Abstract: Pancreatic Ductal Adenocarcinoma (PDAC) has a very dismal patient prognosis, with average patient survival less than 6 months and a 5-year survival less than 5%. Despite aggressive treatment regiments and persistent research efforts, there has been little improvement in PDAC patient survival in the last 50 years\cite{23856911}. This is largely due to late diagnosis, aggressive metastasis, and chemotherapy resistance \cite{28034553} \cite{27821486}. Recent attention in pancreatic cancer research has turned towards targeting the pancreatic tumor microenvironment to better understand molecular signatures of metastasis and therapeutic response\cite{28477742}\cite{25337831}\cite{21605068}. Here, we employ the use of mass spectrometry imaging (MSI) to allow untargeted spatial and chemical profiling of pancreatic cancer tissues. We combine the chemical versatility of MSI in multimodal imaging systems to better understand the pancreatic tumor microenvironment. We hypothesize that the tumor microenvironment plays an important role in contributing to cancer metastasis and chemoresistance in pancreatic cancer. In aim 1, we combine second harmonic generation imaging to understand metabolite signatures of aligned collagen fibers surrounding the tumor boundary that are believed to promote tumor metastasis. In aim 2, we combine fluorescence lifetime imaging (FLIM), which is used to predict chemotherapy efficacy. to identify metabolite signatures of regions of the tumor or patients that are resistant to chemotherapy regiments. Finally, in aim 3, we expand MSI analysis to clinically-banked formalin fixed paraffin embedded (FFPE) samples by understanding how metabolites are preserved during the formalin fixation and paraffin embedding process. If a metabolite is conserved in both spatial distribution and intensity, metabolites studies can be rapidly expanded to large clinical tissue pancreatic cancer inventories available through the UWCCC-Biobank and Laboratory for Optical and Computational Instrumentation (LOCI) for MSI analysis. Additionally, FFPE conserved metabolites could be analyzed in the future for clinical biomarker, diagnostic analysis, and personalized medicine.
Specific Aims:
Aim 1: Combining Mass Spectrometry Imaging and Second Harmonic Generation Imaging. Here, flash frozen human pancreas cancer biopsies are analyzed from m/z 50-1000 to characterize the distribution of hundreds of metabolites in the tissues. Following analysis, second harmonic generation is used to analyze the collagen fiber alignment in these same tissues. Analyses are overlaid spatially and region of interest discriminate analysis is used to reveal metabolites that can statistically distinguish aligned and unaligned regions of pancreatic cancer tumors. Collagen alignment in pancreatic cancer is a negative prognostic of survival\cite{27776346}. MSI enable us to understanding molecular signatures in these aligned regions that may be contributing to decrease survival rates.
Aim 2: Combining Mass Spectrometric Imaging and Optical Metabolic Imaging. Optical Metabolic Imaging measures the dynamic intrinsic fluorescence of metabolites NADH and FAD, which can be used to calculate a readout of tumor drug response using pancreatic cancer organoid systems. MSI is coupled with optical metabolic imaging to provide a readout of the drug and drug metabolites distribution within the organoid, as well as an untargeted analysis of metabolism biomarkers. This can reveal a more comprehensive analysis of the biological response of combination chemotherapy and can confirm the optical metabolic readouts of drug efficacy.
Aim 3: Expanding Mass Spectrometry Imaging to Formalin Fixed Paraffin Embedded Samples: Histopathological examination of tissues is used to stratify patient specimens and provide a diagnosis to help the patient get the best standard of care. While pathology relies heavily on visual inspection of tissues, MSI can provide an unbiased, analytic analysis of biomarkers in the tissue to supplement a pathologist’s diagnosis. Here, we compare MSI metabolite analysis between formalin fixed paraffin embedded and flash frozen pancreatic cancer spheroids. This analysis will allow us to understand what metabolite biomarker distributions can be conserved through the fixation process. Following this analysis, MSI can be expanded to analyze conserved metabolites in large tissue banks and tissue microarrays for prognostic and diagnostic biomarkers analysis in large patient samples sets.
Significance:
PDAC is the most common type of pancreatic cancer and is considered essentially a death sentence. PDAC remains one of the hardest-to-treat cancers because most patients diagnosed at a late, metastatic stage, with only 20% of patients eligible for surgical resection\cite{25170206}. Additionally, treatment options are often ineffective. Despite aggressive treatment regiments and persistent research efforts, there has been little improvement in the 5-year PDAC patient survival in the last 50 years. Currently, even the best treatment options only prolong life by 6 to 12 weeks\cite{24782785}\cite{23856911}. PDAC chemoresistance and aggressive metastatic rates may be a result of the influence of the tumor microenvironment and increased extracellular matrix (ECM) deposition characteristic of PDAC histology. Therapeutic advancements for PDAC have been negligible, likely due to targeting the cancer cells themselves, and not considering the heterogeneity of the tumor microenvironment\cite{24048067}. Additionally although some genetic and environmental factors\cite{19360152} have been implicated in PDAC carcinogenesis, the exact biological mechanisms underlying progression to metastasis remain unclear. It is believed in the tumor microenvironment, consisting of pancreatic stellate cells and cells involved in immunosuppression, can make up to 90% of the tumor volume in pancreatic cancer\cite{26080604}. New alternative therapies targeting the tumor microenvironment are urgently being developed to improve pancreatic cancer outcomes. Increased tumor complexity and heterogeneity suggests the need for personalized medicine to make therapeutic advances in PDAC. With this proposal to expand molecular understanding of PDAC in and its microenvironment, this is an exciting time as we shift our focus to the tumor–stroma relationship to design new therapies for a currently deadly disease. Thus a need to understand the difference in spatial distributions. Nonetheless, increased complexity and heterogeneity suggests the need for personalized medicine to make therapeutic advances in PDAC and likely other cancers\cite{24172537}. With our exponentially increasing understanding of the molecular basis of PDAC and its microenvironment, we are seeing shifts in focus to the tumor–stroma relationship to design new therapies for a currently deadly disease \cite{26747091}.
Mass Spectrometry Imaging Background: Mass spectrometry imaging (MSI) is a powerful tool that enables untargeted analysis of the spatial distribution of a variety of molecular species. It has the capability to image thousands of molecules in a single experiment without labeling\cite{27322466}. Unlike other techniques used understanding in-tissue distributions, including radio and fluorescent labeling, MSI is capable of imaging the spatial distributions of multiple analytes from a single tissue analysis\cite{23603211}. Additionally, fluorescent and radio labeling of compounds can significantly change tissue distributions, may require extensive synthetic and analytical steps, and labeled molecules are indistinguishable from their corresponding biological metabolites and/or degradation products. MSI was first introduced into biological sciences at the end of the 20th century, and has rapidly expanded today integrating itself into many clinical and pharmaceutical applications\cite{25621874}. Matrix assisted laser desorption and ionization (MALDI) is a common mass spectrometry ionization mechanism used for imaging studies. MALDI uses a laser beam to irradiate a matrix-coated tissue section mounted on a mobile x-y stage, and collects an array of mass spectra at a specified interval, which are then compiled to form an image of the distribution of analyte. When the matrix is applied to the tissue, biological molecules co-crystallize into the matrix, which allows them to be ionized and then detected by mass spectrometer. Following spectra collection, software is used to select an individual mass-to-charge (m/z) value, and the intensity in each pixel. These intensities are combined into a heatmap image depicting the relative distribution of that m/z value throughout the sample. In order to determine the identity of a specific m/z value, tandem MS (MS/MS) fragmentation can be performed on ions from each pixel, the fragments can be used to piece together the structure of the MSI unknown molecule. Otherwise, the molecule can be identified based on its intact mass by accurate mass matching to databases of known molecules within a certain mass error range \cite{26859000}, including the comprehensive human metabolome database. MSI is also expanding into multi-modal imaging systems. Because MSI has high chemical specificity for a variety of biological molecules, while other imaging modalities, such as microscopy, typically have high spatial resolution, but not high chemical specificity, complementary modalities are combined to answer new biological questions. The ability of MSI to image the multiple molecules in a single analysis run makes it indispensable to pharmaceutical and clinical applications. Here, we describe novel multi-modal image pairings that allow us to gain structural and chemical insight into pancreatic cancer biological questions that are otherwise unable to be investigated with a single modality system. The research goal of these studies is to combine the chemical specificity of mass spectrometry imaging on with imaging tools to understand cancer metastasis and chemotherapy response in pancreatic cancer.
Aim 1: Combining Mass Spectrometric Imaging and Second Harmonic Generation Imaging
Objective: Investigate metabolic changes in aligned collagen regions in PDAC tissues
Background: Second harmonic generation (SHG)-based microscopy is a label-free modality used for high resolution, quantitative imaging of collagen in a range of biological tissues\cite{24974046}. Based on the triple helical structure and noncentrosymetric center of collagen fibers, the fibers can emit a SHG signal after being pulsed with a femtosecond laser, with double the frequency and half of the wavelength of the original. Second harmonic generation has been previously used to study collagen changes in various cancer types \cite{26716418,24407500,27767180,27588592}. In breast cancer, collagen organizational patterns have been identified that correlate with patient prognosis, described as “tumor associated collagen signatures” (TACS)\cite{21356373}. TACS-3 describes the organization of aligned fibers perpendicular to the tumor boundary. It is believed that these fibers facilitate cancer cell migration and act as metastatic highways for cancer cells to migrate to surrounding tissues \cite{27663743} . Additionally, these same tumor associated signatures were studied in pancreatic cancer and collagen alignment relative to the tumor boundary in PDAC was found to correlate with poor survival\cite{27776346}. The SHG signal also colocalized with E-cadherin and vimentin, indicating that these collagen fibers correlated with metastatic cells. Here, we further investigate metabolism changes occurring in aligned collagen regions in pancreatic cancer\cite{27776346}. Approach: Nine 1mm biopsies each from nine different patients are aligned in gelatin microarrays and flash frozen on dry ice. Gelatin is used because it easily sectioned and does not produce interfering peaks for mass spectrometry imaging. Twelve micron sections are taken for analysis and matrix is applied using an automatic sprayer system. MSI is acquired on a Thermo MALDI orbitrap XL at 60,000 mass resolution, at 75 micron spatial resolution for m/z range from 50 -1000. Following MSI acquisition, sections are washed with 50:50 methanol:water, and then SHG signal is acquired on the Optical Workstation at LOCI. Following acquisition, tissues are stained using hemotoxlin and eosin, and then pathology annotated to identify tumor boundaries. Collagen fiber, width, length, alignment, and straightness measurements are quantified from the SHG signal\cite{25250186}, and Curve Align is used to generate alignment profile by comparing each collagen fiber to the nearest 16 fibers around the pathology annotated tumor boundary. Visual regions of high alignment fibers and low alignment fibers are generated in Curve Align and displayed over the SHG to create regions of interest, with red demonstrating high alignment and yellow demonstrating low alignment\cite{25250186}. Biomarker receiver operating characteristic (ROC) curves are then generated for both univariate (single m/z) or multivariate (multiple m/z) analysis to determine which m/z or combination of m/z values can best distinguish low and high alignment regions surrounding the tumor boundary. From ROC curve, an area under the curve (AUC) values demonstrates which metabolites (m/z values) serves as the best discriminators between the regions outlined. The closer the AUC value is to 1, the better discrimination power it has\cite{24009950}. Expected Results: Metabolic pathways such as the arginine/proline pathway are typically regulated with collagen synthesis, and likely to be upregulated\cite{25474014}\cite{25917076}. Pancreatic cancer tumor microenvironment cells have been shown to secrete increase alanine levels, which is then used by the cancer cells as a fuel source\cite{27509858}. Also, glutamine as a fuel sources is significantly enhanced in high density collagen regions in breast cancer, and therefore expected to be increased in aligned regions of the tumor\cite{27743905,27213771}\cite{28697344}. Dense extracellular matrix can cause vascular collapse to the tumor, therefore metabolites involved in oxygen consumption is expected to be down regulated in these regions\cite{23299539}. Likely, these metabolites will best discriminate regions of high collagen alignment and low collagen alignment in pancreatic tissues. Alternative Approaches/Pitfalls: It is possible that a specific single m/z value will not be able to provide good distinction (AUC>.8) between low and high aligned collagen regions. Therefore, each m/z will also be statistically investigated with colocalized m/z values to increase the predictive power of the model. This may help provide better discriminative power by using multiple biomarkers to conduct the statistical analysis.
Aim 2: Combining Mass Spectrometry Imaging and Optical Metabolic Imaging
Objective: Investigating metabolic changes in patient treatment response by combining non-invasive optical imaging methods with mass spectrometry imaging using human-derived organoids.
Background: Fluorescence lifetime imaging microscopy (FLIM) can measure the rate of decay of a molecule's fluorescence, where the fluorescence lifetime depends on its protein binding state allowing visualization of the metabolic process that are dominant in the cell. The fluorescence lifetime is the average time a molecule spends in an excited state before returning to the ground state. This is used to measure the lifetime of metabolites NAD(P)H and FAD, which are naturally fluorescent coenzyme involved in metabolic processes including glycolysis and oxidative phosphorylation. The lifetime of a metabolite depends on its local environment, and not on its concentration\cite{18042710}. If the lifetime is long, the NAD(P)H is bound to a protein, while if the lifetime is short, the NAD(P)H is unbound. If NADH is bound, the organoid is more likely to be resistant to a chemotherapy than if NADH is free. FAD is the opposite. If FAD is bound to a protein, it has a short lifetime, while if FAD is free, it has a long lifetime. The ratio of NAD(P)H and FAD is combined into a single optical metabolic index (OMI), which serves as a single biomarker of cellular metabolism. OMI is imaged live, and can give dynamic changes of metabolism with respect to time of treatment. It has been used in pancreatic cancer organoids to determine chemotherapy response (\cite{28588148}\cite{26495796} Approach: Here, OMI is applied to organoids, which are grown from primary patient tissues. Organoids are generated by mechanically separating the tumor into small pieces, which are then embedded in collagen or Matrigel for support and cultured in cell media. Organoid size varies between 50-500 um in diameter. Organoids still maintain organ structure, but also can be treated with multiple chemotherapy agents. OMI provides a robust platform to measure drug response. Additionally, for large organoids, drug delivery is mediated by diffusion, and organoids can reproduce structural features of in vivo tumor characteristic. By combining FLIM with MSI, we can track drug distribution in the organoids, along with drug metabolites, or multiple drugs used in a single treatment. Organoids are grown in Matrigel, and centrifuged out of Matrigel, and grown in formalin linked gelatin microwells, with one organoid/well. Each group of organoids is exposed to drug for 7 days, with FLIM taken at Day 3, and Day 7 for each organoid and also in it supporting fibroblasts. Following FLIM, organoids are immediately flash frozen on dry ice, sectioned, and matrix is applied. Each organoid is analyzed with the MALDI Orbitrap XL follow matrix application. For analysis, FLIM provides a supervised classification method: resistant and nonresistant method based on the OMI index. Although useful that we can predict efficacy, especially in a personalized way, it is also important to understand why these regions are resistant to chemotherapy. Coupling with MSI allows us insight into the metabolic processes that might be changing as a result of drug application. FLIM data is overlaid onto MSI data to select regions of interest. Supervised classification of organoids is used to create a machine learning algorithm, with these datasets serving as training data. The machine learning algorithm will mine the data to find patterns that best distinguish resistant and responsive tumor organoids. This algorithm is then applied to new data. Drugs tested include standard of care (gemcitabine, 5FU) and novel drugs (MLN0128, an mTOR inhibitor; and ABT263, a pro-apoptotic drug). Preliminary data analyzed in patient 1 (Figure 2) indicates response to all treatments except 5FU, on days 3 and 7 in organoids and fibroblasts. While FLIM is a great tool for tracking drug predictability, only resectable PDAC tissue is eligible for use, as FLIM imaging requires live tissue for analysis. Therefore, we hope that mass spectrometry imaging will be able to use this knowledge to identify biomarkers for chemoresistance that can be found from tumor biopsies, prior to chemotheapy application, to help determine correct chemotherapy prior to treatment administration. For example, we hope to identify metabolite differences between gemcitabine resistant and responsive organoids that are present prior to treatment. Biomarker ROC curves will be used to determine how well a a single m/z or group of m/z values to determine values with high discriminate capability\cite{24009950}. Additionally, supervised machine learning algorithms will be applied to new unclassified samples. By applying machine learning algorithms, we can use mass spectrometry imaging to expand clinical samples eligibleeligi for metabolite analysis. Expected Results: Resistance to gemcitabine is characterized metabolically by decreased levels of glutamine and proline, and increased levels of creatine, hydroxyproline, and creatinine \cite{24518513}. We expected that metabolites involved in these pathways will be associated with resistance. Additionally, the Tryptophan -Kyneurenine pathway has been used as a pharmacodynamical output for drug response\cite{26751220}. Additionally, the concentrations of NAD(P)H and FAD may also be relevant biomarkers to distinguish resistant and responsive tumor regions. output. However, metabolites associated with new combination therapies have not been well characterized and therefore untargeted analysis anal using mass spectrometry imaging allows new biomarkers to be discovered. Alternative Approaches/Pitfalls: In order to successfully co-register FLIM and MSI, co-registration of the images is very important. While organoids are typically grown in Matrigel, some of the components of Matrigel contain possible metabolites that can cause interference between organoid signal and background signal. Current optimization of collagen or gelatin microwells are being investigated that will not change the FLIM readouts to streamline the co-registration process and for optimal MSI readouts.
Aim 3: Expanding Mass Spectrometry Imaging to Formalin Fixed Paraffin Embedded Samples
Objective: Distinguish metabolites whose distribution and intensity is not modified by the fixation and paraffin embedding processes for MSI Hypothesis: The fixation and paraffin embedding process can affect the intensity and possible spatial distribution of some metabolites in clinically banked samples. Our goal is to identify conserved metabolites that can provide relevant biological information from PDAC FFPE tissues and tissue microarrays to expand the patient numbers needed for metabolite analysis for MSI clinical research. Background: Because clinical tissue banks can store FFPE at room temperature for years, clinical repositories for a variety of diseases exist in the form of FFPE tissues, including biopsies, surgical resection specimens and tissue microarrays. Biopsies are particularly important because they are highly relevant in terms of determining tumor susceptibility to drug for clinical decision-making and tissue microarrays (TMAs) are blocks from different sources. FFPE tissues remain an untapped resource for mass spectrometry imaging due to the belief that small molecules were lost during fixation and processing steps\cite{27414759}. Commercial formalin is a 2-phase fixative, with an initial alcohol fixation phase, followed by a cross-linking phase. Formaldehyde is the crosslinking agent that crosslinks primary amino groups with other nearby nitrogen atoms in protein or DNA. Once fixed, the tissue is washed with varying concentrations of water and ethanol to dehydrate the sample, and then the tissue specimen is embedded in a paraffin wax for long term storage\cite{23248474}. Despite this belief that these washes would get rid of small molecules, some histochemical stains, such as the periodic acid-Schiff for polysaccharides, demonstrate that more than just proteins can be retained even through FFPE tissue processing. In 2015, metabolites from FFPE tissues were first analyzed by gas chromatography coupled with mass spectrometry, where 60 metabolites were identified from FFPE, and 59 of those 60 metabolites were also identified in flash-frozen. This countered the idea that these metabolites were lost during fixations\cite{26348873}. Additionally, in 2015, researchers investigated groups and metabolite properties that allowed the metabolites to survive the fixation and paraffin embedding process and how the concentration of the metabolite changed based on the sample preparation. They identified 80 metabolites, and that about 50% of all species can be detected in FFPE tissue, compared with flash frozen tissue. Commonly retained metabolites include lipids, amino acids, and carbohydrates. Peptides, and energy metabolites were not well conserved for analysis. Additionally, most metabolites showed different expression levels between different prostate cancer patients, as well as differences between sample preparation \cite{26415588}. In 2016, MSI was first applied to formalin fixed paraffin embedded samples, with successful detection of metabolite distribution in clinical tissues. While this accomplishment is huge for the field of mass spectrometry imaging, there has not been characterization of how distributions change based on the FFPE tissue embedding process \cite{28523495}. Here, we developed a method to analyze how to identify metabolites that can be conserved, spatially and intensity, during this process. In order to analyze this group of biomolecules in biologically relevant context, it is important to know that metabolite distribution and intensity is not modified by the fixation and paraffin embedding processes. While preliminary results demonstrate that a range of metabolite can indeed be detected in formalin-fixed paraffin embedded tissues (See Figure 3), it is important for us to explore distributions and intensities that while detected, are not conserved. Approach: To make sure that FFPE tissues and Flash frozen tissues are identical, we will not be using clinical tissues for optimization. Although clinical tissues give us great insight into metabolites conserved, tumor heterogeneity and intratumor heterogeneity add another complexing factor to complicated metabolite analysis. We use a microarray of spheroids (81 identical spheroids/ microarray) to conduct this analysis. Additionally, we use in vitro pancreatic cancer samples to compare sample preparations with 81 biological replicates per a single analysis. This sample, while not clinical tissue, is ideal for conservation analysis, as the sample can be reproducibly grown using all explored sample preparation mechanisms, and therefore sample to sample variability is not an issue. Additionally, 81 biological replicates/condition/analysis ensures that the signals we see are not due to a change error. Experimental variability between spheroids will be evaluated for each analysis, and then between analyses from different samples preparation. Using our novel microarray spheroid platform, where we grow 81 spheroids, each 800 micron in diameter in a crosslinked-gelatin hydrogel, spheroids can be grown and analyzed in the same array. Three arrays will be formalin fixed, while three arrays of spheroids will be flash frozen. Both samples will be sectioned into 10 micron sections – with matched tissue depth, covered with matrix and analyzed using mass spectrometry imaging. Each m/z intensity and spatial distribution is expressed in a heatmap. Following analysis, colocalization analysis will be used to evaluate the intensity and spatial conservation. To evaluate how similar the spatial distributions are between FFPE and Flash Frozen, we will use a Manders’ split coefficient analysis. The Manders’ split coefficient is a value between 0 to 1, which provides information on the colocalization of two channels. While very sensitive, it is important to note that this does not provide information on the relative intensities in each pixel. This tells us that the same pixels from two images have a value above 0, and is used to tell us how well the spatial distributions of the analyte overlap. For intensity analysis, Pearson’s Correlation Coefficient (r) is used to measure the correlation between intensities of each sample preparation for each pixel. The equation for analysis is r = where m1= intensity of m/z in FFPE, ma= mean intensity of M/z in FFPE; n1=intensity of m/z in FF, na= mean intensity of m/z in FF. The scale used for analysis uses a scale from -1 to 1, where 1 is perfect correlation, 0 is no correlation, and -1 is anti-correlation\cite{20653013}. For comprehensive statistical analysis, the Costes analysis will tell us if the Pearson Correlation Coefficient and the Manders’ analysis coefficients are better than pure chance or not. This is done by shuffling the pixels in one of the images, and then reperforming the Manders and Pearson Correlation Analysis\cite{15189895} \cite{24117417}. A P-value of 1.00 means that none of the randomized images had better correlation. Any m/z values that do not meet this requirement will be discarded as noise. Following analysis, each m/z value will have a value for both spatial and intensity for both FFPE and FF conditions. Expected Results: We believe that lipids, amino acids, and carbohydrates will also remain conserved, as in mass spectrometry studies previously done\cite{26415588}. Criteria for conservation of a m/z value for mass spectrometry: Pearson’s R value greater than or equal to .5, a Manders coefficient greater than or equal to .85, and a Costes p-value of 1.0. Following characterization in spheroid analysis, this study will be used to generate a conserved list of metabolites to perform biomarker analysis using FFPE tissue microarrays. Mass spectrometry imaging may be able to provide new metabolite biomarkers that are able to distinguish normal tissue, chronic pancreatitis, and PDAC grade. Tissue microarrays will be similarly prepared, and receiver operator curve analysis and pathology annotation will be able to calculate biomarkers from FFPE tissues. By creating a list of conserved metabolites, the mass spectrometry imaging community and cancer research community can further probe many different FFPE for metabolite markers of chemoresistance and metastatic markers. Pitfalls or alternative methods: Matrix application and mass spectrometry parameters may need to be optimized to successfully extract various classes of metabolites for analysis.