Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Assessment of the onset and progress of apoptosis via Raman spectroscopy Rangan, Shreyas 2016

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

Item Metadata

Download

Media
24-ubc_2016_may_rangan_shreyas.pdf [ 2MB ]
Metadata
JSON: 24-1.0300285.json
JSON-LD: 24-1.0300285-ld.json
RDF/XML (Pretty): 24-1.0300285-rdf.xml
RDF/JSON: 24-1.0300285-rdf.json
Turtle: 24-1.0300285-turtle.txt
N-Triples: 24-1.0300285-rdf-ntriples.txt
Original Record: 24-1.0300285-source.json
Full Text
24-1.0300285-fulltext.txt
Citation
24-1.0300285.ris

Full Text

ASSESSMENT OF THE ONSET AND PROGRESS OF APOPTOSIS VIA RAMAN SPECTROSCOPY by  Shreyas Rangan  B.Tech., M.Tech., Indian Institute of Technology Madras, 2012  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF SCIENCE in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Genome Science and Technology)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  April 2016  © Shreyas Rangan, 2016 ii  Abstract  Apoptotic and necrotic cell death is ultimately the cause of productivity loss in bioreactors used to produce therapeutic proteins. This study investigates the suitability of Raman spectroscopy to detect the onset and types of cell death in Chinese Hamster Ovary (CHO) cells - the most widely used cell type for therapeutic protein production. Apoptotic, necrotic or autophagic CHO cells producing tissue plasminogen activator were compared to uninduced cultures using Raman spectroscopy and principal component analysis (PCA). A fingerprint region was identified where several peaks change in the course of cell death. Further, uninduced cells were compared to cells sorted at different stages of apoptosis, in order to establish how early the onset of apoptosis could be detected. These results move past what has been described in literature, as we have shown that apoptosis-induced cells that score as viable in conventional apoptosis assays appear to have an altered biochemical composition compared to uninduced cells.  Cells from different stages of fed-batch cultures were compared, and the results showed that Raman spectroscopy can be used to monitor the progress of a fed-batch culture. However, further work is required to elucidate the onset of apoptosis in fed-batch cultures.  Future goals include assessing different inducers of apoptosis in order to construct a "library" of biochemical changes during apoptotic cell death, and developing automated classification models such as support vector machines to classify cell death types.  iii  Preface The identification of research goals and design of experiments was done by me in conjunction with my supervisor, Dr. James Piret. Dr. Robin Turner (co-supervisor), Dr. Michael Blades, Dr. Hans Georg Schulze and Dr. Stanislav Konorov provided valuable comments and added insight to the interpretation of data.  Experiments were carried out by me, except the data presented in sections 2.5 and 2.7, which were performed by Sepehr Kamal, a co-op student working in the Piret lab. Data analysis was done by me using MATLAB code written by Dr. Hans Georg Schulze. Cell sorting was performed by Justin Wong, UBCFlow.  The experiments in this thesis did not require ethics approval, and the data has not yet been submitted for publication in peer reviewed journals.  iv  Table of Contents  Abstract .......................................................................................................................................... ii Preface ........................................................................................................................................... iii Table of Contents ......................................................................................................................... iv List of Tables ................................................................................................................................ vi List of Figures .............................................................................................................................. vii List of Abbreviations ................................................................................................................... ix Acknowledgements ........................................................................................................................x Chapter 1: Introduction ................................................................................................................1 1.1 Apoptosis ........................................................................................................................ 1 1.2 Primary Necrosis ............................................................................................................. 5 1.3 Autophagy ....................................................................................................................... 6 1.4 Raman Spectroscopy ....................................................................................................... 8 1.5 Chinese Hamster Ovary Cells ....................................................................................... 15 Chapter 2: Assessment of Apoptosis, Necrosis and Autophagy in CHO Cells Using Raman Spectroscopy .................................................................................................................................18 2.1 Introduction ................................................................................................................... 18 2.2 Cell Culture ................................................................................................................... 20 2.3 Induction of Cell Death ................................................................................................. 20 2.3.1 Apoptosis .................................................................................................................. 20 2.3.2 Primary Necrosis ....................................................................................................... 21 2.3.3 Autophagy ................................................................................................................. 21 v  2.4 Flow Cytometry ............................................................................................................ 21 2.5 Raman Spectroscopy of Apoptotic Cells ...................................................................... 25 2.6 Biochemical Validation of Raman Spectroscopy Results ............................................ 32 2.7 Cell Sorting ................................................................................................................... 33 2.8 Live-Cell Microscopy ................................................................................................... 34 2.9 Raman Spectroscopy of Induced Viable Sorted Cells .................................................. 35 2.10 Raman Spectroscopy of Primary Necrotic, Secondary Necrotic and Autophagic Cells..... ..................................................................................................................................... 40 Chapter 3: Raman Spectroscopy of Fed-Batch Cultures .........................................................46 3.1 Fed-Batch Cell Culture ................................................................................................. 46 3.2 Raman Spectroscopy of Fed-Batch Cultures ................................................................ 47 Chapter 4: Conclusion .................................................................................................................54 References .....................................................................................................................................60 Appendix .......................................................................................................................................66 Appendix A MATLAB code used for preprocessing of Raman data ....................................... 66  vi  List of Tables  Table 1.1 A literature perspective on Raman spectroscopy-based studies of apoptosis and necrosis ........................................................................................................................................... 9  vii  List of Figures  Figure 1.1 Morphological changes typically observed in apoptotic cell death ............................... 4 Figure 1.2 Schematic overview of stages of autophagy indicating autophagosome formation and eventual degradation  ...................................................................................................................... 7 Figure 2.1 Flow cytometry of camptothecin-induced apoptotic cells ........................................... 23 Figure 2.2 Flow cytometry of cells induced for primary necrosis via glucose and oxygen deprivation .................................................................................................................................... 24 Figure 2.3 Mean spectra of uninduced cells vs. apoptosis induced cells ...................................... 28 Figure 2.4 Fingerprint region of uninduced cells vs. apoptosis induced cells .............................. 29 Figure 2.5 Principal component analysis of uninduced cells vs. apoptosis induced cells ............ 30 Figure 2.6 Principal component loadings of uninduced, 24 h and 48 h camptothecin-induced cells. .............................................................................................................................................. 31 Figure 2.7 Fold changes in protein, DNA and RNA in camptothecin treated cells compared to uninduced cells.............................................................................................................................. 33 Figure 2.8 Progression of apoptosis in induced, sorted viable cells compared to uninduced cells.... ........................................................................................................................................... 35 Figure 2.9 Raman spectra of uninduced, induced annexin V/PI -/- and +/- sorted populations ... 37 Figure 2.10 Fingerprint region of uninduced cells vs. apoptosis-induced sorted cells ................. 38 Figure 2.11 PCA reveals a separation between uninduced cells and apoptosis-induced, sorted viable and early apoptotic cells ..................................................................................................... 39 Figure 2.12 Principal component loadings reveal key peaks contributing to the observed separation between uninduced and apoptosis-induced sorted cells .............................................. 40 viii  Figure 2.13 Mean Raman spectra of viable, primary necrotic, secondary necrotic and autophagic cells ............................................................................................................................................... 42 Figure 2.14 Fingerprint region of viable, primary necrotic, secondary necrotic and autophagic cells ............................................................................................................................................... 43 Figure 2.15 PCA reveals a clear distinction between viable, primary necrotic and secondary necrotic cells. ................................................................................................................................ 44 Figure 2.16 Principal component loadings reveal key peaks that contribute to the observed separation between the populations. ............................................................................................. 45 Figure 3.1 Growth curve showing change in viable cell concentration over time and % viable cells in fed batch cultures .............................................................................................................. 47 Figure 3.2 Average, phenylalanine-normalized Raman spectra of fed-batch cultures at days 3, 7, 10 and 12 ....................................................................................................................................... 50 Figure 3.3 Fingerprint region of phenylalanine-normalized spectra from fed-batch cultures ...... 51 Figure 3.4 PCA showed the progression of fed-batch cultures clearly from day 3 to day 7 to day 10; days 10 and 12 cluster together............................................................................................... 52 Figure 3.5 Principal component loadings indicate the peak contributions to the observed separation using PC1 and PC3 ...................................................................................................... 52  ix  List of Abbreviations ATP BSA CHO DAPI DNA DHFR FACS FITC HEPES HMGB-1 IFN IXM LC3 PBS PCA PI RNA SVM TNF tPA  Adenosine triphosphate Bovine serum albumin Chinese hamster ovary 4',6-diamidino-2-phenylindole Deoxyribonucleic acid Dihydrofolate reductase Fluroescence activated cell sorting Fluorescein isothiocyanate 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid  High mobility group box-1 Interferon ImageXpress Micro Light chain-3 Phosphate buffered saline Principal component analysis Propidium iodide Ribonucleic acid Support vector machine Tumor necrosis factor Tissue plasminogen activator  x  Acknowledgements  I am extremely grateful to Dr. James Piret for his enduring faith in me and the constant source of support and encouragement that he has been over the years. I couldn't ask for anything more from a supervisor.  I am thankful to Dr. Robin Turner, Dr. Mike Blades and Dr. Stanislav Konorov for their invaluable scientific insight and for the encouragement and insight they gave me.  I am particularly thankful to Dr. Hans Georg Schulze, without whom this project would have been impossible. He has been one of the best teachers I have ever had, and a good friend.  I would like to acknowledge the members of the Piret and Turner labs and my amazing friends who have supported me throughout this journey.  Special thanks are owed to my parents, whose have supported me throughout my years of education with their wisdom and grace.  Last but not least, I would like to acknowledge Mirza Asadullah Khan 'Ghalib', without whom my life would be incomplete. "Gham-e-hastii kaa 'Asad' kis se ho juz-marg ilaaj; Shamma har rang mein jaltii hai sahar hote tak" 1  Chapter 1: Introduction  Mammalian cell culture is widely used in industrial bioprocesses to produce therapeutic proteins. Cell death is a major limiting factor to maintaining high viable cell densities and, consequently, high protein production. Cell death also complicates product recovery and can affect protein quality due to degradative enzymes released. The predominant form of cell death observed in industrial bioreactors is apoptotic cell death; necrotic and autophagic cell death are observed to a lesser extent[1]. Many studies have modulated apoptosis as a means to extending culture viability and achieving high cell densities and protein titers [2,3,4]. A landmark achievement in this regard was the stable, inducible transfection of two anti-apoptotic genes into CHO cells by Figueroa et al. in 2007 that enabled dramatic improvements in maintenance of culture viability and recombinant protein production in glucose-limited culture conditions [34].  Typical methods used to detect apoptosis are time intensive and require sampling cells from the culture [5,39]. This thesis aims to investigate the potential of Raman spectroscopy as a tool for detecting apoptosis, with the eventual goal of real-time, non-invasive, label-free detection of the onset of cell death in industrial mammalian cell cultures.  1.1 Apoptosis Apoptosis, also called type I cell death, is an evolutionarily conserved [6] mechanism of programmed cell death found in higher eukaryotes that enables multicellular organisms to destroy unwanted or unnecessary cells in a controlled manner. Controlled removal of cells is crucial for various aspects of an organism's normal function, including development, removal of 2  dying cells and combating infection [7]. Apoptosis is typically affected through a family of cysteine proteases known as caspases [8]. There are two major pathways of apoptosis induction - the extrinsic pathway and the intrinsic pathway. The extrinsic pathway is dependent on the interaction of a death ligand (such as FasL or TNFα) with a corresponding 'death receptor' on the cell surface, whereas the intrinsic pathway can be activated by various stress or damage stimuli to the cells. Once activated, the pathways lead to the activation of caspase-8 or -9 (extrinsic or intrinsic pathway, respectively), which in turn process and activate the effector caspases-3, 6 and 7 [7,8].  Morphological changes in cells undergoing apoptosis are well characterized (Fig 1.1). Early on, the cell and nucleus shrink and nuclear chromatin condenses and adheres to the nuclear membrane. The nucleus then condenses and breaks up (karyorexxhis) and the cell membrane begins to form protrusions (blebbing). Cellular components are closely packed into these protrusions that break off from the plasma membrane to form closed structures known as apoptotic bodies. The cells detach from the extracellular matrix in vivo and the apoptotic bodies are phagocytosed by neighbouring cells, such as macrophages and parenchymal cells. Eventual cell death via apoptosis leads to secondary necrosis [9].  Conventional methods to detect apoptosis in mammalian cell culture include:    Caspase activation - Activation of caspases 3, 6, 8 and 9 serves as a marker for the onset of apoptosis and can be detected via western blotting [9,39]. 3   Annexin V/Propidium iodide staining - Annexin V is a protein that binds phosphatidylserine residues. These residues are usually present on the inner surface of the cell membrane, but some 'flip' to the exterior cell membrane surface in early stages of apoptosis. Annexin V conjugated with a fluorescent marker thus serves as a marker for early apoptosis. Propidium Iodide (PI) is a small molecule that intercalates into DNA and then fluoresces, with maximum emission at 617 nm. However, PI is not membrane permeable and only stains DNA after membrane integrity is lost, thus serving as a marker for the late stage of apoptosis. Annexin V/PI staining is frequently coupled with flow cytometry or fluorescence microscopy to assess apoptosis in a population of cells [9,39].   TUNEL assay - The terminal deoxynucelotidyl transferase dUTP nick end labelling (TUNEL) assay enables detection of DNA fragmentation by labeling terminal ends of nucleic acid chains with dUTPs conjugated with a marker. Since this assay requires DNA fragmentation, it is suitable for detection of late stage apoptosis [9,10,39].   Gel electrophoresis - Enables detection of DNA fragmentation; DNA appears as a characteristic "ladder" in gel electrophoresis after fragmentation [9].  Cell death in batch and fed-batch cultures of Chinese Hamster Ovary (CHO) cells is predominantly apoptotic [11]. As noted earlier, most conventional methods of apoptosis detection are time intensive and require sampling from the culture, which precludes the possibility of pharmacological interventions to rescue the culture. A quick, reliable method to 4  detect the onset of apoptosis in industrial bioreactors will afford the possibility of such interventions.     Fig 1.1 Morphological changes typically observed in apoptotic cell death [open source image]. 5  1.2 Primary Necrosis Primary necrosis, also known as type III cell death, is characterized by a gain in cell volume (oncosis), swelling of organelles and plasma membrane rupture, followed by a dissipation of intracellular constituents into the surrounding environment (tissue or culture media). Other characteristic necrotic cellular changes include chromatin digestion, ATP depletion and DNA hydrolysis [3]. Necrosis was considered to be an accidental, uncontrolled form of cell death until recently. However, new evidence shows that the execution of necrotic cell death can be regulated by a set of signal transduction pathways. There is now a clearer distinction between programmed, regulated and accidental cell death - programmed cell death has been defined as [12] cell death occurring in a purely physiological setting, such as during embryonic development; regulated cell death should refer to cases where the progression to cell death can be halted or reversed by genetic or pharmacological interventions up to a point; accidental cell death should be death modes that do not fit into either of the above categories. Based on these definitions, programmed cell death is a subset of regulated cell death [12,13].  Primary necrosis typically proves challenging to distinguish, as most methods rely on showing that cell death was NOT apoptotic. This is usually done on the basis of discriminating cell morphology and size after death, using microscopy or based on forward and side scatter measurements by flow cytometry, that indicate cell size and granularity, respectively. Biochemically, the only marker indicative of primary necrosis is the high mobility group box-1 protein (HMGB-1); this protein is released from cells in primary necrosis but not during apoptosis or even secondary necrosis [14].  6  1.3 Autophagy Autophagic cell death was first described as type II cell death. Since the 1990s, it has become apparent that autophagy mainly serves as a major cell survival mechanism by rerouting essential nutrients such as amino acids to critical cellular processes during periods of starvation [15]. Autophagy begins with the formation of autophagosomes - double membraned structures containing cytoplasm and organelles routed in either targeted or untargeted ways [16]. Autophagosomes fuse with lysosomes to form autolysosomes. Hydrolytic lysosomal enzymes degrade the components captured within the autophagosome and release the generated nutrients back into the cytoplasm [15,16]. This bulk degradation process is called macroautophagy [19] (Fig 1.2).   A further distinction can be made between "induced macroautophagy", and "basal macroautophagy" that plays an essential role in the normal turnover of cellular components [19]. Other forms of autophagy include microautophagy (bypassing the autophagosomes, cytoplasmic constituents are directly engulfed by lysosomes) and chaperone-mediated autophagy that generally corresponds to organelle-specific degradation such as pexophagy (peroxisome degradation).   Autophagy can be detected by a combination of assays, including a key one for microtubule-associated protein light chain-3 (LC3). This LC3-I is found in the cytoplasm, whereas when LC3-I is conjugated to phosphatidylethanolamine it forms LC3-II that is incorporated into the membranes of autophagosomes. LC3 in mammalian cells is found in three isoforms (LC3A, LC3B and LC3C) - of these, an increase in LC3B-II is indicative of increased autophagy [17,35]. 7  An increase in LC3B-II can be assessed using western blots normalized to a loading control (such as β-actin), with a healthy population used as a control to account for basal levels of autophagy. The progress of autophagy can also be monitored in real-time using cells engineered to express green fluorescent protein (GFP)-tagged LC3; in parallel, a snapshot of autophagic state can be obtained using LC3-II antibodies tagged with a fluorescent marker to stain fixed cells [18].        Fig 1.2 Schematic overview of stages of autophagy indicating autophagosome formation and eventual degradation [41]. 8  1.4 Raman Spectroscopy When a photon interacts with molecular bonds, it is typically scattered at the same energy as the incident light by Rayleigh scattering. However, one out of every ~106 to 109 photons is scattered with a lower or higher energy than the incident photon by Stokes or anti-Stokes Raman scattering, respectively. Specific chemical bonds, such as C=C or O-P-O, produce consistently defined changes in the energy of incident photons. This fact is exploited in Raman spectroscopy by using monochromatic light to identify molecules based on the changes in the wavelength of scattered light. Raman spectroscopy primarily uses Stokes scattering, as anti-Stokes scattering is much less frequent (1 in 109 photons) [20].  Raman spectroscopy has been increasingly used to study the biochemical state of cells and tissues since the 1990s. Combined with confocal microscopy, it enables even the detection of the subcellular localization of biomolecules. Compared to conventional methods of detecting changes in intracellular biomolecules, it offers the advantages of single-cell resolution and non-destructive detection [21]. Some avenues where Raman spectroscopy has been employed for cell and tissue studies include:    Distinction between cancerous and non-cancerous cells [21,23]  Differentiation status of embryonic stem cells [21,24]  Physiological characteristics of red blood cells [21]  Effects of drug-cell interaction [25]; intracellular localization of drugs [21,26]  Characterization of bacteria and bacterial spores [21,22] 9  Apoptosis and necrosis have been assessed using Raman spectroscopy since 2004 (Table 1.1). The studies have included various different cell lines and methods of cell death induction. Most studies have focused on single cell analysis of apoptosis, and have shown recurring changes in intracellular lipids and nucleic acids. However, only one study has attempted to discriminate between the stages of apoptosis, with some success. Two studies have analyzed primary necrosis, and have reported decreases in nucleic acids & lipids and increases in protein peaks. Since direct comparisons between primary necrosis and apoptosis are not included, these results in general do not have the ability to distinguish between modes of cell death.  Table 1.1 A literature perspective on Raman spectroscopy-based studies of apoptosis and necrosis. Peak assignments reflect the known predominant biochemical contributor(s) to the peak, though there may be contributions from other biomolecules.  Cell Type(s) Induction Methods &  Biochemical Assessment Spectral Observations & Multivariate Analyses Reference A549 - Human epithelial lung carcinoma line Cell death induced by Ricin & Sulphur Mustard. MTT assay to assess cell viability; Annexin V/PI staining & fluorescence microscopy to distinguish apoptosis & Peak changes evaluated using spectra from pure biochemical components as a reference (DNA, RNA, phosphatidyl choline - lipid, albumin - protein). Protein (760 cm-1, 1004 cm-1)↓ DNA (495 cm-1, 782 cm-1)↓ Lipid (1301 cm-1, 1449 cm-1)↑ Notingher et al., 2004 10  necrosis A549 - Human epithelial lung carcinoma line Apoptosis induced  by etoposide, a topoisomerase-II inhibitor. MTT assay for cell viability; p53 western blot to verify apoptosis induction Spectra of apoptotis-induced and uninduced control cells were fitted to spectra acquired from pure biochemical components using a Classical Least Squares algorithm; this was used to determine relative changes in cellular components.  DNA: -28% (24h) and -87% (48h)↓ RNA: Decrease (% not reported) ↓ Lipids: 43% (48h)↑ Proteins: No significant change      Owen et al., 2006 11  Cell Type(s) Induction Methods &  Biochemical Assessment Spectral Observations & Multivariate Analyses Reference L132 - Human embryonic lung fibroblast cell line Apoptosis induced by 24h 1mM Glyoxal treatment  Raman microphotographs of single cells showed characteristic shrinkage of nucleus, disappearance of lipid bodies and bleb formation   Nucleic acid bands in nucleus -26%↓ RNA bands in cytoplasm -38% ↓ Blebs show an increase in protein and nucleic acid content compared to membranes of uninduced cells. Krafft et al., 2006 MEL-28 - Human melanoma cell line Necrosis induced by glucose and oxygen deprivation. Flow cytometry using propidium iodide and trypan blue exclusion assessed cell viability. Lipids (717, 1300, 1448, 2850 cm-1)↓ RNA (990, 1080, 1100 cm-1) ↓  Proteins (642, 651, 827,1617 cm-1) ↑ DNA (1421 cm-1) ↑ Kunapareddy et al., 2008 12  MDA-MB-231 - Human breast cancer cell line Apoptosis induced by treatment with 300 μM etoposide for 6 h; sampled every 2 h. Induction validated via fluorescence microscopy of cells stained with Annexin V, 6-CFDA and DAPI DNA (788 cm-1) ↑ interpreted as a sign of DNA condensation  Lipids (1301, 1449, 1659 cm-1) ↑ interpreted as accumulation of lipid bodies Zoladek et al., 2011 K562 - Human chronic myelogenous leukemia cell line Apoptosis induced by 300 μM Cytosine arabinoside treatment. Necrosis induced by 100 μM Triton X-100 treatment. Apoptosis:  Membrane lipids (734 cm-1)↑ DNA/RNA (794, 1098 cm-1) ↓ Protein: 1462 cm-1 constant,  1672 cm-1↑ Necrosis: Membrane lipids (734 cm-1) ↓ DNA/RNA (794, 1098 cm-1) ↓ Protein/lipid peaks (1462, 1672 cm-1)↓ PCA distinguished viable, apoptotic and necrotic cells Ong et al., 2012 13  Cell Type(s) Induction Methods &  Biochemical Assessment Spectral Observations & Multivariate Analyses Reference MCF-7 - Human breast cancer cell line Apoptosis induced by20 ng/mL TNF-α and IFN-γ treatment for 48 h. DNA fragmentation assessed using TUNEL staining; cell viability assessed using MTT assay   Increase in Raman signals at 775-875 cm-1; correlated with DNA condensation observed in apoptosis Ladiwala et al., 2013 Saos-2 - Human bone osteosarcoma cell line; SW-1353 - chondro- sarcoma cell line Apoptosis induced by exposure to room temperature for 7 days. Necrosis induced by heat shock at 55 C for  90 min. Inductions verified using Annexin V/PI flow cytometry and fluorescence Saos-2:   Early apoptosis:  795 cm-1↑; 1375 cm-1↑  1003 cm-1↓; 1658 cm-1↓   Late apoptosis: 1047 cm-1 ↓  SW-1353: Early apoptosis: 1047 cm-1 ↓; 1375 cm-1↑ Late apoptosis: Brauchle et al., 2014 14  microscopy; further validation of apoptosis induction using activity assays for  caspase-3 and -6 786 cm-1 ↓; 1437 cm-1↑  PCA somewhat distinguished the stages of apoptosis, and clearly distinguished between apoptosis and necrosis. Stages of apoptosis classified with ~90% accuracy using a Support Vector Machine (SVM) classification model EA.hy926 - HUVECs fused with thioguanine-resistant clone of A549 lung carcinoma cells Apoptosis induced by :  1. Exposure to Fas ligand  2. Treatment with cycloheximide Cluster analysis revealed changes in nucleus, nucleoli, endoplasmic reticulum and cytoplasm between control and apoptosis-induced cells.  Chemical changes in early apoptosis were revealed by integral analysis of band intensities across the entire cell: 785 cm-1 ↑; 720 cm-1↑; 1007 cm-1 ↓  Czamara et al., 2016    15  1.5 Chinese Hamster Ovary Cells Genetically engineered Chinese Hamster Ovary (CHO) cells are the most commonly used mammalian cells for industrial-scale therapeutic protein production [27]. Cell death in industrial bioreactors can be apoptotic or necrotic. The majority of cell death in batch cultures of CHO cells is known to be apoptotic (~80%) [11]. In fed-batch cultures, the situation is more complicated as cell death can occur due to nutrient deprivation, or due to other factors such as pH or osmolality changes, insufficient oxygen transfer due to poor mixing at high cell concentrations, or generation of reactive oxygen species [3].   Various strategies have been considered for apoptosis inhibition in industrial bioreactors; these have been implemented with varying degrees of effectiveness [1, 4]. However, the fact remains that conventional methods of apoptosis detection are time consuming and require sampling from the bioreactor. Methods designed to sustain culture viability by inhibiting apoptosis may negatively affect the performance of cells with respect to protein production. On the other hand, rapid and non-invasive detection of the onset of apoptosis in bioreactors may provide opportunities to control the progression of cell death; an example of this would be to supplement the culture with nutrients in the context of nutrient-deprivation induced apoptosis.  Current strategies for assessing the status of an industrial culture and the onset of apoptosis include:  1. ABER biomass monitor -  Assessment of culture viability based on overall changes in the conductance and dielectric constant of the culture [36]. 16  2.   Ovizio digital holographic microscopy - Combines flow cytometry with microscopy to enable cell counting and viability assessment without destroying cells [37]. 3. Raman spectroscopy - Kaiser optical systems has developed an in situ bioprocess monitoring unit that uses Raman spectroscopy to evaluate total and viable cell densities as well as the concentration of several key metabolites, including glucose, glutamine, lactate and ammonium [38].   While these on-line methods provide an elegant way to monitor the viability and progression of a culture and also give information about the concentration of certain key metabolites whose accumulation or depletion can lead to cell death, they do not provide information about the onset of cell death in a time-frame where remedial action can be taken to try and salvage the culture.  With these considerations in mind, this thesis investigates the suitability of Raman spectroscopy as a means of detecting the onset of apoptosis, with a secondary goal to detect necrosis. The specific objectives are:  1. Establish the differences between uninduced and apoptosis-induced cells using Raman spectroscopy and principal component analysis (PCA) 2. Show that Raman spectroscopy and PCA can distinguish between the stages of apoptosis (early apoptosis, late apoptosis and secondary necrosis) 3. Establish the differences between apoptotic, necrotic and autophagic cells using Raman spectroscopy and PCA 17  4. Investigate the applicability of these results in the context of a fed-batch culture, where the cells undergo apoptotic and necrotic cell death     18  Chapter 2: Assessment of Apoptosis, Necrosis and Autophagy in CHO Cells Using Raman Spectroscopy  Apoptotic, necrotic or autophagic CHO cells producing tissue plasminogen activator were compared to uninduced cultures using Raman spectroscopy and principal component analysis (PCA). A fingerprint region was identified where several peaks change in the course of cell death. Further, uninduced cells were compared to cells sorted at different stages of apoptosis using flow cytometry, in order to establish how early the onset of apoptosis could be detected. These comparisons showed that uninduced cells can be distinguished from cells in different stages of apoptosis both by visually observing the Raman spectra as well as by PCA. These results move past what has been described in literature, as it was shown that apoptosis-induced cells that score as viable in conventional apoptosis assays appear to have an altered biochemical composition compared to uninduced cells.   2.1 Introduction Chinese Hamster ovary cells are the most frequently used cell type for industrial-scale production of therapeutic proteins [27]. Cell death is one of the major limiting factors affecting productivity in industrial cultures. Cell death in fed-batch cultures in bioreactors may be apoptotic or necrotic - the situation is quite complex as apart from death occurring naturally over the course of the culture, cells may also experience stresses such as nutrient limitation or insufficient oxygen transfer at high densities, which can also trigger the onset of apoptotic cell death [3]. Strategies for inhibiting apoptosis to extend culture lifetime and productivity have 19  been employed with some degree of success [1,4]. However, there is a need for the development of methods that can detect the onset of cell death earlier than currently available methods, as this will further enable timely intervention in order to reverse the progression of cell death. Raman spectroscopy has been employed to study biological samples, particularly cells and tissues, since the 1990s. It is particularly useful in detecting the subcellular localization of biomolecules, as it offers single cell resolution and non-destructive analysis of the sample. It has been used effectively in applications as widespread as identifying cancer cells, characterizing bacterial spores, and studying the intracellular localization of drugs [21].  Raman spectroscopy has been used to study apoptotic and necrotic cell death, as elaborated in Table 1.1. Most studies were able to successfully distinguish apoptotic or necrotic cells from viable, exponentially growing cells. However, most studies tended to stop there - only one study, by Brauchle et al., attempted to assess the stages of apoptosis with limited success. The stages of apoptosis were only compared to cells induced for apoptosis that stained negatively for both annexin V and propidium iodide. The cells were not compared to uninduced cells, thus limiting the applicability of the results, particularly in the context of early detection of apoptosis. This section assesses the suitability of Raman spectroscopy as a tool for detecting the onset of cell death, by comparing uninduced cells with apoptosis-induced cells, comparing cells from different apoptosis stages, and finally looking and primary necrotic, secondary necrotic and autophagic cells.    20  2.2 Cell Culture CHO cells cloned to produce human tissue-type plasminogen activator (tPA) were used for all experiments. The cells were derived from a DHFR- CHO cell line (ATCC accession number CRL-9096) by Cangene (Winnipeg, MB). These cells were cultured in a humidified incubator at 37 C, 5% CO2 and 140 rpm using chemically defined growth medium (CD-CHO, GIBCO, Grand Island, NY) supplemented with 25 ng/mL IGF, 4X anti-clumping agent and 4 mM glutamine. Passaging was done at 2.5 x 105 cells/mL, and the cells grown up to 5 x 106 cells/mL; viability of >98%, as assessed by a trypan blue exclusion assay (Cedex, Roche, Basel, Switzerland), was maintained. For cell death induction, cells were harvested when they reached a density of 3.5-4 x 106 cells/mL, in the exponential phase of growth. In each experiment, a sample of >98% viable growing cells was taken at the same time to serve as an “uninduced” control.  2.3 Induction of Cell Death  2.3.1 Apoptosis CHO cells were transferred from a shake flask culture to a T-25 static suspension culture flask at 1x106 cells/mL; the volume was made up to 5 mL by adding fresh media. Apoptosis was induced by the addition of 20 μM camptothecin, then the flask was incubated in a humidified incubator at 37 C and 5% CO2. Camptothecin is a topoisomerase-I inhibitor that introduces breaks in DNA during DNA replication [29,30]. The concentration of camptothecin used was 20 μM, and was determined empirically. Several treatment times were tested; it was observed that 24 h of treatment with camptothecin induced apoptosis while still maintaining a large (~60%) proportion of "viable" cells that enabled us to test for the early onset of apoptosis (Fig. 2.1). 21  Secondary necrosis was induced by treating the cells with camptothecin for 72 h; in this case, culture viability was determined by trypan blue exclusion to drop down to ~0.2%.  2.3.2 Primary Necrosis Primary necrosis was induced in CHO cells by combined glucose and oxygen deprivation [31]. In this case, 5 x 106 cells were transferred to a 15 mL falcon tube then centrifuged at 1000 rpm for 5 min. The supernatant was discarded and PBS carefully layered over the cell pellet without disrupting or dislodging it. The tube was placed in a humidified incubator at 37 C and 5% CO2. Primary necrosis was successfully induced with 6 h of treatment, but 24 h treatment was preferred in order to obtain complete necrosis (<1% viable population) (Fig. 2.2).  2.3.3 Autophagy Autophagy was induced by culturing CHO cells in CD-CHO medium without glutamine. It has been shown previously that glutamine deprivation in this particular CHO cell line induces autophagy [32]. An inoculum of 1 x 106 cells/mL were added to a shake flask containing 20 mL of glutamine-free CD-CHO and incubated in a humidified incubator at 37 C, 5% CO2 and 140 rpm for 72 h. After incubation, a high cell viability (~93%) was maintained but the cell numbers increased only ~30% (~1.3x106 cells/mL after 72 h).  2.4 Flow Cytometry Uninduced and apoptosis- or necrosis-induced cells were centrifuged in 15 mL tubes. The supernatant was discarded and the cells resuspended in annexin V binding buffer (HEPES buffer supplemented with CaCl2 to facilitate annexin V binding). The cells were stained with annexin 22  V-FITC (30 μL for every 1x106 cells; 15 min incubation in the dark at room temperature) and PI (10 ng/mL, 5 min incubation). After incubation, the cells were kept on ice until being loaded into a LSR-II flow cytometer (BD Biosciences, San Jose, CA) for measurement. After data acquisition, its analysis was done using FlowJo (TreeStar Inc., San Carlos, CA). Flow cytometry of apoptosis-induced cells (Fig. 2.1) shows increasing apoptosis with longer camptothecin treatment. The forward- and side-scatter plots show a progression towards decreased forward scatter and increased side-scatter with increasing treatment times. The selected P1 population gate excludes smaller cell debris and larger aggregates of cells. The annexin V-FITC and PI plots show an increasing proportion of early and late apoptotic and then secondary necrotic cells with a corresponding decrease in viable cells. After 48 h of treatment, the proportion of necrotic cells seems to decrease; this is likely due to a large percentage of secondary necrotic cells being broken down into apoptotic bodies or cell debris, and these excluded from the analysis of cells by the flow cytometer gating. Flow cytometry of primary necrosis-induced cells (Fig. 2.2) shows progressive necrosis with longer treatment, with nearly 100% of the population becoming necrotic after 24 h of treatment. The level of annexin V cell staining at all time points is on average 5-10 fold lower than for the apoptotic cells (Fig. 2.1). 23   Fig. 2.1 Flow cytometry of camptothecin-induced apoptotic cells. (A) Untreated control cells, (B) 24 h camptothecin-treated cells and (C) 48 h camptothecin treated cells. The left pane shows forward- and side-scatter plots; the right pane shows annexin V and propidium iodide staining. Gates were first assigned for the viable population in the untreated sample, and then applied to the other treatment conditions. Cells positive for both annexin V and PI in the untreated sample were classified as necrotic.  24   Fig. 2.2 Flow cytometry of cells induced for primary necrosis via glucose and oxygen deprivation. (A) 6 h (B) 12 h and (C) 24 h treatment. The left panes show forward- and side-scatter plots; the right panes show annexin V and propidium iodide staining. Gates were retained from the analysis of apoptotic samples to provide a direct comparison. 25  2.5 Raman Spectroscopy of Apoptotic Cells Uninduced CHO cells were compared to cells induced with camptothecin for 24 and 48 h. For spectroscopy, 1 x 106 cells were centrifuged at 1000 rpm for 5 min, the supernatant was removed and the cells washed twice with 5 mL PBS. The cells were resuspended in 100 μL of methanol and incubated at -20 C for 20 min to fix the cells. The cell suspension was then placed on 0.5 inch diameter gold mirrors (ThorLabs); the methanol was allowed to air dry in a biosafety cabinet.  Raman spectroscopy was performed using an inVia Raman microscope (Renishaw, Gloucestershire, UK) attached to a fluorescence microscope (Leica Microsystems, Wetzlar, Germany). A 785 nm monochromatic laser induced the Raman signals. Once the mirror with fixed cells under the microscope was focused using the bright field image, the laser was turned on in darkness and Raman signals from the samples measured using the map acquisition mode, a form of Raman signal acquisition that obtains signals from a 2-dimensional grid on the surface of the sample. The sample was divided into a 64 x 64 micron grid, such that Raman signals were acquired at 8 micron horizontal and vertical intervals, resulting in 63 spectra per sample. Noise reduction, baseline correction and principal component analysis were performed using MATLAB (MathWorks, Inc., Natick, MA). The MATLAB suite, Automation 8, was created by Dr. Hans Georg Schulze [40]. Spectra were normalized to the peak at 1002 cm-1 to control for cell loading, as it is a stable, narrow, high-intensity peak which is known to only have a contribution from the amino acid phenylalanine.   26  Comparison of mean spectra from uninduced and apoptosis-induced cells (Fig. 2.3) reveals changes in several peaks, many that can be correlated with specific cellular biochemical components. In general, it is important to note that while individual peaks are assigned to specific biochemical components, most peaks tend to have contributions from multiple components (multivalent), and the listed peak assignments reflect the molecules known to make a dominant contribution to the Raman signal. In particular, we identified the following features:  Decreasing apoptotic cell nucleic acid associated peak intensities:   o 667 cm-1 - Thymine and guanine o 725 cm-1 - Adenine o 782 cm-1 - DNA; RNA; phosphodiester bond stretch; nucleic acid ring breathing mode (representative of total nucleic acid content) o 811 cm-1 - RNA o 1094-1102 cm-1 - DNA; phosphodioxy group in nucleic acids; O-P-O backbone stretch o 1230-1240 cm-1 - Phosphate backbone stretch in DNA; RNA o 1330-1345 cm-1 - DNA; adenine; guanine; ring breathing modes in DNA. This region is confounded as it has contributions from several other biomolecules, and as such is not regarded as a signature DNA peak o 1575 cm-1 - Adenine; guanine; ring breathing modes in DNA    27  Increasing intensity in lipid peaks in apoptotic cells:  o 702 cm-1 - Cholesterol o 717-719 cm-1 - Phospholipids; apoptosis-induced spectra show a clear broadening of the peak in this region which is indicative of an increased contribution from phospholipids and corresponding decreased contribution from adenine o 1449-1454 cm-1 - C-H stretch in lipids; phospholipids. This peak is quite broad and the information it represents is confounded as many different cellular components contribute Raman signals in that region.  Since the spectra were normalized to phenylalanine at 1002 cm-1, we do not expect significant changes in peaks corresponding to protein/amino acids (Fig 2.7). Interestingly, changes were observed in some amino acid peaks, particularly at 759 cm-1, purported to correspond to tryptophan. However, this is not a change that is observed in all samples.       28  Fig 2.3 Mean Raman spectra of uninduced cells vs. apoptosis-induced cells reveal changes in several peaks corresponding to specific cellular biochemical components, n=3.          The observed intensity changes in peaks led to the identification of a "fingerprint" region between 650 cm-1 and 815 cm-1 encompassing several peak changes related to changes in DNA, RNA and lipid levels (Fig. 2.4). This region alone allows for clearer visual discrimination between uninduced and apoptosis-induced cells, and is also indicative of the progression of apoptosis (e.g. 48 h-induced spectra have a greater decrease in nucleic acid peaks than the 24 h peaks). 29   A subtle change that becomes apparent in the fingerprint region is increasing intensity in apoptotic cells at 745 cm-1. This peak is correlated with the formation of quadruplex structures by guanine-rich DNA and RNA oligonucleotides [33]. An inverse correlation with other nucleic acid peaks is expected and observed, since these oligonucleotides will be generated upon the cleavage of DNA that occurs after camptothecin treatment. Other than that, a decrease is observed in all nucleic acid-associated peaks in apoptotic cells, and an increase is observed in lipid-associated peaks.  Fig 2.4 Fingerprint region of uninduced cells vs. apoptosis-induced cells related to changes in several key biochemical components. 30  Principal component analysis was used to assess whether the different populations could be distinguished based on analysis of the whole spectra. PCA converts a set of potentially correlated variables into a set of linearly uncorrelated variables, to enable a representation of multi-dimensional data into far fewer dimensions.  PCA of baseline-corrected, vector normalized spectra revealed a clear distinction between uninduced and apoptosis-induced cells (Fig. 2.5).     Fig 2.5 Principal component analysis of uninduced cells and cells induced for apoptosis with camptothecin shows clear distinction between the populations. A small difference is also observed between 24h and 48h induction. 31  An analysis of the principal component loadings (Fig 2.6) indicates the peaks that contribute to the principal component scores represented in Fig 2.5. The loadings also enable us to distinguish what peaks, and consequently to infer what biochemical components, vary similarly or in opposite ways. In particular, there are large contributions from nucleic acid and lipid associated peaks. Opposite variations in peaks corresponding to nucleic acids and lipids indicate that as apoptosis progresses, a reduction in cellular nucleic acids and an increase in cellular lipids is observed in both PC1 and PC2. Contributions from both nucleic acids and lipids are observed in both PC1 and PC2, although the observed contributions vary in opposite ways; for instance, most nucleic acid-associated peaks vary negatively in PC1 but positively in PC2. The region where major differences are observed between PC1 and PC2 is between 1200-1400 cm-1, where there are contributions from both nucleic acids and lipids.       Fig 2.6 Principal component loadings of uninduced, 24h and 48h camptothecin-induced cells show relative contributions of various cellular biomolecules to principal components. 32  2.6 Biochemical Validation of Raman Spectroscopy Results Changes in nucleic acid-associated peaks observed in Raman spectra of camptothecin treated cells when compared to uninduced cells were investigated using biochemical DNA and RNA assays. The validity of normalizing spectra to the phenylalanine at 1002 cm-1 was also tested by measuring changes in cellular protein content before and after induction with camptothecin. For each analysis, 2 x 106 uninduced and apoptosis induced cells were harvested. DNA was extracted using the DNEasy Blood and Tissue kit (Qiagen, Hilden, Germany). RNA was extracted using the RNEasy Plus Mini kit (Qiagen). While DNA extraction was done on the bench, RNA extraction was done in a biosafety cabinet that was thoroughly cleaned using RNAseZap (Thermo Fisher Scientific, Waltham, MA). Nuclease-free water and filtered tips were used for all RNA processing. DNA and RNA were quantified using a NanoDrop (Thermo Fisher). Protein was extracted by lysing the cells in 1 mL of RIPA buffer (Thermo Fisher) and measured using a BCA protein analysis kit (Thermo Fisher).   The changes in nucleic acids and proteins of induced cells were normalized relative to their uninduced counterparts (Fig 2.7). Decreases were observed in both DNA and RNA. DNA decreased by about 15%, compared to a ~50% decrease observed in Raman spectra. RNA decreased by ~35%, compared to a near disappearance of the RNA peak in apoptotic Raman spectra. Protein levels are maintained, further justifying the normalization to phenylalanine as an acceptable control for adjusting cell loading.    33             2.7 Cell Sorting For further experiments, induced cells stained with annexin V and PI were sorted into annexin V/PI -/- (viable, non-apoptotic) and annexin V/PI +/- (apparently viable, early apoptotic) populations using a FACSAria cell sorter (BD Biosciences, San Jose, CA). The goal of this experiment was to discover differences using Raman spectroscopy between uninduced cells and induced cells that still scored as "viable" using conventionally accepted annexin V/PI staining.  The 24 h camptothecin-treated (107) cells were stained with annexin V and PI, as described in Section 2.4. After sorting, approximately 106 cells were recovered from the viable and the early apoptotic populations.  For spectroscopy, the cells were washed twice with PBS and then fixed onto gold mirrors using methanol as described in section 2.5   Fig 2.7 Fold change in proteins, DNA and RNA in camptothecin treated cells compared to uninduced cells. n=7, **p<0.01. 34  2.8 Live-Cell Microscopy Uninduced cells and 24 h camptothecin-treated sorted viable cells were restained with annexin V, PI and Hoechst 33342 and imaged over a 48 h period using an ImageXpress Micro (IXM) system (Molecular Devices, Sunnyvale, CA).  This fluorescent microscope system analyzes cells in a humidified, 37 C, 5% CO2 incubator. This was used to analyze the progression of apoptosis in camptothecin-treated viable cells after removal of the inducer, to establish that these cells undergo apoptosis over time.   After sorting, triplicate 50,000 cells per condition were seeded in a 96-well plate in fresh CD-CHO medium. The stains were incorporated into the medium to ensure continued staining of the cells. Four fields of view per well were imaged with three corresponding filters every 30 min (Fig. 2.8A). The results were quantified using the functions built into the IXM software (Fig. 2.8B). Quantification time is from the time of induction, and so the data starts at 26 h due to the previous 24 h of induction and ~2 h for cell sorting.  It was determined based on counting Hoechst stained nuclei that the uninduced cells continued their cell division, whereas induced cells did not (Fig. 2.8B). Uninduced cells had an initial increase in the proportion of annexin V and PI positive cells, likely an artifact of apoptosis induction in ~10% of the cells during the flow cytometry sorting. However, this proportion decreased as the cells divided, decreasing to below 5%. On the other hand, the induced viable cells consistently increased in their proportions of both annexin V and PI positive cells. This confirmed our hypothesis that these cells continued to undergo apoptosis over time, despite the removal of the inducing agent (Fig. 2.8B). 35   Fig 2.8 Progression of apoptosis in induced, sorted viable cells as compared to uninduced cells. (A) A representative cell undergoing apoptosis over time. (B) Progression of apoptosis in induced viable cells vs. uninduced cells over 48h.  2.9 Raman Spectroscopy of Induced Viable Sorted Cells Raman spectroscopy was performed on uninduced cells and induced viable (annexin V/PI -/- ) and early apoptotic (annexin V/PI +/-) sorted cells (Fig 2.9). Comparison of the mean spectra normalized to phenylalanine revealed changes in several peaks, with the changes in the nucleic acid-associated peaks similar to those observed between uninduced and unsorted apoptotic populations (Fig 2.3):   36  Decreasing intensity in nucleic acid-associated peaks in apoptotic cells:  o 667 cm-1 - Thymine and guanine o 725 cm-1 - Adenine o 782 cm-1 - DNA; RNA; phosphodiester bond stretch; nucleic acid ring breathing mode o 811 cm-1 - RNA o 1094-1102 cm-1 - DNA; phosphodioxy group in nucleic acids; O-P-O backbone stretch o 1230-1240 cm-1 - Phosphate backbone stretch in DNA; RNA o 1303-1315 cm-1 - Adenine, guanine  In general, increasing lipid intensities were not observed, and this could be explained if lipid changes arise during later stages of apoptosis. Some minor changes were observed in protein-associated peaks, but without a consistent trend. Minor increases in the peak intensities of uninduced cells were observed at 576 cm-1 and 596 cm-1, associated with phosphatidylinositol. The expected inversely-correlated nucleic acid peak at 745 cm-1 was observed again.    37     The fingerprint region defined previously captures many of the changes observed in these spectra (Fig 2.10), and shows that spectra from these populations can be visually distinguished without further mathematical analysis. In particular, a decrease is observed in all nucleic acid-associated peaks, including a remarkable near-disappearance of the RNA peak at 811 cm-1.   PCA of uninduced cells compared to induced annexin V/PI -/- or +/- sorted cells reveal separation between the sets of data, and a progression in the PCA distribution as apoptosis progresses (Fig. 2.11). Much of the separation occurs on the basis of the first principal component (PC1), though some separation occurs along the diagonal between PC1 and PC4. There is a clearer separation observed between uninduced cells and annexin V/PI -/- cells than Fig 2.9 Raman spectra of uninduced, induced annexin V/PI -/- and +/- sorted populations reveal changes in various peaks, n=5. 38  between annexin V/PI -/- and +/- cells in both Figs. 2.10 and 2.11, indicating that viable, uninduced cells are in fact more distinct from induced viable cells, than those induced viable cells are from early apoptotic cells. This result shows promise for the early detection of onset of apoptosis, even before the cells become annexin V+.       Fig 2.10 Fingerprint region of uninduced cells vs. apoptosis-induced sorted cells indicate changes in several key biochemical components.  39   Analysis of the principal component loadings for PC1 and PC4 (Fig. 2.12) reveal the peak contributions to the observed PCA separation. In particular, again the contributions from nucleic acid and lipid peaks to both PC1 and PC4, varying oppositely. The inversion observed from 1435 cm-1 to 1470 cm-1 in PC4 indicates the multivalent nature of the peak present in that location, and supports our decision to proceed cautiously when interpreting observed spectral intensity changes in that region.   Fig 2.11 PCA reveals a separation between uninduced cells and apoptosis-induced, sorted viable and early apoptotic cells. A progression is also visible from left to right along PC1 as apoptosis progresses.  40   2.10 Raman Spectroscopy of Primary Necrotic, Secondary Necrotic and Autophagic Cells Raman spectroscopy was performed on primary and secondary necrotic cells to assess spectral changes occurring once cell death has progressed to its end-point. Primary necrosis was induced using oxygen and glucose deprivation for 24 h. The trypan blue measured cell viability after treatment was ~0.1%. Secondary necrosis was induced by treating the cells with camptothecin for 72 h, yielding a cell viability after treatment of ~0.2%. In parallel, autophagic cells were assessed to test whether autophagy can be detected via Raman spectroscopy, and also to analyze whether the loss in nucleic acid signal is a feature of cell death, or an artifact of cessation of cell growth - as established in section 2.7, treatment with camptothecin causes the cells to stop growing. Autophagy was induced by depriving the cells of glutamine; after 72 h of this treatment, the cell number increased only by ~15%, while their viability was maintained at ~93%. All spectra were compared to viable, uninduced cells whose numbers increased by ~18 fold over 72 h.  500 600 700 800 900 1000 1100 1200 1300 1400 1500-0.15-0.1-0.0500.050.10.15  PC1PC4Fig 2.12 Principal component loadings reveal key peaks contributing to the observed separation between uninduced and apoptosis-induced sorted cells. The derivative feature observed at 1002 cm-1 in PC1 arises due to inherent small deviations in Raman shift in the raw data obtained from the spectrometer.  41  Changes were observed in a large number of peaks when comparing the mean spectra of necrotic and autophagic cells to uninduced viable cells (Fig. 2.13). In particular, the key changes were:   Primary necrosis:  o Nucleic acids: Decreased intensity at 667, 725, 782, 811, 1094-1102  and 1235 cm-1 o Lipids: Slightly increased intensity at 702 cm-1 (cholesterol), no peak broadening at 717-725 cm-1  Secondary necrosis:  o Nucleic acids: Strong decrease in intensity at 725, 782, 811, 1094-1102 and 1235 cm-1 o Lipids: Large increase in intensity at 702 cm-1 and 717-725 cm-1; in the latter case, the adenine peak at 725 cm-1 seems to disappear completely and the peak presents most strongly at 717 cm-1. An increased intensity is also observed at 1073 cm-1 (triglyceride-associated) o Proteins: Increasing intensity at 758 cm-1 and 852 cm-1. This is unexpected as the spectra are normalized to phenylalanine, and it is not clear what cellular changes contributed to these features o The adenine and guanine-associated peak at 667 cm-1 is shifted to 673 cm-1 and has a higher intensity than the neighbouring 667 cm-1 peak. It is currently unknown what this peak corresponds to, though by correlation to trends observed in the spectra it would be expected to represent lipids to some extent as it varies 42  similarly to other lipid peaks.  Autophagy:  o Most peaks were more similar to the viable cell spectra, with the exception of the RNA-associated 811 cm-1 peak that increased. This may be due to the fact that some intracellular proteins are consumed during autophagy, and so the increase observed in the RNA signal may be an artifact generated by normalization to phenylalanine at 1002 cm-1 o A slight broadening of the peak at 717-725 cm-1 is observed, indicating slightly increased lipid levels in these cells.  Fig 2.13 Mean Raman spectra of viable, primary necrotic, secondary necrotic and autophagic cells show spectral changes occurring in cell death that allow clear classification of the mode of cell death relative to uninduced, viable cells.  43  Several of the observed spectral changes were once again captured in the fingerprint region (Fig.2.14), providing a simpler set of data that distinguishes the modes of cell death analyzed.   Both primary and secondary necrotic cells show decreasing signals in nucleic acid-associated peaks, though the decrease is greater in secondary necrotic cells. Secondary necrotic cells also show a large increase in lipid signals. Autophagic and viable cells have similar spectra, with slight increases observed in nucleic acid- and lipid-associated peaks in autophagic cells. PCA reveals a clear distinction between primary necrosis, secondary necrosis, and viable cells (Fig. 2.15). Viable and autophagic cells cluster together as they are much more similar to each other than to either primary or secondary necrotic cells. This addressed the concern regarding nucleic acid signal intensities decreasing due to cessation of cell growth - as observed in the 660 680 700 720 740 760 780 800 8200.050.10.150.20.25  ViableAutophagyPrimary NecrosisSecondary NecrosisFig 2.14 Fingerprint region of viable, primary necrotic, secondary necrotic and autophagic cells reveals key biochemical changes that enable simple visual distinction of the populations.  44  spectra and PCA, there is no loss of nucleic acid signal in autophagic cells comparable to that observed in primary and secondary necrotic cells.  Analysis of the principal component loadings reveals that contributions from nucleic acid components dominate PC1, whereas contributions from lipids and the unknown peak at 672 cm-1 dominate PC3 (Fig. 2.16).  Fig 2.15 PCA reveals a clear distinction between viable, primary necrotic and secondary necrotic cells. Autophagic cells cluster with much of the data overlapping with the viable cells.  45   It has been demonstrated that Raman spectroscopy coupled with PCA can successfully distinguish between the modes of cell death. Further, it has been established that uninduced cells can be distinguished from a cell population that is heterogeneously undergoing apoptosis. Finally, it has been shown that Raman spectroscopy can reveal the progression of apoptosis by differentiating between different stages; in particular, uninduced cells can be clearly distinguished from induced "viable" cells, thus going beyond conventional annexin V/propidium iodide-based analysis. The next challenge to address is whether signature peak changes identified from this data can be used to identify naturally occurring cell death in fed-batch cultures.        500 600 700 800 900 1000 1100 1200 1300 1400 1500-0.15-0.1-0.0500.050.10.150.20.25  PC1PC3Fig. 2.16 Principal component loadings reveal key peaks that contribute to the observed separation between the populations. 46  Chapter 3: Raman Spectroscopy of Fed-Batch Cultures  Apoptosis is known to be the predominant mode of naturally occurring cell death observed in fed-batch bioreactor cultures [3]. Raman spectroscopy was used in this part of the study to analyze the population of cells present in a fed-batch culture with the goal of identifying specific spectral signatures that may correlate with the features identified in Chapter 2 as spectra that distinguish cell death processes. This is preliminary work to both test the applicability of our previous results to a fed-batch culture, and to provide insight into the future work required to most effectively assess cell death in fed-batch cultures using Raman spectroscopy.  3.1 Fed-Batch Cell Culture Fed-batch cultures were initiated by seeding 2.5 * 105 cells/mL in 20 mL of CD-CHO in shake flasks. The cultures were performed in a humidified incubator at 37 C, 140 rpm and 5% CO2. From the third day, 2 mL of culture was removed daily and replaced with 2 mL of Efficient Feed A (ThermoFisher). The cells were counted every day from day 3 onwards until the end of the cultures, and the cell viability was assessed using trypan blue exclusion. The cells were analyzed by Raman spectroscopy on days 3, 7 and 12.  Fig. 3.1 shows the average cell concentration and viability from the fed-batch cultures. The cell growth continued until day 10, with little or no stationary phase, and then the death phase followed from day 11 onwards. The viable cell percentage declined gradually until day 8 and then more rapidly.   47   3.2 Raman Spectroscopy of Fed-Batch Cultures Raman spectroscopy of fed-batch cultures was performed as described before, using the map acquisition mode on cell samples fixed with methanol onto gold mirrors. Spectra were averaged from five fed-batch cultures and the peak data normalized to phenylalanine at 1002 cm-1 (Fig. 3.2). One recurring problem was that samples from days 10 and 12 generated a very strong Raman signal, resulting in saturation of the collector and unusable data. This was solved by using 50% laser intensity instead of 100%, in order to lower the Raman signal generated. This problem was also encountered when collecting spectra from secondary necrotic cells (section 2.9), and so it may be that later stages of apoptotic cell death cause the accumulation of one or more cellular metabolites that produce a very strong Raman signal.  The analysis of the averaged, normalized spectra revealed changes in several peaks:  Fig 3.1 (A) Growth curve showing change in viable cell concentration over time and (B) % viable cells in fed batch cultures, n=5. 48  Nucleic acids: o 667 cm-1 - Adenine and guanine - Decreased intensity at days 10 & 12 o 725 cm-1 - Adenine - Increased intensity in days 7-12 along with peak broadening and shifting, indicating greater contributions from lipids. o 782 cm-1 - DNA; RNA; phosphodiester bond stretch - Minor changes; days 3 & 12 were equal and very slightly lower than days 7 & 10 o 811 cm-1 - RNA - Day 3 has lowest intensity; increased intensity at days 7 & 10 and then a slight reduction at day 12 o 1094-1102 cm-1 - DNA; O-P-O backbone stretch - Similar to 667 cm-1 o 1230-1240 cm-1 - Phosphate backbone stretch in DNA; RNA - Day 10 had the highest intensity followed by day 12; days 3 & 7 were equal and slightly lower o 1575 cm-1 - Adenine; guanine; ring breathing modes in DNA - Similar to 667 cm-1  Lipids: o 535 cm-1 - Cholesterol - Increasing intensity with time o 702 cm-1 - Cholesterol - Increasing intensity with time o 717-719 cm-1 - Phospholipids - Clear peak broadening and shift towards 717 cm-1 from 725 cm-1 with time, indicating increasing phospholipid content with time  49  The peaks corresponding to nucleic acids had no particular trend. In general, peaks associated with adenine and guanine decreased as time progressed, and so lower at days 10 & 12 than on days 3 & 7. Conversely, RNA-associated peaks increased as time progressed.  The lipid-assciated peaks had trends similar to those observed in apoptotic cells, insofar as having peak levels that increased with time. Since cells in fed-batch culture undergo apoptotic cell death these increasing lipid levels were likely a consequence of apoptotic cell death in the culture.  Peak increases at days 10 & 12 were also observed at 455 cm-1 and 758 cm-1. It is currently unknown what cellular components 455 cm-1 corresponds to. However, given the trends observed in other regions of the spectrum, it is expected that this peak also corresponds to lipids. 758 cm-1 is purported to correspond to tryptophan; however, since these spectra have been normalized to phenylalanine, changes in other amino acid peaks are unexpected. Since this was observed throughout the apoptotic cell death spectra, it may be that this peak also contains contributions from lipids as the observed change is similar to the trend observed in lipid peaks.  50   Fig 3.2 Average, phenylalanine-normalized Raman spectra of fed-batch cultures at days 3, 7, 10 and 12, n=5.  A view of the fingerprint region (Fig. 3.3) captures both the recurring changes observed in lipid-associated peaks as well as the confounded nucleic acid peak observations. Nonetheless, the progression of the lipid peak changes was monotonic in fed-batch cultures. The change observed in the RNA peak at 811 cm-1 from day 3 to day 7 is similar to the observation in this peak when autophagic samples were compared to uninduced samples (section 2.10). The cells in fed-batch culture are not supplemented with glutamine over time, and it is likely that the cells are beginning to undergo autophagy by day 7 [32]. This would further complicate our analysis of the onset of apoptosis. An analysis of fed-batch samples from days between 3 and 7 may shed more light on the early stages of both autophagy and apoptosis.   Principal component analysis distinguished day 3 from day 7, and days 3 & 7 from days 10 & 12 (Fig 3.4). This result indicates that Raman spectroscopy can be used to evaluate the progression 51  of a fed-batch culture. However, further experiments are required to identify the onset of apoptosis in fed-batch cultures.       Fig 3.3 Fingerprint region of phenylalanine-normalized spectra from fed-batch cultures revealed clear changes in lipid peaks and confounding changes in nucleic acid peaks. 52     Fig 3.5 Principal component loadings indicate the peak contributions to the observed separation using PC1 and PC3.  An analysis of principal component loadings (Fig. 3.5) reveals contributions from both nucleic acid and lipid peaks in both PC1 and PC3. PC1 has a higher intensity at the 1002 cm-1 phenylalanine peak due to PC1 representing spectral intensity variations more strongly than other Fig 3.4 PCA showed the progression of fed-batch cultures clearly from day 3 to day 7 to day 10; days 10 and 12 cluster together. 53  principal components. PC1 nucleic acid and lipid-associated peaks varied oppositely. PC3 also had nucleic acid and lipid peaks varying oppositely but with reversed signs from PC1. In addition, PC3 had the highest intensity contributions from lipid peaks, indicating that those signals dominated in the separation observed in PC3. This indicates that lipid signals can be used to elucidate the progression of fed-batch cultures. PC3 also shows that the peak at 758 cm-1 varies similarly to the lipid peaks throughout the spectra; this provides further evidence that this peak may contain signals from lipids, apart from the reported signal from tryptophan.   The results obtained from fed-batch cultures thus far indicate that it is possible to follow the progression of a fed-batch culture by monitoring it using Raman spectroscopy. In particular, changes observed in lipid signals enable distinction early and late stages of culture. Days 3 and 7 are clearly distinguished from each other and from days 10 and 12. However, further experiments are required to achieve our objective of identifying the onset of apoptosis in fed-batch culture.          54  Chapter 4: Conclusion  CHO cells are the major mammalian cell type used for industrial-scale bioreactor production of therapeutic proteins. The ultimate cause of reduced productivity in these bioreactors is cell death, either apoptotic or necrotic. Conventional biochemical analyses are invasive and time consuming, and generally do not allow detection of the onset of cell death before the point of irreversibility has passed. In this context, this work has shown that Raman spectroscopy can serve as a method of detecting the onset of apoptosis earlier than cell membrane exposure of phosphatidylserine that is detected by annexin V and flow cytometry analysis.  It may also be feasible to implement this detection in a non-invasive and on-line manner such that cell samples would not be needed.  Prior Raman spectroscopy studies have shown that apoptotic cells can be distinguished from viable cells, primarily on the basis of lipid and nucleic acid Raman peaks. This has been reported for a variety of mostly anchorage dependent cancer cell lines. We have established this result for CHO cells induced to undergo apoptosis using camptothecin. In particular, there were clearly observable decreased peaks associated with the total nucleic acids, DNA, RNA and lipid content in apoptotic cells. Principal component analysis of the entire spectra also revealed a clear separation between the Raman spectroscopic results for uninduced and apoptotic cells. Raman spectroscopy can indeed distinguish apoptotic from uninduced CHO cells.  Further, camptothecin treated CHO cells were stained with annexin V & propidium iodide and sorted using FACS to obtain enriched populations of the earliest stages of apoptosis: annexin 55  V/PI -/- and annexin V/PI +/- populations. Raman spectroscopy was performed on these cells and uninduced cells to detect both the progression of apoptosis and any differences between uninduced and apoptosis induced annexin V/PI -/- cells. The results showed that there was indeed a difference between these cells, with induced annexin V/PI -/- cells having decreasing nucleic acid-associated signals when compared to uninduced cells. Annexin V/PI +/- cells, classified as early apoptotic, had further decreased nucleic acid signals. Principal component analysis distinguished all three populations, with a progress along principal component 1 from uninduced to induced -/- cells, to induced +/- cells. This result reinforces the concept that Raman spectroscopy has the potential to detect the onset of apoptosis earlier than many conventional assays. This would be with the potential advantages of being a non-invasive, non-destructive method.   This is also a result beyond what has been reported in literature as the earliest detection of apoptosis using Raman spectroscopy.  A study of primary and secondary necrotic CHO cells revealed that they were clearly distinguishable from viable and autophagic CHO cells. In particular, primary necrotic cells had decreased nucleic acid-associated peaks and small increases in lipid peaks. Secondary necrotic cells had their nucleic acid-associated peaks decrease to be nearly undetectable and had a very large increases in lipid peaks. A previously uncharacterized peak was discovered at 672 cm-1 in the secondary necrotic population. On the other hand, autophagic cells had spectra that were very similar to viable cells, with slight increases in total nucleic acids and RNA that could be attributed to decreased cellular protein content during autophagy in these Phe-normalized spectra. Since the autophagic cells had very low cell growth rates but maintained viability, and yet had spectra that were very similar to viable cells, also addressed our concern that the 56  observed decrease in nucleic acid peaks during apoptosis could be due to the cessation of cell growth during that cell death mode. Principal component analysis clearly distinguished primary and secondary necrotic cell death from viable and autophagic cells, indicating that Raman spectroscopy is a candidate to distinguish modes of cell death.  This is an important result since it is difficult to distinguish primary necrotic cells, other than by measuring the release of HMGB-1 protein from cells [14].  CHO cells grown in fed-batch culture were analyzed to assess whether observed spectral signatures of apoptotic and necrotic cell death were also observed in cells that die during these cultures widely used to produce therapeutic proteins by the biotechnology industry. Cells from days 3, 7, 10 and 12 of fed-batch culture were analyzed. Nucleic acid-associated peaks had multiple different trends in fed-batch culture, which was different from the camptothecin-induced cells where these peaks uniformly decreased. In fed-batch, it was observed that adenine and guanine-associated peaks decreased over time; while such trend was evident for the other nucleic acid peaks that had diverse variations. The lipid-associated peaks increased with time as the fed-batch culture progressed, similarly to the camptothecin-induced cell observations. These lipid signals are a primary candidate for monitoring the progression of fed-batch cultures. Principal component analysis clearly distinguished days 3 & 7 of the cultures from each other and from days 10 & 12. A similar distinction was not observed between days 10 & 12, whose principal component scores overlapped to a large extent. Further experiments are required to determine Raman signals that can indicate the onset of apoptosis in fed-batch cultures.  57  Methanol fixation of cells prior to Raman spectroscopy enables high quality spectroscopy as the cells are immobilized and spectra can be collected over the span of several seconds. However, methanol fixation can also leach lipids out of cells, thus influencing the spectra obtained. This problem is worked around by drying all the methanol used for fixing onto the gold mirrors, thus enabling the detection of lipid content on average, as spectra are collected from a large area on the mirror.  A recurring problem with the Raman spectroscopy of apoptotic cells was saturation of the detector, especially when analyzing secondary necrotic cells. This was also observed during spectroscopy of cells from days 10 & 12 of fed-batch culture, indicating that there may be a cellular metabolite generated during apoptosis that generates a very strong Raman signal. Since the primary peaks trending similarly between the two sets of samples are lipid peaks, it is speculated that the metabolite is likely a lipid.  This is among the first analyses of CHO cells in fed-batch culture using Raman spectroscopy - the only previously reported results are measurement of cell viability & concentration and the concentration of amino acids and other metabolites in the culture. While the overall results had some similarities to reports for apoptotic mainly anchorage dependent cancer cell lines, there were also several differences.  A review of reported peak shifts in literature during apoptosis revealed that the observed changes were sometimes not consistent with one another; this is likely in part due to the complexity of apoptosis that can follow different intracellular signaling pathways. These processes also can depend on the varied stimule used to initiate the cell death cascade and other factors such as the genetic composition of the cells and the cell 58  microenvironment [7]. Depending on how apoptosis proceeds in a given cell or culture, the intracellular composition of the cells may be different, thereby leading to varied Raman signals and peak shifts. Thus, there is a need for creating a "library" of Raman peak shifts in apoptosis induced by different metabolites, in the case of CHO cells those relevant to fed-batch culture stresses such as ammonium, AMP and oxidized glutathione [2].  Ideally, common features could be identified to provide for robust detection of apoptosis for a variety of conditions, including combinations of stresses.     Cell death in general is the ultimate limiting factor to achieving higher cell densities and higher product titers in industrial cultures. However, apoptotic cell death may be preferable to primary necrotic cell death, as primary necrosis is generally uncontrolled and involves the release of a large number of intracellular enzymes and metabolites into the surrounding medium. This can adversely affect both the viability of the culture as a whole, and the quality of the recombinant protein being produced. This uncontrolled release of cellular constituents can also negatively impact downstream protein product purification. Since primary necrosis is difficult to detect; our results show that Raman spectroscopy may provide a method whereby primary necrosis can be detected in cultures. This may enable targeted remedial action, either to rescue the culture or to time the end of the process to limit cellular protein release contamination of the recombinant protein produced.   The longer-term goal of this project is to develop an in situ Raman probe that would act as a real-time sensor of cell state in bioreactors. In addition to constructing a probe capable of collecting Raman signals from a cell culture, this will be improved by the development of automated 59  spectral classification algorithms to classify even subtle peak shifts that represent overall changes in the state of the cells in culture. Possible candidates for such algorithms include support vector machine and neural network-based learning models. Overall, this should include the ability to continuously collect spectra from cultures in bioreactors, to classify the spectra to automatically indicate when a transition occurs to initiate increased cell death.  This will move beyond current commercially available devices that are generally limited to assessing the concentrations of key metabolites in culture along with assessing the total and viable cell concentrations.   While there are many more conditions to be evaluated before a complete picture of the Raman peak shifts during each stage of apoptosis can be developed, the results obtained thus far clearly indicate that Raman spectroscopy is an information-rich candidate for assessing the onset of cell death in CHO cell cultures. The longer-term success of this project should establish Raman spectroscopy as a tool of choice to monitor the state of cell cultures, applicable to both process development and industrially production.  60  References  1. Wong, D.C.F., Wong, K.T.K., Nissom, P.M., Heng, C.K. and Yap, M.G.S. (2006). Targeting early apoptotic genes in batch and fed-batch CHO cell cultures. Biotechnology and Bioengineering 95, pp 350-361. 2. Chong, W.P., Yusufi, F.N., Lee, D.Y., Reddy, S.G., Wong, N.S., Heng, C.K., Yap, M.G. and Ho, Y.S. (2011). Metabolomics-based identification of apoptosis-inducing metabolites in recombinant fed-batch CHO culture media. Journal of Biotechnology 151, pp 218-224. 3. Krampe, B. and Al-Rubeai, M. (2010). Cell death in mammalian cell culture: molecular mechanisms and cell engineering strategies. Cytotechnology 62, pp 175-188. 4. Arden, N. and Betenbaugh, M.J. (2004). Life and death in mammalian cell culture: strategies for apoptosis inhibition. Trends in Biotechnology 22, pp 174-180. 5. Archana, M., Bastian, Yogesh, T.L. and Kumaraswamy, K.L. (2013). Various methods available for detection of apoptotic cells - a review. Indian Journal of Cancer 50, pp 274-283. 6. Zhang, H., Huang, Q., Ke, N., Matsuyama, S., Hammock, B., Godzik, A. and Reed, J. C. (2000). Drosophila pro-apoptotic Bcl-2/Bax homologue reveals evolutionary conservation of cell death mechanisms. The Journal of Biological Chemistry 275, pp 27303-27306. 7. Taylor, R.C., Cullen, S.P. and Martin, S.J. (2008). Apoptosis: Controlled demolition at the cellular level. Nature Reviews Molecular Cell Biology 9, pp 231-241.  61  8. Creagh, E.M., Conroy, H. and Martin, S.J. (2003). Caspase-activation pathways in apoptosis and immunity. Immunology Reviews 193, pp 10-21. 9. Saraste, A. and Pulkki, K. (2000). Morphologic and biochemical hallmarks of apoptosis. Cardiovascular Research 45, pp 528-537. 10. Neogesecu, A., Lorimier, P., Labat-Moleur, F., Drouet, C., Robert, C., Guillermet, C., Brambilla, C. and Brambilla, E. (1996). In situ apoptotic cell labelling by the TUNEL method: improvement and evaluation on cell preparations. Journal of Histochemistry and Cytochemistry 44, pp 959-968. 11. Goswami, J., Sinskey, A. J., Steller, H., Stephanopoulos, G. N. and Wang, D.I.C. (1999). Apoptosis in batch cultures of Chinese hamster ovary cells. Biotechnology and Bioengineering 62, pp 632-640. 12. Kroemer, G. et al. (2009). Classification of cell death: Recommendations of the Nomenclature Committee on Cell Death 2009. Cell Death and Differentiation 16, pp 3-11. 13. Galluzzi, L., Maiuri, M.C., Vitale, I., Zischka, H., Castedo, M., Zitvogel, L. and Kroemer, G. (2007). Cell Death Modalities: Classification and Pathophysiological Implications. Cell Death and Differentiation 14, pp 1237-1243. 14. Krysko, D.V., Berghe, T.V., D'Herde, K. and Vandenabeele, P. (2008). Apoptosis and necrosis: Detection, discrimination and phagocytosis. Methods 44, pp 205-221. 15. Das, G., Shravage, B.V. and Baehrecke, E.H. (2012). Regulation and function of autophagy during cell survival and cell death. Cold Spring Harbor Perspectives in Biology 4.  62  16. He, C. and Klionsky, D. (2009). Regulation mechanisms and signaling pathways of autophagy. Annual Review of Genetics 43, pp 67-93. 17. Taneda, I., Ueno, T. and Kominami, E. (2008). LC3 and Autophagy. Methods in Molecular Biology 445, pp 77-88. 18. Barth, S., Glick, D. and MacLeod, K.F. (2010). Autophagy: assays and artifacts. Journal of Pathology 221, pp 117-124. 19. Mizushima, N. (2007). Autophagy: process and function. Genes and Development 21, pp 2861-2873. 20. Luck, W. A. P. (1990), D. J. Gardiner, P. R. Graves (Eds.): Practical Raman Spectroscopy, mit Beiträgen von H. J. Bowley, D. J. Gardiner, D. L. Gerrard, P. R. Graves, J. D. Louden, and G. Turrell, Springer-Verlag, Berlin, Heidelberg, New York 1989. 157 Seiten, brosch., Preis: DM 86,—. Berichte der Bunsengesellschaft für physikalische Chemie, 94, pp 1047. 21. Huser, T. and Chan, J. (2015). Raman spectroscopy for physiological investigations of tissues and cells. Advanced Drug Delivery Reviews 89, pp 57-70. 22. Maquelin, K., Choo-Smith, L.P., van Vreeswijk, T., Endtz, H.P., Smith, B., Bennett, R., Bruining, H.A. and Puppels, G.J. (2000). Raman spectroscopic method for identification of clinically relevant microorganisms growing on solid culture medium. Analytical Chemistry 72. 23. Neugebauer, U., Clement, J.H., Bocklitz, T., Krafft, C. and Popp, J. Identification and differentiation of single cells from peripheral blood by Raman spectroscopic imaging. Journal of Biophotonics 3, pp 579-587.  63  24. Schulze, H.G., Konorov, S.O., Caron, N.J., Piret, J.M., Blades, M.W. and Turner, R.F. (2010). Assessing differentiation status of human embryonic stem cells noninvasively using Raman microspectroscopy. Analytical Chemistry 82, pp 5020-5027. 25. Saha, A. and Yakovlev, V.V. (2009). Towards a rational drug design: Raman micro-spectroscopy analysis of prostate cancer cells treated with an aqueous extract of Nerium oleander. Journal of Raman Spectroscopy 40, pp 1459-1460. 26. Chernenko, T., Sawant, R.R., Miljkovic, M., Quintero, L., Diem, M. and Torchilin, V. (2012). Raman microscopy for noninvasive imaging of pharmaceutical nanocarriers: intracellular distribution of cationic liposomes of different composition. Molecular Pharmaceutics 9, pp 930-936. 27. Wurm, F.M. (2004). Production of recombinant protein therapeutics in cultivated mammalian cells. Nature Biotechnology 22, pp 1393-1398. 28. Urlaub, G. and Chasin, L.A. (1980). Isolation of Chinese hamster cell mutants deficient in dihydrofolate reductase activity. Proceedings of the National Academy of Sciences 77, pp 4216-4220. 29. Wall, M.E., Wani, M.C. and Cook, C.E. (1966). Plant antitumor agents. 1. The isolation and structure of camptothecin, a novel alkaloidal leukemia and tumor inhibitor from Camptotheca acuminata. Journal of the American Society of Chemistry 83, 3888-3890. 30. Hsiang, Y.H., Hertzberg, R., Hecht, S. and Liu, L.F. (1985). Camptothecin induces protein-linked DNA breaks via mammalian DNA topoisomerase I. Journal of Biological Chemistry 260, 14873-14878. 31. Kunapareddy, N., Freyer, J.P. and Mourant, J.R. (2008). Raman spectroscopic characterization of necrotic cell death. Journal of Biomedical Optics 13(5). 64  32. Jardon, M.A., Sattha, B., Braasch, K., Leung, A.O., Côté, H.C., Butler, M., Gorski, S.M. and Piret, J.M. (2012). Inhibition of glutamine-dependent autophagy increases t-PA production in CHO cell fed-batch processes. Biotechnology and Bioengineering 109, pp 1228-1238. 33. Pagba, C.V., Lane, S.M. and Wachsmann-Hogiu, S. (2011). Conformational changes in quadruplex oligonucleotide structures probed by Raman spectroscopy. Biomedical Optics Express 2, pp 207-217. 34. Figueroa Jr., B., Ailor, E., Osborne, D., Hardwick, J.M., Reff, M. and Betenbaugh, M.J. (2007). Enhanced cell culture performance using inducible anti-apoptotic genes E1B-19K and Aven in the production of a monoclonal antibody with Chinese hamster ovary cells. Biotechnology & Bioengineering 97, pp 877-892. 35. Klionsky, D.J. et al., Guidelines for the use and interpretation of assays for monitoring autophagy. Autophagy  8, pp 445-544. 36. Braasch, K., Nikolic-Jaric, M., Cabel, T., Salimi, E., Bridges, G.E., Thomson, D.J. and Butler, M. (2013). The changing dielectric properties of CHO cells can be used to determine early apoptotic events in a bioprocess. Biotechnology & Bioengineering 110, pp 2902-2914. 37. Ovizio Technical Insight - Novel Digital Holographic Microscopy. Frost & Sullivan 2011. 38. Whelan, J., Craven, S. and Glennon, B. (2012). In situ Raman spectroscopy for simultaneous monitoring of multiple process parameters in mammalian cell culture bioreactors. Biotechnology Progress 28.  65  39. Butler, M., Spearman, M. and Braasch, K. (2014). Monitoring cell growth, viability and apoptosis in Animal Cell Biotechnology – Methods and Protocols (ed. Ralf Poertner), 1104: 169-192. Methods in Molecular Biology. Humana Press. 40. Schulze, H.G. and Turner, R.F. (2015). Development and integration of block operations for data invariant automation of digital preprocessing and analysis of biological and biomedical Raman spectra. Applied Spectroscopy 69, pp 643-664. 41. Meléndez, A. and Levine, B. (2009) Autophagy in C. elegans. WormBook, ed. The C. elegans Research Community, WormBook, doi/10.1895/ wormbook.1.147.1,  http://www.wormbook.org. 66  Appendix Appendix A  MATLAB code used for preprocessing of Raman data % Processing sequence % (Converts sequence to code that calls functions at each of numbered points; user input required only at point 1) %  % Processing % 1. Load a sequence of files or load a map and parse (determined automatically) and get key input information % 2. Do baseline flattening (blf)  % 3. Remove spikes  % 4. Get standard deviation and SNR of data set from subsample % 5. Smooth (determine best method) % 6. Scale to counter smoothing effects % 7. Subtract background % 8. Normalize (select method: spectral max, internal standard, vector % normalization, probability quotient vector normalization, volume exclusion) % 9. QC:compare smoothed originals to smoothed finals using normalized derivatives and PCA scores  %  % Analysis % 10. PCA and check for clustering/uniformity (alert user if clustering present); ID and provide principal components 67  % 11. 2DCOS and provide synchronous and asynchronous maps % 12. Statistics?     % Code   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Specify core results name, intermediate results will be appended to this core name % Example % input filename:    '542ThreeWeeks - Raw Data - 1 - Liquid + coverslip - 1 acc 40 sec acq x20.txt  % core results name: 'D542liq_3week' % Intermediate results matrices, as successive processing steps are performed, will have suffixes appended to this core name.  % E.g. , after loading raw data it becomes 'D542liq_3week_raw',  % after subsequent 15-iteration baseline flattening it becomes 'D542liq_3week_raw_blf15', etc. % Note: uipickfiles may be better, download from Matlabcentral if needed.   clear all close all pauseTime = 0; processNum = 0; BGprocessNum = 0; suffixes = {}; BGsuffixes = {}; 68  warningstrings = {}; BGwarningstrings = {}; warningNumber = 0;   % 1. Get key input information; load data prompt = {'Enter a descriptive core name (e.g. ID_054_liq_24h_3mW) for your data results, for intermediate results, suffixes will be appended to this core name (e.g. ID_054_liq_24h_3mW_raw_blf15 etc.). A suggestion is supplied below. ',...           'Are there background spectra to load and subtract from the analyte spectra? Y/N : ',...           'The number of iterations to use for baseline flattening can be determined automatically if the default is accepted. Otherwise, specify the number of iterations to use : ',...           'Smoothing of spectra uses a Savitzky-Golay filter, the default is a zero-order filter. A second-order filter could be used if desired, resulting in generally less smoothing but better peak retention. Higher orders are not recommended. Enter a different order (e.g. 2) if desired: ',...           'Smoothing of spectra uses the specified Savitzky-Golay filter iteratively and the number of iterations can be determined automatically. If fewer iterations are required, specify them now: '...           'Specify the time to pause between processing steps. This will activate the plotting of intermediate results. Useful if troubleshooting unsatisfactory processing, otherwse use 0.'...           'Specify the saturation level of the current detector (i.e. the maximum spectral value that can be obtained). Spectra with such values, along with other pathological spectra, will be culled before analysis.'...           'Specify the total collection time for each spectrum (seconds).'... 69            'Specify the average number of cosmic ray-induced spikes per pixel for your detector.'};   dlg_title = 'Input for automated spectral data processing'; num_lines = [2, 100]; def = {datestr(date,12),'N','Auto','0','Auto','0','156175','20','0.000342'}; useranswer = inputdlg(prompt,dlg_title,num_lines,def); if isempty(useranswer)     % No input, return to keyboard     clear all     close all     return else     for n = 1:length(useranswer);useranswer{n} = strtrim(useranswer{n});end; end resultsName = [useranswer{1} '_']; figName = [useranswer{1} ' ']; % useranswer{2} assessed below if isempty(str2num(useranswer{3}))          Max_j = 1e128; else     Max_j = str2num(useranswer{3}); end 70  polyorder = str2num(useranswer{4}); if isempty(str2num(useranswer{5}))     iterations  =  1e128; else     iterations = str2num(useranswer{5}); end pauseTime = str2num(useranswer{6}); saturationLevel = 0.99*str2num(useranswer{7}); collectionTime = str2num(useranswer{8}); spikeFreq_expected = str2num(useranswer{9}); % upper range of fraction of pixels affected by spikes spikeFraction_expected = collectionTime*spikeFreq_expected;   % Specify  normalization methods     optionsStr = {'1. Spectral maximum method (generally destroys correlation structure of data, follow-up data analyses of limited use)',...                   '2. Internal standard method (preserves correlation structure of data, follow-up data analyses useful)',...                   '3. Vector normalization method (generally destroys correlation structure of data, follow-up data analyses of limited use)',...                   '4. Probability quotient vector normalization method (NOT good where numerous peaks change, but preserves correlation structure of data, follow-up data analyses possibly useful)',... 71                    '5. Volume exclusion method not available in this arrangement.',...                   '6. None (when no normalization is desired or initially when unsure about the appropriate normalization approach).'};     [normMethod,responseMade] = listdlg('PromptString','Select a normalization method:',...                                         'SelectionMode','single',...                                         'ListString',optionsStr,'ListSize',[999 100], 'InitialValue', 6);     if (normMethod(1) == 2 || normMethod(1) == 5)         prompt = 'What is the peak position of the internal standard or the volume exclusion reference? Specify in terms of x-values, e.g. if the x-values are in pixels, specify approximate pixel number, if in wavenumbers, specify approximate wavenumber: ';         def = {'0'};         temp = inputdlg(prompt,'Provide reference peak position',2,def);         peakPosition = str2num(temp{1});         if (normMethod(1) == 5)             prompt = {'Estimate sample thicknesses also (Y/N)? If no, enter "N" and click OK, otherwise provide/verify input. ',...                        'Enter the numerical aperture (NA) of the microscope objective: ',...                       'Enter the refractive index of the medium: '...                       'Enter the working distance of microscope objective (in micrometers): ',...                       'Enter the size of the physical area of the laser spot on the sample (in micrometers squared): '};                         def = {'Y','0.80','1.335','3300','90'}; 72              instrumentInfo = inputdlg(prompt,'Provide setup information for sample thickness estimation',2,def);         else         end     else         peakPosition = 0;       end   % Load a sequence of files or load a map and parse (determined automatically) loadBoxTitle = 'Select the analyte file(s) to load'; [pathName, CellofFileNames, formatDataLoaded, generatedvarname, suffix]  = f_loadDatafile_s_v2(resultsName, loadBoxTitle, pauseTime); useData = eval(generatedvarname); origStr = pathName(length(pathName)-12:length(pathName)); generatedvarname = genvarname([resultsName suffix]); processNum  = processNum + 1; suffixes{processNum} = suffix; pause(pauseTime) close all   % Check to see if spectra actually contain data indices = useData(:,2:end) > 0; 73  if isempty(indices)     Message = 'Spectra all zeros or negative. Please check your data. Processing aborted.' ;     Title = 'Data integrity';     h = msgbox(Message, Title, 'warn', 'modal');      return; end % Set saturated spectra to zero values [r,c]= find(useData(:,2:end) > saturationLevel); c = unique(c)+1; % First column with Raman shift was removed, add 1 to all columns for n = 1:length(c)     useData(:,c(n)) = 0*useData(:,c(n)); end if pauseTime > 0 % Plot loaded data for inspection     figHandles = f_plotRaman(1, 1, 'Data loaded');     figure(figHandles(1));     plot(useData(:,1),useData(:,2:end));     title([strrep(resultsName, '_', '') ' ' 'saturated spectra zeroed']);      axis tight;     pause(pauseTime) end disp([formatDataLoaded ' into variable:    ' generatedvarname]); close all   74  % Load background, if any. if isempty(useranswer{2}) || useranswer{2} == 'Y' || useranswer{2} == 'y'     loadBoxTitle = 'Select the background file(s) to load';     BGname = [resultsName  'bg'];     BGgeneratedvarname = [resultsName  'bg'];     [pathName, BGCellofFileNames, formatDataLoaded, BGgeneratedvarname, BGsuffix]  = f_loadDatafile_s_v2(BGgeneratedvarname, loadBoxTitle, pauseTime);     BGprocessNum  = BGprocessNum + 1;     BGsuffixes{BGprocessNum} = BGsuffix;     BGuseData = eval(BGgeneratedvarname);     pause(pauseTime)     close all     % Check to see if background spectra actually contain data     indices = BGuseData(:,2:end) > 0;     if isempty(indices)         Message = ['Background all zeros or negative. Please check your data. Processing aborted.'] ;         Title = 'Data integrity';         h = msgbox(Message, Title, 'warn', 'modal');          return;     end     % Set saturated spectra to zero values     [r,c]= find(BGuseData(:,2:end) > saturationLevel); 75      c = unique(c)+1; % First column with Raman shift was removed, add 1 to all columns     for n = 1:length(c)         BGuseData(:,c(n)) = 0*BGuseData(:,c(n));     end       if pauseTime > 0 % Plot loaded data for inspection          figHandles = f_plotRaman(1, 1, 'Background loaded');         figure(figHandles(1));         plot(BGuseData(:,1),BGuseData(:,2:end));         title([strrep(BGgeneratedvarname, '_', '') ' ' 'saturated spectra zeroed']);         axis tight;         pause(pauseTime)     end     disp([formatDataLoaded ' into variable:    ' BGgeneratedvarname]) end      Message = ['Step ' num2str(processNum) ' - data loaded. Working...'] ; Title = 'Processing status'; h = msgbox(Message,Title,'warn','modal');  % return % input(['Number ' num2str(processNum) ' finished']); %%%%%%%%%%%%  -----   Begin processing  -----  %%%%%%%%%%%%%%%%%%%%%%%%%% 76  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%   % 2. Do baseline flattening (blf)  close all suffix2 = ''; if Max_j == 1e128      % Determine number of iterations to use automatically on a subsample of 20 spectra     if size(useData,2) > 20         subsample(:, 1)   = useData(:,1);         subsample(:,2:21) = useData(:,[randi([2,size(useData,2)],1,20)]);     else         subsample  = useData;     end     pixelwidth = useData(2,1) - useData(1,1);     [blf_iterations_used, tempgeneratedvarname, suffix2] = f_autobaselineRemoval_SWiMA_v3(subsample, Max_j, generatedvarname, pauseTime);     if 2*blf_iterations_used*pixelwidth > 50         warningNumber  = warningNumber + 1;         blf_iterations_used = ceil(50/(2*pixelwidth));         new_warningstrings{warningNumber} = ['Too many baseline flattening iterations, window size possibly too large. Iterations limited to ' num2str(blf_iterations_used) '.']; 77          warningstrings = {warningstrings{:}, new_warningstrings{:}};     end     generatedvarname   = tempgeneratedvarname;     if exist('BGuseData','var') ~= 0         BGgeneratedvarname = [BGgeneratedvarname suffix2];     end     Max_j = blf_iterations_used;     clear('subsample');     close all end % Use same parameter settings for analyte spectral baseline flattening [blf_iterations_used, generatedvarname, suffix] = f_autobaselineRemoval_SWiMA_v3(useData, Max_j, generatedvarname, pauseTime); useData = eval(generatedvarname); processNum  = processNum + 1; suffixes{processNum} = [suffix2 suffix]; disp(['Baseline flattened data in variable:    ' generatedvarname]) pause(pauseTime)   % Use same parameter settings for background baseline flattening close all; if exist('BGuseData','var') ~= 0 78      [BGblf_iterations_used, BGgeneratedvarname, BGsuffix] = f_autobaselineRemoval_SWiMA_v3(BGuseData, Max_j, BGgeneratedvarname, pauseTime);     BGuseData = eval(BGgeneratedvarname);     BGprocessNum  = BGprocessNum + 1;     BGsuffixes{BGprocessNum} = [suffix2 BGsuffix];     disp(['Baseline flattened data in variable:    ' BGgeneratedvarname])     pause(pauseTime) end Message = ['Step ' num2str(processNum) ' - baseline flattened. Working...']; Title = 'Processing status'; h = msgbox(Message, Title,'warn','modal');  % input(['Number ' num2str(processNum) ' finished']); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%   % 3. Remove spikes  close all indicesCombi = []; % Alternatives % [generatedvarname, suffix] = f_deSpike_v2(useData, generatedvarname, pauseTime); % [generatedvarname, suffix] = f_spikeRemoval_v5(useData, generatedvarname, pauseTime); % Note: because the upper range of percentage of pixels affected by spikes 79  % is about 1%, the threshold is set to 1%, i.e. no more than 1 pixel in 100 % should be affected. Spikes occurring in data sets with less than 100 % spectra may not be removed. if (size(useData,1) < 3 || size(useData,2) < 3) % cannot use f_spikeRemoval_v5 for fewer than 3 spectra    [generatedvarname, suffix] = f_deSpike_v2(useData, generatedvarname, pauseTime);  else    [generatedvarname, suffix, indicesCombi] = f_spikeRemoval_v5(useData, spikeFraction_expected, generatedvarname, pauseTime); end if sum(sum(indicesCombi)) > 0     warningNumber  = warningNumber + 1;             new_warningstrings{warningNumber} = ['Problem spikes may exist, see indicesCombi for locations. Number of problems: ' num2str(sum(sum(indicesCombi))) '.'];     warningstrings = {warningstrings{:}, new_warningstrings{:}}; end processNum  = processNum + 1; suffixes{processNum} = suffix; useData = eval(generatedvarname); disp(['Data after spike removal in variable:    ' generatedvarname]) pause(pauseTime)   if exist('BGuseData','var') ~= 0 80      if (size(BGuseData,1) < 3 || size(BGuseData,2) < 3)         [BGgeneratedvarname, BGsuffix] = f_deSpike_v2(BGuseData, BGgeneratedvarname, pauseTime);     else         [BGgeneratedvarname, BGsuffix, indicesCombi] = f_spikeRemoval_v5(BGuseData, spikeFraction_expected, BGgeneratedvarname, pauseTime);     end         BGuseData = eval(BGgeneratedvarname);     BGprocessNum  = BGprocessNum + 1;     BGsuffixes{BGprocessNum} = BGsuffix;     disp(['Baseline data after spike removal in variable:    ' BGgeneratedvarname])     pause(pauseTime) end Message = ['Step ' num2str(processNum) ' - cosmic spikes removed. Working...']; Title = 'Processing status'; h = msgbox(Message,Title,'warn','modal');  % input(['Number ' num2str(processNum) ' finished']); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%   % 4. Get standard deviation and SNR of data set from subsample close all; 81  specStddevs  = []; specMaxPeaks = []; SNRs         = []; % Get noise of original data origData = eval(genvarname([resultsName suffixes{1}])); % Raw data for n = 2:size(origData,2)     specStddevs(n)  = f_spectralStddev_v2(origData(:,n)); end % Get max peak values of baseline-flattened and despiked data for n = 2:size(origData,2)     specMaxPeaks(n) = max(origData(:,n));     SNRs(n)         = specMaxPeaks(n)/specStddevs(n); end average_SNR   = mean(SNRs(2:end)); average_Noise = mean(specStddevs(2:end)); processNum  = processNum + 1; suffixes{processNum} = ''; if exist('BGuseData','var') ~= 0     BGprocessNum  = BGprocessNum + 1;     BGsuffixes{BGprocessNum} = ''; end   disp(['In ' genvarname([resultsName suffixes{1}]) ]) 82  disp(['the estimated mean SNR is:   ' num2str(average_SNR) ' and ']) disp(['the estimated mean spectral noise is:   '  num2str(average_Noise)]) Message = ['Step ' num2str(processNum) ' - basic spectral attributes determined. Working...']; Title = 'Processing status'; h = msgbox(Message,Title,'warn','modal');  pause(pauseTime) % input(['Number ' num2str(processNum) ' finished']); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%   % 5. Smooth  close all; [generatedvarname, suffix, smoothingIters_used] = f_smoothAutomated_v2(useData, generatedvarname, iterations, polyorder, pauseTime); processNum  = processNum + 1; suffixes{processNum} = suffix; useData = eval(generatedvarname); disp(['Data after smoothing in variable:    ' generatedvarname]) pause(pauseTime)   if exist('BGuseData','var') ~= 0 83      [BGgeneratedvarname, BGsuffix, BGsmoothingIters_used] = f_smoothAutomated_v2(BGuseData, BGgeneratedvarname, iterations, polyorder, pauseTime);     BGuseData = eval(BGgeneratedvarname);     BGprocessNum  = BGprocessNum + 1;     BGsuffixes{BGprocessNum} = BGsuffix;     disp(['Baseline data after smoothing in variable:    ' BGgeneratedvarname])     pause(pauseTime) end Message = ['Step ' num2str(processNum) ' - smoothing completed. Working...']; Title = 'Processing status'; h = msgbox(Message,Title,'warn','modal');  % input(['Number ' num2str(processNum) ' finished']); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%     % 6. Scale smoothed results close all; noisyData = eval(genvarname([resultsName suffixes{1:3}]));   [generatedvarname, suffix] = f_weightNscale_smoothedData(noisyData, useData, generatedvarname, pauseTime); processNum  = processNum + 1; 84  suffixes{processNum} = suffix; useData = eval(generatedvarname); disp(['Data after scaling in variable:    ' generatedvarname]) pause(pauseTime) if exist('BGuseData','var') ~= 0     BGnoisyData = eval(genvarname([BGname BGsuffixes{1:3}]));        [BGgeneratedvarname, BGsuffix] = f_weightNscale_smoothedData(BGnoisyData, BGuseData, BGgeneratedvarname, pauseTime);     BGuseData = eval(BGgeneratedvarname);     BGprocessNum  = BGprocessNum + 1;     BGsuffixes{BGprocessNum} = BGsuffix;     disp(['Baseline data after scaling in variable:    ' BGgeneratedvarname])     pause(pauseTime) end Message = ['Step ' num2str(processNum) ' - scaling completed. Working...']; Title = 'Processing status'; h = msgbox(Message,Title,'warn','modal');  % input(['Number ' num2str(processNum) ' finished']); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%     85  % 7. Normalize (select method: spectral max, internal standard, vector normalization, probability quotient vector normalization, volume exclusion) close all; if responseMade == 0     % No normalization was selected, further analyses skipped. Return to keyboard     return else         switch normMethod       case 1 % Maximum peak             [generatedvarname, suffix] = f_normInternalStd(useData, generatedvarname, peakPosition, pauseTime);              processNum  = processNum + 1;              suffixes{processNum} = suffix;       case 2 % Internal standard             [generatedvarname, suffix] = f_normInternalStd(useData, generatedvarname, peakPosition, pauseTime);              processNum  = processNum + 1;              suffixes{processNum} = suffix;       case 3 % Vector normalization (really: constant sum)             [generatedvarname, suffix] = f_normConstSum(useData, generatedvarname, pauseTime)              processNum  = processNum + 1;              suffixes{processNum} = suffix;       case 4 % Probability quotient vector normalization (variant of vector normalization) 86              [generatedvarname, suffix, new_warningstrings] = f_normProbQuotient(useData, generatedvarname, specStddevs, pauseTime);             warningstrings = {warningstrings{:}, new_warningstrings{:}};             processNum  = processNum + 1;             suffixes{processNum} = suffix;       case 5 % Volume exclusion normalization (variant of internal standard)             BGData_blf = eval(genvarname([BGname BGsuffixes{1:6}]));             Data_blf = eval(genvarname([resultsName suffixes{1:6}]));             [generatedvarname, suffix, new_warningstrings] = f_normVolumeExcl(useData, Data_blf, BGData_blf, instrumentInfo, generatedvarname, peakPosition, pauseTime);             warningstrings = {warningstrings{:}, new_warningstrings{:}};             processNum  = processNum + 1;             suffixes{processNum} = suffix;       case 6 % No normalization              suffix = 'notnormd_' ;                         generatedvarname = [generatedvarname suffix];             assignin('base', generatedvarname, useData);             processNum  = processNum + 1;             suffixes{processNum} = suffix;     end end useData = eval(generatedvarname);    disp(['Data after normalization in variable:    ' generatedvarname]) 87  Message = ['Step ' num2str(processNum) ' - normalization completed. Working...']; Title = 'Processing status'; h = msgbox(Message,Title,'warn','modal');  pause(pauseTime) % input(['Number ' num2str(processNum) ' finished']); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%   % 8. Subtract background or reference spectrum, if any close all; if exist('BGuseData','var') ~= 0      [generatedvarname, suffix] = f_subRefSpectrum(useData, eval(BGgeneratedvarname), generatedvarname, peakPosition, pauseTime);     processNum  = processNum + 1;     suffixes{processNum} = suffix;     useData = eval(generatedvarname);     disp(['Spectra after background subtraction in variable:    ' generatedvarname])     pause(pauseTime) else     processNum  = processNum + 1;     suffixes{processNum} = 'nobgs_';       generatedvarname = genvarname([generatedvarname suffixes{processNum}]); 88      assignin('base', generatedvarname, useData);     useData = eval(generatedvarname); end Message = ['Step ' num2str(processNum) ' - background removal, if any, completed. Working...']; Title = 'Processing status'; h = msgbox(Message,Title,'warn','modal');  % input(['Number ' num2str(processNum) ' finished']); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%   % 9. QC:compare smoothed originals to smoothed finals using normalized derivatives and PCA scores  pauseTime1 = 1; if size(useData,2) <= 2    % Diagnostic plots     close all     set(0,'Units','pixels');     scrsz = get(0,'ScreenSize');         figHandles = f_plotRaman(2, [1,1], figName);         figure(figHandles(1));tempdata = evalin('base','genvarname([resultsName suffixes{1}])');plot(evalin('base',[tempdata '(:,1)']), evalin('base',[tempdata '(:,2:end)']));axis tight 89          title([figName 'Raw spectra']);         figure(figHandles(2));         for n = 1:size(useData,2)-1             plot(useData(:,1), useData(:,n+1),'color', [(0.75 + 0.25*(-1)^n)*n/size(useData,2) (size(useData,2)-n)/size(useData,2) (1 - 0.5*(-1)^n)*n/(2*n)],'linewidth',2);axis tight                 end         title([figName 'Processed spectra']);          save([pathName '/' resultsName '.mat']);         Message = 'Processing and analyses completed.';         Title = 'Processing status';         h = msgbox(Message,Title,'warn','modal');     return end   iterations = mode(smoothingIters_used); [generatedvarname, suffix, smoothingIters_used] = f_smoothAutomated_v2(origData, genvarname([resultsName suffixes{1}]), iterations, polyorder, pauseTime); QCraw = eval(generatedvarname); QCprocessed = useData; clear(generatedvarname); [QCgeneratedvarname_first, QCgeneratedvarname_last, outliers, new_warningstrings] = f_QualityControl_v4(QCprocessed, QCraw, spikeFraction_expected, pauseTime1, figName); warningstrings = {warningstrings{:}, new_warningstrings{:}}; 90  disp(['Quality control data in variables:    ' QCgeneratedvarname_first ' and ' QCgeneratedvarname_last]) processNum  = processNum + 1; suffixes{processNum} = ''; if pauseTime > 0 || ~isempty(warningstrings)      Message = ['Step ' num2str(processNum) ' - quality control data acquired. Working...'];     Title = 'Processing status';     h = msgbox(Message,Title,'warn','modal');  else     Message = ['Step ' num2str(processNum) ' - quality control data acquired. Working...'];     Title = 'Processing status';     h = msgbox(Message,Title,'warn','modal');  end for n = 1:6     switch n                  case 1          saveas(n, [pathName '/' resultsName suffixes{1} '.fig']);      case 2          saveas(n, [pathName '/' resultsName suffixes{1:processNum} '.fig']);      case 3          saveas(n, [pathName '/' resultsName suffixes{1:processNum} 'surf.fig']);      case 4          saveas(n, [pathName '/' resultsName  'compare_1deriv.fig']); 91       case 5          saveas(n, [pathName '/' resultsName  'compare_2deriv.fig']);      case 6          saveas(n, [pathName '/' resultsName suffixes{processNum} 'clustering.fig']);      otherwise          saveas(n, [pathName '/' resultsName 'unspecified.fig']);             end end % input(['Number ' num2str(processNum) ' finished']); % return %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%   % 10. PCA:  ID and provide principal components % close all; [PCAldgs, PCAscrs, PCAeigs] = princomp(useData(:,2:end)'); PCAeigs = PCAeigs/sum(PCAeigs); sumeigs = 0; PCs_major = 0; while sumeigs < 0.95 && PCs_major < 5     PCs_major = PCs_major + 1;     sumeigs = sumeigs + PCAeigs(PCs_major); end 92  figHandles = f_plotRaman(3, [3,4,5], figName);  figure(figHandles(1));  for n = 1:PCs_major     plot(useData(:,1), PCAldgs(:,n),'color', [(0.75 + 0.25*(-1)^n)*n/size(useData,2) (size(useData,2)-n)/size(useData,2) (1 - 0.5*(-1)^n)*n/(2*n)],'linewidth',1);axis tight; %     pause(pauseTime) end  axis tight; saveas(figHandles(1), [pathName '/' resultsName 'PCAldgs.fig']); figure(figHandles(2)); for n = 1:PCs_major      plot(PCAscrs(:,n), PCAscrs(:,n+1),'*','color', [(0.75 + 0.25*(-1)^n)*n/size(useData,2) (size(useData,2)-n)/size(useData,2) (1 - 0.5*(-1)^n)*n/(2*n)],'linewidth',2); axis tight;  %      pause(pauseTime) end axis tight; saveas(figHandles(2), [pathName '/' resultsName 'PCAscrs.fig']); figure(figHandles(3));  for n = 1:2*PCs_major     plot(100*PCAeigs(1:2*PCs_major),'ok','linewidth',2); end axis tight; saveas(figHandles(3), [pathName '/' resultsName 'PCAeigs.fig']); 93  disp(['Principal component analysis results in :    PCAldgs, PCAscrs, and PCAeigs']) processNum  = processNum + 1; suffixes{processNum} = ''; Message = ['Step ' num2str(processNum) ' - principal component analysis completed. Working...']; Title = 'Processing status'; h = msgbox(Message,Title,'warn','modal');  % input(['Number ' num2str(processNum) ' finished']);     % 11. 2DCOS and provide synchronous and asynchronous maps [TDCOV, PDDCOV, TDCOR, PDDCOR, TDASN, PDDASN, TDDSR, PDDDSR] = f_2DCOS_2DDRS(useData(:,1), 1:size(useData,2)-1, useData(:,2:end)); figHandles = f_plotRaman(2, [6,6], figName); % 2DCOS maps figure(figHandles(1)); fig_axes = findobj(figHandles(1)); axes(fig_axes(max(size(fig_axes)))); surf(TDCOR(2:end,1), TDCOR(1,2:end), TDCOR(2:end,2:end)); shading interp; view(2)     % View azimuth = 0; elevation = 90 --> view from top contour3(TDCOV(2:end,1), TDCOV(1,2:end), TDCOV(2:end,2:end)+max(max(TDCOV(2:end,2:end)))+1, 'k');    94  axis tight box on hold on; axes(fig_axes(max(size(fig_axes)-1))); plot(useData(:,1), useData(:,2),'k'); set(gca,'XTickLabel',[]); set(gca,'YTickLabel',[]); axis tight box on title([figName '2DCOS maps'],'FontName','Times','FontSize',24); axes(fig_axes(max(size(fig_axes)-2))); plot(useData(:,end),useData(:,1),'k'); set(gca,'XTickLabel',[]); set(gca,'YTickLabel',[]); axis tight box on drawnow; saveas(figHandles(1), [pathName '/' resultsName '2DCOS.fig']); % Perturbation domain decompositions figure(figHandles(2));           fig_axes = findobj(figHandles(2)); axes(fig_axes(max(size(fig_axes)))); surf(PDDCOR(2:end,1), PDDCOR(1,2:end), PDDCOR(2:end,2:end)'); ylabel('Perturbation sequence','FontName','Times','FontSize',24); view(2);   % View azimuth = 0; elevation = 90 --> view from top box on shading interp;     95  hold on; contour3(PDDCOV(2:end,1), PDDCOV(1,2:end), PDDCOV(2:end,2:end)'+max(max(PDDCOV(2:end,2:end)))+1, 'k'); axis tight axes(fig_axes(max(size(fig_axes)-1))); plot(useData(:,1), useData(:,2),'k'); set(gca,'XTickLabel',[]); set(gca,'YTickLabel',[]); title([figName '2D perturbation domain maps'],'FontName','Times','FontSize',24); axis tight box on axes(fig_axes(max(size(fig_axes)-2))); plot(1:size(useData,2)-1,'k'); set(gca,'XTickLabel',[]); set(gca,'YTickLabel',[]); axis tight box on drawnow;  saveas(figHandles(2), [pathName '/' resultsName '2DPDD.fig']) disp('2D correlation spectroscopy analyses results in :  TDCOV, TDCOR, PDDCOV, and PDDCOR') processNum  = processNum + 1; suffixes{processNum} = ''; Message = ['Step ' num2str(processNum) ' - 2D correlation spectroscopy analyses completed. Working...']; Title = 'Processing status'; h = msgbox(Message,Title,'warn','modal'); 96  % input(['Number ' num2str(processNum) ' finished']);     % 12. Statistics (load and track 2 or more groups of spectra?) % close all temp = []; temp(:,1) = useData(:,1); temp(:,2) = mean(useData(:,2:end)'); assignin('base', [generatedvarname 'mean'], temp); figHandles = f_plotRaman(2, [1,1], figName);  figure(figHandles(1)); plot(temp(:,1), temp(:,2),'k','linewidth',2);title([figName 'Mean of spectra'],'FontName','Times','FontSize',24);box on;axis tight; saveas(figHandles(1), [pathName '/' resultsName 'mean.fig']);   temp(:,2) = std(useData(:,2:end)'); assignin('base', [generatedvarname 'std'], temp); figure(figHandles(2)); plot(temp(:,1), temp(:,2),'r','linewidth',2);title([figName 'Standard deviation of spectra'],'FontName','Times','FontSize',24);box on;axis tight; saveas(figHandles(2), [pathName '/' resultsName 'STD.fig']);   disp('Basic statistic calculations not available.') processNum  = processNum + 1; suffixes{processNum} = ''; 97  Message = ['Step ' num2str(processNum) ' - basic statistics not performed. Working...']; Title = 'Processing status'; h = msgbox(Message,Title,'warn','modal'); % input(['Number ' num2str(processNum) ' finished']);     % 13. Clean up and save workspace  for n = 6:-1:1     figure(n) end figure(2) disp(['Processing and analyses completed. Results save in: ' resultsName '.mat']) processNum  = processNum + 1; suffixes{processNum} = ''; Message = 'Processing and analyses completed.'; Title = 'Processing status'; h = msgbox(Message,Title,'warn','modal'); if ~(isempty(warningstrings))     dlgname = 'Alert';     for n = 1:max(size(warningstrings))         warningstrings{n} = [num2str(n) ' .' warningstrings{n}];     end     hw = warndlg(warningstrings, dlgname); 98  end   clear('Diff1_dsp','Max_j','def','dlg_title','figHandles','formatDataLoaded','generatedvarname','iterations','loadBoxTitle','n','normMethod','num_lines'); clear('optionsStr','origData','origStr','pauseTime','peakPosition','polyorder','processNum','prompt','responseMade','subsample','suffix','temp','useData','fig_axes'); clear('Message','Title','h','pauseTime','sumeigs','n','Latest_diff1','Diff1dspSG_','QCgeneratedvarname_first','QCgeneratedvarname_last','Raw_diff1','BGgeneratedvarname','BGname'); clear('specMaxPeaks','specStddevs','noisyData','BGnoisyData','tempgeneratedvarname','BGData_blf','BGuseData','Data_blf','indices','new_warningstrings','pixelwidth','suffix2','figName'); save([pathName '/' resultsName '.mat']); clear('resultsName','hw','dlgname');  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

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"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0300285/manifest

Comment

Related Items