Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Exploring the tumour microenvironment with non-invasive Magnetic Resonance Imaging techniques Moosvi, Firas 2020

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

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata


24-ubc_2020_may_moosvi_firas.pdf [ 80.77MB ]
JSON: 24-1.0388623.json
JSON-LD: 24-1.0388623-ld.json
RDF/XML (Pretty): 24-1.0388623-rdf.xml
RDF/JSON: 24-1.0388623-rdf.json
Turtle: 24-1.0388623-turtle.txt
N-Triples: 24-1.0388623-rdf-ntriples.txt
Original Record: 24-1.0388623-source.json
Full Text

Full Text

Exploring the tumour microenvironment withnon-invasive Magnetic Resonance Imaging techniquesbyFiras MoosviMSc, University of Toronto, 2012BSc, University of British Columbia, 2009A THESIS SUBMITTED IN PARTIAL FULFILLMENTOF THE REQUIREMENTS FOR THE DEGREE OFDoctor of PhilosophyinTHE FACULTY OF GRADUATE AND POSTDOCTORALSTUDIES(Physics)The University of British Columbia(Vancouver)February 2020c© Firas Moosvi, 2020The following individuals certify that they have read, and recommend to the Fac-ulty of Graduate and Postdoctoral Studies for acceptance, the thesis entitled:Exploring the tumour microenvironment with non-invasive MagneticResonance Imaging techniquessubmitted by Firas Moosvi in partial fulfillment of the requirements for the degreeof Doctor of Philosophy in Physics.Examining Committee:Stefan Reinsberg, Physics & AstronomySupervisorPiotr Kozlowski, Physics & AstronomySupervisory Committee MemberVesna Sossi, Physics & AstronomyUniversity ExaminerRobert Rohling, Electrical and Computer EngineeringUniversity ExaminerAdditional Supervisory Committee Members:Corree Laule, Department of Pathology & Laboratory MedicineSupervisory Committee MemberAndrew Minchinton, Department of Integrative OncologySupervisory Committee MemberiiAbstractThis thesis comprises development and application of several MRI techniques toimprove our understanding of tumour growth, drug distribution, and drug effectusing pre-clinical tumour models in mice. In the first part of the thesis, a novelhigh molecular weight contrast agent, HPG-GdF is introduced. This molecule isa hyperbranched polyglycerol labeled with an MRI contrast agent (Gd-DOTA) aswell as a fluorescent tag. After injecting the agent into mice within an MRI scanner,contrast-agent kinetics were quantified using a two-parameter linear model andvalidated with quantitative immunohistochemistry via direct fluorescence imagingof HPG-GdF.HPG-GdF was used to assess whether vascular function plays a role in howa chemotherapy (Herceptin) distributes within a tumour. Tumour vessel perme-ability and fractional plasma volume were quantified using the HPG-GdF and norelationship was found between vascular function and presence of drug. HPG-GdFwas then applied to show that Avastin (an antiangiogenic agent) decreased ves-sel permeability in tumours. Using histological methods, a dramatic reduction inhypoxia (oxygen deficiency in tissues) was observed in treated tumours. Unfor-tunately, existing MRI methods to evaluate oxygenation were time-intensive andlacked sensitivity. In the second part of this thesis, we introduce, develop, validate,and apply a new method to assess tumour oxygenation using MRI.Oxygen (O2) is a paramagnetic molecule that shortens the longitudinal relax-ation time (T1) of protons in MRI. This subtle effect has been widely reported in theliterature but its applications in cancer have been limited. Our technique - dynamicoxygen-enhanced MRI (dOE-MRI) - uses T1W signal intensity images acquiredduring a cycling gas challenge (air or oxygen) and independent component analysisiii(ICA). Hypoxia staining with pimonidazole correlated strongly with dOE-MRI val-ues in a murine tumour model (SCCVII) and only weakly in a colorectal xenograftmodel (HCT-116). Finally, we provide compelling evidence that treatment withAvastin improves tumour oxygenation in subcutaneous tumours. With dOE-MRI,the sensitivity and speed of existing techniques was greatly improved. Since ourtechnique requires no injectable contrast agent, special sequences or hardware, weanticipate that this technique can be quickly translated into the clinic.ivLay SummaryIn this thesis we have described the development and application of several mag-netic resonance imaging (MRI) techniques to improve our understanding of tumourin mice. In the first part of the thesis, a new analysis method was outlined to char-acterize tumour blood vessels with a novel bio-compatible contrast agent. In thesecond part of this thesis, we introduced, developed, validated, and applied a newmethod to assess tumour oxygenation using MRI. Our technique is a significant im-provement over other available methods as it uses a statistical technique to extractvery small changes in MRI signal just by inhaling 100% oxygen gas. We used thistechnique to show that treatment with a cancer drug that prunes malformed and ab-normal blood vessels in tumour improves oxygenation levels. Since our techniquerequires no injectable contrast agent, special sequences or hardware, we anticipatethat this technique can be quickly translated into the clinic.vPrefaceThe work presented in this thesis relies on development of MRI sequences, pre-clinical animal experiments, immunohistological staining and analysis, as wellas post-acquisition MRI data analyses in collaboration with multiple researchers.Since different researchers played a part in different aspects of the works, thepreface is divided by chapter to ensure adequate credit is provided to the peo-ple responsible. In general, Dr. Jennifer Baker working in the lab of Dr. AndrewMinchinton at the BC Cancer Research centre provided all of the tumour xenograftbearing mice used in experiments, and contributed extensive biological expertiseand knowledge to the works presented in this thesis. All animal experimental pro-cedures in this thesis were carried out in compliance with the guidelines of theCanadian Council for Animal Care and were approved by the institutional AnimalCare Committee (A17-0042, A16-0105, A13-0053).Chapter 2 - MRI and histology of vascular function in xenografts usingHPG-GdFWork presented in chapter 2 has been published in its entirety in Contrast Media &Molecular Imaging with the manuscript titled “Multi-modal magnetic resonanceimaging and histology of vascular function in xenografts using macromolecularcontrast agent hyperbranched polyglycerol (HPG-GdF)” [1]. The author of thisthesis is listed as the 3rd author on this manuscript.Drs. Baker and Reinsberg initiated a collaboration with Drs. Katayoun Saatchiand Urs Hafeli from the Faculty of Pharmaceutical Sciences. Drs. Saatchi andHafeli had synthesized a new hyper-branched poly-glycerol molecule (HPG-GdF)that was biocompatible and could be used as a high molecular weight MRI contrastviagent. During her masters degree at UBC, Dr. Kelly McPhee began work on thisproject to characterize the molecule. She performed all pilot and phantom experi-ments to test and measure its relaxivity, as well as initial pilot testing, QA to assessthe viability of the molecule as an MRI contrast agent, as well as acquisition andanalysis of pilot animal data. The work described above is not a part of this thesis,but the development and validation was essential and provided a springboard forthis author’s work.Following initial characterization of the molecule, in collaboration with Dr.Baker and Dr. Reinsberg, Dr. McPhee also acquired and analysed animal data from10 mice. Raw and some processed MRI data (notably the calculated concentration-time curves) from this experiment were used in this manuscript by this author. Theraw data was used to develop and present a new method to quantify HPG-GdFconcentration-time curves using a two-parameter linear model. To develop thismethod, several follow-up experiments with a new MRI sequence and improvedanalysis methods was done in collaboration with Drs. Baker and Reinsberg - thisdata is not part of this chapter but is instead presented in Chapters 3 and 4. Dr.Reinsberg provided much guidance and support in the design of the experiment,development of the sequence, and in troubleshooting the myriad problems thatarose during sequence development.Initial draft of the manuscript was prepared primarily by Dr. Baker based onthe final chapter of Dr. McPhee’s MSc. thesis, with this author contributing to thenew MRI methods, followed by the MRI analysis, as well as the results and dis-cussion sections. In developing this manuscript, all analysis code was written bythis author and all parameteric maps were generated using custom code written inPython. Code to read in generic MRI data was developed by Dr. Reinsberg withsome assistance from Andrew Yung, a scientific engineer at the 7T MRI. All histol-ogy data presented in this paper was collected and processed by Dr. McPhee underthe supervision of Dr. Baker, and all final figure preparation including manuscriptsubmission was done by Dr. Baker. Reviewer comments were addressed collabo-ratively with Dr. McPhee, Dr. Baker, Dr. Reinsberg, and this author equally.In this chapter, this author’s novel contributions include:• Development of the two-parameter model to extract aPS and fPVvii• Interpreting model parameters (in collaboration with Drs. Reinsberg andBaker)• Implementing pharmacokinetic modelling of DCE-MRI data using the Tofts,Extended Tofts, and 2 compartment exchange model in pythonChapter 3 - Applications of HPG-GdF: Investigating distribution oftrastuzumab in the tumour microenvironmentWork presented in chapter 3 has been published in its entirety in Clinical & Ex-perimental Metastasis with the manuscript titled “Heterogeneous distribution oftrastuzumab in HER2-positive xenografts and metastases: role of the tumour mi-croenvironment” [2].The author of this thesis is listed as the 4th author on this manuscript. As thismanuscript was part of a much larger study led by Dr. Baker, only the sectionsthat pertain to MRI data and analysis have been reproduced here. This authorcontributed to the experimental design of the study, collected and analyzed all MRIdata presented in this study, and assisted with figure preparation. Dr. Baker wrotethe first draft of this manuscript excluding the MRI methods and results, with Dr.Reinsberg and this author assisting with editing and writing of the methods, results,and discussion sections that pertained to MRI.In this chapter, this author’s novel contributions include:• Application of previously developed methods and models to an experimentwith a biological questionChapter 4 - Applications of HPG-GdF: Assessing vascular normalizationusing an antiangiogenic chemotherapyDr. Baker had the initial idea to use HPG-GdF as a macromolecular contrast agentto assess response of anti-angiogenic agents, designed the experiments, preparedthe tumours, and conducted the interventions. Dr. Baker performed all the histolog-ical analysis including sectioning, staining, imaging, and cropping of the tumours.This author contributed to the experimental design of the study, collected and anal-ysed all MRI data presented in this study, generated all the figures, and wrote allviiiof the text. Dr. Baker and Dr. Reinsberg were present during the imaging data col-lection. Interpretation of results - particularly of histological images - was donecollaboratively by this author, Dr. Reinsberg, and Dr. Baker.In this chapter, this author’s novel contributions include:• Application of previously developed methods and models to an experimentwith a biological questionChapter 5 - Oxygen-enhanced MRIWith the exception of Sections 8.3, 5.5.8, and 5.5.6 material presented in this chap-ter was published in the Journal of Magnetic Resonance in Medicine [3].This author was the lead investigator, responsible for all major areas of MRIdata collection, analysis, manuscript composition. Dr. Baker initially approachedus with the biological need to assess tumour hypoxia non-invasively using MRIand completed the immunohistochemistry staining and analysis for this project. Dr.Reinsberg was the supervisory author on this project and was involved throughoutthe project in concept formation and manuscript composition; he also came up withthe initial idea to apply un-supervised machine learning techniques to our data. Dr.Martin McKeown provided assistance in understanding the utility of ICA in thegiven context.In this chapter, this author’s novel contributions include:• Development and optimization of the dOE-MRI imaging sequence• Development of the cycling gas delivery method• Implementation of ICA applied to 4-dimensional spatial and temporal data• Interpretation of the dOE-MRI maps• Development of a metric for assessing ICA-extracted components• Comparison of dOE-MRI maps to standard correlation maps• Development of a method to compare dOE-MRI between experimentsixChapter 6 - Validation of oxygen-enhanced MRI in animalsSections 6.3.1, 6.3.3 have also been published as part of the first OE-MRI manuscript [3].Sections 6.1.1, 6.2.5, and 6.3.2 pertain to work that was completed after the firstpublication and will be combined with the work in Chapter 7 for a new manuscript.In this chapter, this author’s novel contributions include:• modelling the oxygen response using a lognormal velocity distribution in avasculature network• fitting, extracting, and interpreting model parameters• quantitatively validating the the MRI- and histo-derived oxygenation metrics• exploring the limits of dOE-MRI applicability by deploying across multipletumour modelsChapter 7 - Applications of oxygen-enhanced MRIThis work was done in collaboration with Dr. Baker. This author responsible forall aspects of MRI data collection, analysis, interpretation, manuscript composi-tion. Dr. Baker’s initially expressed the need to assess tumour oxygenation aftermodulating it with an anti-angiogenic compound. Dr. Baker also processed theimmunohistochemistry staining, produced the images, proposed the drug, and wecollaboratively designed the experiments. Dr. Reinsberg was the supervisory au-thor on this project and was involved throughout the project in concept formationand manuscript composition. Both Drs. Baker and Reinsberg were instrumental inediting of the manuscript and providing guidance on data presentation, visualiza-tion. A version of this chapter will be submitted as a manuscript as a stand-aloneMRI intervention paper. Portions of this chapter are also being prepared as partof a larger body of work on increasing radiation sensitivity of tumours using anti-angiogenic agents with Dr. Baker as principal author.In this chapter, this author’s novel contributions include:• applying a VEGF-inhibitor to improve tumour oxygenation and characteriz-ing this change using dOE-MRIx• exploring the differences in oxygenation between tumours implanted subcu-taneously and intramuscularlyChapter 8 - Future WorkAll analysis presented in this chapter was conducted solely by this author and con-tributions for the data collected has already been reported in previous chapters.In this chapter, this author’s novel contributions include:• Implementation of independent vector analysis• Implementation of Group ICA• Analysis of sequential air/O2 switches• Plan to incorporate T∗2 in a multi-gradient echo dOE-MRI sequencexiTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiiList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xviiiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xixGlossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxxiiAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxxv1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Cancer biomarkers and targets . . . . . . . . . . . . . . . . . . . 11.2 Animal models of cancer . . . . . . . . . . . . . . . . . . . . . . 41.3 Need for non-invasive imaging . . . . . . . . . . . . . . . . . . . 51.4 Principles of Magnetic Resonance Imaging . . . . . . . . . . . . 61.4.1 MRI sequences . . . . . . . . . . . . . . . . . . . . . . . 91.4.2 Paramagnetic contrast agents . . . . . . . . . . . . . . . . 111.4.3 Dynamic contrast-enhanced MRI (DCE-MRI) . . . . . . . 131.5 Thesis structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 14xii2 MRI and histology of vascular function in xenografts using HPG-GdF 162.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192.2.1 Mice and tumours . . . . . . . . . . . . . . . . . . . . . 192.2.2 Contrast agents and dosage . . . . . . . . . . . . . . . . . 202.2.3 MRI acquisition . . . . . . . . . . . . . . . . . . . . . . 202.2.4 MRI data analysis . . . . . . . . . . . . . . . . . . . . . 222.2.5 Histology . . . . . . . . . . . . . . . . . . . . . . . . . . 252.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262.3.1 HPG-GdF accumulates in tumour tissue in the extravascu-lar space but does not distribute far from vasculature . . . 262.3.2 Bolus arrival time (BAT) for HPG-GdF as a screen for vi-able tissue . . . . . . . . . . . . . . . . . . . . . . . . . . 262.3.3 Fractional plasma volume (fPV) and apparent permeability-surface area product (aPS) as measures of vascular function 282.3.4 Variable vascular function in HCT116 and HT29 humancolorectal xenografts . . . . . . . . . . . . . . . . . . . . 282.3.5 MRI analysis of HPG-GdF in HCT116 and HT29 xenografts:BAT, fPV and aPS . . . . . . . . . . . . . . . . . . . . . 322.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393 Applications of HPG-GdF: Investigating distribution of trastuzumabin the tumour microenvironment . . . . . . . . . . . . . . . . . . . . 403.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423.2.1 Reagents . . . . . . . . . . . . . . . . . . . . . . . . . . 423.2.2 Mice and tumours . . . . . . . . . . . . . . . . . . . . . 423.2.3 MRI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423.2.4 Immunohistochemistry . . . . . . . . . . . . . . . . . . . 433.2.5 Image acquisition and analysis . . . . . . . . . . . . . . . 433.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44xiii3.3.1 Measures of vascular density, architecture and function donot consistently correlate with heterogeneous patterns oftrastuzumab distribution . . . . . . . . . . . . . . . . . . 443.3.2 Dynamic vascular permeability and blood volume mea-surements do not consistently relate to patterns of trastuzumabdistribution . . . . . . . . . . . . . . . . . . . . . . . . . 453.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 494 Applications of HPG-GdF: Assessing vascular normalization usingan antiangiogenic chemotherapy . . . . . . . . . . . . . . . . . . . . 514.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534.2.1 Mice tumours, and treatment groups . . . . . . . . . . . . 534.2.2 MRI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534.2.3 Immunohistochemistry . . . . . . . . . . . . . . . . . . . 544.2.4 Image acquisition and analysis . . . . . . . . . . . . . . . 544.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 554.3.1 Treatment with B20 reduces tumour hypoxia and HPG-GdF72 hours after treatment . . . . . . . . . . . . . . . . . . 554.3.2 Blood vessel permeability (aPS) and HPG-GdF accumula-tion (fluorescence intensity) decreases in tumours treatedwith B20 . . . . . . . . . . . . . . . . . . . . . . . . . . 564.3.3 HPG-GdF enhancement curves and AUC60 are altered af-ter treatment, but fPV does not change . . . . . . . . . . . 584.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 584.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 635 Oxygen-enhanced MRI . . . . . . . . . . . . . . . . . . . . . . . . . 655.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 655.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665.2.1 Physiology . . . . . . . . . . . . . . . . . . . . . . . . . 665.2.2 Origin of the OE-MRI signal . . . . . . . . . . . . . . . . 67xiv5.3 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 695.4 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 715.4.1 Mice and tumours . . . . . . . . . . . . . . . . . . . . . 715.4.2 MRI data acquisition . . . . . . . . . . . . . . . . . . . . 725.4.3 MRI data analysis . . . . . . . . . . . . . . . . . . . . . 725.4.4 Quality Scores . . . . . . . . . . . . . . . . . . . . . . . 745.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 755.5.1 ICA isolates small changes in T1W signal intensity . . . . 755.5.2 Extracting periodicity of the signal intensity change usinga Fourier transform based approach is less sensitive thanusing ICA . . . . . . . . . . . . . . . . . . . . . . . . . . 755.5.3 dOE-MRI with ICA does not require assumption of a re-sponse function . . . . . . . . . . . . . . . . . . . . . . . 775.5.4 Variability of response in individual oxygen cycles . . . . 785.5.5 Quality of extracted component . . . . . . . . . . . . . . 805.5.6 dOE-MRI can be extended by interleaving other scans be-tween each repetition . . . . . . . . . . . . . . . . . . . . 805.5.7 Exploring other independent components extracted usingICA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 815.5.8 Comparing oxygen responsiveness with dOE-MRI acrossexperiments . . . . . . . . . . . . . . . . . . . . . . . . . 825.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 855.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 876 Validation of oxygen-enhanced MRI in animals . . . . . . . . . . . . 896.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 896.1.1 Theory: Modelling oxygen response . . . . . . . . . . . . 896.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 926.2.1 Animals . . . . . . . . . . . . . . . . . . . . . . . . . . . 926.2.2 Immunohistochemistry . . . . . . . . . . . . . . . . . . . 926.2.3 MR Imaging . . . . . . . . . . . . . . . . . . . . . . . . 936.2.4 dOE-MRI Analysis . . . . . . . . . . . . . . . . . . . . . 936.2.5 Model Fitting . . . . . . . . . . . . . . . . . . . . . . . . 93xv6.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 946.3.1 ICA enabled dOE-MRI detects variable oxygenation in arange of tumour models . . . . . . . . . . . . . . . . . . 946.3.2 Vascular area A and mean velocity v f do not vary acrosstumour models, but σ f differentiates between tumours . . 956.3.3 dOE-MRI maps correspond to matched histology sections 966.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 976.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1047 Applications of oxygen-enhanced MRI . . . . . . . . . . . . . . . . . 1057.1 Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1057.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1057.2.1 VEGF inhibition and bevacizumab . . . . . . . . . . . . . 1067.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1077.3.1 Animals . . . . . . . . . . . . . . . . . . . . . . . . . . . 1077.3.2 Immunohistochemistry . . . . . . . . . . . . . . . . . . . 1077.3.3 MR Imaging . . . . . . . . . . . . . . . . . . . . . . . . 1077.3.4 dynamic oxygen-enhanced MRI (dOE-MRI) analysis . . . 1087.3.5 Experiment Summaries . . . . . . . . . . . . . . . . . . . 1087.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1097.4.1 Subcutaneously implanted SCCVII tumours treated withB20 are more responsive to oxygen than controls . . . . . 1097.4.2 IM tumours have a higher baseline oxygenation level thanSC tumours from the same cell line . . . . . . . . . . . . 1107.4.3 Effects of B20 are dependent on the tumour microenviron-ment and baseline oxygenation . . . . . . . . . . . . . . . 1107.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1127.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1178 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1198.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1198.2 Group ICA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1198.3 Further investigation of the characteristic oxygen response curve . 121xvi8.4 dOE-MRI maps of 10 consecutive air/O2 switches are stable . . . 1228.5 Exploring the link between perfusion and oxygenation . . . . . . 1248.6 Expanding dOE-MRI to include T∗2 . . . . . . . . . . . . . . . . . 1268.6.1 Independent vector analysis . . . . . . . . . . . . . . . . 1278.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130xviiList of TablesTable 2.1 Summary of the quantitative histological histological data forthe HCT116 and HT29 tumours. . . . . . . . . . . . . . . . . 31Table 2.2 Summary of MRI parameters derived from pharmacokinetic mod-elling of the Gadovist contrast agent for both the HCT116 andHT29 tumours. . . . . . . . . . . . . . . . . . . . . . . . . . . 32Table 2.3 Summary of MRI parameters derived from applying a linearmodel to the HPG-GdF contrast kinetics for both the HCT116and HT29 tumours. . . . . . . . . . . . . . . . . . . . . . . . 32Table 7.1 Summary of scan parameters for the experiments used in thisstudy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109xviiiList of FiguresFigure 1.1 Graphical illustration of the hallmarks of cancer as presentedby Hanahan and Weinberg [4]. Many of the targets describedare inaccessible to non-invasive imaging and in this thesis, wefocus on angiogenesis as the target of our imaging methods.Figure used with permission from Elsevier Inc. . . . . . . . . 2Figure 1.2 Schematic of the normal tissue (left) and tumour (right) vascu-lature network. Note the hierarchical structure of oxygenatedblood (red) passing through the arteries, arterioles, and deoxy-genated blood leaving via the venules, veins. In tumours, thisstructure is severely compromised and often, no clear flow pat-terns can be distinguished with many vessels ending in deadends or looping back onto feeding vessels. . . . . . . . . . . . 3Figure 1.3 A) a collection of protons with magnetic moments are repre-sented by arrows pointing in the direction of the magnetic mo-ment, starting from a common starting point (centre). B) Afterswitching on a main magnetic field ~B0, the net magnetic mo-ment ~M slightly aligns with ~B0) because a larger fraction ofspins point in the direction of the main magnetic field (in therotating frame). C) The net magnetization moment is tippedto the transverse axis with an RF pulse ~B1 so the signal canbe measured. Annotations were added to the simulated imagesproduced by Hanson et al.([5]), used with permission from Wi-ley and Sons. . . . . . . . . . . . . . . . . . . . . . . . . . . 8xixFigure 1.4 Dependence of flip angle on the signal from an SPGR sequencefor two tissues with T1 of 2200 and 1760 ms respectively. TheErnst angles for each tissue is marked with dotted lines, and theflip angle that would result in the maximum contrast differenceis marked with a solid black line. . . . . . . . . . . . . . . . . 11Figure 1.5 On the left, the contrast (∆SI) for two tissues with T1s of 2200msand 2090ms for three TRs are plotted along with the Ernst an-gles for each TR. As the TR increases, the total available sig-nal also increases because the magnetization has longer to re-cover. On the right, the curves are normalized by the TR whichis more useful in optimizing sequence parameters. The threechosen repetition times (240ms in black, 120ms in red, and60ms in grey) represent the range of TRs used in the work pre-sented in this thesis. . . . . . . . . . . . . . . . . . . . . . . 12Figure 1.6 Extended Tofts Model . . . . . . . . . . . . . . . . . . . . . 14Figure 2.1 HPG-GdF. The 583 kDa globular Hyperbranched PolyGlyc-erol (HPG) molecules are derivatized with p-NH2-benzyl-DOTA(Macrocyclics) at 20 µg Gd per mg HPG (approximately 300chelates per molecule) and tagged with Alexa Fluor 647 dye, aspreviously described [6]. Figure reused with permission fromWiley and Sons. . . . . . . . . . . . . . . . . . . . . . . . . . 21xxFigure 2.2 MR-derived parameters to measure vascular function using HPG-GdF:bolus arrival time (BAT), fractional plasma volume (fPV) andapparent permeability-surface area product (aPS). (A) Exam-ple of how the control theory procedure was applied to deter-mine BAT on a sample enhancement curve showing change inconcentration of CA as a function of scan time. The green lineis the central line while the red lines indicate the upper andlower control limits (LCL and UCL). (B) HPG-GdF enhance-ment curve from which the fPV is derived as the concentrationat the start relative to the plasma concentration. The aPS isthe slope of the enhancement after the bolus arrival. The con-centration curves are shown as ratio of tissue concentration toblood peak concentration (24 µM). Insets within the plots areschematics of our interpretation of the situation at particulartime-points and are not measured values. (C) The fraction ofMR-measured HPG-GdF BAT-enhancing voxels has a nega-tive association with the proportion of necrotic tissue deter-mined in histological sections. Figure reused with permissionfrom Wiley and Sons. . . . . . . . . . . . . . . . . . . . . . . 23Figure 2.3 (A) T1-weighted RARE images show signal enhancement at40-min that increases with longer exposures (tumours outlinedin red). (B) Quantitative histology shows HPG-GdF extravasa-tion and distribution gradually increasing with time (2-min to7 days). (C) Whole tumour maps of HPG-GdF (red) and vas-culature (blue) show that at early (2 min) timepoints HPG-GdFis primarily overlapped with vasculature, but by 60-min thereis substantial heterogeneity, where some vessels have greateramounts of perivascular HPG-GdF than others. Figure reusedwith permission from Wiley and Sons. . . . . . . . . . . . . . 27xxiFigure 2.4 Individual HT29 tumour (T01-05) slice maps are presented forparameters derived using Gadovist (BAT, AUGC60, Ktrans),HPG-GdF (BAT, fPV, aPS, AUGC60) and histological imag-ing of necrosis. Figure reused with permission from Wiley andSons. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Figure 2.5 fPV and aPS parameter maps for an HT29 tumour (T01) areshown with corresponding histological image depicting CD31-stained vessels (blue) and HPG-GdF native fluorescence (red).The high fPV region has notably greater vascular density, andHPG-GdF is clearly seen overlapping with vessels (black) oraccumulating in the extravascular space. A region with highaPS that does not correspond to high fPV is magnified (A) andcompared with a high fPV region that does not correspond tohigh aPS (B). Figure reused with permission from Wiley andSons. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30Figure 2.6 HPG-GdF distribution in HCT116 and HT29 colorectal tu-mour xenografts at 60-min: histological analysis. (A) Thewhole-slice average fluorescence intensity shows that HPG-GdFaccumulates to a greater degree and distributes further awayfrom vasculature in HT29 tumours. (B) Sample images showHPG-GdF (red) overlaid on CD31 (blue); insets also displaynuclear density (grey). A greater degree of HPG-GdF accumu-lation can easily be appreciated in the HT29 tumours. Figurereused with permission from Wiley and Sons. . . . . . . . . . 33Figure 2.7 Vascular function in HCT116 xenografts. Whole-slice mapsare presented for individual HCT116 tumours (T01-T05, ver-tical columns) for parameters derived from MR imaging ofGadovist (BAT, AUGC60, Ktrans), MR imaging of HPG-GdF(BAT, fPV, aPS, AUGC60) and histological imaging (necro-sis). Figure reused with permission from Wiley and Sons. . . 37xxiiFigure 3.1 A) Magnified region of a BT474 xenograft treated with 10mg/kg trastuzumab for 24 h. Carbocyanine fluorescent dye(cyan) around CD31 stained vessels (blue) indicates patency;non-patent vessels are indicated as red arrows. Trastuzumabextravasates from vessels heterogeneously, with many patentvessels showing no extravascular bound trastuzumab (green ar-rows) even when adjacent patent vessels do have perivasculartrastuzumab. B) An anatomical RARE MR image with thetumour is shown alongside an AUC60 map using Gadovist inMDA-MB-361 tumours. The AUC60 map is compared withslice-matched histology sections of bound trastuzumab (pur-ple), vascular architectural markers αSMA and CIV (both shownin red). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46Figure 3.2 fPV and aPS parameter maps are compared to matched his-tology sections stained for bound trastuzumab (magenta), andfor HER2 (grey), carbocyanine marker of perfusion (cyan) andfor CD31 vasculature (blue). Areas of vascular function (MRI)and trastuzumab (histology) correlation are indicated (orangearrows) in both modalities; example areas of poor matching arealso shown (purple arrows). Stars indicate location of fiducialmarkers for multi-modal slice comparison. . . . . . . . . . . 47Figure 4.1 Histological sections of a control (left) and treated tumour (right)are shown to illustrate the dramatic decreases both in pimonida-zole staining and in accumulation of HPG-GdF in the treat-ment group. . . . . . . . . . . . . . . . . . . . . . . . . . . . 55Figure 4.2 Median pimonidazole staining is markedly reduced for the B20-treated group (Ipimo = 17.2) compared to the controls (Ipimo =21.7). Median HPG-GdF staining intensity is also reducedfor the treated tumours (Ihpg = 12.7) compared to the controls(Ihpg = 11.0). P-values from the Mann-Whitney U test werep = 5× 10−5 (pimo, left) and p = 8× 10−6 (HPG-GdF fluo-rescence, right). . . . . . . . . . . . . . . . . . . . . . . . . . 56xxiiiFigure 4.3 Histological sections shown for 6 control and 6 treated tu-mours, with four slices per tumour. All sections are stainedfor pimonidazole (green) and HPG-GdF (red). . . . . . . . . . 57Figure 4.4 Summary of group differences for all three DCE-MRI param-eters: AUC60, aPS, and fPV. There is a statistically significantreduction in AUC60 and aPS for the treated tumours, but no dif-ference measured for fPV. P-values from the Mann-WhitneyU test for the comparisons were p = 0.03 (AUC60), p = 0.01(aPS), and p= 0.15 (fPV). . . . . . . . . . . . . . . . . . . . 59Figure 4.5 Histological stains of HPG-GdF shown alongside approximateslice matched DCE-MRI parameter map of aPS and the group-averaged contrast enhancement curves of control and treatedtumours. There is an overall reduction in aPS for the treated tu-mours. The mean contrast enhancement curves (group meansin dark blue and red lines; shaded region is the 95% confi-dence interval determined by bootstrapping) also show that thetreated tumours have a higher enhancement slope after contrastagent injection. . . . . . . . . . . . . . . . . . . . . . . . . . 60Figure 5.1 Sigmoidal curve illustrating the relationship between the haemoglobinsaturation (y-axis) and the oxygen tension (x-axis). When theoxygen tension is low, the Hb easily binds O2 and there is arapid rise in oxygen saturation (green arrow). Note that it takesa large increase in oxygen tension to bind the last O2 and sim-ilarly, a large decrease in oxygen tension to release the last O2(purple arrow) [7]. . . . . . . . . . . . . . . . . . . . . . . . 67xxivFigure 5.2 A schematic of the change in the partial pressures of oxygenand carbon dioxide at various points in the body. The pO2 ofinhaled ambient air is 160 mmHg, and this is breathed in to thelungs. Oxygen diffuses out of the alveoli into the surround-ing capillaries and binds to Hb due to the pressure gradient(capillary pO2 is 40 mmHg). This oxygenated blood (pO2 =100mmHg) now enters the heart and is pumped through thebody, with the Hb releasing oxygen through capillaries dueagain to the pressure gradient (tissue pO2 is < 40 mmHg) [8].Textbook content produced by OpenStax Biology is licensedunder a CC-BY 4.0 license. . . . . . . . . . . . . . . . . . . 68Figure 5.3 A schematic representation of our current understanding of theorigin of the OEMRI effect. In normoxic tissue, Hb is al-most fully saturated and any excess breathed O2 cannot bindto the Hb molecule. Consequently, O2 dissolves in the bloodplasma and as the excess oxygen diffuses out into the tissue,it also dissolves in the interstitial tissue fluid resulting in a netT1 decrease. It is hypothesized that in the hypoxic tissue, Hbis not fully saturated with oxygen due to increased tissue de-mands and/or a poorly organized vascular network. The excessbreathed oxygen in this case binds to the Hb molecule and doesnot dissolve in the plasma leading to no change in T1. . . . . 70Figure 5.4 (A) T2W MRI of a tumour xenograft at 7T and (B) the corre-sponding T1W signal-time traces of a single voxel (solid yel-low) and whole-tumour slice ROI (dotted black) during gascycling at two-minute intervals of air (x axis; blue) and O2 (xaxis;yellow). (C) Plot of the four extracted ICA componentsfrom the entire tumour ROI, component c4 (purple) exhibitsthe same temporal features as the oxygen cycling time courseshown along the bottom. All components are normalized, novertical scale is shown. Figure reused with permission fromWiley and Sons. . . . . . . . . . . . . . . . . . . . . . . . . . 76xxvFigure 5.5 Histogram of all the normalized intensities at the frequency ofinterest in all the tumour voxels. The intensities were normal-ized to the value at the frequency of interest extracted from theICA component (5.67), plotted on the x-axis. . . . . . . . . . 77Figure 5.6 The normalized intensities at the frequency of interest as a spa-tial map. Higher values in the spatial map are represented asgreen and low intensity values are dark. . . . . . . . . . . . . 77Figure 5.7 (A) dOE-MRI map of an SCCVII tumour where purple vox-els contribute strongly to the extracted component using ICAin the T1W signal timecourses. Green voxels in the dOE-MRImap have a strong contribution of the inverse extracted com-ponent. Pearson’s r-maps are shown correlating the raw time-signal voxel by voxel with a square wave (B), and an exponen-tial convoluted with a square wave called the hemodynamicresponse function (HDRF) (C). Panels B and C are correlationmaps whereas A is the dOE-MRI map from ICA. Note thelow correlation coefficients (on the order of 10−3) are charac-teristic of the extremely low amplitude of the oxygen cyclingcompared to other competing effects. Figure reused with per-mission from Wiley and Sons. . . . . . . . . . . . . . . . . . 78Figure 5.8 The first row shows the extracted ICA component of a wholetumour. The second row shows the magnitude of the Fourier-transformed signal plotted against frequency in Hz. The high-est frequency is kept and all other values are set to 0.This fil-tered data is inverse Fourier transformed to produce the Fourier-filtered response curve. This response curve is plotted in thethird plot. A correlation map (colour bar ranging from -0.01to +0.01) of the Fourier-filtered response curve with the mean-normalized signal intensity is shown alongside the dOE-MRImap (colour bar ranging from -0.15 to +0.15) for comparison. 79xxviFigure 5.9 The dOE-MRI map including the full dataset of all three cy-cles (A) is compared to each of the three gas cycles separately(B,C,D), and to a map that temporally undersamples by select-ing every third datapoint from the full dataset (E). Voxel-wiseplots of each map are correlated to the full dataset and a lin-ear regression with Pearson’s r is shown. Figure reused withpermission from Wiley and Sons. . . . . . . . . . . . . . . . 80Figure 5.10 All ninety-one dOE-MRI extractions shown with the qualityscores (1=unusable to 5=ideal) and the number of tumoursshown in parentheses. Colours represent the different scoreswith 1 in dark green, 2 in orange 3 in blue, 4 in purple, and 5in lime green. . . . . . . . . . . . . . . . . . . . . . . . . . 81Figure 5.11 dOE-MRI maps and associated component traces of differentlysampled data. To achieve different levels of subsampling, theraw data was spliced and then ICA was applied. The oxy-genation maps look very similar between different subsamplefactors. Temporal resolution and number of points were 4.3 sand 200 points (subsample 1), 8.5 s and 100 points (subsample2), 12.8 s and 67 points (subsample 3), 17.1 s and 50 points(subsample 4), 21.3 s and 40 points (subsample 5), and 25.12 sand 34 points (subsample 6). . . . . . . . . . . . . . . . . . . 82Figure 5.12 Plots of the four components extracted from ICA are shown((||ci|| = 1,∀i) along with the corresponding weighting fac-tor maps (normalized to mean voxel wise mean signal inten-sity). Corrupting influences such as temperature drifts are of-ten present and produce slowly increasing or decreasing trends(for e.g., c1) and breathing artefacts corresponding to short-lived spikes (c3). No explanation could be found for c2. Figurereused with permission from Wiley and Sons. . . . . . . . . 83xxviiFigure 6.1 Schematic representation of the complex vascular network (left)with multiple velocity profiles (arrows) and its equivalent rep-resentation as a single vessel with a distribution of flow profiles(right). The distribution of velocity profiles of the represen-tative single vessel is the lognormal distribution. Schematicrepresentation re-created from Hudson et al. [9]. . . . . . . . 90Figure 6.2 Fit of equation 6.6 to an oxygen response curve with the resid-uals plotted along the x-axis for each point. For this fit, A =0.18, v f = 0.66mm/s, and σ f = 0.58mm/s. . . . . . . . . . . 94Figure 6.3 Top: dOE-MRI maps for four tumour models HCT-116, BT-474, SCCVII, and SKOV3 are shown. Chosen slices are repre-sentative of the mean percent negative dOE-MRI fraction forthe respective tumour model. Bottom: The box-whisker plotshows the quartiles of percent negative dOE-MRI voxels forall imaged tumours. Figure reused with permission from Wi-ley and Sons. . . . . . . . . . . . . . . . . . . . . . . . . . . 95Figure 6.4 Top: Individual fits to the oxygen response curve for each ani-mal and tumour type. Bottom: Box plots showing the medianvalue and quartiles for A, v f , and σ f across the three tumourmodels. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96Figure 6.5 The proportion of negative dOE-MRI voxels is plotted againstthe histological hypoxic fractions with Pearson’s r = 0.76 forSCCVII tumours (r = 0.14 after excluding tumour with highhypoxic fraction) and r = 0.037 for HCT-116 tumours. Eachpoint is a slice average. Figure reused with permission fromWiley and Sons. . . . . . . . . . . . . . . . . . . . . . . . . 97Figure 6.6 SCCVII murine tumours with slice-matched histological im-ages depicting pimonidazole-labeled hypoxia (green) and CD31-stained vasculature (purple) are shown next to the dOE-MRIparameter maps similarly colored with O2-positive (purple)and O2-negative (green) areas. Corresponding ICA extractedcomponents are also shown. Figure reused with permissionfrom Wiley and Sons. . . . . . . . . . . . . . . . . . . . . . 98xxviiiFigure 6.7 HCT-116 human colorectal xenografts with slice-matched his-tological images depicting pimonidazole-labeled hypoxia (green)and CD31-stained vasculature (purple) are shown next to thedOE-MRI parameter maps similarly colored with O2-positive(purple) and O2-negative (green) areas. Corresponding ICAextracted components are also shown. Figure reused with per-mission from Wiley and Sons. . . . . . . . . . . . . . . . . . 99Figure 7.1 A) NCWF maps obtained from Independent Component Anal-ysis (ICA) (dOE-MRI maps) are shown for control tumoursas well as those treated with 5mg/kg B20 and imaged 48 hourslater. Of the 10-16 slices for each animal, a representative slicewas chosen. As indicated by the distribution of purple vox-els, control tumours show considerably less response to oxy-gen than the treated tumours. Additionally, regions markedin green are considered to be hypoxic; these regions were notprevalent in the treated tumours. B) A representative histologyslice from a control and a treated tumour is shown stained withpimonidazole (green) and CD31 (purple). . . . . . . . . . . . 111Figure 7.2 A) Group differences of the normalized mean NCWF are shownin a boxplot. Each dot represents the mean value of a mousewith the controls in blue and treated in green. The differ-ences are statistically significant (p=0.0092) with a large effectsize (Hedge’s g = 1.08). B) Density distributions of all voxelsshows treated tumours shifting towards increased responsive-ness to delivered oxygen (higher NCWF). . . . . . . . . . . . 112Figure 7.3 Calculated tumour volumes from each of the four groups isshown. There were no statistically significant differences intumour volumes amongst any of the groups. . . . . . . . . . . 113xxixFigure 7.4 A) Boxplot with four groups, 5mg/kg B20 treated and controlmice with both SC and IM tumours. Differences between con-trol SC and IM tumours, as well as control SC and treated SCtumours are statistically significant (Mann-Whitney U test; p<0.005, marked as **). There was no significantly differencein treated and control IM tumours. B) Voxel density distri-butions of NCWF for SC (top) and IM (bottom) treated andcontrol tumours. Density distribution (i.e. normalized his-tograms) are shown rather than voxel counts due to unevengroup size. Note the shift of the IM control tumours towards ahigher NCWF value. . . . . . . . . . . . . . . . . . . . . . . 114Figure 7.5 Representative histological sections from 16 total tumours ofall four groups: 5mg/kg B20 treated and control mice withSC and IM tumours. Hypoxia marker pimonidazole stainingis shown in green, and purple indicates the presence of bloodvessels stained by CD31. . . . . . . . . . . . . . . . . . . . . 114Figure 8.1 Comparison of the standard ICA technique and Group ICA.The main difference is in the pre-processing of the MRI datacomprising spatial coordinates (x,y,z) and temporal informa-tion (t). Group ICA datasets are prepared by spatially concate-nating all animals together (n). The output of the ICA tech-niques also differs: in standard ICA each application producesa set of independent components whereas in Group ICA onlya single set of independent components is produced. . . . . . 120Figure 8.2 Extracted oxygen enhancing component from ICA applied tothe spatially concatenated cohort data. . . . . . . . . . . . . 121xxxFigure 8.3 Two animals were selected to explore the effect of the num-ber of components (m) on the dOE-MRI maps. The two an-imals presented here were selected to exhbit the full range ofvariability in extracted components. The low variability exam-ple shows no discernible difference in the extracted componentanywhere from m = 3 to m = 12. The high variability exam-ple shows considerably more noise in the extracted component,but the same overall trend. The corresponding dOE-MRI mapsfor both the low variability and high variability examples showalmost no difference in the oxygenation maps. . . . . . . . . . 122Figure 8.4 Results of a dOE-MRI-based analysis to sequentially analyzeten consecutive air-oxygen switches (D). The averaged dOE-MRImap (A) across all 10 cycles reveals some hyperintense re-gions corresponding to oxygen-responsive areas. The voxel-wise standard deviation (B) and coefficient of variation (C) ofthe ten dOE-MRI maps shows some variability at the top of thetumour as well as on the mid-right of the tumour. . . . . . . . 125Figure 8.5 dOE-MRI maps and DCE-MRI AUC60maps and slice-matchedhistology sections of SCCVII and HCT-116 tumours. Large re-gions marked as purple in the dOE-MRI maps are O2-positiveand also correspond to regions that have high AUC60 values(yellow). Green or O2-negative regions from the dOE-MRImap are often consistent with unperfused regions in the AUC60(black), but there are regions of mismatch. Histology imagesstained with pimonidazole (green) and CD31 (purple) are shownfor corresponding sections. . . . . . . . . . . . . . . . . . . 126Figure 8.6 Schematic of the current and proposed acquisition and analysisfor dOE-MRI with combined R1 and R∗2 imaging. . . . . . . 128xxxiGlossaryAIF Arterial input functionaPS Apparent permeability surface area productαSMA alpha smooth muscle actin - a vascular architectural markerAUC Area under the CurveAUGC60 Area under the gadolinium-concentration curve 60 seconds after injec-tionBAT Bolus arrival timeBBB Blood brain barrierBOLD Blood oxygen level dependent MRIBT474 Cell line derived from a human mammary gland (breast/duct) cancerCA Contrast agentCD31 cluster of differentiation 31CIV Collagen IV - a marker of basal lamina collagen, a vascular architecturalmarkerDa Dalton, a unit of measurement for molecular weightDCE-MRI dynamic contrast-enhanced MRIDCE-US dynamic contrast-enhanced ultrasoundxxxiidOE-MRI dynamic oxygen-enhanced MRIDOTA A chemical compound - also known as tetraxetan - that serves as a chelat-ing agent to prevent toxicity of gadolinium ions.EPR Enhanced permeability and retentionFastICA An ICA algorithm implemented in the scipy.sklearn v0.17.1python package.FLASH Fast low angle shot magnetic resonance imaging.fMRI functional magnetic resonance imaging.fPV Fractional plasma volume - a DCE-MRI parameter extracted using a highmolecular weight agentGd-DTPA Gadolinium ion chelated to Diethylenetriamine Pentaacetic Acid (DTPA),an organic chelating moleculeHb HaemoglobinHER2 Human epidermal growth factor receptor 2HPG-GdF Hyperbranched polyglycerol (HPG)i.p. Intraperitoneal, injections take place in the body cavity of an animali.v. Intravenous, injections are administered through a vein of an animalICA Independent Component AnalysisIM Intramuscular, injections are administered in the muscle of an animalIVA Independent vector analysisKtrans Volume transfer constantkDa kilo dalton, or 1000 daltonsMAb Monoclonal antibodyxxxiiiMCA Macromolecular contrast agents.MDA-MB-361 Cell line derived from a breast cancer that metastasized to thebrainMGE Multi gradient echoMW Molecular weightNCWF Normalized component weighting factor valueOE-MRI Oxygen-enhanced magnetic resonance imagingPF Perfused fractionPS Permeability surface area productRARE RARE sequence is a rapid acquisition with refocused echoes. This se-quence is also known known as fast spin echo (FSE) or turbo spin echo(TSE)ROI Region of interestSC SubcutaneousSNR Signal to noise ratioUSPIO Ultrasmall super-paramagnetic iron oxideVEGF Vascular endothelial growth factorve Volume of extravascular extracellular space per unit volume of tissuevp Blood plasma volume per unit volume of tissuexxxivAcknowledgmentsThis work would not be possible without the tireless commitment and mentoringby my friend and supervisor, Dr. Stefan Reinsberg. Stefan, I could not have donethis without you - you made my journey everything that it was. Dr. Baker was alsoinstrumental in helping run, plan, and provide support for all of the work presentedhere - thank you Jenn! Members of my supervisory committee, Drs. Kozlowski,Laule, and Minchinton were all excellent sources of advice, feedback, guidance,and general project direction. Particular thanks to Dr. Kozlowski for supportingand organizing weekly lab meetings giving me and other trainees to give and gethonest feedback on preliminary work - I learned a tonne through these meetings.I am fortunate enough to have some truly excellent mentors that have guidedme through my PhD program and introduced me to the world of educational lead-ership. Special kudos to Dr. Simon Bates who has been, and continues to be aninvaluable mentor and supporter of my pursuits in teaching and learning. Every-thing I know about teaching large classes has come from Simon - truly, thank you.Both he and Dr. Ido Roll were instrumental in my initial dabbling in physics educa-tion research and Ido took me under his wing and set me up with a solid foundationin the scholarship of teaching and learning (SoTL). I then met a community of like-minded individuals that I truly connected with on a deep and foundational level -SoTL specialists, thank you for the debates, conversations, and most importantly,your friendship.My first big SoTL project was with Drs. James Charbonneau and Chris Ad-dison and it dealt with the idea of measuring interdisciplinarity in undergraduatestudents, particularly in the Science One program. All together we learned so muchin this project and I will forever be thankful for the opportunity to work on such axxxvfun project! Dr. Georg Rieger gave me my first taste of teaching a class, and I willnever forget the immense amount of trust and faith you placed in me. Many formergraduate students were instrumental in my growth and development, and I owe alot to Natasha Holmes, Ellen Schelew, Amanda Parker, Mike Sitwell, Emily Al-tiere, Jonathan Massey, and Jared Stang - you folks were amazing and I truly lookup to you! Past, current, and former members of PHASER, thank you for givingme a place to feel at home discussing everything related to physics education.Finally, last but certainly not least, a huge note of gratitude goes to my parents,brothers, family, and of course my wife Hira, who has supported and humoured methrough my meanderings over the course of my PhD. I love you all.xxxviChapter 1IntroductionCancer is a disease that can occur at any age but mostly affects Canadians fiftyyears and older. Based on estimates from 2010, 49% of men and 45% of womenare expected to develop a type of cancer during their lifetimes and one in fourCanadians are expected to die from cancer-related diseases [10]. One approachto achieving better patient outcomes and reducing toxicity in cancer patients isto improve our understanding of tumour progression and treatment in preclinicaltumour models using biological markers, also called biomarkers. In 2001, theNational Institutes of Health commissioned a working group to create standardsand definitions for what would constitute an effective biomarker. A biomarkeris “a characteristic that is objectively measured and evaluated as an indicator ofnormal biological processes, pathogenic processes, or pharmacologic responsesto a therapeutic intervention” [11]. There is an urgent need for the developmentof new imaging biomarkers to aid in the development of more targeted tumourtherapies [12]. There has been considerable interest in the potential of predictivebiomarkers for early assessment of tumour therapies. In this thesis, we aim todevelop imaging-based biomarkers to explore the tumour microenvironment.1.1 Cancer biomarkers and targetsHanahan and Weinberg catalogued a vast array of factors that contribute to tumourgrowth and provided a framework for understanding of diseases that result in ab-1normal cell growth [4, 13]. These “hallmarks of cancer” are “essential alterations incell physiology that collectively dictate malignant growth in tumours” [13]. Thesehallmarks are summarized in Figure 1.1 and the work presented in this thesis fo-cuses on the 5th hallmark, angiogenesis - the process by which new blood vesselsform.Figure 1.1: Graphical illustration of the hallmarks of cancer as presented byHanahan and Weinberg [4]. Many of the targets described are inaccessi-ble to non-invasive imaging and in this thesis, we focus on angiogenesisas the target of our imaging methods. Figure used with permission fromElsevier Inc.Angiogenesis is the formation of new blood vessels from pre-existing ones isa normal and vital process in the body tightly regulated by various cell signallingpathways and growth factors. In tumours, angiogenesis is a critical step in thegrowth and spread of tumours as new blood vessels are recruited from the existingvascular network to promote rapidly accelerated and abnormal tumour growth [14].2Normal Vasculature Tumour VasculatureFigure 1.2: Schematic of the normal tissue (left) and tumour (right) vascula-ture network. Note the hierarchical structure of oxygenated blood (red)passing through the arteries, arterioles, and deoxygenated blood leavingvia the venules, veins. In tumours, this structure is severely compro-mised and often, no clear flow patterns can be distinguished with manyvessels ending in dead ends or looping back onto feeding vessels.Normally, this process is regulated by several angiogenic and antiangiogenic fac-tors such as αβ integrin, vascular endothelial growth factor (VEGF) and fibroblastgrowth factor [15]. In tumours however, this process is deregulated (Figure 1.2)and excess production of growth factors from rapidly proliferating tumour cellsleads to a drastic increase in angiogenesis. These newly formed vessels are un-stable growth patterns of blood vessels in tumours are often described as abnormalwith a defective and leaky endothelium [16]. Irregular diameters of tumour vessels,abnormal branching patterns and leaky vessel walls all contribute to an increase invessel permeability. It is estimated that a single hole larger than 0.5µm in diameterwould alter the permeability of that vessel significantly enough to result in soluteextravasation to be limited by blood flow [16]. Disorganized and inefficient bloodflow also limits the delivery of macromolecules, such as chemotherapeutic agentsvia the blood. Poor perfusion in the tumour due to a disorganized vascular networkimpairs the delivery of systemic drugs to the whole tumour and ultimately, reducesefficacy.Tumour angiogenesis is extremely important in tumour growth, progression3and metastasis and is a promising target for novel therapies [17]. For instance,measuring tumour angiogenesis has the potential to serve as a highly predictiveprognostic marker for disease outcome and treatment. Measurement of microvesseldensity using histology is generally considered an independent prognostic factor inseveral cancers [18] but has several limitations. Histology requires biopsy samplesand patient comfort aside, biopsies only sample a small fraction of the affectedorgan. The lack of functional information from biopsies as well as the practicalchallenges of obtaining longitudinal biopsy samples make non-invasive imaginga promising technique to complement and potentially reduce unneeded biopsies.Angiogenesis is especially suitable for analysis with MRI due to its exquisite soft-tissue contrast, ability to image deep into the body, and finally its utility in assessingvascular function using injectable contrast agents.1.2 Animal models of cancerDeveloping models of human cancers in mice that predict clinical outcomes is ben-eficial as failure of novel chemotherapy drugs is often not determined until signifi-cant investments of time and money have been made designing and implementingphase I, II and III clinical trials [? ]. Animal models of cancer are essential for drugdevelopment, investigating mechanisms of action of physical phenomena, identify-ing gene targets, and protocol development prior to translating to human patients.According to Ce´spedes et al., an ideal cancer model should have the followingcharacteristics [? ]:1. Share histopathological features with the human tumour,2. Progress through the same stages and result in the same physiological andsystemic effects,3. Share the same genes and biochemical pathways in both tumour initiationand tumour progression,4. Reflect the response of the human tumour to a particular therapy and5. Predict therapeutic efficacy in human clinical assays.4Despite this, it is widely acknowledged that the translation potential of subcu-taneous tumour xenograft models is limited particularly for translating chemother-apies to humans that have been shown effective in mice [? ? ]. Naturally arisingtumours often take years to months or years to grow but mouse xenograft modelsare selected so that experiments can be completed in days or weeks [? ]. The tu-mour vasculature that forms from injected tumour cells is fundamentally differentfrom tumours that arise in vivo as it lack the architectural and cellular complexityoften seen in real tumours [? ]. Vessels in xenograft models typically grow muchfaster, are more chaotic in structure, have a leakier endothelium, and do not havemuch smooth muscle to regulate blood flow [? ? ].The heterogeneity of tumours also makes translation to humans difficult aswithin a specific tumour, several cell subpopulations exist that differ in their mor-phology, growth rate, receptor status and sensitivity to therapeutic potential [? ? ].There are also regional differences in pH, degree of oxygenation, nutrient concen-tration that result in scattered pockets of hypoxia, apoptosis and necrosis through-out the tumour [? ? ? ]. Vascular reorganization, particularly following treatment[?? ], leads to an irregular and shifting tumour microenvironment, a moving targetfor imaging modalities [15]. However, measuring these changes functionally, lon-gitudinally, and non-invasively with imaging techniques has the potential to greatlyimprove our understanding of the tumour microenvironment. Care must be takennot to over-interpret and generalize results from preclinical cancer studies. Sig-nificantly more work needs to be done to obtain clinically relevant animal cancermodels, but in the interim, animals models provide researchers with a valuableplatform for developing novel agents of potential targets.1.3 Need for non-invasive imagingNon-invasive imaging methods are proving indispensable for studying angiogen-esis in vivo as they provide researchers with quantitative information about bloodflow, vascular permeability, vessel density, vessel function and blood volume [19].Imaging modalities such as computed tomography (CT), magnetic resonance imag-ing (MRI), positron emission tomography (PET), single photon emission computedtomography (SPECT) and ultrasound (US), have all been proposed for studying5angiogenesis [15]. Each modality is optimal for probing a particular aspect ofbiomarkers. To study angiogenesis and its effects on tumour growth and treatmentresponse, the tumour environment needs to be probed using minimally invasiveimaging techniques. Nuclear medicine techniques such as PET and SPECT employradiotracers that can be measured at picomolar concentrations but at a significantlylower spatial resolution. DCE-MRI and DCE-CT offer similar perfusion measure-ments (rate of leakage and leakage space) as both rely on the administration of acontrast agent that diffuses from the vasculature. DCE-CT is advantageous as it hasa direct linear relationship between the contrast agent concentration and the imageintensity (attenuation numbers, given by Houndsfield Units) [20]. The disadvan-tage of CT however is that it requires ionizing radiation and iodinated contrastagents used in CT have been shown to have worse safety profiles compared to MRcontrast agents [21]. MRI can also be used to measure additional information suchas diffusion, tissue oxygenation, spectroscopy, chemical exchange and magnetiza-tion transfer. In this thesis, several MRI techniques will be explored in a bid toimprove our understanding of the tumour microenvironment. We begin with somebasic principles of MRI.1.4 Principles of Magnetic Resonance ImagingIn biological specimens, water is by far the most abundant molecule in the bodyand the hydrogen atoms (1H) in water are central to MR imaging. Other MR-active nuclei include 13C, 19F, 23Na and 31P, but these are rare and not often used.The molecular mass of a water molecule (H2O) is approximately 18 g/mol and itsdensity is 1 g/mL so in 1L, there are approximately 3×1025 molecules of water.At the atomic level, each molecule of water consists of one oxygen atom (eightprotons, neutrons, and electrons) as well as two hydrogen nuclei (a neutron andproton). An intrinsic quantum mechanical property of fundamental particles suchas the proton, neutron, and electron is that they posses angular momentum. Thereare two types of angular momenta, spin and orbital angular momentum. The protonand neutron possess only spin angular momentum but electrons also possess orbitalangular momentum. For electrons the two angular momenta nearly always cancelout in the lowest energy state of a chemically stable molecule such as water [22].6The hydrogen nucleus has an odd number of protons (n=1) so there is a net spinangular momentum. In fact, the only sources of angular momentum in the groundstate are the molecular rotation and the nuclear spins associated with the proton andneutrons [22]. Nearly all of the MR signal being measured in the body is derivedfrom the hydrogen nuclei in water.To summarize, each proton in the hydrogen nucleus has spin angular momen-tum, is charged and thus has a net magnetic moment. Though there is no analogyto this from a classical physics perspective, one can imagine the net magnetic mo-ment of a proton as a close cousin to the classical situation of the magnetic fieldgenerated by a loop of current in a wire. We will model the hydrogen atom witha net magnetic moment as a small bar magnet spinning on its own axis (with anarrow vector representing the direction and strength of the magnetic moment) andrely on classical physics to describe the principle of magnetic resonance imaging.If a spinning bar magnet is placed in an external magnetic field, the magnetic mo-ment vector of the bar magnet will precess, or rotate about the new magnetic fieldwith a frequency known as the Larmor frequency:~ω = γ ~B0 (1.1)The proportionality factor γ is the gyromagnetic ratio and is nuclei-dependentand for protons, γ = 42MhZ/T . For convenience it is useful to change our ref-erence frame to a rotating reference frame so the the magnetic moment vector isstationary on a (rotating) cartesian axis.The quantity of interest in MRI is the net magnetic moment ~M, and this is thesummation of all individual magnetic moments present in the hydrogen nuclei. ~Mis the quantity that is measured and ultimately leads to the images produced. Fig-ure 1.3A shows a schematic of the situation; for visualization, individual magneticmoments from the protons are localized to originate from the same central point.Since the water molecules are tumbling around due to thermal motion, the protonmagnetic moments are oriented randomly they are pointed in nearly every direc-tion and there is no net magnetic moment (Figure 1.3A). If we now put these watermolecules into an MRI scanner and switch on a main magnetic field of strength~B0 = 7 Tesla, there is a slight tendency of protons to align with the main magnetic7A) B) C)⃗M = 0 ⃗M = Mz = M0⃗B0 = 0 ⃗B0 > 0 ⃗B0 > 0⃗B1 > 0Mx,y = 0Mx,y > 0Mz < M0Figure 1.3: A) a collection of protons with magnetic moments are repre-sented by arrows pointing in the direction of the magnetic moment,starting from a common starting point (centre). B) After switching ona main magnetic field ~B0, the net magnetic moment ~M slightly alignswith ~B0) because a larger fraction of spins point in the direction of themain magnetic field (in the rotating frame). C) The net magnetizationmoment is tipped to the transverse axis with an RF pulse ~B1 so the sig-nal can be measured. Annotations were added to the simulated imagesproduced by Hanson et al.([5]), used with permission from Wiley andSons.field~B (along z axis, see Figure 1.3B), and a net magnetization vector ~M is present.~M aligns with ~B0 and the longitudinal component Mz = M0 while the transversecomponent Mx,y = 0 (in the x-y plane). ~M is many orders of magnitude smallerthan the external magnetic field so the MR signal cannot be measured when it isaligned with the external main magnetic field ~B0). Applying a radiofrequency (RF)pulse ~B1 at the Larmor frequency results in a torque applied to ~M, causing it to ‘tip’down into the transverse (x-y) plane (Figure 1.3C).Interacting nuclei exchange energy with both the surrounding environment(spin-lattice interaction) as well as neighbouring nuclei (spin-spin interaction), and~M relaxes back to its equilibrium value. The time constant of the recovery of Mzto its equilibrium value ~M0 is characterized by the time T1,Mz =M0(1− e−t/T1) (1.2)T1 s after the RF pulse, the magnetization value has recovered to ≈ 63% (1-8e−1) of its equilibrium value. Prior to the ~B1 pulse, the transverse component ofthe initial magnetization Mx,y was 0. Following the ~B1 pulse, Mx,y decays from itsmaximum value of M0 to 0 through the interactions between nuclei and is charac-terized by the time constant T2 (also called spin-spin relaxation).Mxy =M0e−t/T2 (1.3)Although T1 and T2 values are affected by various factors including field-strength, and local environmental factors such as temperature, proton concentra-tion, and molecular mobility. Differences in T1 and T2 values are used to generatecontrast between different tissues. For example, in a study conducted with ten vol-unteers at 1.5T, the spleen (T1 = 919 ms), liver (T1 = 616 ms), muscle (T1 = 785ms), fat (T1 = 239 ms), and renal cortex (T1 = 919 ms) all had measurably differentT1 values [23]. Contrast between tissues can be generated by weighting images tohighlight differences between T1, T2, and proton density. In the next section, thesequence of choice for T1-weighted images is described.1.4.1 MRI sequencesT1-weighted images can be produced using both a spin echo pulse sequence aswell as a gradient echo. In a spin echo sequence, the magnetization is first flippeddown to the transverse plane with a 90◦ RF pulse. Then the magnetization in thetransverse plane is allowed to gradually de-phase and subsequently, a 180◦ RFpulse is applied causing the spins to re-phase. Once the spins re-phase, an echo isproduced and the time from the initial 90◦ to the eventual echo is the echo time, TE .The 180◦ refocusing pulse is applied TE /2 after the initial 90◦ pulse. Specifyingthe TE allows control of image weighting - either T1, T2 or mixed weighted. Therepetition time TR is another important sequence parameter that controls how muchmagnetization, or signal is available to be flipped down to the transverse axis. Therepetition time is the time between subsequent 90◦ RF pulses - the longer the TR,the more the longitudinal magnetization recovers and is available to be tipped downto the transverse plane at the next 90◦ RF pulse.In a gradient echo sequence, only one RF pulse is needed to flip the magnetiza-tion to the transverse plane, and unlike the spin echo sequences, gradients are used9to de-phase and re-phase the transverse magnetization. The interpretation of TE issimilar in a gradient echo and also measures the time between the 90◦ pulse andthe echo produced after gradient re-phasing. Compared to the spin echo sequence,the use of gradients permits significantly shorter echo times and repetition times -TE and TR respectively. This is why, in typical DCE-MRI experiments, gradientechoes are preferred because signal acquisition is significantly faster allowing rapidand dynamic imaging. To ensure there is no transverse magnetization after eachrepetition of the gradient echo sequence, spoiling is needed to suppress creationof spin and stimulated echoes. The signal from a spoiled gradient echo (SPGR)sequence with spoiling and after steady state has been achieved is given by [? ]:S= ksinα(1− e−TR/T1)(1− (cosα)e−TR/T1)e−TE/T ∗2 (1.4)where k is a proportionality constant that also includes the proton density. Inpractice, if TE is small and close to T∗2, this term (as it often is in SPGR sequences)approaches unity and is negligible. The signal characteristics of a spoiled gradientecho (SPGR) sequence is also shown in Figure 1.4 with a specific T1 of 2200 ms(typical T1 values in tumours studied in this thesis) and with a TR of 120 ms.The maximum signal intensity under these conditions is given by the Ernstangle, obtained by setting the derivative dS/dα = 0:α = arccos(e−TR/T1) (1.5)It is important to note that the Ernst angle maximizes the signal at a particularflip angle given a specific intrinsic T1 but often it is necessary to optimize the signaldifference between two (or more) species with different T1 values. The differencein signal intensities is the contrast (∆SI) between two tissue types. The maximumcontrast difference for two tissues with T1 values of 2200 and 1760 ms is markedwith a solid black line in Figure 1.4. In Figure 1.5, the contrast (∆SI) and theTR-normalized contrast (∆SI /√TR) is shown as described by Busse [? ]. Fromthis figure - particularly the TR-normalized contrast - we note that choice of TR islargely independent of optimizing contrast and there exists an optimal choice offlip angle regardless of the TR. This is important because very often, other factors100 20 40 60 80Flip Angle ( ) = 2200 msT1 = 1760 msMaximum contrast difference Ernst AnglesSignal characteristics of an SPGR sequence (TR = 120 ms)Figure 1.4: Dependence of flip angle on the signal from an SPGR sequencefor two tissues with T1 of 2200 and 1760 ms respectively. The Ernstangles for each tissue is marked with dotted lines, and the flip angle thatwould result in the maximum contrast difference is marked with a solidblack line.such as SNR, temporal, and spatial resolution constrain the minimum TR.For the sequences used in this thesis, competing factors were considered in-cluding imaging time, T1-weighting, image intensity, SNR, image contrast, tem-poral, and spatial resolutions; appropriate compromises and trade-offs were madeto balance the trade-offs to produce maximum benefit. Another tool at our dis-posal is the paramagnetic contrast agents as they are often used to increase the T1contrast between different species. The following sections outline how dynamiccontrast-enhanced MRI or DCE-MRI is used in the imaging of cancer.1.4.2 Paramagnetic contrast agentsParamagnetism is defined as the intrinsic tendency for a material to become mag-netized when placed within a magnetic field. By far the most common elementused as a contrast agent (tracer) in MRI is Gadolinium as it is strongly paramag-110 20 40 60 80Flip Angle ( )0.0000.0020.0040.0060.008Contrast (SI)TR=60msTR=120msTR=240msMaximizing contrast with SPGR  T1 from 2200 to 2090 ms0 20 40 60 80Flip Angle ( )0.0000.0050.0100.015Normalized contrast  (SI / T R)TR=60msTR=120msTR = 240msMaximizing normalized contrast with SPGR  T1 from 2200 to 2090 msFigure 1.5: On the left, the contrast (∆SI) for two tissues with T1s of 2200msand 2090ms for three TRs are plotted along with the Ernst angles foreach TR. As the TR increases, the total available signal also increasesbecause the magnetization has longer to recover. On the right, the curvesare normalized by the TR which is more useful in optimizing sequenceparameters. The three chosen repetition times (240ms in black, 120msin red, and 60ms in grey) represent the range of TRs used in the workpresented in this thesis.netic due to its seven unpaired electrons. Because electrons are much smaller thanprotons but have the spins, they have a significantly higher gyromagnetic ratio. Theunbalanced electrons in the gadolinium shell or bonding orbital result in a strongnet magnetic moment, which interacts with hydrogen nuclei and dramatically re-duces the longitudinal relaxation time T1 (and to a lesser extent T2). Unfortunately,free Gadolinium ions are toxic so they need to be attached to an organic chelatingagent [24].The ability for a contrast agent to affect the T1 relaxation time is given by itsrelaxivity r1, obtained from the following equation:1T1=1T10+ r1[Gd] (1.6)where T1o is the initial T1, prior to the influence of the paramagnetic contrastagent, r1 is the relaxivity of the contrast agent in units of (mM·)−1, and [Gd] isthe contrast agent concentration. It is important to note that all contrast agentsshorten both T1 and T2 but whether their dominant influence is on the transverserelaxation time (T2) or the longitudinal relaxation time (T1) is expressed by therelative strengths of r1 and r2.121.4.3 Dynamic contrast-enhanced MRI (DCE-MRI)Through the use of a paramagnetic contrast agent, DCE-MRI techniques increasecontrast between species whose T1 and T2 times are otherwise very similar. How-ever the true value of DCE-MRI comes from extracting physiologically relevantinformation from the body. In applications of cancer imaging, DCE-MRI has beenextremely successful in diagnostics, treatment monitoring, assessing severity ofpathologies, distingushing between tumour models and types, improving our un-derstanding of tumour metastases, and development of drugs.Health Canada has approved eight gadolinium-based contrast agents for use inhumans and they have molecular weights less than 1 kDa that readily traverses theendothelium but not the cell membrane [25]. This property allows modelling of thevascular dynamics of the tumour but because the contrast agent is small, perfusionand permeability cannot be decoupled without extremely fast imaging and accurateknowledge of the arterial input function (AIF)[26]. Choosing a kinetic model to fitthe data requires some prior knowledge about the organ or system in question. Forinstance, the blood-brain barrier in the brain dramatically alters the contrast agentkinetics. Similarly, in leaky tumours the extravascular contrast agents typicallyused in DCE-MRI leak out (and back in) of vasculature considerably faster than inother tissues. Sourbron et al. postulate that choice of a tracer kinetic model shouldprovide a link between relevant physiological parameters and measured data [26].The most widely used model in DCE-MRI is the extended Toft’s model, whichis valid in highly perfused tissues and weakly vascularized tissues with a well-mixed extravascular extracellular space (ve) [27]. Figure 1.6 provides a graphicaldescription of this two compartment model, and its mathematical representation is:C(t) = vp ·AIF(t)+Ktranse−tKtransve ∗AIF(t) (1.7)where vp is the plasma volume, Ktrans is the volume transfer constant, and theAIF(t) is the arterial input function which needs to be measured independently ofthe contrast agent kinetics in the tissue.In this thesis, DCE-MRI modelling using a traditional small molecule agent(Gd-DTPA is used only briefly in Chapter 2 and the AIF used in that modelling wasmeasured and published by a former lab member [28]. Nevertheless, the concepts13vpPSveContrast AgentBlood VesselInterstitial VolumeCellsFigure 1.6: Graphical description of the Extended Tofts Model. An arterialinput function (AIF) governs the introduction of the tracer (grey circles)in the vascular compartment (pink) via a bolus injection. Contrast agentmolecules exchanges with the extravascular extracellular space (ve, in-terstitial volume in blue) at a rate given by PS, the permeability-surfacearea product.and introduction to DCE-MRI are relevant for several portions of the thesis.1.5 Thesis structureIn Chapter 2 a new macromolecular contrast agent is described and its value indescribing the tumour microenvironment was explored. A two-parameter linearmodel was applied to the contrast agent enhancement curve and measures of vesselpermeability and fractional plasma volume were obtained. These parameters werethen used to distinguish between two tumour models. In Chapter 3, this techniquewas applied to determine whether molecule size played a role in the distributionof a high molecular weight anti-cancer drug (trastuzumab). We showed that nei-ther vessel permeability nor fractional plasma volume corresponded to presence ofbound drug (determined via histological staining), indicating other barriers limitdistribution of trastuzumab in vivo. In Chapter 4 we set out to determine whetherour new contrast agent could assess changes in vessel permeability after treatment14with an anti-angiogenic drug. We discovered not only that vessel permeability isindeed reduced after treatment, but also that hypoxia dramatically decreased aftertreatment, as predicted by the vascular normalization hypothesis [29]. This led usto develop a new method for assessing tumour oxygenation in vivo using MRI.In Chapter 5, we outlined how a blind source separation technique increased thesensitivity of existing methods. The technique was validated in Chapter 6 withhistological staining, and we demonstrated utility of a new parameter to separateoxygenation replenishment in different tumour models. Finally in Chapter 7 weshowcase a typical application of the technique: detection of tumour oxygenationimprovements after administering an anti-angiogenic agent. We also showed thatthe tumour implant site has a large bearing on the tumour microenvironment, andno oxygenation improvements are observed if the baseline oxygenation is high. InChapter 8, interesting observations are presented that may be useful starting pointsfor future work in this field.15Chapter 2Multi-modal magnetic resonanceimaging and histology of vascularfunction in xenografts usingmacromolecular contrast agenthyperbranched polyglycerol(HPG-GdF)2.1 IntroductionThe vascular network in tumour tissue is abnormal, often resulting in vessels thathave variable flow rates and high permeability relative to blood vessels in normaltissue [16]. Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI)is a useful tool for non-invasively assessing tumour vasculature by imaging andmeasuring concentrations of CA delivered to tumours by the vessels [30–32]. Theappeal of a dynamic, non-invasive approach for measuring tumour vascular func-tion in the clinic is clear. Such data is applicable in the field of assessing treatmentresponse for vascular targeting therapies [30–32]. In addition to utility as a treat-16ment biomarker, vascular function data may be able to predict which tumours arelikely to respond to therapy [33–35], or which regions or tumours have greater het-erogeneity in their microenvironment [36, 37]. The issue of its limited access foranticancer drugs in solid tumours is significant [38], particularly given the efforts tocreate nanoparticle therapeutics that target the tumour via the EPR effect [39–41].A translatable imaging protocol and suitable contrast agent that yields meaningfuland reproducible biomarkers of vascular function could be widely useful in theseareas of cancer research.Low MW Gd(III)-based, chelated CAs such as Gadovist (MW = 605 Da) ex-hibit short half-lives and rapid renal clearance [42]. With the exception of braintissue containing a functional blood-brain barrier, low MW Gd agents currently inclinical use diffuse across the vascular endothelium in most normal and neoplastictissues. Therefore, a well-established shortcoming of low MW contrast agents isthe difficulty of attributing local signal enhancement specifically to either vascularperfusion or vessel permeability. The ability to characterize physiologically rele-vant biomarkers of vascular function is desirable for studying the effects of anti-angiogenic treatments in tumours (described in a comprehensive review regardingDCE-MRI and anti-vascular therapies in cancer [30]).Development and application of macromolecular contrast agents (MCAs) at-tempt to improve upon DCE-MRI assessment of vascular function by relying onthe primarily intravascular nature of MCAs. High MW contrast agents and thera-peutics unable to diffuse across the endothelium selectively extravasate from largepores and inter-endothelial cell gaps that characterize the abnormal vessels in thetumour microenvironment [16, 43]. MCAs commonly used in cancer research in-clude albumin, dextran polymers, and dendrimers conjugated with DTPA/DOTA-Gd3+ chelates as described by Tang et al. [44]. Size of MCAs is often reportedboth in molecular weight (kDa) and longest diameter (nm) where available. MCAsunder development range considerably, but dextran-based agents have been pro-posed from 15 kDa-3820 kDa [31]. Particles less than 5 nm in size have beenfound to leak rapidly from tumour vasculature, whereas those in the 5-8 nm rangeare limited to leaking from hyperpermeable vessels; those greater than 8 nm arethought to have minimal leakage [45, 46]. The most commonly used MCAs in pre-clinical research are albumin based (90 kDa); however, these are not translatable17to the clinic due to immunogenicity concerns [47].For intravascular MCAs, signal enhancement is linked to contrast agent con-centration within the tumour blood vessels at early time points after injection.Rates of MCA leakage from hyperpermeable vessels may then be modelled toevaluate permeability, since extravascular accumulation of the agents will manifestas increased enhancement in repeat images [47, 48]. This analysis is dependenton the assumption that the MCAs have unidirectional flow and do not leak backinto the plasma, and that the concentration in the plasma is constant, creating apermeability-limited environment.In this study we investigated a multi-modal, high MW contrast agent, HPG-GdF:hyperbranched polyglycerol molecules doubly labeled with Gd-DOTA and a fluo-rescent marker. HPGs are soluble, globular, asymmetrical, have low immunogenic-ity and are highly biocompatible molecules with low polydispersity [6, 49, 50].HPG has been previously tested as a human serum substitute [51] and as a drugdelivery vehicle due to its versatility as a chemical, such that drugs, Gd chelates,fluorescent and radiolabels may all be attached to it [52]. Many other MCAs, in-cluding Gd-albumin, are highly viscous, which can limit the applicable dose [53].The MW of HPG-GdF can be adjusted for different applications, and a biodegrad-able version of HPG is also available for potential use [52]. The HPG-GdF de-scribed in this study is 583 kDa and 8-10nm in diameter; synthesis of HPG-GdFhas previously been described [6]. A significant advantage of HPG-GdF is thatit is a multi-modal agent, with both fluorescent and Gd-chelate labels that permithistological validation of observations made using MRI, including determining thedegree to which the agent extravasates from the vasculature. Previous studies havealso used 111In as a SPECT label with utility for biodistribution studies [6]. Inthis work, we employed comprehensive histological methods to investigate the mi-croregional location of HPG-GdF in two human colorectal xenograft models, andused this information to interrogate observations made non-invasively using DCE-MR imaging of the same agent in the same tumours.182.2 Methods2.2.1 Mice and tumoursFemale NOD-SCID mice were bred and housed in institutional animal facilities;experiments in this study were approved by the Animal Care Committee of the Uni-versity of British Columbia. Fiducial markers were constructed of PE-50 polyethy-lene tubing (inner diameter, 0.58mm) and filled with paraffin wax and saline, cre-ating an MR-visible interface [35]. Marker tubes were implanted subcutaneouslyin the sacral region of mice, in a craniocaudal orientation, 2 days prior to subcu-taneous implantation of tumours. Both histology and MR slices are imaged in theplane perpendicular to the marker tube to minimize angular differences betweenserial MR image slices obtained over multiple sessions, and for correspondingcryosections in histological processing. HCT116 or HT29 human colorectal car-cinoma cells obtained from the American Type Culture Collection (ATCC) wereimplanted near the fiducial tubes such that the tumours grew around the tubes. Tu-mours were used when diameters reached 8-12 mm. Mice were anaesthetized withisoflurane for the duration of imaging sessions. Animals were positioned supine onthe custom surface coil apparatus fitted with a lid lined by a temperature-controlled,water-filled heating blanket. Body temperature and respiration rate were monitoredthroughout imaging. Following their final scan animals were administered a 35 µLintravenous dose of 0.6 mg/mL carbocyanine (DioC7(3); Molecular Probes, Eu-gene, OR, USA) in 75% dimethylsulfoxide as a fluorescent dye indicator of vesselperfusion in histological measures, and euthanized 5-min after injection. Givenenough time, carbocyanine freely diffuses throughout the tumour tissue but witha limited exposure time (less than 5 minutes), it serves as a good vessel perfusionmarker [80]. Some animals received HPG-GdF and tumours were collected at earlytime points for histological analysis only, with no MR-imaging. Tumours wereembedded and frozen vertically in optimum cutting temperature medium (OCT;Tissue-TEK) using their fiducial markers for guidance.192.2.2 Contrast agents and dosageHyperbranched polyglycerol (Figure 2.1): 583 kDa HPG-GdF was synthesized atthe University of British Columbia (D. Brook’s laboratory, Department of Chem-istry) with a narrow polydispersity of PDI = 1.01 by ring-opening multibranch-ing polymerization of glycidol using dioxane as the reaction medium, accordingto a published procedure [49]. HPG was derivatized with p-NH2-benzyl-DOTA(Macrocyclics, Dallas, TX, USA) at 20µg Gd per mg HPG and tagged with AlexaFluor 647 (Invitrogen Life Technologies, Burlington, ON, Canada) as previouslydescribed (20). The biological half-life of HPG-Gd (with no fluorescent tag) inmice has previously been examined in biodistribution studies and was reported as32.6 h (20). HPG-GdF was administered as a 6 µL/g bolus dose from 100 mg/mL(0.2 mM) using an intravenous (i.v.) catheter; therefore, the administered molardose of HPG was 1.2 nmol/g and that of the chelated Gd(III) was 240-360 nmol/g(determined according to an estimated 200-300 chelates per HPG molecule (20)).Assuming a blood volume of 5% of mouse body weight, peak blood concentrationof HPG-GdF is 24µM. This value has been used to normalize relative tissue con-centrations (Figure 2.2(B)) and to calculate the fractional plasma volume (fPV; seesection 2.2.4). The relaxivity of HPG-GdF has previously been measured and re-ported to be 1075 mM−1s−1, which is approximately 300 times greater than that ofGadovist (3.58 mM−1 s−1) [6]. Gadovist (Bayer Healthcare, Toronto, ON, Canada;607.4 Da) was administered by i.v. catheter as a 5 µL/g bolus dose from 60 mMsolution, for an administered molar dose of Gadovist of 300 nmol/g. All in-scanneri.v. injections were performed using a power injector at a rate of 1 mL/min. An ex-tended fluid line connected to the i.v. catheter secured to the tail vein of the micepermitted remote initiation of injections outside the scanner room. Injections in-cluded a small volume (<40 µL) of heparinized saline followed by the contrastagent, which was then followed by a 20 µL flush of heparinized saline.2.2.3 MRI acquisitionAll MRI experiments were performed at the UBC MRI Research Centre on a 7TBruker BioSpec 70/30 scanner at room temperature with a combination of volume(transmit)/surface (receive) coil. Each imaging session began with axial RARE T2-20!"#$%&'()*Alexa 647Gd-DOTAFigure 2.1: HPG-GdF. The 583 kDa globular Hyperbranched PolyGlycerol(HPG) molecules are derivatized with p-NH2-benzyl-DOTA (Macro-cyclics) at 20 µg Gd per mg HPG (approximately 300 chelates permolecule) and tagged with Alexa Fluor 647 dye, as previously de-scribed [6]. Figure reused with permission from Wiley and Sons.weighted images for morphological reference and precise alignment of imagingplane (RARE factor 8, effective TE = 42.98 ms, TR = 4250ms, FA = 178◦). Insubsequent imaging sessions, slice location and orientation were adjusted to matchprevious sessions.In the first scanning session, T1 and flip angle maps were acquired prior to theDCE-MRI experiment, followed by another T1 measurement. T1 measurementsand flip angle mapping were performed using a multi-slice FLASH variable flipangle experiment (FLASH TR/TE = 500/2.75 ms, FA = 10-200◦ in increments of2110◦, and 215◦). Data were fit simultaneously for T1 and the B1 scaling factor map.DCE-MRI data was collected at 2.24 s time resolution (FLASH; TR/TE = 35/2.75ms; FA = 40; NR = 1200). T1 and DCE-MRI experiments all had identical geom-etry (matrix = 128×64; three slices; voxel size= 0.33×0.297×1.5 mm3; 2.5 mmslice separation). A follow-up T1 measurement was performed (FLASH; TR/TE= 35/2.75; FA = 10◦, 20◦, 30◦, 40◦, 50◦, 60◦, 80◦, 100◦, 120◦) and the B1 scal-ing factor map from the baseline acquisition was used to determine post-contrastT1 (assuming that B1 scaling does not change due to contrast injection). Differ-ence maps of relaxation rates ∆R1 = 1/T1(post-contrast) - 1/T1(precontrast) wereconstructed for a measure of contrast agent concentration. Animals received twoDCE-MRI scans 24–48h apart, with Gadovist administered and scanned at the firstsession and HPG-GdF administered and scanned at the subsequent session withtumours collected for histological processing at about 60-min post administrationof their HPG-GdF. For qualitative assessment of contrast agent distribution T1-weighted RARE images (RARE factor 4, effective TE = 7.5 ms, TR = 1300 ms, FA= 180◦) were acquired after contrast agent administration.2.2.4 MRI data analysisRegions of interest (ROIs) were drawn on T2-weighted RARE images to outlinethe tumour using ImageJ (NIH), and all other MR analysis was performed usingMATLAB (MathWorks, R2009a) and Python. T1 and flip-angle maps were calcu-lated from variable flip-angle data with a slice-profile correction based on simula-tions described by Parker et al. [54]. The same method was extended to providetime-dependent T1 and concentration-time in DCE data series. Areas under thecurves (AUCs) were numerically integrated starting from the bolus arrival time,or, for tumour-averaged AUC, starting from the common injection time point andextended to the indicated time points (1 and 37 min). Bolus arrival time (BAT) isthe time when detectable signal enhancement begins for a voxel (Figure 2.2(A))due to contrast agent arrival. BAT maps from HPG-GdF and Gadovist obtainedfrom the same HT29 xenograft imaged 24h apart show that HPG-GdF is slower toarrive; both CAs arrive most quickly at the tumour margins (Figure 2.2(A)). Basedon the control-chart decision criterion and the Western Electric decision rules from22Figure 2.2: MR-derived parameters to measure vascular function usingHPG-GdF: bolus arrival time (BAT), fractional plasma volume (fPV)and apparent permeability-surface area product (aPS). (A) Example ofhow the control theory procedure was applied to determine BAT on asample enhancement curve showing change in concentration of CA asa function of scan time. The green line is the central line while the redlines indicate the upper and lower control limits (LCL and UCL). (B)HPG-GdF enhancement curve from which the fPV is derived as the con-centration at the start relative to the plasma concentration. The aPS isthe slope of the enhancement after the bolus arrival. The concentrationcurves are shown as ratio of tissue concentration to blood peak con-centration (24 µM). Insets within the plots are schematics of our inter-pretation of the situation at particular time-points and are not measuredvalues. (C) The fraction of MR-measured HPG-GdF BAT-enhancingvoxels has a negative association with the proportion of necrotic tissuedetermined in histological sections. Figure reused with permission fromWiley and Sons.23MATLAB’s statistical toolbox [55], voxel enhancement was detected as a positivechange from baseline signal for three consecutive timepoints (frames) (i) in thesame direction, (ii) starting 5 s before the time of injection or later and (iii) for atleast 10% of all timepoints following initial change of signal away from baseline.Therefore, voxels with a finite BAT and for which at least 10% of the followingintensities were classified as enhancing by the BAT criteria were called enhancingvoxels, whereas all other voxels were labeled as non-enhancing.A former lab member developed a set of criteria based on control theory aspart of their masters thesis [1]. A change from baseline was determined to haveoccurred when any one of the following inclusion criteria were met:1. any point fell outside of 3 SDs from the average baseline concentration, or2. two of three consecutive points fell outside of the 2 SD limit on the sameside of the mean, or3. four of five consecutive points fell outside of the 1 SD line, on the same sideof the centre line, or4. eight consecutive points all fell on one side of the centre line.In the illustrated example (Figure 2.2(A)), a change from the mean was identi-fied at Frame 31 and the two following scans by Rule (1); therefore, the BAT pointwas selected to be the 30th data point (t=67.2s). The centre line (mean of baseline)and upper and lower control limits (±3 SDs) are drawn for reference.Pharmacokinetic modelling of DCE-MRI dataGadovist: The extended Tofts model was used and three parameters resulted: Ktrans,ve, and vp [26]. The arterial input function (AIF) for this model was determinedpreviously by our laboratory using a projection-based method [28] and is a four-mouse population average.HPG-GdF A two-parameter linear model was applied [56]. Two parameterswere used to characterize the MCA time curves: (1) the rapid increase at the timeof injection (related to the fPV) and (2) the slope of the later enhancement (aPS)(Figure 2.2(B)). The relative plasma volume could be determined from the ratio24of concentrations in the voxel of interest to the concentration in whole blood inthe seconds after injection, since extra-vascular spread of the agent was negligibleat this time. The slope of the concentration time curve following contrast arrivalis proportional to the permeability-surface area product (PS) when the assumptionof a permeability-limited environment is valid. However, in the highly variabletumour microenvironment, extravasation of contrast agent may deplete the intra-vascular concentration appreciably in conditions of high permeability, which wouldincrease the relative contribution of perfusion to the composite measure of PS. Tostress that the interpretation of this slope value depends on the assumption of apermeability-limited environment, we term the slope the apparent permeability-surface area product (aPS) [57, 58].2.2.5 HistologyCryosections of 10 µm were obtained along the plane perpendicular to the fidu-cial marker at depths corresponding to MR imaged slices. Sections were imagedfor DiOC7(3) and HPG-GdF native fluorescence and fixed in acetone-methanolfor 10min prior to staining and re-imaging for CD31 and Hoechst 3342, labelingvascular endothelium and cell nuclei, respectively. Sections were imaged as pre-viously described [59] using a system of tiling adjacent microscope fields of viewsuch that images of entire tumour cryosections were captured at a resolution of1.5 µm/pixel. Using both fiducial and anatomical landmarks, histological sectionswere chosen to match the MR slices. Using ImageJ [60] and user-supplied algo-rithms, digital images were superimposed and manually cropped to tumour tissueboundaries; staining artifacts and necrosis were also removed for some analyses.Positive fluorescence for CD31 and DiOC7(3) images was obtained by applyinga threshold, with neighbouring positive pixels grouped as “objects”. The aver-age distance of tissue to the nearest vascular object was reported as a repeatablemeasure of vascular density. Perfused vessel fraction (PF) was calculated as theproportion of CD31-positive objects that had at least 20% overlap with positiveDiOC7(3) or HPG-GdF pixels on the overlaid image. Data for individual tumourswere displayed as mean ± SEM values. HPG-GdF extravasation was assessed byits distance from blood vessels: pixels from the HPG-GdF fluorescence image were25sorted according to their distance from vascular objects and the average HPG-GdFfluorescence intensity was reported.2.3 Results2.3.1 HPG-GdF accumulates in tumour tissue in the extravascularspace but does not distribute far from vasculatureAveraged data from whole HT29 tumour xenograft images obtained using MRI andhistology showed accumulation of HPG-GdF over time (Figure 2.3(A)), as previ-ously described [6]. HPG-GdF fluorescence was detectable within or very nearto CD31-labeled tumour vessels as early as 2-min following contrast agent injec-tion (Figure 2.3(B), (C)). More HPG-GdF fluorescence accumulated in histologicaltumour sections over time, but very little agent was observed at distances fartherthan 40 µm from the vasculature, even at 7 and 15 days (Figure 2.3(B), (C)). By60 min, there was extravascular accumulation of HPG-GdF around some vessels,but considerable inter-vessel heterogeneity was observed. Some vessels showed noHPG-GdF fluorescence (Figure 2.3(C)). Figure 2.3(C) also shows that HPG-GdFaccumulates over several days but does so heterogeneously, and does not distributethrough tumour tissue even after prolonged exposures.2.3.2 Bolus arrival time (BAT) for HPG-GdF as a screen for viabletissueMaps of BAT overlaid on T1-RARE images for both Gadovist and HPG-GdF areshown in Figure 2.4, rows 1 and 2. A pattern of faster contrast enhancement at thetumour margins was consistent for both contrast agents. While Gadovist eventu-ally distributes to all the tissue, many voxels fail to enhance with HPG-GdF withinthe 37-min imaging period. Comparison of BAT-enhancing voxels with a histo-logical image delineating viable versus necrotic tissue (Figure 2.4, Rows 2 and 3)shows that the non-enhancing voxels consistently corresponded to large areas oftumour necrosis. A negative association was seen between the histological necro-sis fraction and the fraction of enhancing voxels for HPG-GdF (Figure 2.2(C)). Forsubsequent analysis of vascular function, only voxels enhancing with HPG-GdF26Figure 2.3: (A) T1-weighted RARE images show signal enhancement at 40-min that increases with longer exposures (tumours outlined in red). (B)Quantitative histology shows HPG-GdF extravasation and distributiongradually increasing with time (2-min to 7 days). (C) Whole tumourmaps of HPG-GdF (red) and vasculature (blue) show that at early (2min) timepoints HPG-GdF is primarily overlapped with vasculature, butby 60-min there is substantial heterogeneity, where some vessels havegreater amounts of perivascular HPG-GdF than others. Figure reusedwith permission from Wiley and Sons.27using the BAT criteria were evaluated.2.3.3 Fractional plasma volume (fPV) and apparentpermeability-surface area product (aPS) as measures ofvascular functionHPG-GdF-enhancing voxels were further characterized for their plasma volume(fPV) by calculating the magnitude of the rapid signal increase after injection andfor their extravascular accumulation (apparent permeability-surface area product,aPS) by measuring the slope of the enhancement curve after the initial increase forthe duration of the scan (0-2000 s). The patterns of high and low fPV and aPSwere often similar to each other, though there are notable differences. A linearregression analysis comparing fPV with aPS for whole-slice averages yields an R2of 0.11, suggesting they are independent of each other. As a detailed example,tumour HT01 has a region of high aPS and low fPV (Figure 2.5(A)), as well asa region exhibiting the opposite, with high fPV and low aPS (Figure 2.5(B)). Thecorresponding histological section seemed to validate these observations, where theregion with high fPV has a greater density of CD31-stained vessels and the regionwith greater aPS has more HPG-GdF in the extravascular compartment. Whilehistological data enables a detailed view of HPG-GdF accumulation, the actual rateof extravasation may only be determined by the dynamic MR data, as illustrated bythe schematic enhancement curve (Figure 2.5(C)). The corresponding Ktrans mapderived from Gadovist concentrations shows high values in both the high aPS andhigh fPV regions of this tumour (Figure 2.4, Column 1). Therefore, both fPV andaPS played important roles contributing to overall tumour vascular function, hadmeasurable intra-tumour heterogeneity and produced data that was distinct fromand more informative than DCE-MRI derived parameters for Gadovist.2.3.4 Variable vascular function in HCT116 and HT29 humancolorectal xenograftsHPG-GdF accumulated to a greater degree in HT29 colorectal xenografts rela-tive to HCT116, and this was seen at all distances from vessels (Figure 2.6(A)).HCT116 and HT29 colorectal xenografts grow at similar rates in mouse modelsbut exhibit distinct vascular function parameters (data summarized in Table 2.1).28Figure 2.4: Individual HT29 tumour (T01-05) slice maps are presented forparameters derived using Gadovist (BAT, AUGC60, Ktrans), HPG-GdF(BAT, fPV, aPS, AUGC60) and histological imaging of necrosis. Figurereused with permission from Wiley and Sons.29Figure 2.5: fPV and aPS parameter maps for an HT29 tumour (T01) areshown with corresponding histological image depicting CD31-stainedvessels (blue) and HPG-GdF native fluorescence (red). The high fPVregion has notably greater vascular density, and HPG-GdF is clearlyseen overlapping with vessels (black) or accumulating in the extravas-cular space. A region with high aPS that does not correspond to highfPV is magnified (A) and compared with a high fPV region that does notcorrespond to high aPS (B). Figure reused with permission from Wileyand Sons.30Overall, HT29 tumours possessed a greater density of vessels (CD31 average dis-tance was 82.3± 3.0 µm for HCT116 and 40.5± 1.5 µm for HT29, p < 0.05).However many vessels were unlabelled for the fluorescent dye (carbocyanine) usedas a histological perfusion marker; the density of perfused vessels was similar be-tween the two xenograft models (CD31 vessels labeled for carbocyanine, PF, was47.8±5.4% for HCT116 and 36.0±2.1% for HT29, p> 0.05). The density of ves-sels labeled for HPG-GdF fluorescence was much greater in HT29 tumours (HPG-GdF+ve CD31, average distance was 172±10 µm for HCT116 and 88.9±6.3 µm;p < 0.05). In addition to a greater density of HPG-GdF-positive vessels, the highMW contrast agent was able to accumulate to a greater degree in the extra-vascularspace around vessels in HT29 cells. This effect can be seen in the histologicalimages of the compared tumour models collected 60-min post HPG-GdF adminis-tration (Figure 2.6(B)).Histological Quan-tityUnits HCT116 HT29 p-valueCD31(Average distance)µm 82.3±3.0 40.5±1.5 0.008Carbocyanine PF % 47.8±5.4 36.0±2.1 0.11Carbocyanine-positive and CD31(Average distance)µm 137±13 95.0±11.5 0.11HPG-GdF PF % 19.9±2.0 23.4±1.2 0.31HPG-GdF-positiveand CD31 (Averagedistance)µm 172±10 88.9±6.3 0.008HPG-GdF (Averageintensity)a.u. 0.405±0.031 0.828±0.123 0.008Necrosis Fraction % 52.4±3.6 37.7±5.1 0.032Table 2.1: Summary of the quantitative histological histological data for theHCT116 and HT29 tumours.312.3.5 MRI analysis of HPG-GdF in HCT116 and HT29 xenografts:BAT, fPV and aPSAdministration of neither Gadovist nor HPG-GdF was useful in detecting the dif-ference in vascular function between HCT116 and HT29 tumours using the ini-tial area under the gadolinium concentration curve (AUGC60). The slice-averagedmeans were not significantly different between tumour groups (data summarized inTables 2.2 and 2.3). Similarity between models was also observed qualitatively inthe parameter maps with AUGC60 or Ktrans overlaid on T1-RARE images shown inFigs. 2.4 and 2.7. Both tumour models show higher AUGC60 values in the tumourmargins for both contrast agents.Gadovist quantity units HCT116 HT29 p valueAUGC60s mM·s 5.09±0.83 7.68±1.66 0.31AUGC37min mM·s 317±23 395±70 0.56vp % 9.92±1.87×10−3 9.03±3.27×10−3 0.42ve % 0.247±0.024 0.246±0.013 0.77Ktrans min−1 8.41±1.56×10−4 1.16±0.039×10−3 >0.99R2 (p75) - 0.892± .017 0.913± .018 0.31BAT seconds 69.5±4.2 62.7±2.3 0.31BAT-EV % 98.9±0.6 99.2±0.5 0.63Table 2.2: Summary of MRI parameters derived from pharmacokinetic mod-elling of the Gadovist contrast agent for both the HCT116 and HT29tumours.HPG-GdF quantity units HCT116 HT29 p valueIAUC60s mM·sec 6.41±1.71E−03 9.62±1.64E−03 0.15AUC37min mM·sec 0.384±0.097 0.532±0.050 0.31BAT seconds 83.3±5.0 83.6±7.0 0.53BAT-EV % 52.2±4.8 81.8±5.0 0.016fPV fraction 1.07±0.33E−02 0.864±0.27E−02 0.22aPS seconds−1 2.38±1.00E−08 5.81±1.4E−08 0.06Table 2.3: Summary of MRI parameters derived from applying a linear modelto the HPG-GdF contrast kinetics for both the HCT116 and HT29 tu-mours.However, an MR-measured difference between HCT116 and HT29 vascular32Figure 2.6: HPG-GdF distribution in HCT116 and HT29 colorectal tumourxenografts at 60-min: histological analysis. (A) The whole-slice av-erage fluorescence intensity shows that HPG-GdF accumulates to agreater degree and distributes further away from vasculature in HT29tumours. (B) Sample images show HPG-GdF (red) overlaid on CD31(blue); insets also display nuclear density (grey). A greater degree ofHPG-GdF accumulation can easily be appreciated in the HT29 tumours.Figure reused with permission from Wiley and Sons.33function was seen with the BAT data for HPG-GdF. The fraction of enhancingvoxels was significantly lower in the HCT116 relative to the HT29 tumours (BAT-enhancing voxels for HPG-GdF in HCT116 was 52.2±4.8% and was 81.8±5.0%in HT29, p = 0.016 ) (Table 2.3). This corresponded to the histological data,which also shows HCT116 tumours as having a greater proportion of necrotic tis-sue (necrosis fraction for HCT116 was 52.4±3.6% vs. 37.7±5.1% in HT29, p=0.03). Of the HPG-GdF BAT-enhancing voxels, the averaged fPV and aPS param-eters were also determined (Table 2.3), and no difference was seen with the plasmavolume between tumour models (fractional fPV for HCT116 is 2.67±0.33×10−4and is 2.16±0.27×10−4 for HT29). aPS of HPG-GdF was the only MR-derivedbiomarker able to detect the significant difference in vascular function betweenthese two tumour models. Complete data sets for HCT116 tumours are shown inFigure DiscussionThe use of MCAs for imaging and describing tumour vascular function is a widelypursued area of research. Many studies have investigated a range of sizes, wheresmaller molecules extravasate and distribute through tissue too quickly to be con-sidered as sensitive blood pool agents, and larger molecules often accumulatetoo slowly for adequate signal detection [26, 44, 59]. In this work we describeHPG-GdF as an MR-visible MCA of 583 kDa, for which we have derived usefulbiomarkers that have sensitivity in measuring vascular function in preclinical stud-ies. The agent HPG-GdF is not intended to be used in patients directly primarilybecause of differences in human and tumour xenograft vessel morphology, as wellas the unknown safety, efficacy, and biocomaptibility considerations of the agent inhumans. However, despite this HPG-GdF may be very useful in developing drugsand understanding the distribution of existing drugs in preclinical experiments.While the signal-to-noise ratio (SNR) from HPG-GdF concentration-time curvesis not adequate to fit a complex pharmacokinetic model to determine parameterssuch as vp, ve and Ktrans, enhancement was measurable in most tumour voxels. Thenumber of Gd chelates on the HPG-GdF is approximately 300 per carrier molecule,which is high relative to many albumin chelates, the most common MCA in pre-34clinical use, which typically have about 20 chelates per carrier [47]. In additionto a greater number of Gd per carrier, the HPG-GdF accumulates in the perivas-cular space and fails to distribute more than a few micrometers from most vesselswithin a short DCE imaging timeframe (Figure 2.3). This accumulation could pro-vide a greater concentration for detection of a local, permeable tumour vessel andavoids the risk of conflating this with the distribution and accumulation elsewherein tumours. These attributes of HPG-GdF may make it more sensitive and morespecific than a smaller MCA such as albumin-Gd-DTPA. A minimum DCE imag-ing time from the presented data would appear to be about 15 min for the analysesdescribed, which is longer than the suggested 5-min ideal for practical clinical util-ity [48]. It is possible that greater signal could also be obtained by decreasingthe time resolution from the 2.24 s used in these studies, since HPG-GdF remainslargely intravascular.HPG-GdF contains the Gd-DOTA complex, which is thermodynamically highlystable and kinetically inert. The dissociation constant for Gd-DOTA (logK = 24.7)is much higher than those for Ca-DOTA (logK = 17.23) or Mg-DOTA (logK =11.92), hence neither Ca nor Mg can transmetallate Gd from the Gd-DOTA com-plex [61]. Biodegradation of HPG-GdF is therefore likely minimal, occurringthrough enzymatic attack of the end group, similar to that seen for polyethyleneglycol [62]. However, the long half-life and relatively slow excretion of HPGsthrough the reticuloendothelial system of the liver suggest that the potential fortoxicity should be investigated and monitored. A biodegradable version of HPGhas recently been synthesized and might circumvent these potential concerns, andwill be a focus in future DCE-MRI studies investigating the use of HPGs as MR-visible MCAs [52].Limited distribution through the interstitium is also most probably the result ofits significant size. We have observed HPG-GdF to be much more restricted in itsdistribution than is suggested by the diffusion coefficients (DHPG = 3.7×10−7 cm2s−1 and DGadovist = 3.6×10−6 cm2 s−1), as the larger agent reaches only 10-15 µmaway from vessels within an hour whereas Gadovist reaches all areas of a tumour.HPG-GdF molecules experience greater obstacles to their movement through theheterogeneous tumour microenvironment due to their size, but also possibly due tosteric hindrances, non-specific binding or sequestration [38].35The large size of HPG-GdF is a significant advantage. An approximately expo-nential increase in sensitivity for detection of permeability has been reported withincreasing MCA size (19). If extravasation of HPG-GdF only occurs in vessels thatare permeable, and the agent does not distribute through the extravascular space todistances away from vessels, then MR-visible HPG-GdF enhancement suggeststhe presence of vasculature, and accumulation of the agent over time indicates thelocal presence of a hyper-permeable vessel.A typical enhancement curve for HPG-GdF is shown in Figure 2.2. The size ofthe initial step-like enhancement is interpreted as the fPV, and the slope of signalchange for the remaining enhancement is reported as the aPS, reflecting the ex-travascular accumulation of HPG-GdF (Figure 2.2). These parameters are not cor-related with each other, with either value potentially significantly higher or lowerthan the other in the same voxels (Figure 2.5). A pattern of faster enhancementat the tumour margins is consistent for both contrast agents, and, while Gadovisteventually arrives in all the tissue, including regions of necrosis, many voxels thatcorrespond to necrotic regions fail to enhance with HPG-GdF within the 37-minimaging session. We found that HPG-GdF BAT-enhancing voxels correspond toareas without necrosis: see Figs. 2.4 and 2.7 for parameter maps of all BAT-enhancing voxels compared with histologically validated necrotic regions. There-fore, we identified perfused regions of tumour tissue for further vascular functionanalysis using the straightforward approach of selecting voxels that had positiveBATs for their HPG-GdF enhancement curves. Selection of viable tissue and theexclusion of necrotic tissue is a desirable approach for controlling for inter-tumourheterogeneity, since MRI data is most often described as a whole-tumour aver-age. In the event that different tumours possess variable necrotic burdens, or if thenecrotic fraction increases over time with treatment, selective analysis of viabletissue can control for this variability.Qualitative comparisons of well-matched, whole-slice images from multiplemodalities emphasize the utility of screening for BAT-enhancing voxels to se-lect for perfused and viable tissues. Without this information, analysis may berestricted to regions of tissues at the tumour margins where hot spots of vascular-ization are observed histologically, and tissue is assumed to be viable in the MRimage [56, 63]. In the colorectal xenografts examined here, the amount of necrosis36Figure 2.7: Vascular function in HCT116 xenografts. Whole-slice maps arepresented for individual HCT116 tumours (T01-T05, vertical columns)for parameters derived from MR imaging of Gadovist (BAT, AUGC60,Ktrans), MR imaging of HPG-GdF (BAT, fPV, aPS, AUGC60) and his-tological imaging (necrosis). Figure reused with permission from Wileyand Sons.37varied considerably, and while necrosis is typically located more in the core thanthe tumour margins, there is substantial inter-tumour heterogeneity with respect tothe amount and location of necrosis. Notably, while the values for Ktrans usingGadovist are on average higher around the entire tumour rim, the fPV and aPSvalues are more heterogeneously arranged around the rim and within the interiorof the tumour. Our more comprehensive approach, validated by observations ofviable tissue and perfused vessels within tumour interiors in histological sections,is a significant improvement over selectively assessing limited regions or hotspotsfor histological or MR quantitative analysis.Traditional DCE-derived parameters such as AUC and Ktrans are compositemeasures influenced to varying degrees by vascular surface area, permeability andblood flow. Our histological data supports a significant range in the propensity forHPG-GdF to extravasate, suggesting that assumptions of perfusion or permeability-limited conditions are not applicable to all areas of the tumour microenvironment.Hence, we cannot conclude that aPS is exclusively proportional to permeability-surface area (PS) in the whole microenvironment. For example, although theamount of HPG-GdF leaking into the extravascular space is dependent on the ves-sel being permeable to large molecules, the amount of accumulation, and thereforethe amount of signal enhancement, may still be impacted by the concentrationof contrast agent within the vessel. Longitudinal gradients can occur locally inpermeable vessels as the contrast agent leaks out, despite plasma concentrationsremaining consistent in overall systemic circulation [64, 65]. Thus, the aPS is ameasure of vascular function that has contributions from permeability as well as,when permeability is very high, perfusion. These limitations in the physiologi-cal interpretation of aPS are similar to those that apply to Ktrans and AUC for allcontrast-enhanced modelling.Parameters that are based on the amount and proportion of contrast agent en-hancement are dependent on fewer assumptions than are pharmacokinetic model-derived parameters such as Ktrans. Biophysical signals are dependent on the manyvariables necessarily involved in obtaining MR images, such as the scanner, imag-ing sequence, RF coil and analysis techniques. Interpreting magnitudes of changein highly variable tumours may be more reliable than assuming invariable phar-macokinetic attributes or unchanging tumour microenvironments, and may make38simplified biomarkers such as fPV and aPS more applicable to clinical studies con-ducted across multiple centres [30].2.5 ConclusionHPG-GdF is a largely intravascular MCA that selectively extravasates from hyper-permeable tumour vessels, accumulating in the perivascular regions without dis-tributing through the tumour interstitium. The high concentration of Gd chelatesper carrier molecule in combination with its excellent solubility makes HPG-GdFdetection possible despite the relatively low plasma fraction within tumours. Bycarefully comparing the vessel parameters of small and large molecule contrastagents in the same tumour, as well as comprehensively assessing the location ofMCA within the tumour relative to vasculature and necrosis, we conclude that BAT,fPV and aPS biomarkers derived from HPG-GdF enhancement provide a sensitiveand specific approach to measuring tumour vascular function. When assessingthe significant differences in vascular function of two tumour models (HCT116and HT29), HPG-GdF was useful in detecting a difference (aPS) while no changewas measured with parameters derived from Gadovist. Thus we conclude thatHPG-GdF and these analysis techniques are appropriate in the evaluation of tu-mour angiogenesis and response to treatment for preclinical research.39Chapter 3Applications of HPG-GdF:Investigating distribution oftrastuzumab in the tumourmicroenvironment3.1 IntroductionTreatment options for patients with the aggressive HER2-positive form of breastcancer continue to improve, though metastatic breast cancer remains a largely in-curable disease. Brain metastases are of particular importance in HER2-positivebreast cancer patients. With improved treatments and prolonged survival, the in-cidence of brain metastases as the first evidence of relapse has increased [66, 67].This effect has been attributed to the the blood-brain barrier (BBB) creating a sanc-tuary site by preventing drug access [68, 69]. Antibody-based therapeutics such astrastuzumab are proposed to have difficulty crossing the BBB and therefore brainmetastases evade drug activity [66, 70, 71]. In addition to the BBB, specific char-acteristics of the tumour microenvironment, beginning with a relative paucity offunctional vessels, can thwart the access of drugs to their targets such that a pop-ulation of under-exposed cells may survive and repopulate the tumour [38]. Our40collaborators have shown that many small molecule cytotoxics have limited tumourtissue penetration in both in-vitro and in-vivo model systems due to difficulties withdrug supply, flux through tissue, consumption or sequestration by cancer cells closeto blood vessels [59, 72–75].Macromolecular compounds such as MAb face particular extravascular dis-tribution difficulties due to their high molecular weights and target-binding affin-ity [41, 76]. The relatively slow distribution of MAbs has been attributed to thebinding site barrier hypothesis, where in the case of high affinity binding of theMAb to antigen, MAb distribution is limited by its binding in the presence ofample antigen [77]. Data from our collaborators’ lab (Dr. Andrew Minchinton)in the vector over-expressed MDA-435-LCC6 HER2 model showed that despitethe ability of trastuzumab to distribute far from vessels in tumours with relativelyeven HER2 distribution, there persisted HER2-positive tissue with poor access totrastuzumab after peak plasma exposures [78]. Staining heterogeneity is very strik-ing at the vessel level, as trastuzumab is able to extravasate from only a subpop-ulation of perfused vessels. Other groups have also demonstrated a limitation onthe accumulation of trastuzumab in tumours, focusing on net tumour accumula-tion [41, 76, 79].The ability of a drug to access and affect all its target cells is intuitively cru-cial for treatment success. The relatively slow distribution of MAbs through theinterstitium in solid tumours is recognized, but the long half-life of MAbs and pro-longed plasma exposure is expected to ensure adequate time to access all the tissuesin most treatment scenarios. However, we have found that in addition to distribut-ing slowly, MAbs experience additional barriers, leaving some areas of tissue withinadequate drug exposure. The aim for this study was to use HER2-positive tu-mour and metastases models of cancer to determine potential limits to trastuzumabaccess that may represent a mechanism of resistance to targeted therapies.413.2 Methods3.2.1 ReagentsTrastuzumab and bevacizumab (Roche, Genentech) were provided by the BritishColumbia Cancer Agency pharmacy; dilutions to 1-2 mg/ml were prepared in ster-ile 0.9% NaCl before intra-peritoneal (i.p.) injection. Trastuzumab antibodies foruse in combination with bevacizumab were tagged with fluorescent labels accord-ing to Alexa Fluor 546 Protein Labeling Kit (ThermoFisher) instructions. Hypoxiamarker pimonidazole (Hypoxyprobe) was administered at 60 mg/kg as an i.p. in-jection 2 h prior to tissue harvest. Fluorescent dye DiOC7(3) (Molecular Probes),0.6 mg/ml dissolved in 75% (v/v) dimethyl sulfoxide/25% sterile H2O, was ad-ministered intravenously as a marker of vessel perfusion 5-min prior to tissue har-vest [80].3.2.2 Mice and tumoursFemale NOD-SCID mice weighing 20-28 g between 8 and 16 weeks of age werebred and maintained in our institutional pathogen free animal facility. Mice wereimplanted with 60 day 17-β -estradiol pellets (Innovative Research of America)subcutaneously 3 days before implantation of BT474 or MDA-MB-361 tumours.Tumors were implanted as single cell suspensions (2-10×106 cells) into the sub-cutaneous sacral region or into the inguinal mammary fat pads. Metastases mod-els were implanted as single cell suspensions (2-5×105 cells per implant) as i.v.or i.p. injections for SKOV3 or BT474 models, or as intra-cardiac injection forMDA-MDA-MB-231-BR-HER2 models, with animals euthanized a maximum of23 days after implant.3.2.3 MRIMRI experiments were performed at the UBC MRI Research Centre on a 7TBruker Biospec 70/30 scanner at room temperature with a combination volume(transmit)/surface (receive) coil. DCE-MRI data was collected as previously de-scribed [1]. Gadovist (Bayer Healthcare) was administered by i.v. catheter as a5 µL/g bolus dose from 60 mM solution. Macromolecular contrast agent hyper-42branched polyglycerol (HPG-GdF, 583 kDa) was synthesized as previously de-scribed [6, 49] and administered as a 6 µL/g bolus dose from 100 mg/mL (0.2mM). Regions of interest (ROI) were drawn on T2-weighted RARE images to out-line the tumour using ImageJ (NIH) and all other MR analysis was performed us-ing Python. Area Under the Curves (AUC) for Gadovist was determined from thecommon injection time point to 60 s. A two-parameter linear model was applied tocharacterize HPG-GdF signal-intensity curves for fractional plasma volume (fPV)determined by the rapid increase at time of injection and for apparent permeabilitysurface area product (aPS) calculated as the slope of later enhancement, as de-scribed earlier in section 2.2. Both MR and histological modalities imaged slicesin the plane perpendicular to an implanted fiducial marker tube to minimize angulardifferences between MR and histological image slices [35].3.2.4 ImmunohistochemistryThe general immunohistochemical procedure used has been previously reported [78].Briefly, 10 µm tumour cryosections were air-dried, imaged for native DiOC7(3) orAlexa 546-tagged trastuzumab fluorescence, and fixed in 50% (v/v) acetone/methanolfor 10-min at room temperature. Trastuzumab was visualized in sections usingAlexa 546 goat anti-human secondary antibody (Invitrogen) HER2 was subse-quently stained with 2.2×10−3 mg/mL trastuzumab as a primary detection an-tibody and goat anti-human Alexa 546 secondary. Additional staining was per-formed using antibodies to PECAM/CD31 (BD PharMingen), pimonidazole (hy-poxprobe) collagen IV (CIV, Gene Tex) and αSMA (Abcam). Visualization ofprimary detection antibodies was done using Alexa fluorescence secondary anti-bodies of appropriate species using 488 nm, 647 nm and 750 nm wavelengths.Nuclear density was stained using Hoechst 33342 (Thermofisher) and imaged at380 nm.3.2.5 Image acquisition and analysisSections were imaged as previously described [59] using a system of tiling adjacentmicroscope fields of view at a resolution of 0.75 µm/pixel. Using ImageJ [60] anduser-supplied algorithms, images were superimposed and manually cropped to tu-43mour tissue boundaries with staining artifacts and necrosis removed. False colourimages were constructed in ImageJ by converting greyscale images to colour andoverlaying selected layers: trastuzumab (magenta), HER2 (blue or grey), Hoechst33342 (grey), CD31 (blue), carbocyanine (cyan), pimonidazole (green) and αSMAor CIV (red). Positive fluorescent staining is reported as average intensity (range1-255) for pimonidazole, trastuzumab and HER2. Perfused vascular density wasdetermined by applying a threshold to CD31 and DiOC7(3) images, with neigh-boring positive pixels grouped as “objects”; CD31 objects with a minimum 20%overlap with DiOC7(3) objects were determined to be perfused vessels. All imagepixels were sorted based on their nearest perfused vessel and the average distanceis reported as a repeatable measure of vascular density.3.3 Results3.3.1 Measures of vascular density, architecture and function do notconsistently correlate with heterogeneous patterns oftrastuzumab distributionSignificant inter-vessel heterogeneity in trastuzumab distribution is seen in ortho-topic BT474 xenografts, with neighbouring patent vessels often showing vari-able amounts of trastuzumab bound to perivascular cells (Figure 3.1A), similarto previous findings in MDA-435-LCC6 vector-overexpressing HER2-positive tu-mours [78]. Non-patent vessels (red arrows in Figure 3.1A) never have extravascu-lar trastuzumab, suggesting intermittent perfusion is not a significant mechanismfor reduced trastuzumab access. The same inter-vessel heterogeneity is seen inMDA-MB-361 tumours where vascular function dynamics were further character-ized by imaging the presence of Gadovist (Figure 3.1B). Regions with greatestvascular function (largest AUC60 with Gadovist) do not consistently correspondto areas of greater trastuzumab distribution (red arrows in Figure 3.1B); this isalso demonstrated quantitatively where matched slices were plotted for amount ofbound trastuzumab and AUC60 (Gadovist). The same sections were stained for vas-cular architectural markers αSMA and CIV (both shown in red), neither of whichexhibit a pattern of distribution similar to the presence or absence of trastuzumab44(Figure 3.1B).Some regions are high for Gadovist uptake, reflecting high vascular function,but these do not consistently correspond with regions of high trastuzumab distribu-tion in histology sections. No correlation was found between Gadovist AUC andtrastuzumab in matched image slices obtained from multiple tumours. Similarly,neither αSMA nor CIV are present to a greater degree proximal to vessels with orwithout trastuzumab bound to perivascular cells.3.3.2 Dynamic vascular permeability and blood volumemeasurements do not consistently relate to patterns oftrastuzumab distributionThe histological measure of perfusion using carbocyanine is a useful indication ofvessel patency, however it is static and therefore its interpretation is limited. Therole of vascular function on trastuzumab distribution was further investigated usingdynamic contrast-enhanced MRI (DCE-MRI) of a high molecular weight contrastagent, HPG-GdF (MW 583 kDa) (Figure 3.2). As previously described, repeatimaging of contrast agent presence in the tumours is analysed and the initial ap-pearance of HPG-GdF reflects fPV and its accumulation over time indicates theapparent permeability surface area (aPS) [1]. Tumours were excised immediatelyafter imaging; corresponding histological sections are compared to the DCE-MRIderived parameter maps. BT474 tumours have microregionally variable levels ofboth fPV and aPS, each exhibiting regions of distinction. Regions of high fPV arewell matched by histological images of carbocyanine that indicate areas of veryhigh perfusion but some of these regions do not have any significant accumula-tion of trastuzumab. The reverse can also be found, where regions with relativelylow fPV correspond to high trastuzumab. Similarly, there are some significanttrastuzumab accumulation areas that have relatively low aPS while some high aPSvalues correspond to areas with relatively low trastuzumab. Examples of good orbad correlation between vascular function and trastuzumab distribution are high-lighted with arrows and suggest that neither of the MRI-derived parameters consis-tently or adequately explain microregional distribution of trastuzumab.45B  Vascular function and structure; 10 mg/kg trastuzumab in MDA-361 xenograft, 24h500 µm150 µm0 10 20 30 4081012141618AUC_GadovisttrastuzumabCD31 Her2 trastuzumab perfusion 150 µmCIVαSMAtrastuzumabHer2 trastuzumabGadovist, AUCA  Inter-vessel heterogeneity; 10 mg/kg trastuzumab in BT474 xenograft, 24h αSMA CIVtrastuzumab0  10  20  30  40  50  60  70  80  90  100 AUC60sec Gadovist (arbitrary units)Figure 4Figure 3.1: A) Magnified region of a BT474 xenograft treated with 10 mg/kgtrastuzumab for 24 h. Carbocyanine fluorescent dye (cyan) aroundCD31 stained vessels (blue) indicates patency; non-patent vessels areindicated as red arrows. Trastuzumab extravasates from vessels hetero-geneously, with many patent vessels showing no extravascular boundtrastuzumab (green arrows) even when adjacent patent vessels do haveperivascular trastuzumab. B) An anatomical RARE MR image withthe tumour is shown alongside an AUC60 map using Gadovist inMDA-MB-361 tumours. The AUC60 map is compared with slice-matched histology sections of bound trastuzumab (purple), vascular ar-chitectural markers αSMA and CIV (both shown in red).46Figure 3.2: fPV and aPS parameter maps are compared to matched histol-ogy sections stained for bound trastuzumab (magenta), and for HER2(grey), carbocyanine marker of perfusion (cyan) and for CD31 vascu-lature (blue). Areas of vascular function (MRI) and trastuzumab (his-tology) correlation are indicated (orange arrows) in both modalities; ex-ample areas of poor matching are also shown (purple arrows). Starsindicate location of fiducial markers for multi-modal slice comparison.473.4 DiscussionUsing tumour mapping analysis and DCE-MRI we examined the impacts of tumourblood vessel architecture, vascular function and the tumour microenvironment onthe patterns of distribution of trastuzumab in primary and metastatic models. Ourcollaborators have observed that even when trastuzumab does have access to tu-mour tissues, the pattern of distribution is highly heterogeneous, similar to theirprevious work in a vector-overexpressing HER2-positive breast cancer model [78].Other groups demonstrating a limitation on the accumulation of trastuzumab inpreclinical tumours have focused on net tumour accumulation and suggest that amajor limitation to MAb distribution is their difficulty distributing through the solidtumour tissue in the extravascular space [41, 76, 79]. In humans, MAbs have a longhalf-life and are administered for many months. The pharmacokinetic parametersof MAbs are therefore thought to likely accommodate relatively slow distributionof MAbs through the interstitium [41, 81]. However, 3D tissue-disc data from theMinchinton lab shows trastuzumab is able to diffuse relatively well through tumourtissues in vivo in conditions of poor convective flow similar to an environment withhigh interstitial fluid pressure in vivo [2]. Neither is trastuzumab distribution lim-ited by tight binding to antigen proximal to vessels as suggested by the binding-site barrier hypothesis [77], as it reaches distances > 150µm from sources within24h despite high HER2 expression [2]. This distance is similar to the diffusionlimit for oxygen, therefore most tumour tissues are within this 150 µm range of ablood vessel and could be expected to have adequate exposure. These data suggestthat neither the binding site barrier hypothesis nor high interstitial fluid pressureadequately explain microregional variation in trastuzumab binding. Given the in-complete access of trastuzumab seen in our models, other vessel- and tissue-levelbarriers appear to limit its distribution in the tumour microenvironment.Vascular maturity and architecture has also been attributed to poor drug ac-cess [82], however, no correlation between markers for αSMA or CIV and therelative distribution of trastuzumab has been found. Higher doses and longer expo-sures do lead to higher numbers of vessels with perivascular trastuzumab, suggest-ing that some degree of intermittent perfusion or very poorly permeable vesselsimpact the observed microregional heterogeneity. Non-perfused vessels have not48been observed to have trastuzumab on perivascular cells, which suggests inter-mittent perfusion is an unlikely mechanism for observed inter-vessel heterogene-ity. Dynamic measurements of perfusion and permeability derived from DCE-MRimaging of high molecular weight contrast agent HPG-GdF (583 kDa) suggest theintuitive relationship between vascular function and drug access is important, butdoes not consistently explain the heterogeneous patterns of drug distribution seen.Areas of highly perfused tissue marked by carbocyanine in histological sectionscorrespond to areas of high perfusion and/or permeability in MRI. However, thesetumour areas of high vascular function described in both imaging modalities mayor may not have bound trastuzumab. Similarly, regions with substantial amountsof bound trastuzumab do not necessarily correspond to areas of high perfusion orpermeability measured using aPS and fPV. HPG-GdF-derived parameters fPV andaPS arise under conditions that are closer to a permeability-limited regime thanstandard low-molecular weight contrast agents. However, in the highly chaoticvascular environment of tumours we can still expect these MR-derived parametersof vascular function to not be completely independent of each other.New strategies that are able to more effectively target and kill cancer cells, par-ticularly metastatic disease, are urgently required. Efforts to vary antigen-bindingaffinities and the size of MAb fragments have been explored to improve their dis-tribution through the extravascular compartment [41, 76]. Our data highlight theimportance of also considering access of antibodies to target tumour tissues andwhether microregional distribution of MAb therapeutics may be affected whencombined with vascular damaging agents or when targeting occult metastases. Im-provements to MAb anti-cancer activity could be made in selection of combinationtherapies and the design of treatment schedules, as well as in the design of noveltargeted drugs. Efforts to examine these phenomena in the clinic would be of sig-nificant interest.3.5 ConclusionsIn this chapter we showed evidence for poor distribution and access of trastuzumabin preclinical tumours through direct visualization of bound drug, with partic-ular implications for metastatic tumours. DCE-MRI imaging parameters such49as vascular permeability and fractional plasma volume do not relate to pat-terns of trastuzumab distribution. Similarly, no correlations were found betweentrastuzumab distribution and histological metrics such as vascular density and func-tion or vascular architectural markers αSMA and CIV. This suggests that the tu-mour microenvironment and tissue- and vessel-level barriers to drug distributioncould effectively limit access of the drug to all its target cells. These effects appearto be more important than slow interstitial distribution resulting from high inter-stitial fluid pressures or high binding affinity to HER2. The non-invasive imagingtechniques to probe the tumour microenvironment described here could also beapplied to investigate the distribution of other drugs.50Chapter 4Applications of HPG-GdF:Assessing vascular normalizationusing an antiangiogenicchemotherapy4.1 IntroductionIn the previous chapter we used HPG-GdF to help shed some light on the distribu-tion of Trastuzumab within a tumour. Though we were unable to use measures ofvascular function to explain its distribution, we did show that vessel permeabilityand plasma volume did not consistently relate to drug distribution. In this chap-ter we aim to use parameters from DCE-MRI using HPG-GdF to probe vascularfunction after treatment of SCCVII tumour xenografts with bevacizumab, an an-tiangiogenic drug that normalizes the tumour vasculature by inhibiting VEGF.Through various signalling pathways, cancerous cells within tumours co-opthost vessels to obtain nutrients for rapid growth and creation of new blood vessels(angiogenesis) [29]. As the tumour expands a new vascular network is formedbut is structurally and functionally abnormal on account of leaky and tortuousblood vessels [16], absent or loosely attached pericytes that provide structural sup-51port for cells, and abnormal morphology of the endothelial cells lining the ves-sels. Aberrant angiogenesis, poor blood perfusion, and a chaotic vascular networkall play a role to limit oxygen delivery to cells, contribute to acidosis and for-mation of hypoxic regions in tumours [29]. Hypoxic regions within tumours areextremely problematic as those cells are resistant to both radiation and many cyto-toxic chemotherapies. Additionally, the resulting tumour microenvironment posesstrong barriers to effective drug delivery and consequently, efficacy.A strategy to mitigate such adverse conditions of drug delivery is to attemptto normalize the vascular network by pruning away newly formed vessels, lim-iting creation of new vessels, and ultimately, spreading out the same amount ofnutrients over fewer vessels. As the structural components of the vasculature im-prove through reduced vessel leakiness and increased network organization, deliv-ery of chemotherapies is more efficient [39]. This effect has been termed ‘VascularNormalization‘ and does not just rely upon increased uptake of drugs and oxygenbut also on improving the delivery of the drugs to a larger population of the tu-mour [29]. Considerable evidence has been presented in the literature in supportof the vascular normalization window [29, 83, 84]. MRI techniques such as dy-namic susceptibility contrast MRI, arterial spin labeling, and DCE-MRI have allbeen used to obtain surrogate parameters of the effects of vascular normalization.Unfortunately, modelling of the kinetics of low molecular gadolinium-based agentsoften makes it difficult to separate blood vessel permeability and capillary surfacearea without resorting to more complex models such as the two compartment ex-change model, extremely fast temporal resolution, and accurate knowledge of thearterial input function. This limitation motivated this work to determine whether ahigh molecular weight agent such as HPG-GdF could be used for measuring vesselpermeability without the confounding effects of blood flow.The binding of vascular endothelial growth factors (VEGF) to the VEGF re-ceptor is a key driver of angiogenesis. Bevacizumab - marketed clinically as‘Avastin c©’ - is a monoclonal antibody that binds VEGF extracellularly, preventingthe interaction of the VEGF molecule with its receptors and inhibits angiogenesis.Amongst others, bevacizumab has been used clinically to treat breast, colon, col-orectal, lung, brain, ovarian and cervical cancers [85]. VEGF ablation has beenshown to at least temporarily reduce vascular permeability and increase tumour52oxygenation in some models [30]. We hypothesized that treatment with a VEGFinhibitor will result in a decrease of vessel permeability measured with DCE-MRIusing HPG-GdF.4.2 Methods4.2.1 Mice tumours, and treatment groupsTwelve female NOD-SCID mice were implanted with murine squamous cell car-cinoma SCCVII tumours (5×105 cells in 50 µl serum-free media; cells providedby Dr. J Evans). Mice were imaged when the largest tumour diameters reached avolume of approximately 500 mm3, positioned supine on the custom surface coilapparatus and anaesthetized with isoflurane for the duration of imaging sessionsuntil euthanasia.The treated group comprised six of the twelve animals and were administered5mg/kg mouse anti-VEGF antibody (B20-4.1.1., Genentech) 72 hours prior to theimaging session. Throughout the imaging session, a small animal monitoring sys-tem (SA Instruments Inc., Stony Brook, NY, USA) was used to monitor respirationrate and body temperature. A continuous airflow heater was used to maintain tem-perature at 36.5± 1◦C. All animals were injected with 60 mg/kg pimonidazole hy-drochloride (HypoxyProbe) i.p. 60 minutes prior to imaging to label hypoxic cellsand were euthanized within 15-min of imaging completion. Tumours were embed-ded and frozen in optimum cutting temperature medium (OCT; Tissue-TEK).4.2.2 MRIMRI experiments were performed at the UBC MRI Research Centre on a 7TBruker Biospec 70/30 scanner at room temperature with a combination volume(transmit)/surface (receive) coil. Macromolecular contrast agent hyperbranchedpolyglycerol HPG-GdF, 500 kDa) was synthesized as previously described [6, 50]and administered as a 6 µL/g bolus dose from 100 mg/mL (0.2 mM). DCE-MRIdata was collected at a time resolution of 4.27 s (FLASH; TR/TE = 66.7/2.67s; FA= 40◦, NR = 250). The matrix size was 128 x 64 x 6 with a field-of-view of 38.4x 19.2 x 6 mm, resulting in an in-plane spatial resolution of 0.3 x 0.3 mm and a53slice-thickness of 1 mm. Regions of interest (ROI) were drawn on T2-weightedRARE images to outline the tumour using ImageJ (NIH, Maryland, USA) and allother MR analysis was performed using Python. A two-parameter linear modelwas applied to characterize HPG-GdF signal-intensity curves for fPV determinedby the rapid increase at time of injection and for aPS calculated as the slope of laterenhancement similar to what was described in section 2.2. Area Under the Curve(AUC) for HPG-GdF was determined using the signal intensity curves from theinjection time point to 60s. Non-enhancing voxels (i.e. those with an AUC60 < 0)were excluded from all analyses. Both MR and histological modalities imagedslices in the plane perpendicular to an implanted fiducial marker tube to minimizeangular differences between MR and histological image slices [35].4.2.3 ImmunohistochemistryFrozen and OCT-embedded tumours were cryosectioned into 10µm-thick slicesand air-dried. Sections were first imaged for native DiOC7(3) or Alexa 647nm-tagged HPG-GdF fluorescence, and fixed in 50% (v/v) acetone/methanol for 10minutes at room temperature. Additional staining was performed using antibodiesto PECAM/CD31 (BD PharMingen) and pimonidazole (hypoxprobe). Visualiza-tion of primary detection antibodies was done using Alexa fluorescence secondaryantibodies of appropriate species using 488 nm and 546 nm wavelengths. Nucleardensity was stained using Hoechst 33342 (Thermofisher) and imaged at 380 nm.4.2.4 Image acquisition and analysisSections were imaged using a system of tiling adjacent microscope fields of viewat a resolution of 0.75 µm/pixel [59]. Using ImageJ [60] and user-supplied algo-rithms, images were superimposed and manually cropped to tumour tissue bound-aries with staining artifacts and necrosis removed. False color images were con-structed in ImageJ by converting greyscale images to color and overlaying selectedlayers: HPG-GdF (red), Hoechst 33342 (grey), CD31 (blue), carbocyanine (cyan),pimonidazole (green). Positive fluorescent staining for each slice is reported as amean, and the median group intensity I (range 0-255) is reported for pimonidazole(Ipimo), HPG-GdF (Ihpg).544.3 Results4.3.1 Treatment with B20 reduces tumour hypoxia and HPG-GdF 72hours after treatmentTreatment with the VEGF inhibiting drug B20 resulted in a dramatic reductionin hypoxia in SCCVII tumour xenografts. Immunohistological staining with pi-monidazole allowed for visualization of these group differences with a representa-tive example shown in Figure 4.1. In the untreated group, large regions of the tu-mours are hypoxic though there is considerable heterogeneity in baseline hypoxiastatus between tumours and between slices of the same tumour. In the treatedgroup, pimonidazole staining is disparate and not concentrated in local regions ofpoor oxygenation. Treated tumours also had relatively lower levels of bound pi-monidazole and this was consistent across tumours and between slices of the sametumour. These staining patterns are representative, but there is considerably moreheterogeneity in pimonidazole staining in the control tumours.Figure 4.1: Histological sections of a control (left) and treated tumour (right) areshown to illustrate the dramatic decreases both in pimonidazole staining andin accumulation of HPG-GdF in the treatment group.55Quantitative analysis of the immunohistolgoical stains in the images was con-ducted to assess whether these observations existed at the group level. A Mann-Whitney U test indicated that the median pimonidazole staining intensity wasgreater for the control group (Ipimo = 21.7) than the treated group (Ipimo = 17.2),U = 47, p = 5× 10−5. Median HPG-GdF staining intensity is also reduced forthe treated tumours (Ihpg = 12.7) compared to the controls (Ihpg = 11.0), Mann-Whitney U = 52, p = 8× 10−6. Data for individual tumours as well as the groupdifferences are shown in Figure 4.2.Figure 4.2: Median pimonidazole staining is markedly reduced for the B20-treated group (Ipimo = 17.2) compared to the controls (Ipimo = 21.7).Median HPG-GdF staining intensity is also reduced for the treated tu-mours (Ihpg = 12.7) compared to the controls (Ihpg = 11.0). P-valuesfrom the Mann-Whitney U test were p = 5× 10−5 (pimo, left) andp= 8×10−6 (HPG-GdF fluorescence, right).4.3.2 Blood vessel permeability (aPS) and HPG-GdF accumulation(fluorescence intensity) decreases in tumours treated with B20Apparent permeability-surface area product (aPS) is a parameter that capturesvessel leakiness with DCE-MRI using the large molecular weight contrastagent HPG-GdF. Figure 4.4 shows the group differences and the median value ofthe B20-treated group (aPSMd = 1×10−4) was significantly lower than the control56Control TreatedSlicesTumoursHPG-GdFPimonidazoleFigure 4.3: Histological sections shown for 6 control and 6 treated tumours, withfour slices per tumour. All sections are stained for pimonidazole (green) andHPG-GdF (red).group (aPSMd = 6.5×10−5); Mann-Whitney U = 3, p= 0.01.Histological staining of HPG-GdF fluorescence matched relatively well withareas of high aPS values as shown in Figure 4.5, however there are also areaswhere aPS is moderate or high but HPG-GdF fluorescence is not present in highconcentrations. Group differences in HPG-GdF accumulation were also presenthistologically as HPG-GdF fluorescence intensity was markedly decreased in theB20-treated tumours (Ihpg = 12.7) compared to the controls (Ihpg = 11.0); a Mann-57Whitney U test indicated the difference was significant, U = 52, p = 9× 10−6.Aggregate voxel distributions of aPS values in Figure 4.2 also shows a clear re-duction in aPS in the treatment group. Figure 4.5 shows the aPS parametric mapsfrom DCE-MRI alongisde approximate slice matched histology sections from thesame tumours. Patterns observed in the quantitative analysis of the data are alsoevident in MRI and histology images: following treatment with B20, there is amarked reduction in blood vessel permeability measured by aPS and a reductionin HPG-GdF accumulation histologically measured by native fluorescence of thecontrast agent.4.3.3 HPG-GdF enhancement curves and AUC60 are altered aftertreatment, but fPV does not changeGroup HPG-GdF enhancement curves are shown in Figure 4.5 and match the re-sults obtained from a quantitative analysis of the parameteric maps: treated tu-mours have a relatively flat profile and plateau whereas the control tumours showa steady increase of enhancement arising from the leakage of HPG-GdF. AUC60is a parameter that captures the relative enhancement of voxels within the first60 seconds after injection. The control tumours had a median AUC60−Md = 5.5and treated tumours had a median AUC60−Md = 4.0 and the reduction was statisti-cally significant (Mann-Whitney U = 6, p= 0.03). The fPV values for the control(fPVMd = 0.12) and treated tumours (fPVMd = 0.09) were not significantly differ-ent (Mann-Whitney U = 3, p= 0.15).4.4 DiscussionIn this study we have demonstrated that vessel permeability can be assessed usingthe apparent permeability-surface area product, aPS. This is consistent with similarreports of extracting permeability parameters with DCE-MRI using albumin-basedmacromolecular contrast agents [57, 86], dendrimer-based contrast agents [87] andothers [48]. Typically, contrast agents used in DCE-MRI are less than 1 kDa inmolecular weight and freely extravasate from vessels and enhance surroundingtissues. These low molecular weight agents are essential for many applicationsincluding cancer detection, evaluating vascular characteristics, and measuring tu-58Figure 4.4: Summary of group differences for all three DCE-MRI parame-ters: AUC60, aPS, and fPV. There is a statistically significant reductionin AUC60 and aPS for the treated tumours, but no difference measuredfor fPV. P-values from the Mann-Whitney U test for the comparisonswere p= 0.03 (AUC60), p= 0.01 (aPS), and p= 0.15 (fPV).59Figure 4.5: Histological stains of HPG-GdF shown alongside approximate slicematched DCE-MRI parameter map of aPS and the group-averaged contrastenhancement curves of control and treated tumours. There is an overall re-duction in aPS for the treated tumours. The mean contrast enhancementcurves (group means in dark blue and red lines; shaded region is the 95%confidence interval determined by bootstrapping) also show that the treatedtumours have a higher enhancement slope after contrast agent injection.60mour microenvironment changes longitudinally [88]. However, there is a strongcoupling of blood flow and vessel permeability in standard DCE-MRI models pa-rameter interpretation varies based on the tumour microenvironment [89]. Forinstance, in regions of high permeability, the movement of the contrast agent islimited by blood flow so Ktrans primarily measures blood flow. When appliedin regimes of low permeability, the small molecular weight contrast agent can-not extravasate from vessels so Ktrans primarily measures vessel permeability [90].Contrast agent size affects measures of vascular permeability and low molecularweight agents result in underestimations of vessel permeability [87]. For applica-tions where distinguishing between blood flow and vessel permeability is impor-tant, low molecular weight agents cannot be used unless more complex pharma-cokinetic models are used.The principal advantage of high molecular weight contrast agents is their in-travascular nature so blood flow is removed as a contributing factor of the measure-ment. There is also an inverse relationship between size of the contrast agent andMR signal enhancement because larger molecules diffuse slower and extravasateless. The reduction in signal enhancement is partially compensated for by the in-creased relaxivity of the larger agent arising from more Gd-chelates attached tothe molecule and a lower tumbling rate [31]. We have previously shown that therelaxivity of HPG-GdF is 300 times larger than standard DCE-MRI contrast agentsbut the agent extravasates only a few micrometers from nearby vessels [1], leadingto an overall reduction in signal enhancement. Consequently, the larger the con-trast agent, the less signal enhancement there is in the MR images. At a molecularweight of 583 kDa, HPG-GdF is considerably larger than common biocompatiblehigh molecular weight agents that range from 20-92 kDa [31] and to our knowl-edge, is the largest agent with which a permeability measure has been obtained.This study provides another application for the high molecular weight agentHPG-GdF in exploring the tumour microenvironment and the effects of vascularnormalization. AUC60 of the signal intensity represents presence of HPG in theblood 60s after injection was also significantly lower in the treated mice whichindicates the antiangiogenic treatment altered the tumour vasculature. There isconsiderably less variability in the enhancement curves of the treated tumours asshown in Figure 4.5,providing strong evidence for B20 normalizing the vascula-61ture. The vascular normalization hypothesis suggests that after treatment with anantiangiogenic therapy, the overall vessel permeability and the number of vesselswill decrease. As summarized in Figure 4.4, our study has provided evidence fora decrease in vessel permeability (reduced aPS) as well as a decrease in accumu-lation of HPG-GdF (reduced AUC60). Additionally, we also showed a dramaticreduction in hypoxia (measured by pimonidazole staining) in the treated group.A limitation of this study is that the model cell line used was a murine tumourmodel which may have limited potential for translation [91]. However, this tumourmodel is ideal for validating non-invasive MRI as it has no necrosis present due toits vascular architecture. Necrosis is a potential confounding factor in macromolec-ular contrast agent imaging studies as large areas of necrosis and empty space resultin pooling of the agent in non-viable areas. Consequently in SCCVII tumours, anyfluorescence present on cryosections has extravasated from nearby vessels and canbe quantified as shown in Figure 4.2. Figure 4.3 shows low inter-tumour slice vari-ability in average staining intensity of HPG-GdF fluorescence, but considerableheterogeneity exists between tumours and not all tumours responded consistently.On aggregate, it is evident that the treated tumours have significantly less HPGaccumulation. Figure 4.3 also shows regions stained with both HPG-GdF and pi-monidazole.There is no clear explanation for why a macromolecule such as HPG-GdF ac-cumulates in a region where oxygen is regionally distributed in such low concen-trations. One possible explanation is that in longer imaging sessions tumors exhibitvarying oxygenation patterns and since pimonidazole and HPG-GdF were admin-istered over two hours apart, the tumour microenvironment had changed. Periodsof oxygen-starvation and re-oxygenation cycling in tumors has been termed cy-cling hypoxia and can arise for intermittent periods based on temporary vesselocclusions, or may be permanent, chronic changes in hypoxia [92, 93]. However,as has been recently demonstrated, it is not as simple as classifying tumours in abinary fashion [94]. To enable a more thorough exploration of this phenomenon,more sophisticated methods are needed including a non-invasive method of as-sessing oxygenation status in vivo validated with multiple stains of hypoxia suchas pimonidazole and EF5 administered sequentially with a delay to capture anychanges that occur during the delay time. Additionally, slice-matched histology62sections and MR images were beyond the scope of this study and this would beneeded to explore patterns of HPG-GdF staining and regions of high aPS or fPV.An important difference in this study compared to our initial experiments withHPG-GdF [1] is that here we analyzed only the signal intensity enhancement ratherthan relying on an arterial input function to compute a contrast agent concentra-tion. Largely due to constraints in available signal enhancement due to the highmolecular weight contrast agent, sufficient signal was not present in all areas of thetumour for pharmacokinetic modelling or computation of contrast-agent concentra-tion. Though this simplification is not ideal as it assumes signal intensity changeslinearly with contrast-agent concentration, it is valid when contrast-agent concen-trations are low [95] and relative group differences are being measured (rather thanabsolute changes in individual tumours). This and other challenges associated witha lack of standardization in analysis methods to use high molecular weight contrastagents poses a significant challenge for its use in the clinic as a tool to assess cancertherapies. Nevertheless, validation with histological images is a positive sign thatthe measurements of aPS and fPV reflect characteristics of the tumour microenvi-ronment. Though we did not use a low molecular weight agent for comparison inthis study, we have previously shown (summarized in Table 2.3) that DCE-MRI pa-rameters from Gadovist (a low molecular weight agent) did not show a differencein vascular function between HT29 and HCT116, but a difference was measuredwhen using HPG-GdF.4.5 ConclusionWe have shown that aPS and AUC60 decreases after administration of B20. Loweraccumulation of HPG-GdF following treatment was confirmed by histology via di-rect HPG-GdF fluorescence imaging. Blood vessels of treated tumours appear tobe less permeable, and tumour hypoxia was markedly reduced as a result of theanti-angiogenic drug. Unfortunately, existing MRI techniques do not permit mea-surement of oxygenation in vivo despite an unmet clinical need [96]. Our resultsindicate that HPG-GdF has potential applications in assessing effects of antiangio-genic agents on blood vessel permeability and function. Future work will includea panel of tumours with more diverse tumour microenvironments to evaluate its63utility in more clinically relevant tumour models. While it is unlikely that macro-molecular contrast agents will become standard clinically, they are a very usefultool to improve our understanding of the tumour microenvironment and to developnovel drugs for cancer.64Chapter 5Oxygen-enhanced MRI5.1 IntroductionHypoxia is a well-established component of the tumour microenvironment, arisingmost often as tumour cell proliferation outpaces the growth of new vasculature.Tumour hypoxia is an indicator of poor prognosis and is responsible for tumour re-sistance to radiotherapy and some chemotherapies, but is also a potentially usefultarget for novel anti-cancer drugs [97]. Assessing tumour hypoxia in the clini-cal setting is challenging largely due to the invasive nature of biopsy-dependenttechniques and the limited capacity and high expense of the more favoured, non-invasive PET imaging of hypoxia tracers [96]. The utility of screening patientsfor hypoxia was demonstrated retrospectively in trials of the hypoxic cytotoxintirapazamine, where those patients with greater PET-imaged hypoxia experiencedgreater benefit [98]. However, subsequent trials of drugs targeting hypoxia, includ-ing those for evofosfamide that failed to show clinical benefit, have not used hy-poxia imaging to stratify patients. A practical, widely applicable, and non-invasiveimaging method is urgently required as a biomarker to monitor tumour hypoxia inmany contexts, and is crucial to the development and clinical evaluation of futurehypoxia-targeting drugs.655.2 Theory5.2.1 PhysiologyThe primary mode of oxygen delivery to tissue is the haemoglobin (Hb) moleculeas it carries and delivers 98% of the oxygen in the body. Over 250 million Hbmolecules are found in a typical red blood cell and each Hb molecule has four bind-ing sites for oxygen molecules. The binding affinity for O2 drastically increasesfor subsequent oxygen molecules that bind to Hb as the conformation of the Hbmolecule changes to increase binding affinity for the next oxygen, a phenomenoncalled cooperativity. Similarly, when the local environment of the Hb moleculechanges such that O2 needs to be released, the reverse conformational changes oc-cur; a proportionately lower drop in oxygen tension is then required to release thenext O2 molecule. The dissociation of oxygen from haemoglobin molecules is welldescribed by the oxygen-haemglobin dissociation curve (Figure 5.1).Upon inspiration of atmospheric air (pO2 = 160 mmHg), gas exchange in thepulmonary capillary beds occurs in the alveoli of the lungs (Figure 5.2). Incom-ing venous blood with a low oxygen tension (pO2 = 40 mmHg) is oxygenated ashaemoglobin molecules readily bind available oxygen. As the oxygenated bloodleaves the alveoli and moves through the systemic arteries, it has an oxygen tensionof 100 mmHg. The oxygenated blood then travels from the arteries to the systemiccapillary bed and the local oxygen tension drops from 100 to 40 mmHg. Simulta-neously, while the Hb molecule undergoes a structural change releasing a moleculeof O2 from its first binding site. The second release of the oxygen molecule occurswhen the tension drops to 26mmHg [7]. In a population of Hb molecules fully sat-urated in tissue (SO2 > 95%), the approximate ∆pO2 required to release the first O2molecule is 60 mmHg (100-40 mmHg). The third O2 molecule is released whenthe pO2 drops from 26 to 19 mmHg, and the final O2 molecule is released whenthe pO2 drops to 12 mmHg [7]. Practically however, it is important to note that Hbmolecules never release all four of the bound oxygen in vivo. This important fea-ture of Hb (cooperativity) ensures that small changes in pO2 have the appropriateeffect in the appropriate place. For instance, in the lung tight binding is requiredso Hb can bind the O2 needed to supply all the tissues. Therefore, small changes66Figure 5.1: Sigmoidal curve illustrating the relationship between thehaemoglobin saturation (y-axis) and the oxygen tension (x-axis). Whenthe oxygen tension is low, the Hb easily binds O2 and there is a rapid risein oxygen saturation (green arrow). Note that it takes a large increasein oxygen tension to bind the last O2 and similarly, a large decrease inoxygen tension to release the last O2 (purple arrow) [7].should not affect the release of O2. Conversely, small changes in pO2 in capillarybeds should result in quick release of O2 so it can easily diffuse to oxygen-starvedtissues.5.2.2 Origin of the OE-MRI signalOxygen is a paramagnetic molecule because it has two unpaired electrons and it iswidely reported that the dominating effect in the OE-MRI signal is a T1 decreaseafter concentrated oxygen gas (100% O2) is breathed in [99, 100]. The excess oxy-gen travels through the blood stream dissolved in plasma and diffuses through the67Figure 5.2: A schematic of the change in the partial pressures of oxygen andcarbon dioxide at various points in the body. The pO2 of inhaled ambientair is 160 mmHg, and this is breathed in to the lungs. Oxygen diffusesout of the alveoli into the surrounding capillaries and binds to Hb dueto the pressure gradient (capillary pO2 is 40 mmHg). This oxygenatedblood (pO2 = 100mmHg) now enters the heart and is pumped throughthe body, with the Hb releasing oxygen through capillaries due again tothe pressure gradient (tissue pO2 is < 40 mmHg) [8]. Textbook contentproduced by OpenStax Biology is licensed under a CC-BY 4.0 license.vessel walls and dissolves in interstitial tissue fluid (Figure 5.3). The net increase indissolved oxygen results in a dramatic and measurable decrease in T1. This changeis reversed soon after the patient is switched back to breathing atmospheric air asexcess oxygen is expelled or metabolized. Perfused tumour regions (i.e regions thatalready have a high Hb-O2 saturation) will see a measurable decrease in T1. Theperfused regions that do not show a decrease in T1 must therefore be hypoxic [99].Importantly, OE-MRI does not yield any information about unperfused regions andin that region, there are likely to be pockets of viable (but hypoxic) tissue. Thoughoutside the scope of this study, blood-oxygen level dependent (BOLD) imagingmay provide insight into these regions by exploiting the paramagnetic propertiesof deoxyhaemoglobin. Dunn et al. explored the use of deoxyhaemoglobin as acontrast agent by coupling BOLD imaging with a modification to the inhaled gas68(carbogen) in intracranial rat tumours [? ]. This study [? ], and others sincethen [108, 109, 111? ] have established that T∗2 increases as tumour oxygenationimproves using a variety of breathing gases. Section 8.6 outlines this technique inmore detail and how it could be used with dOE-MRI to obtain a more completepicture of the tumour microenvironment.The oxygen status of healthy tissue is fairly well regulated in normal tissueand every cell in the body is at most 150µm away from a blood vessel. In tu-mours however, the vascular network is chaotic and the growth patterns of vesselsare abnormal leading to a defective and leaky endothelium [16]. Irregular diam-eters of tumour vessels, abnormal branching patterns and porous vessel walls allcontribute to an increase in vessel permeability and pockets of hypoxic tissue. Al-though pO2 and pCO2 are by far the largest factors in determining Hb staturation,related factors such as pH and temperature also play a role. Furthermore, these hy-poxic regions are heterogeneous, transient, and drastically differ between tumourmodels. In a mammary adenocarcinoma mouse tumour model, Sorg et al. usedspectral imaging with an implanted window chamber to show that upon breathing100% oxygen, the Hb saturation in the tumour vascular network increases from20-30% up to 70-80% while the Hb saturation in the normal vascular network doesnot change appreciably [101].5.3 MotivationThe T1-shortening property of oxygen dissolved in fluid has been known since1955 [102] and pioneering work by Young et al. showed that oxygen acts as aparamagnetic contrast agent by demonstrating its ability to reduce T1 upon inhala-tion [103]. Inhalation of 100% oxygen has also been shown to elicit strong T1 ef-fects in the kidney[104], spleen[105] and the poorly oxygenated retina [106]. Sub-sequent oxygen-enhanced MRI (OE-MRI) efforts have included either acquisitionof quantitative T1 maps before and after oxygen breathing, or acquiring dynamicT1-weighted (T1W) signal intensity images and calculating ∆T1 during periods ofoxygen inhalation. The subtle but measurable influence of tissue oxygenation onT1 in tumours has been reported by O’Connor [23, 99, 107, 108], Mason [109–111], Gallez [112], and others [100, 105, 113, 114]. However, due to the changes69?Figure 5.3: A schematic representation of our current understanding of theorigin of the OEMRI effect. In normoxic tissue, Hb is almost fullysaturated and any excess breathed O2 cannot bind to the Hb molecule.Consequently, O2 dissolves in the blood plasma and as the excess oxy-gen diffuses out into the tissue, it also dissolves in the interstitial tissuefluid resulting in a net T1 decrease. It is hypothesized that in the hypoxictissue, Hb is not fully saturated with oxygen due to increased tissue de-mands and/or a poorly organized vascular network. The excess breathedoxygen in this case binds to the Hb molecule and does not dissolve inthe plasma leading to no change in T1 that arise as oxygen dissolves in the plasma and interstitial fluid being quitesmall, T1 maps have poor sensitivity and application of OE-MRI techniques in can-cer has yielded mixed success. OE-MRI continues to suffer from low SNR and ithas not found routine clinical use largely because isolating small signal changesdue to dissolved O2 is a challenge [99, 109].Typical imaging times for existing OE-MRI methods range from 20-45 minutesoften making it impractical for easy inclusion in experimental protocols. An MRItechnique measuring tumour oxygenation that is sensitive, fast, flexible, repeat-able, and non-invasive has the potential to significantly impact the clinical fields ofradiation biology and hypoxia drug targeting. In this study, we present a new dy-70namic OE-MRI (dOE-MRI) method that allows extraction of very small dynamicsignal changes in T1W images by inducing step changes in the inspired oxygenthrough a repeated, cycling gas challenge. To isolate the signal component thatmatches cycling gas, a blind source separation approach called independent com-ponent analysis (ICA) is used to analyze MR images as first proposed by McKeownat al [115]. ICA is a form of blind source separation algorithm that separates theadditive signals on the basis of the statistical independence of individual compo-nents [116]. With the application of a cycling oxygen challenge and processing thedata using ICA, our dOE-MRI approach represents a significant improvement inthe sensitivity and application of MRI for measuring tumour oxygenation, makingit more practical for wide application.5.4 MethodsThis section of the thesis introduces and describes the dOE-MRI technique anddiscusses technical details of the technique. Its characteristics are described in sec-tions 5.5.1,5.5.3,5.5.7. An analysis of robustness is done in section 5.5.5, repeata-bility in section 5.5.8, and variability in section 5.5.4. Ninety-one ICA extractionswere done in a total of eighty-one (N=81; some mice had two distinct tumours)mice that were scanned in a variety of experimental conditions. Multiple tumourlines are explored in Chapter 6 and a tumour treatment study is discussed in Chap-ter 7. Below methods for the dOE-MRI technique are discussed, and subsequentchapters will refer to these details and provide amendments as needed.5.4.1 Mice and tumoursFemale NRG (NOD rag gamma) mice were implanted with murine squamous cellcarcinoma (SCCVII; 5x105 cells in 50 µl serum-free media; cells provided by Dr.J. Evans) in the dorsal subcutaneous region. These mice and tumours are shownin all figures in this chapter except figure 5.10, which contains the aggregate dataacross all dOE-MRI experiments). Mice were anaesthetized with isoflurane for theduration of imaging sessions until euthanasia, and were positioned supine on thecustom surface coil apparatus. Throughout the imaging session, a small animalmonitoring system (SA Instruments Inc., Stony Brook, NY, USA) was used to71monitor respiration rate, varying between 80-100 breaths per minute, and bodytemperature, maintained at 36.8± 0.5◦C using a continuous airflow heater. Allanimals were injected with 60 mg/kg pimonidazole hydrochloride (HypoxyProbe)30 min prior to imaging to label hypoxic cells and were euthanized within 15 minof imaging completion. Tumours were embedded and frozen in optimum cuttingtemperature medium (OCT; Tissue-TEK).5.4.2 MRI data acquisitionAll MRI experiments were performed at the UBC MRI Research Centre on a 7TBruker BioSpec 70/30 scanner at room temperature with a volume transmit coiland custom surface receive coil. Each imaging session began with pilot axial andcoronal T2-W scans for tumour localization and slice prescription. Eight contigu-ous axial slices (1 mm thickness) were acquired with an in-plane field of view of3.84 cm×1.92 cm and a matrix size of 128×64. Dynamic oxygen-enhanced MRI(dOE-MRI) scans were acquired with a 2D multi-slice FLASH-based sequencewith TE /TR = 2.67/66.7, α = 40◦, temporal resolution of 4.3 s with 198 repetitionsfor a total scan time of about 14 min. The spatial resolution and geometry for allscans in the imaging session were matched and an experienced operator outlinedthe tumour on each slice of the anatomy MR images to construct the region ofinterest (ROI) for each animal.Gas challenge during MRI: tumour-bearing mice began the dOE-MRI gas chal-lenge breathing medical air and were switched between 100% oxygen and medicalair in two-minute intervals. This paradigm continued for three cycles over a totalof fourteen minutes; gases were switched manually and each switch took about fiveseconds to complete.5.4.3 MRI data analysisdOE-MRI maps: A suite of in-house software was developed based on the tech-nique described by Hyvarinen [116]. Specifically the python machine learninglibrary scikit-learn, sklearn.decomposition.FastICA, was used [117].The FastICA algorithm is applied to serially acquired T1W images and the output isa paired set of components and weighting factors for each voxel in the dataset. Ex-72tracted independent components are not ordered and while the component selectioncan be automated, in this study an observer was assigned to select the appropriatecomponent (Figure 5.4). The number of independent components for each imag-ing session was chosen by the operator and ranged from 4-9 to ensure the cyclicbehaviour of the T1W signal intensity corresponding to the gas challenge appearedin only one component. The dOE-MRI maps were obtained by dividing the ICAweighting-factor maps by the mean signal-intensity maps to obtain a spatial mapfor the strength of a particular voxel’s contribution to the component of interest(c4 in Figure 5.4). In these dOE-MRI maps, voxels are coloured to indicate theamount by which a given pixel intensity timecourse is modulated by the oxygen-related component. The green-white-purple colour spectrum depicts the degree towhich voxels respond to the cycled gas challenge. Purple indicates O2-positivevoxels whose timecourse exhibits a higher and more positive contribution from thecorresponding ICA component, representing an increase in T1W signal intensity inresponse to the supplied 100 % oxygen. O2-negative voxels that show a decreasein T1W signal intensity with a negative contribution from the corresponding ICAcomponent under 100 % oxygen breathing are depicted as green. Regions whoseT1W signal intensity timecourses responds only weakly or not at all to the gas chal-lenge are shown in white hues. Fraction of voxels that are negative on dOE-MRImaps were correlated with the histological hypoxic fraction using Pearson’s r.OE-MRI without ICA: To assess whether or not ICA was necessary to createoxygenation maps, the MR signal intensity data was correlated with three mod-elled paradigms: 1) a square wave which corresponds to the concentration of de-livered oxygen; 2) a synthetic hemodynamic response function (HDRF) created byconvolving a square wave with an exponential (τ = 0.32ms); 3) Fourier filteredresponse curve (details below). The value of τ for the HDRF was determined bymanually fitting a portion of the ICA-extracted oxygen-enhancing component tothe HDRF with a range of τ values. Correlations were calculated voxel by voxelusing:r =Σni=1(xi− x¯)(yi− y¯)√Σni=1(xi− x¯)2Σni=1(yi− y¯)2(5.1)where x is the model paradigm and y is the T1W signal intensity timecourse. The73resulting correlation maps are estimates of the strength of the input paradigms withthe acquired signal intensity.Fourier filtered response curve: Frequency-based filtering was applied to the ex-tracted ICA component to isolate the periodicity in the signal intensity response.Fourier filtering was applied by first Fourier transforming the extracted ICA com-ponent for an animal. Next, all but the highest frequency response was filtered outand the magnitude of the signal was plotted against the frequency. Finally, an in-verse Fourier transform was applied to the filtered data and this was considered tobe the periodic response curve for the animal. This extracted response curve wasthen correlated with the mean-normalized signal intensity voxel by voxel as de-scribed in the preceding paragraph to produce the “Fourier transform” correlationmap.Extracting the periodic signal using Fourier transforms only: A standard signalprocessing method was deployed voxel-by-voxel to extract the periodicity in thesignal intensity data. First, the MR signal intensity data was normalized to themean intensity in each voxel. Then, this normalized data was Fourier transformedand the intensity at the frequency of interest (selected by Fourier transforming theICA-extracted component) was visualized voxel-by-voxel.5.4.4 Quality ScoresA coarse qualitative scoring system was developed to evaluate the quality of theICA component from all 91 extractions. The scored extractions are shown in fig-ure 5.10. This is similar to the Gleason score that radiologists use to predict theaggressiveness of prostate cancer by looking at tissue biopsies. The levels of thescoring system and the associated explanations were:• Unusable: These extractions were removed from consideration as they donot correspond to the cycling paradigm in a meaningful way.• Poor: In these cases, the extracted component corresponds to the cyclingparadigm, but either due to imaging noise, motion, or notably different phys-iological response, the extracted component is noisy.• Satisfactory: A distinct pattern matching the cycling gas paradigm is present74and recognizable.• Good: These extractions contain trends other than the cycling paradigm, andthere is some variability in the “humps”• Ideal/perfect: These extractions are ideal and show the presence of a strongcycling component with very little noise, and approximately equal “humps”5.5 Results5.5.1 ICA isolates small changes in T1W signal intensityAn example signal intensity vs. time curve is shown for a whole slice ROI com-pared with a single voxel (Figure 5.4B). A mean signal intensity increase is seenfor both the whole slice and the individual voxel during each of the oxygen peri-ods of the cycle, however the magnitude of ∆SI for the individual voxel is about10% and the noise is of the same order of magnitude. The slice-averaged time-course has much less noise compared to the individual voxel, but the size of theeffect is significantly reduced, the contrast to noise is similarly poor, and all spatialheterogeneity in the response is lost in the averaging process.To retain spatial information and improve the sensitivity of the technique, ICAwas applied to the same dataset and in this example, four independent componentswere extracted (Figure 5.4C). Each individual independent component is scaledsuch that its norm is one (||ci||= 1,∀i). Only one extracted component follows thestep function of the oxygen challenge and positively identifies an effect of oxygenbreathing (c4). Speculations for the source of the other components is provided inFigure Extracting periodicity of the signal intensity change using aFourier transform based approach is less sensitive than usingICAFigure 5.5 shows the normalized intensities of the frequency of interest in the tu-mour. The normalized mean across all the voxels was 0.07, nearly eighty times750 200 400 600 800Time (seconds)−1001020Δ SI (%)AirO2AirO2AirO2AirSlice ROI averageSingle Voxel0 200 400 600 800Time (seconds)c4c3c2c1AirO2AirO2AirO2Air A. T2-weighted ImageB. Signal Intensity TracesC. ICA Components1mmFigure 5.4: (A) T2W MRI of a tumour xenograft at 7T and (B) the corre-sponding T1W signal-time traces of a single voxel (solid yellow) andwhole-tumour slice ROI (dotted black) during gas cycling at two-minuteintervals of air (x axis; blue) and O2 (x axis;yellow). (C) Plot of the fourextracted ICA components from the entire tumour ROI, component c4(purple) exhibits the same temporal features as the oxygen cycling timecourse shown along the bottom. All components are normalized, novertical scale is shown. Figure reused with permission from Wiley andSons.lower than the value at the frequency of interest extracted from the ICA compo-nent (5.67). This indicates the SNR at the voxel level is not high enough to use theFourier filtering method to extract the periodicity of the signal intensity signal. Fig-ure 5.6 shows the spatial patterns of the intensities. Comparing Figure 5.6 and 5.8shows that the regions of high intensity at the frequency of interest are correspondstrongly to the highly O2-positive (purple) regions.760.0 0.1 0.2 0.3Intensity at F = 0.0037 Hz / 5.67  (Intensity at freq for FT of ICA component)0200400600Count of voxelsFigure 5.5: Histogram of all the normalized intensities at the frequency ofinterest in all the tumour voxels. The intensities were normalized to thevalue at the frequency of interest extracted from the ICA component(5.67), plotted on the x-axis.Figure 5.6: The normalized intensities at the frequency of interest as a spatialmap. Higher values in the spatial map are represented as green and lowintensity values are dark.5.5.3 dOE-MRI with ICA does not require assumption of a responsefunctionTo determine whether dOE-MRI maps obtained with a model-free ICA approach(Figure 5.7A) are comparable to maps assuming mathematical models of the re-sponse, alternative oxygenation status correlation maps were constructed (Fig-ure 5.7B and C). In Figure 5.7B and C, two example mathematical models - asquare wave and the estimated hemodynamic response function (HDRF) - are cor-related to the voxel-by-voxel raw time signal. Regions most correlated with theinput paradigm remained purple in both alternative maps generated from modelledresponse functions. Figure 5.8 shows the result of the Fourier filtering processas well as the resulting correlation map for individual slices of a tumour. The77values of the correlation map from the Fourier-filtered response curve were cor-related with the dOE-MRI map with Pearson’s r = 0.72. Particularly when usingthe HDRF, the alternative oxygenation map showed very similar patterns in theregions demarcated as O2-positive and O2-negative. However, the map generatedfrom correlating a square wave led to consistent underestimation of oxygenationrelative to the model-free dOE-MRI map.B. Square Wave C. HDRF0.003-0.0030A. dOE-MRI Map0O2-PositiveO2-Negative1mm Figure 5.7: (A) dOE-MRI map of an SCCVII tumour where purple voxelscontribute strongly to the extracted component using ICA in the T1Wsignal timecourses. Green voxels in the dOE-MRI map have a strongcontribution of the inverse extracted component. Pearson’s r-maps areshown correlating the raw time-signal voxel by voxel with a square wave(B), and an exponential convoluted with a square wave called the hemo-dynamic response function (HDRF) (C). Panels B and C are correla-tion maps whereas A is the dOE-MRI map from ICA. Note the lowcorrelation coefficients (on the order of 10−3) are characteristic of theextremely low amplitude of the oxygen cycling compared to other com-peting effects. Figure reused with permission from Wiley and Sons.5.5.4 Variability of response in individual oxygen cyclesA full dOE-MRI sequence involved three cycles of oxygen but to assess the po-tential for shortening the sequence we also separately applied ICA to each of thethree oxygen cycles independently. Separate dOE-MRI maps, as well as voxel-wise correlation plots of a representative SCCVII tumour, are shown in Figure 5.9with Pearson’s rall−1=0.74,rall−2=0.86,rall−3=0.84. Pearson’s r ranged from 0.79to 0.87 for a similar analysis in a representative HCT-116 tumour.The stability of the independent component extraction was assessed by un-dersampling the full timecourse threefold prior to application of ICA, and a high78--Figure 5.8: The first row shows the extracted ICA component of a whole tu-mour. The second row shows the magnitude of the Fourier-transformedsignal plotted against frequency in Hz. The highest frequency is keptand all other values are set to 0.This filtered data is inverse Fouriertransformed to produce the Fourier-filtered response curve. This re-sponse curve is plotted in the third plot. A correlation map (colour barranging from -0.01 to +0.01) of the Fourier-filtered response curve withthe mean-normalized signal intensity is shown alongside the dOE-MRImap (colour bar ranging from -0.15 to +0.15) for comparison.correlation between the dOE-MRI maps from full and three-fold undersampledtimecourses is observed (Figure 5.9E; Pearson’s r = 0.84). In Section 5.5.6, fur-ther undersampling up to a factor of six is shown with minor differences in thedOE-MRI map.79A. 3 Cycles B. Cycle 1 C. Cycle 2 D. Cycle 3 E. 3x Undersampledr=0.74 r=0.86 r=0.84 r=0.84dOE-MRI Map from 3 cyclesdOE-MRI Map from Cycle 1dOE-MRI Map from Cycle 2dOE-MRI Map from Cycle 3dOE-MRI Map from 3x undersampledO2PositiveO2Negative01mmFigure 5.9: The dOE-MRI map including the full dataset of all three cycles(A) is compared to each of the three gas cycles separately (B,C,D), andto a map that temporally undersamples by selecting every third data-point from the full dataset (E). Voxel-wise plots of each map are corre-lated to the full dataset and a linear regression with Pearson’s r is shown.Figure reused with permission from Wiley and Sons.5.5.5 Quality of extracted componentAcross 91 extractions, figure 5.10 shows the quality of the extraction process foreach dataset. A quality score of 1 was given only five times, and a score of 2 wasgiven sixteen times (out of 91). Only data with a quality score of 1 was discardedfrom analysis, and in the studies presented in this thesis, no data was discarded.A quality score of five, four, and three was awarded thirty-three, twenty-two, andfifteen times, respectively.5.5.6 dOE-MRI can be extended by interleaving other scans betweeneach repetitionTo investigate the subsampling limit of ICA, Figure 5.11 shows the dOE-MRI mapsfor progressively more severe undersampling. O2-positive and O2-negative regionsare repeatable for all maps (plot in Figure 5.11) until subsample 5, where only 40of the 200 available data points were used. The original data was acquired at a tem-poral resolution of 4.3 s but at subsample 5, the effective temporal resolution goesto 21.3 s. In other words, temporal resolution could be sacrificed for SNR simplyby averaging. More usefully, the additional time can be repurposed to interleave80Quality Score: 1 (N=5)Quality Score: 2 (N=16)Quality Score: 3 (N=15)Quality Score: 4 (N=22)Quality Score: 5 (N=33)Figure 5.10: All ninety-one dOE-MRI extractions shown with the qualityscores (1=unusable to 5=ideal) and the number of tumours shownin parentheses. Colours represent the different scores with 1 in darkgreen, 2 in orange 3 in blue, 4 in purple, and 5 in lime green.other scans for multi-parametric imaging within an dOE-MRI scan. Details of thisare discussed in section Exploring other independent components extracted using ICAFigure 5.12 shows the extracted independent components and their correspondingweighting factor maps for an application of ICA on a OE-MRI scan. Extractedcomponents typically have a mix of high and low frequency responses and may81Figure 5.11: dOE-MRI maps and associated component traces of differentlysampled data. To achieve different levels of subsampling, the raw datawas spliced and then ICA was applied. The oxygenation maps lookvery similar between different subsample factors. Temporal resolutionand number of points were 4.3 s and 200 points (subsample 1), 8.5 sand 100 points (subsample 2), 12.8 s and 67 points (subsample 3),17.1 s and 50 points (subsample 4), 21.3 s and 40 points (subsample5), and 25.12 s and 34 points (subsample 6).include temperature drifts, breathing artefacts and other motion. For instance, c4is clearly the component of interest here as the cycling pattern is not present inany other component. We speculate that c1 corresponds to a temperature driftover the course of the scan and c3 is likely related to a breathing motion artefact.Component 2 is a relatively weak spurious signal fairly low in magnitude withno obvious spatial or temporal pattern. Additional physiological monitoring datais needed for a more thorough analysis of the other independent components andwhether they can aid our understanding of the mechanism of action.5.5.8 Comparing oxygen responsiveness with dOE-MRI acrossexperimentsIt is necessary to characterize entire tumours in a compound fashion for compar-ison between groups or studies. Existing quantification methods have involvedcomputing fractions of positively responding and negatively responding voxels asa surrogate for oxygen responsiveness in tumours with dOE-MRI. However, thissemi-quantitative metric relies on a binary classification of voxels as either O2positive or O2-negative. Calculating fractions is sufficient to broadly categorize82O2PositiveO2Negative0-0.25 0.25Figure 5.12: Plots of the four components extracted from ICA are shown((||ci|| = 1,∀i) along with the corresponding weighting factor maps(normalized to mean voxel wise mean signal intensity). Corruptinginfluences such as temperature drifts are often present and produceslowly increasing or decreasing trends (for e.g., c1) and breathing arte-facts corresponding to short-lived spikes (c3). No explanation could befound for c2. Figure reused with permission from Wiley and Sons.tumours as oxygenated or not but consequently, rich information about the level ofresponse is lost. Here we present an improved quantification method of dOE-MRIdata that captures the level of response and furthermore, allows direct comparisonof data acquired at varying temporal resolutions.Since each extracted ICA component is scaled such that its norm is one(||ci|| = 1,∀i), weighting factor maps are only directly comparable between scansif dOE-MRI images are acquired at the same sampling frequency over the dura-tion of the cycling oxygen (14 minutes). However, holding the sampling frequencyfixed over different experiments is not practical as imaging tumours of differentsizes require modification of the field of view and consequently, temporal and spa-tial resolutions. A scaling s factor must be applied to scale component map valuesso they can be compared between scans of different temporal resolutions,s=√Nre f√N. (5.2)A reference sampling frequency should be chosen and in this study, it waschosen to be 0.24s−1 (corresponding to N=198 images over the cycling oxygen).Gas delivered to mice switches between room air and 100% oxygen in two83minute cycles, for a total of 14 minutes. ICA extracts the tissue response to thedelivered oxygen, and four periods of room air interspersed with three periods ofoxygen can be modelled mathematically using a general Heaviside function:y(t) =+b if t ≥ 4T7−b if t < 4T7 , (5.3)where T is the total imaging time and T/7 is the time for a single segment ofthe gas challenge. The Fast ICA algorithm places a condition on the norm of theextracted component y(t), √N∑i=1∣∣∣y(ti)∣∣∣2 = 1. (5.4)Simplifying the expression above for our y(t) (the Heaviside function), wehave:1 =√b2N∑i=112b= N−12 (5.5)With this, we can compute the scaling factor directly using bre f =198 repeti-tions as the reference:s=bNbre fs=√198√N(5.6)This scaling factor was applied to all dOE-MRI maps used in this study toretain information about the level of response to the supplied oxygen across scanswith different sampling frequencies.845.6 DiscussionIn this study, we presented an improved method for OE-MRI that employs two syn-ergistic techniques to achieve higher speed and greater sensitivity. First, a repeatedgas challenge was used to probe tissue response by introducing an independent sig-nal modulation unrelated to spurious contributions such as temperature drifts andmotion. A repeating gas challenge improved the detection sensitivity of small am-plitude signal changes that are typical of oxygen-enhanced MRI. Second, a repeat-ing signal modulation enabled further improved sensitivity through the use of ICA,a signal processing technique to isolate source signals - T1W changes due solely tothe cycling oxygen - without knowledge of the tissue response (Figure 5.4). Whileit is possible to generate correlation maps of the oxygen cycling paradigm withT1W signal changes that appear very similar to dOE-MRI maps, an a-priori as-sumption of a response function is required for this approach (Figures 5.7 and 5.8).In theory, Fourier-filtering may be useful in isolating the periodicity of the oxygenresponse on a voxel-by-voxel manner. The overall mouse response may then beobtained by averaging the frequency response across the entire tumour. However,the individual voxel time courses are quite noisy and the Fourier filtering processis prone to fail, particularly in regions where no oxygen dissolves in the plasmato reduce T1. Similarly, Fourier transforming the signal intensity data voxel-by-voxel and visualizing the spatial map of the intensity at the frequency of interest(Figure 5.6) suggests the SNR in individual voxels is too low to apply this tech-nique. Admittedly, the Fourier-filtered response curve extracted after performingICA is a somewhat contrived exercise. However, this analysis very clearly showsthat even in the best-case scenario - applying a Fourier Transform on the extractedICA curve - it is not appropriate to model the oxygen response with a sinusoid.Furthermore, presupposing a particular oxygen response function biases the iden-tification of responding O2-positive voxels (Figures 5.7 and 5.8) underscores theneed for a model-free approach to extract the oxygen-responding component.A robustness analysis of all ninety-one extractions is shown in figure 5.10.Though component selection is currently a manual process, Figure 8.3 (appear-ing in the last chapter of this thesis) clearly shows that the number of componentsselected has almost no bearing on the dOE-MRI maps produced. The technique85appears quite robust as over 75% of the tumours imaged had a quality score of“satisfactory” or higher. Only five datasets were excluded from subsequent analy-sis (quality score of 1). Typical reasons for a score of 1 (unusable or no extractedcomponent) were excessive motion or poor SNR but in some cases it was difficultto explain poor extraction. Further work is needed to explore whether the poor ex-traction is due to an actual physiological response (or lack thereof), histologicallydistinct tumour microenvironment, excessive motion, or MR signal/noise consid-erations. Indeed, it may be interesting to further investigate the tumours/animalsthat do not appear to respond to the oxygen stimulus in a consistent and expectedway.However, the unambiguous match of the identified components for a vast ma-jority of the extracted components with the periods of the gas cycles increasesthe confidence that the small T1W signal changes result from increased oxygendissolved in the plasma and interstitial tissue fluids. Maps from other extractedcomponents (Figure 5.12) exhibit spatial patterns that could provide clues to thesignal sources but associating meaning to them is challenging and would requireadditional data. For example, even moderate shifts in temperature could drive ameasurable change in T1W signal during the timecourse, and a physiological mon-itoring system that is temporally synchronized to the MR acquisition could illu-minate this confounding variable. Nevertheless, we have established reliability ofthe technique by comparing maps from each cycle of the gas challenge to the mapincorporating data from all three oxygen-cycles and have found no significant dif-ferences. In fact, the strong correlations between dOE-MRI maps from each of theindividual cycles of the gas challenge (Figure 5.9) show that it is technically feasi-ble to assess tissue oxygenation within 6 minutes. Performing this analysis againwith three-fold (Figure 5.9) and six-fold (Section 5.5.6) temporally under-sampleddata suggests that there is sufficient SNR to successfully extract the oxygen respon-sive component with even a subset of the data. Though we have shown that thereis sufficient SNR to undersample and/or interleave, this is only applicable to thecoil we are using, at a particular field strength and in a particular tumour model.It would be prudent for investigators to explore the limitations of undersamplingin other situations by initially acquiring the full dataset in a pilot experiment andsubsequently analyzing subsets of it to evaluate whether there is sufficient SNR.86dOE-MRI offers a versatile technique where the duration of the cycles and gaschallenge, temporal resolution and desired signal-to-noise can be modified basedon the imaging objectives, which could include investigating intermittent perfu-sion or intervention-mediated changes in the tumour microenvironment. Of note,supplying excess oxygen to hypoxic tumour cells over time has the potential for in-creasing the baseline oxygen concentration, effectively reducing the hypoxic frac-tion and altering the tumour microenvironment [100]. This would result in voxelsbecoming more oxygen responsive over progressive oxygen cycles and would de-pend on the tumour characteristics as well as the duration of the oxygen challenge.This was not observed on the time scales in our study when using ICA to extractchanges in T1W signal intensity just due to the gas challenge. Should it arise inother contexts it could possibly be mitigated by extending the air-breathing part ofthe cycle, or by extracting that as a separate component using ICA. The potentialfor creating a hyperoxia steady state by modulating oxygen duration is discussedfurther by Losert et al. [118].Depending on the application of dOE-MRI, quantitative O2-positive and O2-negative fractions can be obtained from dOE-MRI maps as shown in this study, bydeploying group ICA techniques [119], or setting significance thresholds using at-test [120] and computing z-scores [121]. In a promising study, White et al. hasshown that OE-MRI may be very relevant in developing prognostic factors to pre-dict tumour response to hypofractionation by stratifying tumours that may benefitfrom oxygen breathing during irradiation [110]. Featherstone et al. have recentlyexplored pre-clinical datasets using feature-extraction and clustering analysis andthis may prove fruitful in understanding the behavior of subregions within a tu-mour microenvironment [122]. Future work to evaluate the utility of dOE-MRIwill ultimately depend on its context-dependent validation as a relevant measure oftumour hypoxia to dynamically characterize the clinically relevant oxygen statusof tumours, relating this information to treatment sensitivities and outcomes.5.7 ConclusionsIn this study we extend existing oxygen-enhanced MRI techniques by adding acycling element to the respiratory challenge and using a blind-source separation87signal processing technique (ICA) to extract the oxygen responsive componentand responding voxels. In the following chapters, we advance development andapplication of the technique and further refine it. In chapter 6 we attempt to val-idate dOE-MRI measurements in vivo using pimonidazole, a histological tumourhypoxia marker. To further explore the oxygen dynamics in tumours we also char-acterized the oxygen replenishment curve as originally proposed by Losert et al. inthe brain [118]. In chapter 7 we deploy the technique to assess tumour oxygenationchanges following treatment with bevacizumab, an antiangiogenic agent.88Chapter 6Validation of oxygen-enhancedMRI in animals6.1 IntroductionRecently our group has proposed a new technique,dOE-MRI to assess tumour oxy-genation in vivo using MRI [3]. In this chapter we first extend the technique tocharacterize the extracted oxygen enhancing component used in DCE-US experi-ments. This model is then fit to the extracted ICA components to assess the feasibil-ity of using the fit parameters as biomarkers of oxygen response. Finally comparedOE-MRI maps of tumour xenografts to slice-matched histological sections.6.1.1 Theory: Modelling oxygen responseA vascular network is a collection of multiple vessels including large arteries feed-ing smaller arterioles that ultimately deliver oxygen and nutrients to tissue via cap-illaries. This complex vascular tree is modelled by assuming a fractal branchinggeometry with the distribution of vessel sizes and flow rates given by a lognormalvelocity distribution [123]:P(v) =1σ f√2pi· 1v· exp(−12(ln(v)−u fσ f)2)(6.1)89Where P(v) is the probability density function of the velocity v, µ f and σ f arethe mean and standard deviation of the distribution for the normally distributed ran-dom variable ln(v). The mean and standard deviation of a lognormal distributionare known quantities,Mean(v) = exp(µ f +σ2f2)(6.2)Var(v) = exp(2µ f +σ2f) · (exp(σ2f )−1) (6.3)AVascular Network Representative Single VesselFigure 6.1: Schematic representation of the complex vascular network (left)with multiple velocity profiles (arrows) and its equivalent representa-tion as a single vessel with a distribution of flow profiles (right). Thedistribution of velocity profiles of the representative single vessel is thelognormal distribution. Schematic representation re-created from Hud-son et al. [9].Figure 6.1 shows a schematic of the vascular network modelled as a represen-tative single vessel with a lognormal distribution of velocity profiles. Therefore theflow function for a vascular network is given by,90F(z, t) =A2· erfc(ln(z/t)−u fσ f√2)(6.4)Where A is defined as the total vascular area, z represents the spatial displace-ment of oxygen through the blood stream, µ f is the mean velocity and σ f is thestandard deviation of the velocity distributions. The velocity v was transformed forconvenience to z/t to facilitate integration over the slice thickness z.This lognormal flow profile model has been used to describe the replenishmentof microbubbles injected into the blood stream in DCE-US [9]. The generalizedrepresentation of the replenishment time-intensity signal requires two components:an ultrasound beam-specific profile that is weighted in the z-direction due to a non-uniform slice profile and the flow profile. The expression is,S(t) =∫VB(x,y,z) ·F(x,y,z, t) ·dV (6.5)where B(x,y,z) = 1 in MRI because of relatively uniform slice profiles. Inte-grating over the imaging plane (x,y) and assuming a slice thickness of 1mm (z),the final model for fitting dOE-MRI data becomes,S(t)=[A2(zerfc(ln(z/t)−µ f√2σ f)− teσ2f /2+µ f erf(σ2f +µ f − log(z/t)√2σ f))]z=1mmz=0(6.6)Where A is defined as the total vascular area, µ f is the mean velocity and σ fis the standard deviation of the velocity distributions. Since this is a generalizedmodel that relies solely on the lognormal flow profiles of blood vessels, it is alsoapplicable for dOE-MRI data. Excess inhaled oxyen that dissolves in the plasmaacts as a contrast agent that enters the imaging plane. The model makes no assump-tion about the type of contrast agent, mechanism of enhancement, nor kinetics ofthe agent in describing the “replenishment” of signal so it was applied to a portionof the ICA-extracted oxygen-enhancing component.916.2 Methods6.2.1 AnimalsFemale NRG (NOD rag gamma) mice were implanted with murine squamous cellcarcinoma SCCVII tumours (5×105cells in 50 µl serum-free media; cells providedby Dr. J Evans) or with human colorectal carcinoma HCT-116, human ovarian car-cinoma SKOV3 or human breast carcinoma BT-474 tumours (each as 10× 106cells in 50 µl serum-free media; cell lines obtained from the American Type Cul-ture Collection) and were imaged when the largest tumour diameters reached ap-proximately 8-10 mm. All mice were injected with 60 mg/kg pimonidazole hy-drochloride (HypoxyProbe) 30-min prior to imaging to label hypoxic cells andwere euthanized within 15-min of imaging completion. Mice were anaesthetizedwith isoflurane using 1.5-2.0% isoflurane for the duration of MR imaging sessionsuntil euthanasia, and were positioned supine on the custom surface coil apparatus.Throughout the imaging session, a small animal monitoring system (SA Instru-ments Inc., Stony Brook, NY, USA) was used to monitor respiration rate, varyingbetween 80-100 breaths per minute, and body temperature, maintained at 36.8 ±0.5◦C using a continuous airflow heater. Tumours were embedded and frozen inoptimum cutting temperature medium (OCT; Tissue-TEK) with their largest diam-eter 8-10 mm.6.2.2 ImmunohistochemistryCo-planar MRI slices and histological sections were obtained by imaging per-pendicular to the longest tumour axis in MRI and serial-step 10 µm cryosec-tions were cut at 0.5-mm intervals in the same plane. Slides were then fixedin acetone-methanol for 10-min and whole sections were immunohistochemicallystained [124] for CD31 (visualized using secondaries labeled with Alexa 647nm)to label blood vessels, and for pimonidazole (HypoxyProbe-1; visualized usingsecondary labeled with Alexa 546nm) to label hypoxic cells. Sections were thenstained using Hoechst 33342 (bisbenzimide) to label all cell nuclei. Whole-tumoursections were imaged using a robotic fluorescence microscope (Zeiss AxioimagerZ1), a cooled, monochrome CCD camera (Retiga 4000R; QImaging), a motorized92slide loader and x-y stage (Ludl Electronic Products) and customized ImageJ soft-ware [60]. Adjacent microscope fields of view were tiled such that images of entiretumour cryosections were captured at a resolution of 1.5 µm/pixel. Using anatom-ical landmarks and accumulated thicknesses of serial-step sections as estimates ofdistances from the edges of whole tumours, sections were chosen to match the MRslices. ImageJ and user-supplied algorithms were used to super impose digital im-ages which were then manually cropped to tumour tissue boundaries with stainingartifacts removed. A threshold was applied to images to identify positive pimonida-zole staining, and the number of positive pixels was determined as a percentage ofthe total number of pixels in the tumour image. Overlaid greyscale images wereconverted to false colour for visualization with pimonidazole as green and CD31as magenta.6.2.3 MR ImagingAs previously described in Section dOE-MRI AnalysisAs previously described in Section Model FittingEquation 6.6 was fit to the first 100 seconds of the ICA component trace afteroxygen was first delivered to the mouse. Model fitting was done with the LMFITpackage [125] using the Levenberg-Marquardt method. A free fit parameter wasadded corresponding to a vertical offset to account for the ICA component startingbelow 0. For each fit, three parameters describe the shape of the oxygen response:A (vascular area), µ f (mean velocity) and σ f (standard deviation) of the velocitydistributions. A sample fit using equation 6.6 is shown in Figure 6.2.93Figure 6.2: Fit of equation 6.6 to an oxygen response curve with the residualsplotted along the x-axis for each point. For this fit, A = 0.18, v f =0.66mm/s, and σ f = 0.58mm/s.6.3 Results6.3.1 ICA enabled dOE-MRI detects variable oxygenation in a rangeof tumour modelsTumours of human and murine origin and comprising a variety of tumour microen-vironments were imaged, including fast growing, highly vascularized murine squa-mous cell (SCCVII) and human ovarian carcinomas (SKOV3), slower growing andwell vascularized human breast cancer (BT-474), as well as a relatively fast grow-ing but more poorly vascularized human colon colorectal carcinoma (HCT-116).The inter-model heterogeneity of the tumours is reflected in the mean fraction ofnegative voxels in the dOE-MRI maps, which were 46 ± 6% for BT-474, 36±3%for HCT-116, 31±5% for SCCVII, and 14±4% for SKOV3 tumours. Consider-able intra-tumour heterogeneity is also observed within some models, particularlythe BT474. dOE-MRI maps representing the mean fraction of negative voxels are94shown for each tumour type in Figure 6.3.A. BT474 B. HCT-116 C. SCCVII D. SKOV31mm O2PositiveO2Negative0-0.15 0.15Figure 6.3: Top: dOE-MRI maps for four tumour models HCT-116, BT-474,SCCVII, and SKOV3 are shown. Chosen slices are representative ofthe mean percent negative dOE-MRI fraction for the respective tumourmodel. Bottom: The box-whisker plot shows the quartiles of percentnegative dOE-MRI voxels for all imaged tumours. Figure reused withpermission from Wiley and Sons.6.3.2 Vascular area A and mean velocity v f do not vary acrosstumour models, but σ f differentiates between tumoursEquation 6.6 was fit to the oxygen response kinetics for three tumour models (SC-CVII, HCT-116, and BT474). Figure 6.4 shows the raw data and the fit for eachanimal. Group averages for each of the three parameters A, v f and σ f are shown inbox plots. The parameter σ f discriminated between tumour types in this analysisand σ f for SCCVII tumours (σ f = 0.30± 0.09 mm/s) was statistically significantlylower than the HCT-116 tumours (σ f = 0.52 ± 0.12 mm/s; p = 0.047) as well asthe BT474 tumours (σ f = 0.76 ± 0.19 mm/s; p = 0.044). The effect size for bothcomparisons was quite small: Hedge’s g = 0.05 for SCCVII vs. HCT-116 tumoursand g = 0.02 for SCCVII vs. BT474. No significant difference was found whencomparing σ f of HCT-116 and BT474 tumours. High intratumour variability waspresent in all tumours, but particularly in the BT474 tumour models. The SKOV3tumours were not considered as part of this analysis due to low sample size (N=3).95Figure 6.4: Top: Individual fits to the oxygen response curve for each animaland tumour type. Bottom: Box plots showing the median value andquartiles for A, v f , and σ f across the three tumour models.6.3.3 dOE-MRI maps correspond to matched histology sectionsTumour tissue cryosections obtained to match MR imaging slices were stained forvasculature (CD31) and regions of pimonidazole-labeled hypoxia and are com-pared side-by-side; Figures 6.6 and 6.7 provide five examples for each of SC-CVII and HCT-116 tumour models for detailed review. Generally, in correspond-ing dOE-MRI maps for both tumour models O2-positive voxels align with themost oxygenated regions of histology sections, where pimonidazole labeling isabsent, however many areas of mismatch are also observed. More consistent isthat O2-positive voxels do not typically correspond to tissues identified as hypoxicin the histology sections (i.e. labeled with pimonidazole). In general, the morenecrotic HCT-116 tumours have fewer O2-positive regions and significantly moreO2-negative regions in the dOE-MRI maps, compared to the SCCVII tumours thathave no necrosis. Pimonidazole labeling is heterogeneously dispersed within re-96gions of viable tissue containing tumour blood vessels for both SCCVII tumours,Figure 6.6, and HCT-116 tumours, which typically have greater amounts of necro-sis, Figure 6.7. Figure 6.5 shows the fraction of negative dOE-MRI voxels cor-related with the histological hypoxic fraction. For SCCVII tumours (n=9) therewas good correlation (Pearson’s r = 0.76) when using all tumours but it is impor-tant to note that one particular tumour had a much higher histological hypoxic andnegative dOE-MRI voxel fraction. Excluding this tumour and its three slices, thecorrelation dropped to r = 0.14. The correlation in the HCT-116 tumours (n=6)was similarly poor, with r = 0.037.Figure 6.5: The proportion of negative dynamic oxygen-enhanced MRI(dOE-MRI) voxels is plotted against the histological hypoxic fractionswith Pearson’s r = 0.76 for SCCVII tumours (r = 0.14 after excludingtumour with high hypoxic fraction) and r= 0.037 for HCT-116 tumours.Each point is a slice average. Figure reused with permission from Wileyand Sons.6.4 DiscussionUsing the dOE-MRI technique, we found that an oxygen-enhancing componentwas extracted successfully in all imaged animals across a range of tumour modelsand environments (Figure 6.3). In this study we also modelled the oxygen responsecurves in the tumour vasculature by deploying models developed in DCE-US and97ICA Component TracedOE-MRI Pimonidazole1mmO2PositiveO2Negative0-0.15 0.15Figure 6.6: SCCVII murine tumours with slice-matched histological imagesdepicting pimonidazole-labeled hypoxia (green) and CD31-stained vas-culature (purple) are shown next to the dOE-MRI parameter maps sim-ilarly colored with O2-positive (purple) and O2-negative (green) ar-eas. Corresponding ICA extracted components are also shown. Figurereused with permission from Wiley and Sons.98PimonidazoleICA Component dOE-MRI Map1mmO2PositiveO2Negative0-0.15 0.15Figure 6.7: HCT-116 human colorectal xenografts with slice-matched his-tological images depicting pimonidazole-labeled hypoxia (green) andCD31-stained vasculature (purple) are shown next to the dOE-MRIparameter maps similarly colored with O2-positive (purple) and O2-negative (green) areas. Corresponding ICA extracted components arealso shown. Figure reused with permission from Wiley and Sons.99adapting them for dOE-MRI. Fitting equation 6.6 to the oxygen response curvesresults in three physiologically relevant parameters though their applicability withoxygen as a contrast agent has not yet been established. Additional work is neededto explore how interpretation of these parameters may change when the contrastagent is intravascular. Nevertheless, a phenomenological fit of this model to ourdata provides interesting insights and use of the model over a standard exponentialfunction is supported by the Akaike information criterion (data not shown). Un-surprisingly, the vascular area A and the mean velocity v f do not appear to varybetween the three tumour models studied(Figure 6.4). The third parameter σ f isthe standard deviation of the velocity distribution and is related to the structuralorganization and morphology of the vascular network [9]. Thus, we hypothesizedthat the oxygen response curves and in particular, σ f may capture informationabout the vascular organization in different tumour models. The SCCVII tumour isa very aggressive and fast growing tumour [126], which implies a chaotic vascularnetwork with a low degree of fractality (propensity to branch in a fractal patternwith large arteries branching to smaller arterioles, leading to extremely small cap-illaries). This behaviour would result in a narrow range of vessel sizes and velocityprofiles in our lognormal velocity profile, and manifests in a low σ f value for SC-CVII tumours relative to the other tumour lines. While only σ f appears to beuseful in delineating tumour models, v f and A may be useful metrics to quantifythe oxygen response curves of tumours after drug interventions.A limitation of OE-MRI is the difficulty in interpreting areas that do not show areduction in T1 as they may be either dead tissues that are unperfused and not oxy-genated or living, viable tissues that are perfused but not oxygenated due to pooroxygen content of the supplying vessels. The latter population are of greater inter-est to the oncology community as it is these hypoxic but viable cells that have sig-nificant influence on treatment outcomes [127]. Though the dOE-MRI techniquepresented here did not successfully correlate with histological hypoxic fractions,there is promise in the technique to map oxygenation in tumours in the absence ofnecrosis (Figure 6.5). Additionally, it is challenging to compare to in vivo mea-sures of oxygenation with pimonidazole staining intensity. Further work is neededto refine the techniques so that dOE-MRI measures better correlate with histolog-ical indices of hypoxia. A quantitative comparison in the HCT-116 tumour line100showed poorer association with dOE-MRI oxygenation status likely attributableto the much higher amounts of necrosis typical of the HCT-116 model relative toSCCVII (Figs. 6.6 and 6.7). Mitigations to this limitation have been explored else-where and generally require a perfusion mask or T ∗2 - either technique can be addedto the method proposed here to exclude necrosis and further improve sensitivity ofthe technique.Application of existing OE-MRI techniques across a range of tumour modelswith varying perfusion characteristics has yielded mixed success without mask-ing for perfused tissue. For instance, O’Connor reported that in the highly per-fused 786-0-R tumour lines, 85-96% of all imaged tumour voxels were deemedto be oxygen-enhancing [99]. In those tumours, there was a good correlationbetween histological hypoxic fraction and oxygen refractory voxels. However,in the more weakly perfused SW620 tumours where only 76% of the voxels areoxygen-enhancing, there were no significant correlations with the histological hy-poxic fraction. These issues were resolved by combining OE-MRI with DCE-MRIas a perfusion mask to select only perfused voxels for oxygenation assessment,thereby distinguishing between the viable hypoxic environment and necrotic deadtissues and improving the specificity and sensitivity of OE-MRI data. Using anIAUGC60 map from DCE-MRI as a mask to obtain Oxy-R fractions O’Connor etal. showed good correlation with the histological hypoxic fraction [99]. In morerecent work, Little et al. showed oxygen enhancement in tumours with a histolog-ical hypoxic fraction as high as 43% [108] and this translated very well to a studyof six renal cell carcinoma patients. Linnik et al. reported excellent correlationbetween percentage of “negative AUCOE” (O2-negative) voxels and percentage ofhypoxic areas in the highly vascular preclinical U87MG tumour xenografts [100].A second approach for differentiating between viable but hypoxic regions and un-perfused dead tissues, is to combine OE-MRI with T ∗2 W acquisition and the BOLDeffect to classify regions [108–110, 128, 129] that show both effects. Excess oxy-gen in the blood will induce changes in Hb saturation, which alter the T2* resultingin a robust measure of areas with functioning vasculature. Conceivably, the savedacquisition time achieved with under-sampling T1W dOE-MRI suggests that T ∗2 Wimages could also be acquired to concurrently assess the blood oxygen level de-pendent (BOLD) response.101Within the relatively short 14-minute imaging time, both the HCT-116 and SC-CVII tumours show only minor changes in the oxygenation maps between cycles(Figure 5.7). In longer imaging sessions, or during administration of an interven-tion, these same tumours may exhibit varying oxygenation patterns between cycles.Periods of oxygen-starvation and re-oxygenation in tumours have been termed in-termittent hypoxia and can arise due to temporary vessel occlusions [92, 94]. Re-cent work on measuring intermittent hypoxia in patients using R∗2 [130] shows thatinterest in this phenomenon continues but the importance of intermittent hypoxiain tumours is unclear largely due to poor availability of techniques to measure it inthe clinic [93]. The relatively short imaging time for dOE-MRI makes assessingtemporal oxygenation changes possible within a timescale on the order of minutesby comparing correlation maps generated from sequential cycles.Typically, histological validation of MR data is done by collapsing rich his-tology data into a single metric, such as a hypoxic fraction, with whole-tumouror single-slice average comparisons. While this is sometimes a useful validationapproach, it may not reflect the highly heterogeneous patterns of hypoxia that areknown to vary spatially and temporally, even within the same tumour, as well asbetween tumour types as highlighted in Figures 6.5-6.7. Further complications areencountered with respect to validation of tools to assess hypoxia considering thathypoxia is not simply a binary metric. Instead, tumour oxygenation exists as aspectrum beginning with some tissues that may be normoxic, at levels similar toneighbouring normal tissues of origin, and can continue decreasing through lev-els of hypoxia to near anoxia where cells are still viable but are no longer able toproliferate. Eventually cells die in the absence of oxygen, and when this occurs inlarge numbers there can be significant regions of necrosis in solid tumours. A rangeof oxygenation levels are likely to be present in the highly heterogeneous microen-vironments of all solid tumours, but what is of interest to the oncology communityis clinically relevant hypoxia [96]. This refers to measurable tumour oxygenationlevels that are biomarkers of physiologically meaningful phenomena, includingpatient prognosis or tumour sensitivity to treatments, such as immunotherapy orradiotherapy. The relevant oxygenation status for any biomarker of interest may in-clude levels spanning from moderate to severely hypoxic. Pimonidazole has beendemonstrated as a clinically relevant marker of hypoxia, but poor or inconsistent102correlation with pimonidazole, as we have seen in our imaged tumours, does notexclude other measures of tumour oxygenation from potential utility.Measurable pimonidazole-adduct formation occurs when the O2 tension in thevicinity drops below 10 mmHg [131] but in dOE-MRI, O2-positive voxels are ex-tracted as excess oxygen dissolves in the plasma and interstitial tissue fluid to de-crease T1. Voxels where T1 has significantly increased has previously been corre-lated to poorly perfused regions and likely corresponds to hypoxic regions wherethe excess oxygen is picked up by deoxyhemoglobin molecules [100, 128, 132].The exact mechanism for a T1 increase as a result of oxygen inhalation has notyet been confirmed [100, 109], however, based on careful work of Silvennoninet al., characterizing behavior of T1 in fresh bovine blood [133], we speculatethe corresponding T1W signal decrease may arise due to the conversion of de-oxyhemoglobin to hemoglobin in the perfused vessels of hypoxic regions . O2-positive regions in dOE-MRI maps are generally in good agreement with well per-fused areas of histology images for both HCT-116 and SCCVII tumours, as shownin Figures 6.6 and 6.7. Voxels exhibiting signal reduction with the O2 stimulusin dOE-MRI maps (O2-negative, green) typically correspond with histology (pi-monidazole, green) but not all pimonidazole-labeled regions appear as O2-negativevoxels. Similarly, CD31-stained tumour regions are not exclusively O2-positive inthe dOE-MRI maps because not all tumour vessels are perfused. In fact many per-fused blood vessels are only intermittently perfused and consequently, the measure-ment of hypoxia is time-sensitive. Mismatches between dOE-MRI and histologymay be attributed in part to the different sensitivities and detection thresholds formeasuring hypoxia and oxygenation in the dOE-MRI and histology-based modal-ities, as well as potential mismatch between the timing of pimonidazole-labelingand dOE-MRI data acquisition. A current limitation of validating dOE-MRI mapsis that visual matching of purple regions in dOE-MRI maps and “non-green” re-gions in histology sections requires some mental gymnastics and a trained eye.Collapsing the rich spatial information to a tumour or slice mean is not an optionso future work is needed to fully validate this relationship in a rigorous manner. Aproposed experiment and analysis pipeline would require robust matching of MRIand histological sections followed by deformable image registration to map MRIregions onto histology sections with image re-sampling as needed. This would103permit microregional analyses and quantitative validation of dOE-MRI oxygena-tion maps accounting for spatial heterogeneity.6.5 ConclusionsIn this chapter, we compared dOE-MRI oxygenation with pimonidazole stainedhistology images and showed that O2-positive regions in dOE-MRI maps gener-ally correspond to pimo-negative regions in histological sections. The versatilityof the technique was apparent due to its applicability in multiple tumour mod-els; though, the presence of large necrotic areas in tumours poses some challengeswhen comparing oxygenation status with pimonidazole staining. We also appliedthe lognormal velocity profile model from DCE-US to oxygen response curves ob-tained from ICA and dOE-MRI and the parameter σ f is useful in quantifying thevascular morphology of tumours. dOE-MRI assesses tumour oxygenation fairlyreliably in vivo and the next application is to see whether the technique is sensitiveto oxygenation changes brought on by chemotherapies.104Chapter 7Applications of oxygen-enhancedMRI7.1 PrefaceA new term has been introduced in this chapter to compare what has previouslybeen referred to as “dOE-MRI values”. The normalized component weighting fac-tor value is now shortened to be NCWF.7.2 IntroductionDynamic oxygen-enhanced MRI (dOE-MRI) has recently been proposed by ourgroup to assess tumour oxygenation in vivo using MRI [3]. This techniquemeasures T1-weighted changes in tissues in response to a cycling oxygen chal-lenge, with the responsive signals detected using Independent Component Analy-sis (ICA). Briefly, oxygenation can be assessed in vivo by administering a simplegas challenge (switching between two-minute periods of medical air and 100%O2) during the dOE-MRI scan and then using ICA to extract the tissue response.ICA is a blind-source separation algorithm that separates multiple signal sourcesby maximizing statistical independence of individual components [116]. Variousflavours of oxygen-enhanced MRI (OE-MRI) have been proposed but all essen-tially leverage the paramagnetic properties of inhaled oxygen. White et al. have105shown that OE-MRI may be very relevant in developing prognostic factors to pre-dict tumour response to hypofractionation by stratifying tumours that may benefitfrom oxygen breathing during irradiation [110]. In this study, we apply dOE-MRIto mice bearing tumour xenografts to assess the effect of a common antiangio-genic agent. We hypothesized that dOE-MRI with group-ICA can detect Vascularendothelial growth factor (VEGF) ablation-induced changes to oxygenation of SC-CVII tumours 48 hours following treatment.7.2.1 VEGF inhibition and bevacizumabVEGF is a key regulator of angiogenesis as the VEGF molecule is rate-limitingin normal and pathological growth of blood vessels. Bevacizumab is a mono-colonal antibody that binds VEGF and inhibits growth of blood vessels [134]and inhibits tumour growth. It is marketed for clinical use for tumours underthe name Avastin c© (Genentech Inc., California, USA) and its effects and out-comes have been widely reported [135–138]. Tumour xenografts receiving mono-colonal antibodies to VEGF have consistently shown reductions in quantitativeperfusion parameters from a variety of non-invasive imaging modalities includingDCE-US [139], DCE-MRI [140], and micro PET [141]. For dynamic contrast-enhanced ultrasound (DCE-US), Wang et al. reported a significant reduction inpeak enhancement, area under the curve, relative blood volume, and relative bloodflow just 24 hours after a treatment of 10 mg/kg of bevacizumab [139]. Sev-eral dynamic contrast-enhanced MRI (DCE-MRI) animal and human studies haveshown reductions in Ktrans, vp, and ve after administration of antiangiogenic agents,including bevacizumab [142]. Tumour growth delay data also clearly indicates thatgrowth is slowed after treatment with bevacizumab [142] and other similar VEGF-inhibitors [30].It is well established that after treatment with antiangiogenic agents, the nor-malized vasculature is characterized by a reduction in vessel permeability, tortu-osity, greater coverage by pericytes and a more normal basement membrane [29].These morphological changes often result in functional changes as well includingdecreased interstitial fluid pressure, increased tumour oxygenation, and improveddelivery of nutrients (and drugs) [29]. Bevacizumab is an expensive drug with106many adverse effects [135] so selective targeting of patients that are most likelyto benefit from it should reduce unnecessary toxicity of ineffective therapies andimprove its overall cost-effectiveness [137]. In this study, we present an applica-tion of dOE-MRI to assess tumour oxygenation before and after treatment withbevacizumab.7.3 Methods7.3.1 AnimalsFemale NRG (non-obese diabetic rag gamma) mice were implanted with murinesquamous cell carcinoma (SCCVII; 5×105 cells in 50 µl serum-free media; cellsprovided by Dr. J. Evans) in the dorsal subcutaneous region. Tumours were im-aged when their largest diameters reached approximately 8-10 mm. All mice wereinjected with 60 mg/kg pimonidazole hydrochloride (HypoxyProbe) 30 min priorto imaging to label hypoxic cells and were euthanized within 15 min of imagingcompletion. Mice were anaesthetized with isoflurane using 1.5-2.0% isoflurane forthe duration of MR imaging sessions until euthanasia, and were positioned supineon the custom surface coil apparatus. Throughout the imaging session, a small an-imal monitoring system (SA Instruments Inc., Stony Brook, NY, USA) was usedto monitor respiration rate, varying between 80-100 breaths per minute, and bodytemperature, maintained at 36.8±0.5◦C using a continuous airflow heater. Tu-mours were embedded and frozen in optimum cutting temperature medium (OCT;Tissue-TEK).7.3.2 ImmunohistochemistryAs previously described in Section MR ImagingAs previously described in Section 5.4.2. All scans were acquired with the samespatial resolution and geometry and an experienced operator outlined the tumouron each slice of the RARE image to construct the region of interest (ROI) for eachanimal and then transferred to all other scans. Tumour volumes were measured by107multiplying individual voxel volume (0.3× 0.3× 0.1 mm3) and the count of totalnumber of voxels included within the manually drawn ROI.7.3.4 dOE-MRI analysisAs previously described in Section 5.4.3. Briefly, in these dOE-MRI maps, voxelsare coloured to indicate the amount by which a given pixel intensity time courseis modulated by the oxygen-related component. To compare dOE-MRI maps be-tween mice with different temporal resolutions, a scaling factor was applied asdiscussed previously (see section 5.5.8). Final normalized dOE-MRI maps wereobtained by dividing each pixel of the component map for each animal with themean signal-intensity over time of the corresponding pixel in the dOE-MRI scan.Mean normalized component weighting factor (NCWF) are reported as a markerfor tumour oxygenation with high values indicating increased oxygenation whilenegative values suggest decreased oxygenation or increased levels of hypoxia.Mann-Whitney U non-parametric tests are used to assess the difference betweenexperimental groups and Hedge’s g was calculated to determine effect size whenp< Experiment SummariesGeneral methods common to both experiments have been described above, beloware implementation details for each of the two separate experiments presented inthis study.Experiment 1: Evaluating the utility of dOE-MRI to assess oxygenationimprovements after anti-VEGF ablation therapyAnimals: Seventeen (17) mice were implanted for this experiment with eight leftuntreated and nine mice treated with 5mg/kg mouse anti-VEGF antibody (B20-4.1.1, Genentech) 48 hours prior to imaging.MRI: Axial dOE-MRI scans were acquired with 90 repetitions using a 2D FLASHbased sequence with TE /TR=2.67/133 ms, flip angle α=40◦, 16 slices each 1mmthick, FOV of 3.84cm x 2.16 cm, encoding matrix of 128x72, and a temporal res-olution of 9.6s for a total scan time of about 14 minutes.108Experiment 2: Assessing oxygenation changes in intramuscular andsubcutaneous tumours Anti-VEGF ablation therapy in IM vs. SC tumoursAnimals: Thirteen (13) mice were used in this experiment. 3 mice were implantedwith SCCVII in the dorsal subcutaneous (SC) region. 10 mice were implanted inboth the dorsal subcutaneous (SC) as well as in the hind limb intramuscular (IM)region. To account for the accelerated rate of growth for tumours implanted intra-muscularly, one-fifth of the cells were implanted IM (1x105 cells in 50µl serum-free media). Separate ROIs were drawn on the T2W images to outline both the SCand IM tumours. 5 mice were treated with 5mg/kg B20 24 hours prior to imagingand 8 were untreated controls.MRI: Coronal images were acquired to enable simultaneous imaging of both tu-mours in the same field of view. All dOE-MRI scans were acquired using a 2DFLASH based sequence with TE = 2.67 ms, spatial resolution 0.3× 0.3× 1 mm3and flip angle α = 40◦. To accommodate additional slices and image both tumoursin the same scan while maintaining spatial resolution, not all imaging parametersin the dOE-MRI scan could be fixed. Table 7.1 summarizes the key differences inthe acquisition parameters for all dOE-MRI sequences used in experiment 2.Experiment Mice TR/ms Slices Repetitions Temporal resolution (ms)1 17 133 16 90 9.62 5 133 16 110 8.52 8 83 10 140 6.7Table 7.1: Summary of scan parameters for the experiments used in thisstudy.7.4 Results7.4.1 Subcutaneously implanted SCCVII tumours treated with B20are more responsive to oxygen than controlsVisual inspection of dOE-MRI parameter maps in Figure 7.1 show that treatmentof SCCVII tumours with the anti-angiogenic agent B20 resulted in an increasedoxygenation compared to untreated controls. Tumours in this experiment were109treated 48 hours prior to imaging. Group differences are shown in Figure 7.2 as astandard box plot; Eight control tumours had a mean NCWF value of 0.037±0.011and nine treated tumours had a mean of 0.094±0.037. This difference was statisti-cally significant (Mann-WhitneyU = 11, p= 0.0092) and the effect size was largewith Hedge’s g= 1.08. Considerable heterogeneity was observed between miceand within a single slice (Figure 7.1). A histogram of voxels within tumour ROIsfor all mice is shown in Figure 7.2B.7.4.2 IM tumours have a higher baseline oxygenation level than SCtumours from the same cell lineTumours implanted in the hind limb (IM) received only 20% of the cells comparedto the subcutaneous tumours to account for faster growth of IM tumours. Fig-ure 7.3 shows no statistically significant difference in tumour volumes for any ofthe groups, control or treated, IM or SC. Across 10 animals and 23 tumours intotal, the mean across all groups was V = 424± 41mm3 (mean ± standard errorof the mean). dOE-MRI was applied to mice bearing both SC and IM tumours toevaluate whether tumour implant site affects baseline oxygenation. As shown inFigure 7.4, baseline NCWF values in IM tumours were significantly higher for IMtumours compared to SC tumours (Mann-Whitney U = 1.0, p = 0.003). The effectsize was large with Hedge’s g= Effects of B20 are dependent on the tumour microenvironmentand baseline oxygenationTo investigate whether the effects of antiangiogenic agents are dependent on dif-ferent tumour microenvironments, mice were implanted with both IM and SC tu-mours. Tumours in this experiment were treated 24 hours prior to imaging. Toensure the SC tumours from mice with a double implant were similar to mice withonly a single implant, three mice were implanted only with subcutaneous tumours.No differences were found (data not shown), so the SC tumours from this experi-ment were pooled together.When comparing between SC baseline and treated groups, there was a statisti-cally significant difference (Mann-Whitney U = 3.0, p= 0.007) observed in mean110(A) dOE-MRI MapsO2-positiveO2-negativeControls5 mg/kg B20CD31 Pimonidazole(B) HistologyFigure 7.1: A) NCWF maps obtained from ICA (dOE-MRI maps) are shownfor control tumours as well as those treated with 5mg/kg B20 and im-aged 48 hours later. Of the 10-16 slices for each animal, a representativeslice was chosen. As indicated by the distribution of purple voxels, con-trol tumours show considerably less response to oxygen than the treatedtumours. Additionally, regions marked in green are considered to behypoxic; these regions were not prevalent in the treated tumours. B)A representative histology slice from a control and a treated tumour isshown stained with pimonidazole (green) and CD31 (purple).111Figure 7.2: A) Group differences of the normalized mean NCWF are shownin a boxplot. Each dot represents the mean value of a mouse with thecontrols in blue and treated in green. The differences are statisticallysignificant (p=0.0092) with a large effect size (Hedge’s g = 1.08). B)Density distributions of all voxels shows treated tumours shifting to-wards increased responsiveness to delivered oxygen (higher NCWF).NCWF values for control (0.073 ±0.009) vs. B20-treated tumours (0.119 ±0.013)with a large effect size: Hedge’s g= 1.77 (Figure 7.4). However, no measurabledifference in oxygenation was observed for the IM tumours after B20 treatment(Mann-Whitney U = 11.0, p= 0.42). Mean NCWF from B20-treated IM tumourswas 0.137±0.018, and control IM tumours was 0.146±0.017. Histological images(Figure 7.5) also suggests that control IM tumours have significantly less pimonida-zole staining compared to SC controls. Furthermore, reduction in pimonidazolestaining between control and treated IM tumours is much lower compared to SCtumours. This provides clear evidence that the effects of B20 are dependent on thetumour microenvironment, and dOE-MRI is sensitive to these differences.7.5 DiscussionHypoxic tumour cells may arise due to proliferation of cancer cells outpacing thegrowth of new vasculature, increasing the separation between blood vessels and112Figure 7.3: Calculated tumour volumes from each of the four groups isshown. There were no statistically significant differences in tumour vol-umes amongst any of the groups.forcing cells beyond the diffusion distance of oxygen from the blood supply. Analternative route to hypoxia is the consequence of poor flow, which can result inhypoxic tumour tissue due to depleted oxygen supply in the flowing blood or dueto intermittent flow of poorly formed vessels. Tumour microenvironments are dy-namic and highly heterogeneous, with variable vascular function and patterns ofhypoxia. However, the importance of hypoxia in tumours is well validated andundisputed; a meta-analysis of tumours from multiple origins found a consistentrelationship between greater hypoxia in tumours and poorer outcomes with radio-therapy [96]. While many techniques exist for measuring hypoxia in tumours, nonehave become clinical routine mainly owing to their expense, poor sensitivity andspecificity, limited availability or their invasive nature [143, 144]. For example,Eppendorf electrodes are sensitive and specific, and are a gold standard of measur-ing tissue hypoxia, but they are highly invasive, limited by sampling, and are not113Figure 7.4: A) Boxplot with four groups, 5mg/kg B20 treated and controlmice with both SC and IM tumours. Differences between control SCand IM tumours, as well as control SC and treated SC tumours are sta-tistically significant (Mann-Whitney U test; p <0.005, marked as **).There was no significantly difference in treated and control IM tumours.B) Voxel density distributions of NCWF for SC (top) and IM (bottom)treated and control tumours. Density distribution (i.e. normalized his-tograms) are shown rather than voxel counts due to uneven group size.Note the shift of the IM control tumours towards a higher NCWF value.Figure 7.5: Representative histological sections from 16 total tumours of allfour groups: 5mg/kg B20 treated and control mice with SC and IMtumours. Hypoxia marker pimonidazole staining is shown in green, andpurple indicates the presence of blood vessels stained by CD31.114widely available. Patients in the ARCON (accelerated radiotherapy, carbogen andnicotinamide) trial with hypoxic tumours (assessed retroactively using pimonida-zole labeling of biopsy samples) demonstrated a treatment benefit [145]. However,immunohistochemical analyses of markers of hypoxia such as pimonidazole areinvasive and do not sample the entire tumour. A more widely applicable hypoxia-measuring tool that overcomes the availability, expense, invasiveness, sensitivityand specificity hurdles would be of high value for stratifying patients in hypoxia-targeted trials, prognostic imaging as well as for monitoring response to treatment.In this study we have demonstrated the utility of dOE-MRI to assess tumouroxygenation changes after administration of B20. Though this is not the first reportof tumour oxygenation improvements after antiangiogenic drug treatment usingMRI methods, we provide clear evidence this change is measurable using only theswitching of inhaled gas and no injectable contrast agents. Using electron para-magnetic resonance and an injectable paramagnetic tracer (triarylmethyl radicalderivatives), Matsumoto et al. mapped increases in the partial pressure of oxygenin tumours after administration of the VEGF inhibiting antiangiogenic agent calledsunitinib. Two to four days following antiangiogenic treatment with sunitinib, theyreported a transient improvement in oxygenation [146]. Lemasson et al. obtainedestimates of blood oxygen saturation using an ultrasmall super-paramagnetic ironoxide (USPIO) contrast agent in a rat gliosarcoma model. They reported that aftersustained administration (>9 consecutive days) of the maximum dose of sorafenib,oxygen saturation in treated tumours reduced compared to untreated control tu-mours [147]. In this case, the dose and treatment schedule resulted in excessivepruning of the tumour vasculature to the point where drug and nutrient deliverybecame limited and tumour oxygenation actually reduced [148]. In our experi-ments, with the dose (5 mg/kg) and treatment schedule (single administration 24or 48 h prior to imaging), we expected a vascular normalization effect correspond-ing to an increase in oxygenation [148–150]. Despite heterogeneity in amount andlevels of oxygen response in the SC tumours, we observed an overall increase inoxygenation after administering B20 (figures 7.1, 7.2, and 7.4).In this study we provided evidence that the location of tumour cell implantshas a large impact on the microenvironment of the resulting solid tumour. Tu-mour kinetics and responses to chemotherapies differ depending on the implant115site and the effect of tumour implantation site is often not considered when assess-ing drug effect using in vivo animal studies [151]. Hubbard et al. compared thetumours derived from the VX2 rabbit carcinoma line implanted intramuscularlyand intra-abdominally and discovered animals with IM tumours had significantlyhigher levels of calcium compared to animals with intra-abdominal tumours [152].This result was attributed to the differences in venous drainage of the two sites. Inanother example, Malave et al. studied the Lewis-lung carcinoma model and re-ported a lower implant success rate for tumours implanted in the flank comparedto the foot pad, and a higher rate of metastatic nodules for tumours in the flank(indicating a reduced tumour-host immune response) [153]. Tumour implant sitealso alters growth kinetics and dramatically different tumour doubling times (indays) has been reported for implants in mouse tail (1.7), foot (1.6), chest (1.2), andleg (0.6) [154]. For the same amount of cells implanted, our experience is that IMtumours typically grow much faster, have a more stable vascular architecture, lesshypoxia (pimo staining), and increased vessel density (CD31 %). We attemptedto control for this growth rate by injecting fewer cells for the IM tumours. Ourtumour volume measurements indicate there was no significant difference betweenthe control IM and SC tumours despite the SC tumours implanted with 5 times theamount of cells implanted. We also showed that dOE-MRI can assess increasedbaseline oxygenation in control tumours when the tumour site is changed from thedorsal subcutaneous region to the hind limb: IM tumours have less pimonidazolestaining compared to SC tumours (Figure 7.5). Ultimately we showed that follow-ing treatment of IM tumours with B20, no change in oxygenation was observed,likely because IM tumours were already well-oxygenated. It is important to notethat our technique is not able to distinguish between increased oxygen concentra-tion in the plasma (or interstitial tissue fluid) and increased blood volume. Boththese effects will result in an increase of dissolved oxygen in the plasma and willlook similar in dOE-MRI maps. To decouple these two effects, an alternative con-trast mechanism (such as T∗2 or an exogenous contrast agent) is required.To fully establish the utility of dOE-MRI to assess tumour oxygenation non-invasively using inhaled oxygen or air, several subsequent experiments should beconducted. Some ambiguities remain with a T1-based dOE-MRI method as itsreflection of dissolved O2 concentrations is complicated by complex, non-linear116relationships with hemoglobin saturation and its effects on T∗2, and with vascularperfusion and blood flow that may confound interpretation of existing dOE-MRIsignal. These are further complicated by the array of physiological possibilitiesthat may be visualized by a change in T1. In future development of this method,the impact of T1-weighting should be explored to ensure effects of interventionsare not manifesting due to changes in tumour microenvironment that alter T1.A decrease in T1 suggests an oxygen-responsive area, while non-responding andnegative-responding regions may represent a variety of physiologies. These areasmay be completely unperfused and even necrotic, or they may be poorly perfusedbut viable, hypoxic tissues. Conversion of deoxygenated hemoglobin (present inhypoxic regions) to hemoglobin in the presence of oxygen results in a reduction ofT2 and T∗2 but the T1 remains largely unchanged. Further exploration of interme-diate regions (i.e. not hypoxic or well-oxygenated) using dOE-MRI and couplingit with BOLD-MRI is warranted to fully classify all areas of the tumour. An unex-plored application of dOE-MRI is its potential to monitor treatment efficacy longi-tudinally as the contrast mechanism used is completely reversible. This opens upthe possibility to do treatment interventions within a single imaging session withperfectly co-registered tumour volumes to allow for assessing oxygenation changesat the level of a single voxel. Nevertheless, dOE-MRI has tremendous potential forassessing tumour oxygenation as a non-invasive imaging method that is urgentlyneeded in the clinic.7.6 ConclusionsThrough this work we have shown that subcutaneously implanted SCCVII tumourstreated with B20 and imaged 48h later are more oxygenated than control tumours.Additionally, we have established dOE-MRI as a tool to assess baseline oxygena-tion level and demonstrated that IM tumours are significantly more oxygenatedthan SC tumours implanted in the same mice. Finally we provided evidence thatlocation of the tumour implant site has a large effect on therapy outcome as themore oxygenated IM tumours did not respond to treatment with B20. It is our ex-pectation that after further refinement and expansion, this technique will becomeaccessible and available in the clinic to screen cancer patients prior to chemo- or ra-117diotherapy prescription, and be useful for developing new hypoxia-targeting drugs.118Chapter 8Future Work8.1 IntroductionThe work presented in this chapter provides a blueprint to continue the dOE-MRIproject and explore several avenues of research. The level of maturity of eachsection varies and all attempts have been made to motivate and guide the readerthrough the methods, results, and their implications.8.2 Group ICAOne of the principal limitations of ICA is that it “does not naturally generalize toa method suitable for drawing inferences about groups of subjects” [119]. In thisthesis, we developed a quantification method to mitigate this issue by comparingcomponent weighting factor strength after normalization (Section 5.5.8). Calhounet al., has reviewed several methods for analyzing multiple subjects within a cohortusing ICA have been proposed [119]. The approach that is most relevant for thedata collected for the experiments presented in this thesis (section 7.3.5) is spa-tial concatenation [155]. Briefly, cohort data for Group ICA was constructed byconcatenating all 16 slices from the 17 subjects together in the z-dimension. Fig-ure 8.1 compares Group ICA to the standard ICA technique described in Chapter 5.The same deflation-based FastICA (python package scikit.sklearn v0.17.1)was used to analyze the data. To ensure the cyclic behaviour of the T1 weighted119signal intensity corresponding to the gas challenge appeared in only one compo-nent, the number of independent components was set to 9. Application of ICAto the spatially concatenated data produced a single oxygen-enhancing componentthat matched the temporal pattern of the gas cycling paradigm in all 17 animals(Figure 8.2). This component appears to be smoother than typical extracted oxy-gen enhancing components because it represents the group response rather than theindividual features that exist in each mouse that responds somewhat differently.[D (x, y, z, t)]n= 10n= 1ICA Group ICAD (x, y, z,n, t)Spatial, TemporalAnimalsApply ICA separately Apply ICA once10 sets of  independent components1 set of  independent componentsSpatial ConcatenationFigure 8.1: Comparison of the standard ICA technique and Group ICA. Themain difference is in the pre-processing of the MRI data comprisingspatial coordinates (x,y,z) and temporal information (t). Group ICAdatasets are prepared by spatially concatenating all animals together (n).The output of the ICA techniques also differs: in standard ICA eachapplication produces a set of independent components whereas in GroupICA only a single set of independent components is produced.Upon selection of the single oxygen enhancing component, reshaping the resul-tant weighting-factor maps to the original matrix size provided inter-subject com-parable data. Final normalized dOE-MRI maps were obtained by dividing eachpixel of the component map for each animal with the mean signal-intensity overtime of the corresponding pixel in the dOE-MRI scan. Corresponding dOE-MRIare comparable to the methods presented in Chapter 7 and conclusions of the B20effect still hold with this analysis method. One major disadvantage of applyingICA only once to multiple animals is that the extracted component averages outany individual features.120Figure 8.2: Extracted oxygen enhancing component from ICA applied to thespatially concatenated cohort data.8.3 Further investigation of the characteristic oxygenresponse curveIn this thesis a limited number of tumours models were studied (SCCVII, BT-474, and HCT-116) so it was difficult to generalize whether modelling of the oxy-gen response curve is characteristic of the tumour environment or model. In sec-tion 6.3.2 we observed that σ f in the SCCVII tumours discriminated between therapidly growing SCCVII tumours and the other two models. Further work shouldinclude characterizing additional tumour models with varying tumour microenvi-ronments, and also comparison of the oxygen signal decay curves (i.e. when thegas is switched back to room air) as well as the enhancement curves.Another interesting feature of the ICA extraction process is that for our ap-plication, the number of components prescribed rarely matters in separating outthe oxygen-responsive component. dOE-MRI maps from ICA applied serially af-ter varying the number of components (m) from 3 to 12 are shown in Figure 8.3.The component of the signal that corresponds to the T1W signal intensity increasedue to the cycling oxygen is quite prominent and independent of the number ofcomponents selected. This analysis is presented to ensure that the number of com-ponents selected had no bearing on the extracted ICA component. Two examplesare shown in Figure 8.3 that were selected to exhbit the variability (low on the leftof Figure 8.3 and high on the right) in extracted components in a single experiment.121In both cases, the corresponding dOE-MRI maps show the same O2-positive andO2-negative regions. Interestingly, in the high variability case, when m = 10, theICA algorithm was not able to extract the oxygen-enhancing component. Explor-ing how the presence or absence of noise affects the extraction process would beuseful in further development of the technique. There may be implications worthexploring in how the variability in the extracted component actually provides in-formation about the underlying tumour microenvironment.Low variability High variability Figure 8.3: Two animals were selected to explore the effect of the numberof components (m) on the dOE-MRI maps. The two animals presentedhere were selected to exhbit the full range of variability in extractedcomponents. The low variability example shows no discernible differ-ence in the extracted component anywhere from m= 3 to m= 12. Thehigh variability example shows considerably more noise in the extractedcomponent, but the same overall trend. The corresponding dOE-MRImaps for both the low variability and high variability examples showalmost no difference in the oxygenation maps.8.4 dOE-MRI maps of 10 consecutive air/O2 switches arestableThe presence of hypoxia in tumours is known to exhibit microregional and tem-poral heterogeneity. The process of cells going through periods of oxygen-122starvation and then subsequently being re-oxygenated has been termed cyclic hy-poxia [92, 94, 156]. A leading cause of cycling hypoxia is the variable flux of redblood cells through the abnormal tumour vasculature on time scales of minutes tohours. There is some evidence of cycling hypoxia in our data as there are clearmismatches between pimonidazole staining and perfusion markers (Figure 8.5).The SCCVII tumour model has specifically been shown to be afflicted by cyclinghypoxia as early as 1986 [157]. It is therefore not surprising that dOE-MRI issensitive to these changes in this tumour model, and provides the first non-invasivedata suggesting cycling hypoxia on the relatively short timeline of less than 15 min-utes. The clinical importance of intermittent hypoxia is unclear largely due to pooravailability of techniques to measure it in humans [93]. Recent work on measuringcycling hypoxia in patients using R∗2 [130] shows that interest in this phenomenoncontinues.Despite the presence of cycling hypoxia in many regions of the SCCVII tu-mours, O2-enhancing regions in dOE-MRI maps are generally in agreement withwell perfused areas of histology images. In particular, concentrated oxygen-responsive regions within a dOE-MRI map correspond to highly vascular, perfusedregions in matching histology images. Voxels anti-correlated with the O2 stimulus(O2 refractory, green) typically correspond with pimonidazole staining but thereare instances of mismatch (Figures 6.6, 6.7, and 8.5). In addition to cycling hy-poxia, there may be possible mismatch between the sensitivities of pimonidazoleand dOE-MRI and other oxygen sensing modalities (described in [96]). Success ofdOE-MRI will ultimately depend on its validation as a clinically useful measure oftumour hypoxia.To assess this phenomenon, a pilot study was carried out to determine whetherthis cycling hypoxia can be measured with dOE-MRI. In a small pilot cohort ofSCCVII xenografts mice (n = 2), rather than the standard protocol of three cy-cles of air-oxygen switches, ten consecutive air-oxygen switches were used. ICAwas applied to each cycle of the sequences separately (as described in section 5.9).Figure 8.4 shows the results from a tumour including a dOE-MRI map (8.4A), astandard deviation map (8.4B), and a coefficient of variation map (8.4C). The stan-dard deviation and coefficient of variation maps are surrogate measures of cyclinghypoxia as they highlight regions of the tumour that show the highest degree of123deviation or variation over the ten cycles. It is expected that if cycling hypoxiaexists, these measures would show regions where it is apparent. If cycling hypoxiais not present, then the two parameter maps would be largely feature-less. In thiscase, there appears to be little evidence for cycling hypoxia as the dOE-MRI mapsare fairly consistent from cycle to cycle. Unfortunately, these results remain un-validated because we were not able to assess cycling hypoxia using histology. Oneway to assess cycling hypoxia using histology is to use two separate hypoxia stains(such as EF5 and pimonidazole) and inject them intravenously 30-60 minutes apart.Imaging the two stains and mapping the differences could confirm the presence orabsence of cycling hypoxia. A subsequent experiment where the dOE-MRI anal-ysis is repeated in more animals alongside the dual hypoxia histological markerswould help confirm this finding. As suggested by Dr. Jeff Dunn in private com-munications, another way of exploring cycling hypoxia within a single imagingsession is to reduce blood pressure to the tumour. Since the malformed vessel ar-chitecture of the tumour are unlikely to have substantial flow regulation, this shouldreduce blood flow to the tumour and result in an increase of hypoxic regions. Mea-suring the oxygenation before and after temporary blood pressure reduction shouldprovide us with an acute change in hypoxia that our technique could measure.8.5 Exploring the link between perfusion andoxygenationOne SCCVII and one HCT-116 tumour-bearing mouse were catheterized and in-jected with 30-mM solution of Gd-DTPA for DCE-MRI at a rate of 1 mL/min usinga power injector at a dose of 5 µL/g.Signal intensity timecourse from the DCE-MRI data was first normalizedto the mean signal intensity pre-injection. A numerical integration techniquethat relies on quadratic polynomials to approximate functions (Simpson’s Rule,scipy.integrate.simps) was used to calculate the area under the first 60 seconds(AUC60) of the normalized signal intensity enhancement curve. A binary ground-truth perfusion map was constructed by classifying all voxels with AUC60 > 0 asperfused and everything else as unperfused.Where dOE-MRI and DCE-MRI scans were acquired in the same SCCVII124D) dOE-MRI map for each air-oxygen cycleFigure 8.4: Results of a dOE-MRI-based analysis to sequentially analyze tenconsecutive air-oxygen switches (D). The averaged dOE-MRI map (A)across all 10 cycles reveals some hyperintense regions correspondingto oxygen-responsive areas. The voxel-wise standard deviation (B) andcoefficient of variation (C) of the ten dOE-MRI maps shows some vari-ability at the top of the tumour as well as on the mid-right of the tumour.and HCT-116 tumour-bearing mice, maps of oxygenation status were compared toAUC60 perfusion maps, as shown in Figure 8.5. Mean AUC60 for the well-perfusedSCCVII tumour was 22 ± 16 %·s and for the comparatively poorly perfused HCT-116 tumour was 7± 7 %·s. Well-oxygenated O2-positive regions generally corre-spond to perfused, high AUC60 areas in both SCCVII and HCT-116 tumours. Alarge patch of necrosis, as identified in histological section, in the HCT-116 tu-mour was also extremely poorly perfused; such large patches of necrosis were notpresent in the SCCVII tumour.125Figure 8.5: dOE-MRI maps and DCE-MRI AUC60maps and slice-matchedhistology sections of SCCVII and HCT-116 tumours. Large regionsmarked as purple in the dOE-MRI maps are O2-positive and also cor-respond to regions that have high AUC60 values (yellow). Green orO2-negative regions from the dOE-MRI map are often consistent withunperfused regions in the AUC60 (black), but there are regions of mis-match. Histology images stained with pimonidazole (green) and CD31(purple) are shown for corresponding sections.8.6 Expanding dOE-MRI to include T∗2For a variety of reasons, the past 5-10 years have seen a gradual resurgence ofoxygen-enhanced MRI and renewed interest to refine and better understand themechanism of action. One important strategy to elucidate the mechanism of oxy-gen as a contrast agent is to rely on the BOLD effect. In 2002, Dunn et al. modifiedthe inhaled gas in rats with intracranial tumours and used BOLD imaging to assesschanges in tumour oxygenation [? ]. More recently, Little et al. have shown thatsimultaneous acquisition of T1 and T∗2 images improves the specificity of oxygen-enhanced MRI [108]. Here we outline how our technique using ICA can be ex-panded to include T∗2 imaging, and what the additional information will be usedfor.T∗2 imaging would utilize the Blood Oxygen Level Dependent (BOLD) effect,which can measure shifts in hemoglobin saturation through changes in T∗2 andtherefore assess tumour perfusion without the need for injectable contrast agents.Applying an oxygen challenge also shifts the haemoglobin saturation and, thus, theT∗2 signal. The expected behaviour of a joint change in T1 and T∗2 in response to126a gas challenge, and how this can be interpreted to reflect tumour oxygenation isbased on data from recent work by Little [108] and Waterton et al. [158].The altered T∗2 provides a robust measure of areas with functioning vasculature.Subsequently, dOE-MRI maps can be masked using the ∆T∗2 maps to exclude un-perfused regions resulting in a completely endogenous technique to assess tumouroxygenation. OE-MRI T1-weighted signal more directly reflects oxygen amountsin plasma and tissues and is more applicable for measuring tumour oxygenationas it relates to radiotherapy. Without sacrificing the information obtained fromT1-weighted signal intensity in our current work, we propose to extract both T1-weighted signal intensity and T∗2 simultaneously using a dynamic, multi-gradientecho in place of a dynamic FLASH sequence. The cycling gas challenge in combi-nation with ICA improvement to T1-weighted oxygen-enhanced imaging will alsobe applicable to T∗2.A multi-gradient echo (MGE) sequence is ideal to extend our current T1 baseddOE-MRI technique to also acquire dynamic T∗2 weighted data. This is becauseinitial echoes from an MGE sequence are T1 weighted and as the echo time in-creases, the images become more T∗2 weighted. The T1w dOE-MRI map can becalculated from the signal intensity of the first gradient echo image (minimal echotime TE≈2.25 ms). The T∗2 dOE-MRI map can be created by applying ICA to themono-exponentially fitted multi-gradient echo data at each repetition. Figure 8.6outlines our approach to obtain T1- and T∗2-based dOE-MRI maps from a singlemulti-gradient echo sequence. While the temporal resolution of the multi-gradientecho technique is lower, the data quality of the T1w dOE-MRI map is not compro-mised until subsampling exceeds six times the original temporal resolution whencompared to that obtained with a FLASH sequence (see Figure 5.5.6).8.6.1 Independent vector analysisICA can be extended if the data being analyzed is multi-dimensional beyond tem-poral and spatial coordinates, for instance, T1 and T∗2 weighted images. Indepen-dent vector analysis (IVA) is the technique that permits the increased statisticalpower of two independent parameters acquired simultaneously. It is a vector-basedblind-source estimation first used in fMRI applications [159]. The problem solved127Figure 8.6: Schematic of the current and proposed acquisition and analysisfor dOE-MRI with combined R1 and R∗2 IVA is that observations that are vector quantities (i.e. data consisting of one T1-weighted signal and one T∗2 value derived from the exponential fit) are explainedby a mixture of source vectors [159]:xi =L∑jai j ◦ s j (8.1)where ◦ indicates an element-wise product. An algorithm suggested by Rafiqueet al. in 2016 solves the implementation problem and makes it available as Fas-tIVA [160]. It would be worthwhile to explore IVA as an extension to ICA oncethe T1 and T∗2 weighted data is acquired.8.7 ConclusionsIn this thesis we have presented two non-invasive imaging-based techniques toprobe the tumour microenvironment. In Chapter 2, HPG-GdF, a novel contrastagent was described and its contrast kinetics quantified using two parameters: aPSand fPV. Chapters 3 and 4 discuss applications of the technique we described in128two drugs: trastuzumab and bevacizumab. Though the same contrast agent wasused in both applications, the research questions were entirely different and wefirst tried to determine whether vascular function could explain the limited distri-bution of trastuzumab in tumours. With bevacizumab, we sought to measure thepermeability change of tumour vessels to assess whether treatment normalizes thetumour vasculature. Overall, the utility of HPG-GdF in evaluating the effects ofcancer drugs is promising but further work needs to be done to assess its applica-bility in multiple tumour models.Motivated by histological data showing a reduction in hypoxia after treatmentwith bevacizumab, in Chapter 5 we presented the dOE-MRI method which pro-vides significant improvements to the speed, and applicability of existing OE-MRItechniques. Traditional quantitative T1-mapping techniques have longer imagingtimes and are impractical for OE-MRI due to SNR and time constraints. dOE-MRIwith ICA is clinically translatable as the sequence acquisition is relatively shortand most centres already have access to dynamic T1W MRI acquisitions that manypatients already routinely receive.In Chapters 5 and 6, we showed that small changes in T1W signal intensityarising from cycling respiratory challenges can be separated robustly using ICA.In Chapter 7 dOE-MRI was used to show that treatment with bevacizumab im-proved tumour oxygenation, and that the location of tumour implants has a bearingon the vascular network that forms. dOE-MRI is an exciting, non-invasive andwidely available technique for assessing tumour oxygenation that could provide acrucial tool in the field of radiation oncology and in the development of treatmentstargeting the tumour microenvironment. We look forward to the opening of excit-ing avenues by dOE-MRI in the areas of drug development, patient stratification,and treatment monitoring.129Bibliography[1] Baker JHE, McPhee KC, Moosvi F, Saatchi K, Ha¨feli UO, Minchinton AI,et al. Multi-modal magnetic resonance imaging and histology of vascularfunction in xenografts using macromolecular contrast agent hyperbranchedpolyglycerol (HPG-GdF). Contrast Media & Molecular Imaging. 2015Aug;11(1):77–88. → pages vi, 24, 42, 45, 61, 63[2] Baker JHE, Kyle AH, Reinsberg SA, Moosvi F, Patrick HM, Cran J, et al.Heterogeneous distribution of trastuzumab in HER2-positive xenograftsand metastases: role of the tumor microenvironment. Clinical &Experimental Metastasis. 2018 Sep;19(48):1–15. → pages viii, 48[3] Moosvi F, Baker JHE, Yung A, Kozlowski P, Minchinton AI, ReinsbergSA. Fast and sensitive dynamic oxygen-enhanced MRI with a cycling gaschallenge and independent component analysis. Magnetic Resonance inMedicine. 2018 Oct;. → pages ix, x, 89, 105[4] Hanahan D, Weinberg RA. Hallmarks of Cancer: The Next Generation.Cell. 2011 Mar;144(5):646–674. → pages xix, 2[5] Hanson LG. Is quantum mechanics necessary for understanding magneticresonance? Concepts in Magnetic Resonance: Part A. 2008 Jul;p.329–340. → pages xix, 8[6] Saatchi K, Soema P, Gelder N, Misri R, McPhee K, Baker JHE, et al.Hyperbranched Polyglycerols as Trimodal Imaging Agents: Design,Biocompatibility, and Tumor Uptake. Bioconjugate Chemistry. 2012Feb;23(3):372–381. → pages xx, 18, 20, 21, 26, 43, 53[7] Go´mez-Cambronero J. The Oxygen Dissociation Curve of Hemoglobin:Bridging the Gap between Biochemistry and Physiology. Journal ofChemical Education. 2001 Jun;78(6):757. → pages xxiv, 66, 67130[8] OpenStax, Gas Exchange across Respiratory Surfaces. OpenStax CNX;2016. → pages xxv, 68[9] Hudson JM, Karshafian R, Burns PN. Quantification of flow usingultrasound and microbubbles: a disruption replenishment model based onphysical principles. Ultrasound in medicine & biology. 2009Dec;35(12):2007–2020. → pages xxviii, 90, 91, 100[10] Canadian Cancer Statistics 2018; 2018. Available from: → page 1[11] Biomarkers Definitions Working Group . Biomarkers and surrogateendpoints: preferred definitions and conceptual framework. ClinicalPharmacology & Therapeutics. 2001 Mar;69(3):89–95. → page 1[12] van der Meel R, Gallagher WM, Oliveira S, O’Connor AE, Schiffelers RM,Byrne AT. Recent advances in molecular imaging biomarkers in cancer:application of bench to bedside technologies. Drug discovery today. 2010Feb;15(3-4):102–114. → page 1[13] Hanahan D, Weinberg RA. The hallmarks of cancer. Cell. 2000Jan;100(1):57–70. → page 2[14] Folkman J. What is the evidence that tumors are angiogenesis dependent?Journal of the National Cancer Institute. 1990 Jan;82(1):4–6. → page 2[15] Laking GR, West C, Buckley DL, Matthews J, Price PM. Imaging vascularphysiology to monitor cancer treatment. Critical reviews inoncology/hematology. 2006 May;58(2):95–113. → pages 3, 5, 6[16] McDonald DM, Baluk P. Significance of blood vessel leakiness in cancer.In: Cancer research. Comprehensive Cancer Center, CardiovascularResearch Institute, University of California-San Francisco, 513 ParnassusAvenue, San Francisco, CA 94143, USA.; 2002. p.5381–5385. → pages 3, 16, 17, 51, 69[17] Miles KA, Charnsangavej C, Lee FT, Fishman EK, Horton K, Lee TY.Application of CT in the investigation of angiogenesis in oncology.Academic radiology. 2000 Oct;7(10):840–850. → page 4[18] Kather JN, Marx A, Reyes-Aldasoro CC, Schad LR, Zo¨llner FG, Weis CA.Continuous representation of tumor microvessel density and detection ofangiogenic hotspots in histological whole-slide images. Oncotarget. 2015Aug;6(22):19163–19176. → page 4131[19] McDonald DM, Choyke PL. Imaging of angiogenesis: from microscope toclinic. Nature Medicine. 2003 Jun;9(6):713–725. → page 5[20] Cuenod CA, Fournier L, Balvay D, Guinebretie`re JM. Tumor angiogenesis:pathophysiology and implications for contrast-enhanced MRI and CTassessment. Abdominal imaging. 2006 Mar;31(2):188–193. → page 6[21] Hasebroock KM, Serkova NJ. Toxicity of MRI and CT contrast agents.Expert opinion on drug metabolism & toxicology. 2009 Apr;5(4):403–416.→ page 6[22] Levitt MH. Spin Dynamics. Basics of Nuclear Magnetic Resonance. JohnWiley & Sons; 2001. → pages 6, 7[23] O’Connor JPB, Naish JH, Jackson A, Waterton JC, Watson Y, Cheung S,et al. Comparison of normal tissue R1 and R∗2 modulation by oxygen andcarbogen. Magnetic Resonance in Medicine. 2009 Jan;61(1):75–83. →pages 9, 69[24] De Leo´n-Rodrı´guez LM, Martins AF, Pinho MC, Rofsky NM, Sherry AD.Basic MR relaxation mechanisms and contrast agent design. Journal ofMagnetic Resonance Imaging. 2015 Sep;42(3):545–565. → page 12[25] Walker-Samuel S, Leach MO, Collins DJ. Evaluation of response totreatment using DCE-MRI: the relationship between initial area under thegadolinium curve (IAUGC) and quantitative pharmacokinetic analysis.Physics in Medicine and Biology. 2006 Jul;51(14):3593–3602. → page 13[26] Sourbron SP, Buckley DL. On the scope and interpretation of the Toftsmodels for DCE-MRI. Magnetic Resonance in Medicine. 2011Mar;66(3):735–745. → pages 13, 24, 34[27] Sourbron SP, Buckley DL. Classic models for dynamic contrast-enhancedMRI. NMR in Biomedicine. 2013 Aug;26(8):1004–1027. → page 13[28] Moroz J, Wong CL, Yung AC, Kozlowski P, Reinsberg SA. Rapidmeasurement of arterial input function in mouse tail from projectionphases. Magnetic Resonance in Medicine. 2013 Feb;71(1):238–245. →pages 13, 24[29] Jain RK. Normalization of tumor vasculature: an emerging concept inantiangiogenic therapy. Science (New York, NY). 2005Jan;307(5706):58–62. → pages 15, 51, 52, 106132[30] O’Connor JPB, Jackson A, Parker GJM, Roberts C, Jayson GC. Dynamiccontrast-enhanced MRI in clinical trials of antivascular therapies. NatureReviews Clinical Oncology. 2012 Feb;9(3):167–177. → pages16, 17, 39, 53, 106[31] Barrett T, Kobayashi H, Brechbiel M, Choyke PL. Macromolecular MRIcontrast agents for imaging tumor angiogenesis. European Journal ofRadiology. 2006 Dec;60(3):353–366. → pages 17, 61[32] Leach MO, Brindle KM, Evelhoch JL, Griffiths JR, Horsman MR, JacksonA, et al. Assessment of antiangiogenic and antivascular therapeutics usingMRI: recommendations for appropriate methodology for clinical trials. TheBritish Journal of Radiology. 2003;76 Spec No 1:S87–91. → page 16[33] De Bruyne S, Van Damme N, Smeets P, Ferdinande L, Ceelen W, MertensJ, et al. Value of DCE-MRI and FDG-PET/CT in the prediction of responseto preoperative chemotherapy with bevacizumab for colorectal livermetastases. British Journal of Cancer. 2012 May;106(12):1926–1933. →page 17[34] Kelly RJ, Rajan A, Force J, Lopez-Chavez A, Keen C, Cao L, et al.Evaluation of KRAS Mutations, Angiogenic Biomarkers, and DCE-MRI inPatients with Advanced Non-Small-Cell Lung Cancer Receiving Sorafenib.Clinical Cancer Research. 2011 Mar;17(5):1190–1199.[35] Bains LJ, Baker JHE, Kyle AH, Minchinton AI, Reinsberg SA. DetectingVascular-Targeting Effects of the Hypoxic Cytotoxin Tirapazamine inTumor Xenografts Using Magnetic Resonance Imaging. InternationalJournal of Radiation Oncology*Biology*Physics. 2009 Jul;74(3):957–965.→ pages 17, 19, 43, 54[36] O’Connor JPB, Rose CJ, Jackson A, Watson Y, Cheung S, Maders F, et al.DCE-MRI biomarkers of tumour heterogeneity predict CRC livermetastasis shrinkage following bevacizumab and FOLFOX-6. BritishJournal of Cancer. 2011 Jun;105(1):139–145. → page 17[37] Alic L, van Vliet M, van Dijke CF, Eggermont AMM, Veenland JF,Niessen WJ. Heterogeneity in DCE-MRI parametric maps: a biomarker fortreatment response? Physics in Medicine and Biology. 2011Mar;56(6):1601–1616. → page 17[38] Minchinton AI, Tannock IF. Drug penetration in solid tumours. NatureReviews Cancer. 2006 Aug;6(8):583–592. → pages 17, 35, 40133[39] Jain RK. Delivery of molecular and cellular medicine to solid tumors.Advanced Drug Delivery Reviews. 2001 Mar;46(1-3):149–168. → pages17, 52[40] Iyer AK, Khaled G, Fang J, Maeda H. Exploiting the enhancedpermeability and retention effect for tumor targeting. Drug discoverytoday. 2006 Sep;11(17-18):812–818.[41] Chauhan VP, Stylianopoulos T, Boucher Y, Jain RK. Delivery of molecularand nanoscale medicine to tumors: transport barriers and strategies.Annual review of chemical and biomolecular engineering.2011;2:281–298. → pages 17, 41, 48, 49[42] Weinmann HJ, Brasch RC, Press WR, Wesbey GE. Characteristics ofgadolinium-DTPA complex: a potential NMR contrast agent. AJRAmerican journal of roentgenology. 1984 Mar;142(3):619–624. → page 17[43] Hashizume H, Baluk P, Morikawa S, McLean JW, Thurston G, Roberge S,et al. Openings between defective endothelial cells explain tumor vesselleakiness. The American journal of pathology. 2000Apr;156(4):1363–1380. → page 17[44] Tang J, Sheng Y, Hu H, Shen Y. Macromolecular MRI contrast agents:Structures, properties and applications. Progress in Polymer Science. 2013Mar;38(3-4):462–502. → pages 17, 34[45] Kobayashi H, Brechbiel MW. Dendrimer-based nanosized MRI contrastagents. Current pharmaceutical biotechnology. 2004 Dec;5(6):539–549. →page 17[46] Sato N, Kobayashi H, Hiraga A, Saga T, Togashi K, Konishi J, et al.Pharmacokinetics and enhancement patterns of macromolecular MRcontrast agents with various sizes of polyamidoamine dendrimer cores.Magnetic Resonance in Medicine. 2001 Dec;46(6):1169–1173. → page 17[47] Ogan MD, Schmiedl U, Moseley ME, Grodd W, Paajanen H, Brasch RC.Albumin labeled with Gd-DTPA. An intravascular contrast-enhancingagent for magnetic resonance blood pool imaging: preparation andcharacterization. Investigative radiology. 1987 Aug;22(8):665–671. →pages 18, 35[48] Turetschek K, Preda A, Novikov V, Brasch RC, Weinmann HJ,Wunderbaldinger P, et al. Tumor microvascular changes in antiangiogenic134treatment: Assessment by magnetic resonance contrast media of differentmolecular weights. Journal of Magnetic Resonance Imaging.2004;20(1):138–144. → pages 18, 35, 58[49] Kainthan RK, Muliawan EB, Hatzikiriakos SG, Brooks DE. Synthesis,Characterization, and Viscoelastic Properties of High Molecular WeightHyperbranched Polyglycerols. Macromolecules. 2006Oct;39(22):7708–7717. → pages 18, 20, 43[50] Saatchi K, Gelder N, Gershkovich P, Sivak O, Wasan KM, Kainthan RK,et al. Long-circulating non-toxic blood pool imaging agent based onhyperbranched polyglycerols. International Journal of Pharmaceutics. 2012Jan;422(1-2):418–427. → pages 18, 53[51] Kainthan RK, Janzen J, Kizhakkedathu JN, Devine DV, Brooks DE.Hydrophobically derivatized hyperbranched polyglycerol as a human serumalbumin substitute. Biomaterials. 2008 Apr;29(11):1693–1704. → page 18[52] Shenoi RA, Lai BFL, Imran ul haq M, Brooks DE, Kizhakkedathu JN.Biodegradable polyglycerols with randomly distributed ketal groups asmulti-functional drug delivery systems. Biomaterials. 2013Aug;34(25):6068–6081. → pages 18, 35[53] Imran ul haq M, Lai BFL, Chapanian R, Kizhakkedathu JN. Influence ofarchitecture of high molecular weight linear and branched polyglycerols ontheir biocompatibility and biodistribution. Biomaterials. 2012Dec;33(35):9135–9147. → page 18[54] Parker GJ, Barker GJ, Tofts PS. Accurate multislice gradient echo T(1)measurement in the presence of non-ideal RF pulse shape and RF fieldnonuniformity. Magnetic Resonance in Medicine. 2001May;45(5):838–845. → page 22[55] Shewhart WA. Economic Control of Quality of Manufactured Product.ASQ Quality Press; 1931. → page 24[56] Pathak AP, Artemov D, Ward BD, Jackson DG, Neeman M, BhujwallaZM. Characterizing extravascular fluid transport of macromolecules in thetumor interstitium by magnetic resonance imaging. Cancer Research. 2005Feb;65(4):1425–1432. → pages 24, 36[57] Daldrup-Link HE, Okuhata Y, Wolfe A, Srivastav S, Øie S, Ferrara N, et al.Decrease in tumor apparent permeability-surface area product to a MRI135macromolecular contrast medium following angiogenesis inhibition withcorrelations to cytotoxic drug accumulation. Microcirculation (New York,NY : 1994). 2004 Jul;11(5):387–396. → pages 25, 58[58] Dafni H, Landsman L, Schechter B, Kohen F, Neeman M. MRI andfluorescence microscopy of the acute vascular response to VEGF165:vasodilation, hyper-permeability and lymphatic uptake, followed by rapidinactivation of the growth factor. NMR in Biomedicine.2002;15(2):120–131. → page 25[59] Kyle AH, Huxham LA, Yeoman DM, Minchinton AI. Limited tissuepenetration of taxanes: a mechanism for resistance in solid tumors. ClinicalCancer Research. 2007 May;13(9):2804–2810. → pages 25, 34, 41, 43, 54[60] Collins TJ. ImageJ for microscopy. BioTechniques. 2007 Jul;43(1Suppl):25–30. → pages 25, 43, 54, 93[61] Baranyai Z, Bru¨cher E, Iva´nyi T, chimica RKH, 2005. ComplexationProperties of N, N’, N”, N”’-[1, 4, 7, 10-Tetraazacyclododecane-1, 4, 7,10-tetrayltetrakis (1-oxoethane-2, 1-diyl)] tetrakis [glycine](H4dotagl) . . . .Concepts in Magnetic Resonance: Part A. 2005;. → page 35[62] Kawai F. Microbial degradation of polyethers. Applied Microbiology andBiotechnology. 2002 Jan;58(1):30–38. → page 35[63] Li KL, Wilmes LJ, Henry RG, Pallavicini MG, Park JW, Hu-Lowe DD,et al. Heterogeneity in the angiogenic response of a BT474 human breastcancer to a novel vascular endothelial growth factor-receptor tyrosinekinase inhibitor: Assessment by voxel analysis of dynamiccontrast-enhanced MRI. Journal of Magnetic Resonance Imaging. 2005Oct;22(4):511–519. → page 36[64] Erickson K, Braun RD, Yu D, Lanzen J, Wilson D, Brizel DM, et al. Effectof longitudinal oxygen gradients on effectiveness of manipulation of tumoroxygenation. Cancer Research. 2003 Aug;63(15):4705–4712. → page 38[65] Dewhirst MW, Ong ET, Braun RD, Smith B, Klitzman B, Evans SM, et al.Quantification of longitudinal tissue pO2 gradients in window chambertumours: impact on tumour hypoxia. British Journal of Cancer. 1999Apr;79(11-12):1717–1722. → page 38[66] Seal MD, Speers CH, O’Reilly S, Gelmon KA, Ellard SL, Chia SK.Outcomes of women with early-stage breast cancer receiving adjuvanttrastuzumab. Current Oncology. 2012 Aug;19(4). → page 40136[67] Bria E, Cuppone F, Fornier M, Nistico` C, Carlini P, Milella M, et al.Cardiotoxicity and incidence of brain metastases after adjuvanttrastuzumab for early breast cancer: the dark side of the moon? Ameta-analysis of the randomized trials. Breast Cancer Research andTreatment. 2007 Jul;109(2):231–239. → page 40[68] Kaplan MA, Ertugrul H, Firat U, Kucukoner M, I˙nal A, Urakci Z, et al.Brain metastases in HER2-positive metastatic breast cancer patients whoreceived chemotherapy with or without trastuzumab. Breast Cancer. 2014Jan;22(5):503–509. → page 40[69] Lai R, Dang CT, Malkin MG, Abrey LE. The risk of central nervoussystem metastases after trastuzumab therapy in patients with breastcarcinoma. Cancer. 2004;101(4):810–816. → page 40[70] Murrell DH, Hamilton AM, Mallett CL, van Gorkum R, Chambers AF,Foster PJ. Understanding Heterogeneity and Permeability of BrainMetastases in Murine Models of HER2-Positive Breast Cancer ThroughMagnetic Resonance Imaging: Implications for Detection and Therapy.Translational Oncology. 2015 Jun;8(3):176–184. → page 40[71] Stemmler HJ, Kahlert S, Siekiera W, Untch M, Heinrich B, Heinemann V.Characteristics of patients with brain metastases receiving trastuzumab forHER2 overexpressing metastatic breast cancer. Breast (Edinburgh,Scotland). 2006 Apr;15(2):219–225. → page 40[72] Kyle AH, Baker JHE, Gandolfo MJ, Reinsberg SA, Minchinton AI. TissuePenetration and Activity of Camptothecins in Solid Tumor Xenografts.Molecular Cancer Therapeutics. 2014 Nov;13(11):2727–2737. → page 41[73] Huxham LA, Kyle AH, Baker JHE, Nykilchuk LK, Minchinton AI.Microregional effects of gemcitabine in HCT-116 xenografts. CancerResearch. 2004 Sep;64(18):6537–6541.[74] Kyle AH, Huxham LA, Chiam ASJ, Sim DH, Minchinton AI. Directassessment of drug penetration into tissue using a novel application ofthree-dimensional cell culture. Cancer Research. 2004Sep;64(17):6304–6309.[75] Kyle AH, Minchinton AI. Measurement of delivery and metabolism oftirapazamine to tumour tissue using the multilayered cell culture model.Cancer chemotherapy and pharmacology. 1999;43(3):213–220. → page 41137[76] Jain RK, Stylianopoulos T. Delivering nanomedicine to solid tumors.Nature Reviews Clinical Oncology. 2010 Sep;7(11):653–664. → pages41, 48, 49[77] Juweid M, Neumann R, Paik C, Perez-Bacete MJ, Sato J, van Osdol W,et al. Micropharmacology of monoclonal antibodies in solid tumors: directexperimental evidence for a binding site barrier. Cancer Research. 1992Oct;52(19):5144–5153. → pages 41, 48[78] Baker JHE, Lindquist KE, Huxham LA, Kyle AH, Sy JT, Minchinton AI.Direct Visualization of Heterogeneous Extravascular Distribution ofTrastuzumab in Human Epidermal Growth Factor Receptor Type 2Overexpressing Xenografts. Clinical Cancer Research. 2008Apr;14(7):2171–2179. → pages 41, 43, 44, 48[79] Lee CM, Tannock IF. The distribution of the therapeutic monoclonalantibodies cetuximab and trastuzumab within solid tumors. BMC cancer.2010 Jun;10(1):255. → pages 41, 48[80] Trotter MJ, Chaplin DJ, Olive PL. Use of a carbocyanine dye as a markerof functional vasculature in murine tumours. British Journal of Cancer.1989 May;59(5):706–709. → pages 19, 42[81] Thurber GM, Dane Wittrup K. A mechanistic compartmental model fortotal antibody uptake in tumors. Journal of theoretical biology. 2012Dec;314:57–68. → page 48[82] Goel S, Duda DG, Xu L, Munn LL, Boucher Y, Fukumura D, et al.Normalization of the Vasculature for Treatment of Cancer and OtherDiseases. Physiological Reviews. 2011 Jul;91(3):1071–1121. → page 48[83] Viallard C, Larrive´e B. Tumor angiogenesis and vascular normalization:alternative therapeutic targets. Angiogenesis. 2017 Nov;20(4):409–426. →page 52[84] Martin JD, Seano G, Jain RK. Normalizing Function of Tumor Vessels:Progress, Opportunities, and Challenges. Annual review of physiology.2019 Feb;81:505–534. → page 52[85] Genentech RG. Avastin (Bevacizumab) [package insert]; 2019. → page 52[86] Turetschek K, Floyd E, Shames DM, Roberts TP, Preda A, Novikov V,et al. Assessment of a rapid clearance blood pool MR contrast medium138(P792) for assays of microvascular characteristics in experimental breasttumors with correlations to histopathology. Magnetic Resonance inMedicine. 2001 May;45(5):880–886. → page 58[87] de Lussanet QG, Langereis S, Beets-Tan RGH, van Genderen MHP,Griffioen AW, van Engelshoven JMA, et al. Dynamic contrast-enhancedMR imaging kinetic parameters and molecular weight of dendritic contrastagents in tumor angiogenesis in mice. Radiology. 2005 Apr;235(1):65–72.→ pages 58, 61[88] Padhani AR. Dynamic contrast-enhanced MRI in clinical oncology:current status and future directions. Journal of Magnetic ResonanceImaging. 2002 Oct;16(4):407–422. → page 61[89] Gerstner ER, Sorensen AG, Jain RK, Batchelor TT. Advances inneuroimaging techniques for the evaluation of tumor growth, vascularpermeability, and angiogenesis in gliomas. Current opinion in neurology.2008 Dec;21(6):728–735. → page 61[90] Tofts PS, Brix G, Buckley DL, Evelhoch JL, Henderson E, Knopp MV,et al. Estimating kinetic parameters from dynamic contrast-enhancedT(1)-weighted MRI of a diffusable tracer: standardized quantities andsymbols. Journal of Magnetic Resonance Imaging. 1999Sep;10(3):223–232. → page 61[91] Mak IW, Evaniew N, Ghert M. Lost in translation: animal models andclinical trials in cancer treatment. American journal of translationalresearch. 2014;6(2):114–118. → page 62[92] Dewhirst MW. Relationships between cycling hypoxia, HIF-1,angiogenesis and oxidative stress. Radiation Research. 2009Dec;172(6):653–665. → pages 62, 102, 123[93] Michiels C, Tellier C, Feron O. Cycling hypoxia: A key feature of thetumor microenvironment. Biochimica et Biophysica Acta (BBA) - Reviewson Cancer. 2016 Aug;1866(1):76–86. → pages 62, 102, 123[94] Bayer C, Shi K, Astner ST, Maftei CA, Vaupel P. Acute Versus ChronicHypoxia: Why a Simplified Classification is Simply Not Enough.International Journal of Radiation Oncology*Biology*Physics. 2011Jul;80(4):965–968. → pages 62, 102, 123139[95] Heilmann M, Kiessling F, Enderlin M, Schad LR. Determination ofpharmacokinetic parameters in DCE MRI: Consequence of nonlinearitybetween contrast agent concentration and signal intensity. Investigativeradiology. 2006 Jun;41(6):536–543. → page 63[96] Horsman MR, Mortensen LS, Petersen JB, Busk M, Overgaard J. Imaginghypoxia to improve radiotherapy outcome. Nature Reviews ClinicalOncology. 2012 Nov;9(12):674–687. → pages 63, 65, 102, 113, 123[97] Wilson WR, Hay MP. Targeting hypoxia in cancer therapy. NatureReviews Cancer. 2011 Jun;11(6):393–410. → page 65[98] Rischin D, Hicks RJ, Fisher R, Binns D, Corry J, Porceddu S, et al.Prognostic significance of [18F]-misonidazole positron emissiontomography-detected tumor hypoxia in patients with advanced head andneck cancer randomly assigned to chemoradiation with or withouttirapazamine: a substudy of Trans-Tasman Radiation Oncology GroupStudy 98.02. Journal of Clinical Oncology. 2006 May;24(13):2098–2104.→ page 65[99] O’Connor JPB, Boult JKR, Jamin Y, Babur M, Finegan KG, Williams KJ,et al. Oxygen-Enhanced MRI Accurately Identifies, Quantifies, and MapsTumor Hypoxia in Preclinical Cancer Models. Cancer Research. 2016Feb;76(4):787–795. → pages 67, 68, 69, 70, 101[100] Linnik IV, Scott MLJ, Holliday KF, Woodhouse N, Waterton JC, O’ConnorJPB, et al. Noninvasive tumor hypoxia measurement using magneticresonance imaging in murine U87 glioma xenografts and in patients withglioblastoma. Magnetic Resonance in Medicine. 2013Jun;71(5):1854–1862. → pages 67, 69, 87, 101, 103[101] Sorg BS, Hardee ME, Agarwal N, Moeller BJ, Dewhirst MW. Spectralimaging facilitates visualization and measurements of unstable andabnormal microvascular oxygen transport in tumors. Journal of BiomedicalOptics. 2008;13(1):014026. → page 69[102] Chiarotti G, Cristiani G, Giulotto L. Proton relaxation in pure liquids andin liquids containing paramagnetic gases in solution. Il Nuovo Cimento.1955 May;1(5):863–873. → page 69[103] Young IR, Clarke GJ, Bailes DR, Pennock JM, Doyle FH, Bydder GM.Enhancement of relaxation rate with paramagnetic contrast agents in NMR140imaging. The Journal of Computed Tomography. 1981 Dec;5(6):543–547.→ page 69[104] Jones RA, Ries M, Moonen CTW, Grenier N. Imaging the changes in renalT1 induced by the inhalation of pure oxygen: a feasibility study. MagneticResonance in Medicine. 2002 Apr;47(4):728–735. → page 69[105] Tadamura E, Hatabu H, Li W, Prasad PV, Edelman RR. Effect of oxygeninhalation on relaxation times in various tissues. Journal of MagneticResonance Imaging. 1997 Jan;7(1):220–225. → page 69[106] Berkowitz BA, McDonald C, Ito Y, Tofts PS, Latif Z, Gross J. Measuringthe human retinal oxygenation response to a hyperoxic challenge usingMRI: eliminating blinking artifacts and demonstrating proof of concept.Magnetic Resonance in Medicine. 2001 Aug;46(2):412–416. → page 69[107] O’Connor JPB, Naish JH, Parker GJM, Waterton JC, Watson Y, Jayson GC,et al. Preliminary Study of Oxygen-Enhanced Longitudinal Relaxation inMRI: A Potential Novel Biomarker of Oxygenation Changes in SolidTumors. International Journal of Radiation Oncology*Biology*Physics.2009 Nov;75(4):1209–1215. → page 69[108] Little RA, Jamin Y, Boult JKR, Naish JH, Watson Y, Cheung S, et al.Mapping Hypoxia in Renal Carcinoma with Oxygen-enhanced MRI:Comparison with Intrinsic Susceptibility MRI and Pathology. Radiology.2018 Sep;288(3):739–747. → pages 69, 101, 126, 127[109] Zhao D, Pacheco-Torres J, Hallac RR, White D, Peschke P, Cerda´n S, et al.Dynamic oxygen challenge evaluated by NMR T1 and T∗2 - insights intotumor oxygenation. NMR in Biomedicine. 2015 Jun;28(8):937–947. →pages 69, 70, 103[110] White DA, Zhang Z, Li L, Gerberich J, Stojadinovic S, Peschke P, et al.Developing oxygen-enhanced magnetic resonance imaging as a prognosticbiomarker of radiation response. Cancer Letters. 2016 Sep;380(1):69–77.→ pages 87, 101, 106[111] Hallac RR, Zhou H, Pidikiti R, Song K, Stojadinovic S, Zhao D, et al.Correlations of noninvasive BOLD and TOLD MRI with pO2 andrelevance to tumor radiation response. Magnetic Resonance in Medicine.2014 May;71(5):1863–1873. → page 69141[112] Jordan BF, Magat J, Colliez F, Ozel E, Fruytier AC, Marchand V, et al.Mapping of oxygen by imaging lipids relaxation enhancement: A potentialsensitive endogenous MRI contrast to map variations in tissue oxygenation.Magnetic Resonance in Medicine. 2012 Sep;70(3):732–744. → page 69[113] McGrath DM, Naish JH, O’Connor JPB, Hutchinson CE, Waterton JC,Taylor CJ, et al. Oxygen-induced changes in longitudinal relaxation timesin skeletal muscle. Magnetic Resonance Imaging. 2008Feb;26(2):221–227. → page 69[114] Kershaw LE, Naish JH, McGrath DM, Waterton JC, Parker GJM.Measurement of arterial plasma oxygenation in dynamic oxygen-enhancedMRI. Magnetic Resonance in Medicine. 2010 Dec;64(6):1838–1842. →page 69[115] McKeown MJ, Makeig S, Brown GG, Jung TP, Kindermann SS, Bell AJ,et al. Analysis of fMRI data by blind separation into independent spatialcomponents. Human brain mapping. 1998;6(3):160–188. → page 71[116] Hyva¨rinen A, Oja E. Independent component analysis: algorithms andapplications. Neural Networks. 2000 May;13(4-5):411–430. → pages71, 72, 105[117] Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B. Scikit-learn:Machine learning in Python. Journal of Machine Learning Research. 2011Oct;12:2825–2830. → page 72[118] Losert C, Peller M, Schneider P, Reiser M. Oxygen-enhanced MRI of thebrain. Magnetic Resonance in Medicine. 2002 Jul;48(2):271–277. →pages 87, 88[119] Calhoun VD, Liu J, Adali T. A review of group ICA for fMRI data andICA for joint inference of imaging, genetic, and ERP data. NeuroImage.2009 Mar;45(1 Suppl):S163–72. → pages 87, 119[120] Greicius MD, Srivastava G, Reiss AL, Menon V. Default-mode networkactivity distinguishes Alzheimer’s disease from healthy aging: evidencefrom functional MRI. Proceedings of the National Academy of Sciences ofthe United States of America. 2004 Mar;101(13):4637–4642. → page 87[121] McKeown MJ, Jung TP, Makeig S, Brown G, Kindermann SS, Lee TW,et al. Spatially independent activity patterns in functional MRI data duringthe stroop color-naming task. Proceedings of the National Academy of142Sciences of the United States of America. 1998 Feb;95(3):803–810. →page 87[122] Featherstone AK, O’Connor JPB, Little RA, Watson Y, Cheung S, BaburM, et al. Data-driven mapping of hypoxia-related tumor heterogeneityusing DCE-MRI and OE-MRI. Magnetic Resonance in Medicine. 2018Apr;79(4):2236–2245. → page 87[123] Qian H, Bassingthwaighte JB. A class of flow bifurcation models withlognormal distribution and fractal dispersion. Journal of theoreticalbiology. 2000 Jul;205(2):261–268. → page 89[124] Kalra J, Baker J. Multiplex Immunohistochemistry for Mapping the TumorMicroenvironment. Methods in molecular biology (Clifton, NJ).2017;1554(1):237–251. → page 92[125] Newville M, Otten R, Nelson A, Ingargiola A, Stensitzki T, Allan D, et al.lm6t/lm6t-py 0.9.13. Zenodo. 2019 Jun;. → page 93[126] Khurana D, Martin EA, Kasperbauer JL, O’Malley BW, Salomao DR, ChenL, et al. Characterization of a spontaneously arising murine squamous cellcarcinoma (SCC VII) as a prerequisite for head and neck cancerimmunotherapy. Head & neck. 2001 Oct;23(10):899–906. → page 100[127] Horsman MR, Overgaard J. The impact of hypoxia and its modification ofthe outcome of radiotherapy. Journal of radiation research. 2016 Aug;57Suppl 1:i90–i98. → page 100[128] Burrell JS, Walker-Samuel S, Baker LCJ, Boult JKR, Jamin Y, Halliday J,et al. Exploring ∆R2∗ and ∆R1 as imaging biomarkers of tumoroxygenation. Journal of Magnetic Resonance Imaging. 2013Jan;38(2):429–434. → pages 101, 103[129] Yang DM, Campbell III JW, Gerberich JL, Zhou H, Mason RP. AssessingTumor Hypoxia Based on Hypoxic Gas Breathing. In: Proceedings of the26th Annual Meeting of ISMRM. (632), Paris; 2018. p. 1–3. → page 101[130] Panek R, Welsh L, Baker LCJ, Schmidt MA, Wong KH, Riddell AM, et al.Noninvasive Imaging of Cycling Hypoxia in Head and Neck Cancer UsingIntrinsic Susceptibility MRI. Clinical Cancer Research. 2017Aug;23(15):4233–4241. → pages 102, 123143[131] Gross MW, Karbach U, Groebe K, Franko AJ, Mueller-Klieser W.Calibration of misonidazole labeling by simultaneous measurement ofoxygen tension and labeling density in multicellular spheroids.International Journal of Cancer. 1995 May;61(4):567–573. → page 103[132] Remmele S, Sprinkart AM, Mu¨ller A, Tra¨ber F, von Lehe M, Gieseke J,et al. Dynamic and simultaneous MR measurement of R1 and R∗2 changesduring respiratory challenges for the assessment of blood and tissueoxygenation. Magnetic Resonance in Medicine. 2012 Aug;70(1):136–146.→ page 103[133] Silvennoinen MJ, Kettunen MI, Kauppinen RA. Effects of hematocrit andoxygen saturation level on blood spin-lattice relaxation. MagneticResonance in Medicine. 2003 Feb;49(3):568–571. → page 103[134] Ferrara N, Hillan KJ, Gerber HP, Novotny W. Discovery and developmentof bevacizumab, an anti-VEGF antibody for treating cancer. Naturereviews Drug discovery. 2004 May;3(5):391–400. → page 106[135] Keating GM. Bevacizumab: a review of its use in advanced cancer. Drugs.2014 Oct;74(16):1891–1925. → pages 106, 107[136] Pavlidis ET, Pavlidis TE. Role of bevacizumab in colorectal cancer growthand its adverse effects: a review. World journal of gastroenterology. 2013Aug;19(31):5051–5060.[137] Barnett JC, Alvarez Secord A, Cohn DE, Leath CA, Myers ER, HavrileskyLJ. Cost effectiveness of alternative strategies for incorporatingbevacizumab into the primary treatment of ovarian cancer. Cancer. 2013Oct;119(20):3653–3661. → page 107[138] Ku¨mler I, Christiansen OG, Nielsen DL. A systematic review ofbevacizumab efficacy in breast cancer. Cancer treatment reviews. 2014Sep;40(8):960–973. → page 106[139] Wang H, Hristov D, Qin J, Tian L, Willmann JK. Three-dimensionalDynamic Contrast-enhanced US Imaging for Early AntiangiogenicTreatment Assessment in a Mouse Colon Cancer Model. Radiology. 2015Nov;277(2):424–434. → page 106[140] O’Connor JPB, Carano RAD, Clamp AR, Ross J, Ho CCK, Jackson A,et al. Quantifying antivascular effects of monoclonal antibodies to vascularendothelial growth factor: insights from imaging. Clinical CancerResearch. 2009 Nov;15(21):6674–6682. → page 106144[141] Nagengast WB, de Vries EG, Hospers GA, Mulder NH, de Jong JR,Hollema H, et al. In vivo VEGF imaging with radiolabeled bevacizumab ina human ovarian tumor xenograft. Journal of Nuclear Medicine. 2007Aug;48(8):1313–1319. → page 106[142] Yang J, Liao C, Liu Y, Yang G, Ke T, Ding Y, et al. MR imagingbiomarkers evaluating vascular normalization window after anti-vesseltreatment. Oncotarget. 2018 Feb;9(15):11964–11976. → page 106[143] Colliez F, Gallez B, Jordan BF. Assessing Tumor Oxygenation forPredicting Outcome in Radiation Oncology: A Review of StudiesCorrelating Tumor Hypoxic Status and Outcome in the Preclinical andClinical Settings. Frontiers in Oncology. 2017 Jan;7(312):638. → page 113[144] Baumann M, Krause M, Overgaard J, Debus J, Bentzen SM, Daartz J, et al.Radiation oncology in the era of precision medicine. Nature ReviewsCancer. 2016 Mar;16(4):234–249. → page 113[145] Kaanders JHAM, Wijffels KIEM, Marres HAM, Ljungkvist ASE, PopLAM, van den Hoogen FJA, et al. Pimonidazole binding and tumorvascularity predict for treatment outcome in head and neck cancer. CancerResearch. 2002 Dec;62(23):7066–7074. → page 115[146] Matsumoto S, Batra S, Saito K, Yasui H, Choudhuri R, Gadisetti C, et al.Antiangiogenic agent sunitinib transiently increases tumor oxygenation andsuppresses cycling hypoxia. Cancer Research. 2011Oct;71(20):6350–6359. → page 115[147] Lemasson B, Christen T, Serduc R, Maisin C, Bouchet A, Le Duc G, et al.Evaluation of the relationship between MR estimates of blood oxygensaturation and hypoxia: effect of an antiangiogenic treatment on agliosarcoma model. Radiology. 2012 Dec;265(3):743–752. → page 115[148] Jain RK. Normalizing Tumor Microenvironment to Treat Cancer: Bench toBedside to Biomarkers. Journal of Clinical Oncology. 2013Jun;31(17):2205–2218. → page 115[149] Chauhan VP, Stylianopoulos T, Martin JD, Popovic´ Z, Chen O, KamounWS, et al. Normalization of tumour blood vessels improves the delivery ofnanomedicines in a size-dependent manner. Nature Nanotechnology. 2012Apr;7(6):383–388.145[150] Huang Y, Yuan J, Righi E, Kamoun WS, Ancukiewicz M, Nezivar J, et al.Vascular normalizing doses of antiangiogenic treatment reprogram theimmunosuppressive tumor microenvironment and enhance immunotherapy.Proceedings of the National Academy of Sciences of the United States ofAmerica. 2012 Oct;109(43):17561–17566. → page 115[151] Arjona AA, Alvarez E. Tumor site implantation and animal modelselection in oncology. vol. 835 of Cancer Drug Discovery andDevelopment. Totowa, NJ: Humana Press; 2006. → page 116[152] Hubbard WC, Hough AJ, Johnson RM, Oates JA. The site of VX2 tumortransplantation affects the development of hypercalcemia in rabbits.Prostoglandis. 1980 Jun;19(6). → page 116[153] Malave´ I, Blanca I, Fuji H. Influence of inoculation site on development ofthe Lewis lung carcinoma and suppressor cell activity in syngeneic mice.Journal of the National Cancer Institute. 1979 Jan;62(1):83–88. → page116[154] Hill SA, Denekamp J. Site dependent response of tumours to combinedheat and radiation. The British Journal of Radiology. 1982Dec;55(660):905–912. → page 116[155] Calhoun VD, Adali T, Pearlson GD, Pekar JJ. A method for making groupinferences from functional MRI data using independent componentanalysis. Human brain mapping. 2001 Nov;14(3):140–151. → page 119[156] Bayer C, Vaupel P. Acute versus chronic hypoxia in tumors: Controversialdata concerning time frames and biological consequences. Strahlentherapieund Onkologie : Organ der Deutschen Rontgengesellschaft [et al]. 2012Jul;188(7):616–627. → page 123[157] Chaplin DJ, Durand RE, Olive PL. Acute hypoxia in tumors: implicationsfor modifiers of radiation effects. International journal of radiationoncology, biology, physics. 1986 Aug;12(8):1279–1282. → page 123[158] O’Connor JPB, Robinson SP, Waterton JC. Imaging tumour hypoxia withoxygen-enhanced MRI and BOLD MRI. The British Journal of Radiology.2019 Mar;92(1095):20180642. → page 127[159] Lee JH, Lee TW, Jolesz FA, Yoo SS. Independent vector analysis (IVA):multivariate approach for fMRI group study. NeuroImage. 2008Mar;40(1):86–109. → pages 127, 128146[160] Rafique W, Naqvi SM, Chambers JA. Mixed source prior for the fastindependent vector analysis algorithm. In: IEEE Sensor Array andMultichannel Signal Processing Workshop SAM. IEEE; 2016. p. 1–5. →page 128147


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items