : ) = Nsv ;num_pore_post = Nsv ( n_pore ) ;%Track maxmer[Temp, p fn_o l i _p rod ] = max(S( rxn_ind , 1 : n_pore ) ) ;i f ( p fn_o l i _p rod > maxmer )maxmer = p fn_o l i _p rod ;end%Recalcu la te p ropens i t i e s f o r t h i s subvolumea_sv = ca lc In t rxnPropens (Nsv , n_pore ,K, K_pfn_ins ,Kg) ;d_sv = ca lcD i f fP ropens (Nsv , n_pore , D_Pfn_jmer , D_Pfn_eff , D_GzB_eff , l ) ;%Update propens i t y matr icesA( sv , : ) = a_sv ;D( sv , : ) = d_sv ;%Update ra te mat r i xa0 = sum( a_sv ) ;d0 = sum( d_sv ) ;s0 = a0 + d0 ;R( sv , : ) = [ a0 d0 s0 ] ;%Track t ime of pore fo rmat ioni f ( ( num_pore_post − num_pore_prior ) > 0)T_pore_form = [ T_pore_form t ] ;Num_pore_formed = Num_pore_formed + 1;end%Calcu la te next event t ime f o r t h i s c e l l based on updated s ta tetau = −log ( rand ) / s0 ;%Working i n absolu te time , so next event t ime i s added to t ime t ha t159%already occuredtnex t = t + tau ;%Update event t r ee[Q, Qarray ] = updateQ (1 , tnex t ,Q, Qarray ) ;e lse%D i f f u s i on%D i f f u s i v e propens i t y f o r subvolumed_sv = D( sv , : ) ;%To ta l d i f f u s i v e p ropens i t i e sd0 = R( sv , 2 ) ;%species t ha t d i f f u sesspecies_ind = f i n d (cumsum( d_sv ) > rand∗d0 , 1 , ’ f i r s t ’ ) ;%Determine which adjacent subvolum