Stratification of chemotherapy-treated stage III colorectal cancer patients using multiplexed imaging and single-cell analysis of T-cell populations

Colorectal cancer (CRC) has one of the highest cancer incidences and mortality rates. In stage III, postoperative chemotherapy benefits <20% of patients, while more than 50% will develop distant metastases. Biomarkers for identification of patients at increased risk of disease recurrence following adjuvant chemotherapy are currently lacking. In this study, we assessed immune signatures in the tumor and tumor microenvironment (TME) using an in situ multiplexed immunofluorescence imaging and single-cell analysis technology (Cell DIVETM) and evaluated their correlations with patient outcomes. Tissue microarrays (TMAs) with up to three 1 mm diameter cores per patient were prepared from 117 stage III CRC patients treated with adjuvant fluoropyrimidine/oxaliplatin (FOLFOX) chemotherapy. Single sections underwent multiplexed immunofluorescence staining for immune cell markers (CD45, CD3, CD4, CD8, FOXP3, PD1) and tumor/cell segmentation markers (DAPI, pan-cytokeratin, AE1, NaKATPase, and S6). We used annotations and a probabilistic classification algorithm to build statistical models of immune cell types. Images were also qualitatively assessed independently by a Pathologist as 'high', 'moderate' or 'low', for stromal and total immune cell content. Excellent agreement was found between manual assessment and total automated scores (p < 0.0001). Moreover, compared to single markers, a multi-marker classification of regulatory T cells (Tregs: CD3+/CD4+FOXP3+/PD1-) was significantly associated with disease-free survival (DFS) and overall survival (OS) (p = 0.049 and 0.032) of FOLFOX-treated patients. Our results also showed that PD1- Tregs rather than PD1+ Tregs were associated with improved survival. These findings were supported by results from an independent FOLFOX-treated cohort of 191 stage III CRC patients, where higher PD1- Tregs were associated with an increase overall survival (p = 0.015) for CD3+/CD4+/FOXP3+/PD1-. Overall, compared to single markers, multi-marker classification provided more accurate quantitation of immune cell types with stronger correlations with outcomes.

An atlas of inter- and intra-tumor heterogeneity of apoptosis competency in colorectal cancer tissue at single-cell resolution

Cancer cells’ ability to inhibit apoptosis is key to malignant transformation and limits response to therapy. Here, we performed multiplexed immunofluorescence analysis on tissue microarrays with 373 cores from 168 patients, segmentation of 2.4 million individual cells, and quantification of 18 cell lineage and apoptosis proteins. We identified an enrichment for BCL2 in immune, and BAK, SMAC, and XIAP in cancer cells. Ordinary differential equation-based modeling of apoptosis sensitivity at single-cell resolution was conducted and an atlas of inter- and intra-tumor heterogeneity in apoptosis susceptibility generated. Systems modeling at single-cell resolution identified an enhanced sensitivity of cancer cells to mitochondrial permeabilization and executioner caspase activation compared to immune and stromal cells, but showed significant inter- and intra-tumor heterogeneity.

Emergent Languages from Pretrained Embeddings Characterize Latent Concepts in Dynamic Imagery

Recent unsupervised learning approaches have explored the feasibility of semantic analysis and interpretation of imagery using Emergent Language (EL) models. As EL requires some form of numerical embedding as input, it remains unclear which type is required in order for the EL to properly capture key semantic concepts associated with a given domain. In this paper, we compare unsupervised and supervised approaches for generating embeddings across two experiments. In Experiment 1, data are produced using a single-agent simulator. In each episode, a goal-driven agent attempts to accomplish a number of tasks in a synthetic cityscape environment which includes houses, banks, theaters and restaurants. In Experiment 2, a comparatively smaller dataset is produced where one or more objects demonstrate various types of physical motion in a 3D simulator environment. We investigate whether EL models generated from embeddings of raw pixel data produce expressions that capture key latent concepts (i.e. an agent’s motivations or physical motion types) in each environment. Our initial experiments show that the supervised learning approaches yield embeddings and EL descriptions that capture meaningful concepts from raw pixel inputs. Alternatively, embeddings from an unsupervised learning approach result in greater ambiguity with respect to latent concepts.

Abstract 2676: Hyperplexed immunofluorescence analysis (Cell DIVETM) of immune-related tumor heterogeneity in stage III colorectal cancer

Colorectal cancer (CRC) has one of the highest Worldwide incidences and mortality rates. Genotoxic chemotherapy following surgery in stage III patients confers treatment benefit to less than 20% of the patients, with more than 50% of stage III patients going on to develop distant metastases. Currently, there are no predictive biomarkers that can identify which stage III patients will recur, which patients will benefit from chemotherapy and which should be redirected towards alternative therapeutic interventions. A major challenge in identifying such a universal biomarker is that CRC is a heterogeneous disease with multiple subtypes. In the current study, we assessed clinically-relevant immune cell populations in the tumour microenvironment (TME) of stage III tumours using a novel hyperplex in situ immunofluorescence imaging technology (Cell DIVETM, GE Healthcare, Issaquah, WA).

Tissue microarrays (TMAs) with up to three 1mm diameter cores per patient were prepared from 139 stage III CRC patients treated with adjuvant FOLFOX chemotherapy. Single sections (5 µm) were iteratively stained with Cy3- and Cy5-conjugated antibodies for immune cell markers as well as markers of cell death and metabolism. The images underwent illumination correction, DAPI-based registration and autofluorescence removal. After image quality control corrections, single cell segmentation was performed using a combination of DAPI [nuclear], pan-cytokeratin [epithelial], NaKATPase [membrane] and S6 [cytoplasmic] segmentation markers and an average of ~3,000 stromal cells and ~ 4,000 epithelial were segmented per tumour core.

A machine learning-based algorithm for immune cell classification and quantification was used to analyse the immune markers CD45, CD3, CD4, CD8, FOXP3 and PD1 to identify: cytotoxic T cells, T helper cells, regulatory T cells and potential relevance of immune checkpoint therapy. In the tumour tissues, the median proportion of CD3+ segmented cells was ~8%. Classified immune cells were counted within epithelial and stromal regions, with patients categorised as Low, Intermediate and High (based on &lt;25th, 25th - 75th and &gt;75th percentile, respectively) for each cell type. Preliminary survival analyses show that patients with ‘CD8 High' intratumoural cytotoxic T cells have better Disease-Free Survival compared to ‘CD8 Low' patients in this FOLFOX-treated cohort. By combining single-cell data with clinicopathological patient data, we aim to identify immune-, cell death- and metabolism-related signatures that can predict benefit from adjuvant FOLFOX chemotherapy for Stage III CRC patients.

Citation Format: Xanthi Stachtea, Andreas Lindner, Manuela Salvucci, Sanghee Cho, Anup Sood, Elizabeth McDonough, Alberto Santamaria-Pang, John Graf, Philip Dunne, Mark Lawler, Jochen Prehn, Fiona Ginty, Daniel Longley. Hyperplexed immunofluorescence analysis (Cell DIVETM) of immune-related tumor heterogeneity in stage III colorectal cancer [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 2676.

Quantitative patterns of motor cortex proteinopathy across ALS genotypes

Degeneration of the primary motor cortex is a defining feature of amyotrophic lateral sclerosis (ALS), which is associated with the accumulation of microscopic protein aggregates in neurons and glia. However, little is known about the quantitative burden and pattern of motor cortex proteinopathies across ALS genotypes. We combined quantitative digital image analysis with multi-level generalized linear modelling in an independent cohort of 82 ALS cases to explore the relationship between genotype, total proteinopathy load and cellular vulnerability to aggregate formation. Primary motor cortex phosphorylated (p)TDP-43 burden and microglial activation were more severe in sporadic ALS-TDP disease than C9-ALS. Oligodendroglial pTDP-43 pathology was a defining feature of ALS-TDP in sporadic ALS, C9-ALS and ALS with OPTN, HNRNPA1 or TARDBP mutations. ALS-FUS and ALS-SOD1 showed less cortical proteinopathy in relation to spinal cord pathology than ALS-TDP, where pathology was more evenly spread across the motor cortex-spinal cord axis. Neuronal pTDP-43 aggregates were rare in GAD67+ and Parvalbumin+ inhibitory interneurons, consistent with predominant accumulation in excitatory neurons. Finally, we show that cortical microglia, but not astrocytes, contain pTDP-43. Our findings suggest divergent quantitative, genotype-specific vulnerability of the ALS primary motor cortex to proteinopathies, which may have implications for our understanding of disease pathogenesis and the development of genotype-specific therapies.

Understanding heterogeneous tumor microenvironment in metastatic melanoma

A systemic analysis of the tumor-immune interactions within the heterogeneous tumor microenvironment is of particular importance for understanding the antitumor immune response. We used multiplexed immunofluorescence to elucidate cellular spatial interactions and T-cell infiltrations in metastatic melanoma tumor microenvironment. We developed two novel computational approaches that enable infiltration clustering and single cell analysis-cell aggregate algorithm and cell neighborhood analysis algorithm-to reveal and to compare the spatial distribution of various immune cells relative to tumor cell in sub-anatomic tumor microenvironment areas. We showed that the heterogeneous tumor human leukocyte antigen-1 expressions differently affect the magnitude of cytotoxic T-cell infiltration and the distributions of CD20+ B cells and CD4+FOXP3+ regulatory T cells within and outside of T-cell infiltrated tumor areas. In a cohort of 166 stage III melanoma samples, high tumor human leukocyte antigen-1 expression is required but not sufficient for high T-cell infiltration, with significantly improved overall survival. Our results demonstrate that tumor cells with heterogeneous properties are associated with differential but predictable distributions of immune cells within heterogeneous tumor microenvironment with various biological features and impacts on clinical outcomes. It establishes tools necessary for systematic analysis of the tumor microenvironment, allowing the elucidation of the "homogeneous patterns" within the heterogeneous tumor microenvironment.

Stromal-Based Signatures for the Classification of Gastric Cancer

Treatment of metastatic gastric cancer typically involves chemotherapy and monoclonal antibodies targeting HER2 (ERBB2) and VEGFR2 (KDR). However, reliable methods to identify patients who would benefit most from a combination of treatment modalities targeting the tumor stroma, including new immunotherapy approaches, are still lacking. Therefore, we integrated a mouse model of stromal activation and gastric cancer genomic information to identify gene expression signatures that may inform treatment strategies. We generated a mouse model in which VEGF-A is expressed via adenovirus, enabling a stromal response marked by immune infiltration and angiogenesis at the injection site, and identified distinct stromal gene expression signatures. With these data, we designed multiplexed IHC assays that were applied to human primary gastric tumors and classified each tumor to a dominant stromal phenotype representative of the vascular and immune diversity found in gastric cancer. We also refined the stromal gene signatures and explored their relation to the dominant patient phenotypes identified by recent large-scale studies of gastric cancer genomics (The Cancer Genome Atlas and Asian Cancer Research Group), revealing four distinct stromal phenotypes. Collectively, these findings suggest that a genomics-based systems approach focused on the tumor stroma can be used to discover putative predictive biomarkers of treatment response, especially to antiangiogenesis agents and immunotherapy, thus offering an opportunity to improve patient stratification. Cancer Res; 76(9); 2573-86. ©2016 AACR.

Optimized multiplex immunofluorescence single-cell analysis reveals tuft cell heterogeneity

Intestinal tuft cells are a rare, poorly understood cell type recently shown to be a critical mediator of type 2 immune response to helminth infection. Here, we present advances in segmentation algorithms and analytical tools for multiplex immunofluorescence (MxIF), a platform that enables iterative staining of over 60 antibodies on a single tissue section. These refinements have enabled a comprehensive analysis of tuft cell number, distribution, and protein expression profiles as a function of anatomical location and physiological perturbations. Based solely on DCLK1 immunoreactivity, tuft cell numbers were similar throughout the mouse small intestine and colon. However, multiple subsets of tuft cells were uncovered when protein coexpression signatures were examined, including two new intestinal tuft cell markers, Hopx and EGFR phosphotyrosine 1068. Furthermore, we identified dynamic changes in tuft cell number, composition, and protein expression associated with fasting and refeeding and after introduction of microbiota to germ-free mice. These studies provide a foundational framework for future studies of intestinal tuft cell regulation and demonstrate the utility of our improved MxIF computational methods and workflow for understanding cellular heterogeneity in complex tissues in normal and disease states.

Abstract A71: Tumor HLA Class I expression influences immune cells infiltration in metastatic melanoma tumor microenvironment

The variable clinical success with immune checkpoint inhibitors in the treatment of solid tumors challenges us to further understand the regulatory mechanisms of immune cells infiltration – a critical step of tumor elimination – in a highly heterogenous tumor microenvironment (TME). Systemic analyses of the tumor-immune cell interface and interactions are of particular significance with direct clinical impact. We employed a multiplexed immunofluorescence (MxIF) method, a novel technology that allows in situ quantitative single cell characterization with high order protein multiplexing on whole tumor sections, to elucidate the key components in TME governing the immune cells' composition and function in metastatic melanoma lymph node metastases. Here, we show that the expression of tumor cell Human Leukocyte Antigen 1 (HLA-1) is highly heterogeneous both within the TME and between patients. The level of tumor HLA-1 expression directly correlates with the magnitude of cytotoxic T lymphocytes (CTLs) infiltration in the TME. In addition, decreased tumor HLA-1 expression is associated with accumulation of CD20+ cells at the edge of the tumor as well as an increased population of CD4+FOXP3+ cells. The association between tumor HLA-1 expression and immune cell distribution is highly consistent yet heterogenous across the entire tumor mass. Moreover, using quantitative single cell data, we developed algorithms for computational and spatial modeling of tumor-immune cell interfaces, allowing statistical comparison of cellular interactions in the context of tumor heterogeneity. Spatial analysis demonstrates that tumor HLA-1 expression favors CTL invasion, while lack of tumor HLA-1 expression results in CTL evasion. Furthermore, in a cohort of 166 Stage III melanoma patients, we confirm that CTLs infiltration is only present in regions with high tumor HLA-1 expressions, and that patients with tumors high in both tumor HLA-1 and CTLs have significantly improved progression free survival (PFS) compared with those with tumors low in HLA-1 and/or CTLs. Our study demonstrates that the heterogeneous tumor HLA-1 expression results in various immune cell distribution patterns in the TME, directly contributing to the varied antitumor immunities and ultimately influencing tumor outcomes. As part of the antigen presentation machinery, HLA-1 expression on tumors is necessary for CTLs infiltration and is indispensable for T cell mediated tumor elimination with direct clinical benefits. We also establish a novel platform for visualization and spatial representation of the cellular heterogeneity within the TME, providing data processing and modeling algorithms necessary for understanding the cellular interplay mediating tumor surveillance, which will ultimately improve therapeutic efficacy and clinical outcomes.

Citation Format: Yiyi Yan, Alexey Leontovich, Thomas Flotte, Michael Gerdes, Fiona Ginty, Alberto Santamaria-Pang, Anup Sood, Keyur Desai, Chrystal Chadwick, Rong Zhang, Svetomir N. Markovic. Tumor HLA Class I expression influences immune cells infiltration in metastatic melanoma tumor microenvironment. [abstract]. In: Proceedings of the AACR Special Conference on Tumor Immunology and Immunotherapy; 2016 Oct 20-23; Boston, MA. Philadelphia (PA): AACR; Cancer Immunol Res 2017;5(3 Suppl):Abstract nr A71.

Abstract P1-06-02: Impact of heterogeneity of DCIS on immune cell infiltrations

Background: Ductal carcinoma in situ (DCIS) accounts for at least 20% of breast cancers. Factors associated with recurrence of DCIS or progression to invasive carcinoma are not well delineated. The goals of the current study were to profile the epithelial and immune cells using the MultiOmyx hyperplexed immuno-fluorescent based analyses. This was coupled with semi-automated algorithms to characterize the inter-relationships between cell populations within individual DCIS lesions.

Patients and Methods: Analysis for 15 antibody markers (EGFR, Her2, Her4, S6, pMTOR, PCAD, CD44v6, NaKATPase, SLC7A5, CD4, CD8, CD20, CD68, and CD10) was performed on a single FFPE section containing 10-20 distinct ducts from 13 cases of DCIS. Briefly, approximately 40 fields of view (FOV) from digitized sections containing DCIS or normal tissue were sequentially (cyclically) stained for the 15 markers. Each cycle entailed staining with 2-3 markers followed by imaging, dye inactivation, and re-staining. DAPI was used for nuclear demarcation and for registration of the images, while S6, pan-cadherin, Na+K+ATPase and pan-cytokeratin were used for epithelial segmentation. K-means clustering was used to determine patterns of co-expression of markers at the single cell, duct, and patient levels. These clusters were then correlated with immune marker expression by tumor infiltrating lymphocytes (TILs) by marker type (CD4, CD8, and CD20) and tumor compartment (stromal versus intraepithelial).

Results: Analysis of the epithelial component in each of 13 cases of DCIS (n= 415 ducts) revealed 8 distinct expression patterns (clusters) using a panel of 7 markers (EGFR, Her2, Her4, pmTOR, CD44v6, SLC7A5, and CD10). The frequency and distribution of clusters, annotated at the single cell level, showed that 4 DCIS's were dominated (&gt;80%) by a single cell phenotype represented by cluster groups 3 and 7 (high Her2), cluster 6 (High Her4 and SLC7A5 and low Her2), or cluster 4 (non-descript). In 5 pts, the pattern was more heterogeneous consisting of mixture of cell populations with 50-70% of the cells belonging to cluster 1 (moderate to high levels for all markers except EGFR and CD10). The remaining pts had a strong representation of cluster 4 and 5 (CD44v6 and phospho-mTOR) cells. The distribution of both intra-epithelial and stromal TILs in DCIS cases were either consisted of a mixed B-cell (CD20+) and T-cell response (n=4), or one dominated by T-cells. Cluster 2 (High EGFR and CD10) was associated with a largely T-cell response (rs = 0.83, P value = 0.0004), while Cluster 7 (strong HER2) was associated with a B-cell response (rs = 0.68, P value = is 0.009).

Conclusions: Analysis 15 markers and use of K-means clustering algorithm, shows prominent inter-tumoral (but not intra-tumoral) heterogeneity in DCIS. Furthermore, epithelial cell specific clusters (high HER2 or EGFR) were associated with distinct B or T cell infiltration by TILs. Additional ongoing studies will determine the clinical significance of the clusters with respect to recurrence of DCIS and development of invasive carcinomas.

Citation Format: Badve S, Gökmen-Polar Y, Harris AL, Sui Y, Sevinsky C, Santamaria-Pang A, Ginty F, Tan PH, Gerdes MJ. Impact of heterogeneity of DCIS on immune cell infiltrations [abstract]. In: Proceedings of the 2016 San Antonio Breast Cancer Symposium; 2016 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2017;77(4 Suppl):Abstract nr P1-06-02.

Multiplexed immunofluorescence delineates proteomic cancer cell states associated with metabolism

The phenotypic diversity of cancer results from genetic and nongenetic factors. Most studies of cancer heterogeneity have focused on DNA alterations, as technologies for proteomic measurements in clinical specimen are currently less advanced. Here, we used a multiplexed immunofluorescence staining platform to measure the expression of 27 proteins at the single-cell level in formalin-fixed and paraffin-embedded samples from treatment-naive stage II/III human breast cancer. Unsupervised clustering of protein expression data from 638,577 tumor cells in 26 breast cancers identified 8 clusters of protein coexpression. In about one-third of breast cancers, over 95% of all neoplastic cells expressed a single protein coexpression cluster. The remaining tumors harbored tumor cells representing multiple protein coexpression clusters, either in a regional distribution or intermingled throughout the tumor. Tumor uptake of the radiotracer 18F-fluorodeoxyglucose was associated with protein expression clusters characterized by hormone receptor loss, PTEN alteration, and HER2 gene amplification. Our study demonstrates an approach to generate cellular heterogeneity metrics in routinely collected solid tumor specimens and integrate them with in vivo cancer phenotypes.

Abstract 1709: Quantification of biologically relevant vascular phenotypes in human prostate cancer: automated image analysis using hyperplexed immunofluorescence

Although anti-angiogenic therapy has emerged as a leading modality in treating human cancer, further improvements in the duration and frequency of clinical response of various human cancers remain important clinical needs. Maturity of the tumor vasculature has been identified as one of the major determinants of response of cancers to anti-angiogenic treatments. We have developed and optimized the application of a quantitative, high throughput, hyper-plexed, fluorescence imaging technology to refine our understanding of the complexity of vascular phenotypes in various human malignancies. The technology was used to quantify tumor blood vessels and expression of thirteen proteins of known roles in blood vessel biology in single sections from archival primary tumor tissues from 64 prostate cancer patients. CD31 was used to segment vascular objects in each image. CD31 and CD34 endothelial cell staining, SMA pericyte staining, and collagen IV basement membrane staining were used to classify detected vessels using K-means cluster analysis. Segmented vessels were clustered into 2-20 cluster sets, and the reproducibility of vessel classification of each cluster set was determined using the consensus clustering algorithm. A six cluster set that reflected biologically relevant tumor vascular subsets with high consensus clustering concordance was selected for further analysis. Clusters consistent with different stages of vessel development were obtained, including clusters with CD34 high/SMA low and CD34 low/SMA high profiles, reflecting immature and mature vascular phenotypes. Additional clusters representing phenotypes consistent with transitional vascular developmental stages were also identified. Nine additional proteins involved in angiogenesis were also quantified in each vessel and expression profiles for each cluster were determined. The enrichment of blood vessel clusters was then analyzed for each patient. This revealed differential patterns of vascular maturity phenotypes in the prostate cancer tissues analyzed. We have demonstrated that high-throughput, quantitative characterization of vascular maturity phenotypes is feasible in human cancer tissue specimens. Such immunofluorescence hyper-plex profiling of vascular-related proteins has the potential to illuminate the complex biology of tumor angiogenesis and to enable novel approaches for patient tailoring in clinical trials of anti-angiogenic therapeutics.

Citation Format: Chris Sevinsky, Alberto Santamaria-Pang, Jingyu Zhang, Christina Lowes, Dipen Sangurdekar, Beverly Falcon, Qing Li, Bronek Pytowski, Laura Benjamin, Jeremy Graff, Fiona Ginty, Aejaz Nasir, Mark T. Uhlik. Quantification of biologically relevant vascular phenotypes in human prostate cancer: automated image analysis using hyperplexed immunofluorescence. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 1709. doi:10.1158/1538-7445.AM2015-1709

A novel, automated technology for multiplex biomarker imaging and application to breast cancer

AIMS

Multiplexed immunofluorescence is a powerful tool for validating multigene assays and understanding the complex interplay of proteins implicated in breast cancer within a morphological context. We describe a novel technology for imaging an extended panel of biomarkers on a single, formalin-fixed paraffin-embedded breast sample and evaluating biomarker interaction at a single-cell level, and demonstrate proof-of-concept on a small set of breast tumours, including those which co-express hormone receptors with Her2/neu and Ki-67.

METHODS AND RESULTS

Using a microfluidic flow cell, reagent exchange was automated and consisted of serial rounds of staining with dye-conjugated antibodies, imaging and chemical deactivation. A two-step antigen retrieval process was developed to satisfy all epitopes simultaneously, and key parameters were optimized. The imaging sequence was applied to seven breast tumours, and compared with conventional immunohistochemistry. Single-cell correlation analysis was performed with automated image processing.

CONCLUSIONS

We have described a novel platform for evaluating biomarker co-localization. Expression in multiplexed images is consistent with conventional immunohistochemistry. Automation reduces inconsistencies in staining and positional shifts, while the fluorescent dye cycling approach dramatically expands the number of biomarkers which can be visualized and quantified on a single tissue section.

Characterization of glioblastoma (GBM) vasculature and protein expression of surrounding tumor cells on single FFPE sections with a multicycle multiplexed in situ immunofluorescent staining technology

2097

Background: GBM is the most common brain tumor in humans and has a dismal prognosis. Although antiangiogenic therapy (bevacizumab) is an option for GBM, there is still unmet need to understand tumor pathophysiology and predictive biomarkers. We built a tissue based multiplexed immunofluorescent assays and developed algorithms to identify and quantify tumor vasculature, that enabled quantification, visualization, and colocalization of multiple protein in surrounding tumor cells at single cell and subcellular levels. This assay provides unique opportunity to explore tumor heterogeneity of tissue morphology and their relationships to vasculature, and is a novel tool for biomarker and treatment discovery. Methods: Tissue micro arrays (TMAs) were constructed from 141 GBM patients. Fluorescent dye labeled antibodies against 18 biomarkers were sequentially applied on single sections of these TMAs. Metrics were built to identify vessels, quantify distance of tumor cells to vessels, and analyze expression profiles of biomarkers, including signaling molecules in EGFR, PI3K/AKT, TGF-beta pathways, and hypoxia marker Glut1, in proximity to blood vessels. Results: CD31 was successfully used to identify blood vessels in GMB. Vessel segmentation and quantification were performed on all of the images. Biomarker profiling in the context of blood vessels demonstrated different patterns in close proximity to vessels, with some biomarkers showing increased levels (e.g. SMA, EGFR, pS6), some showing decreased levels (e.g. p4EBP), and others remain the same (FOXO3a, S6). Quantification of biomarkers showed heterogeneous expression within the same sample and across the cohort. In addition, co-localization of the above markers was visualized and demonstrated on single cell and subcellular levels. Conclusions: We were able to use a novel fluorescent multiplexing technology (MultiOmyx) to study GBM biology. This technology allowed the simultaneous analyses of multiple biomarkers of GBM, and provides new insights on the relationship of markers to each other, tumor heterogeneity and angiogenesis.

Quantitative single cell analysis of cell population dynamics during submandibular salivary gland development and differentiation

Epithelial organ morphogenesis involves reciprocal interactions between epithelial and mesenchymal cell types to balance progenitor cell retention and expansion with cell differentiation for evolution of tissue architecture. Underlying submandibular salivary gland branching morphogenesis is the regulated proliferation and differentiation of perhaps several progenitor cell populations, which have not been characterized throughout development, and yet are critical for understanding organ development, regeneration, and disease. Here we applied a serial multiplexed fluorescent immunohistochemistry technology to map the progressive refinement of the epithelial and mesenchymal cell populations throughout development from embryonic day 14 through postnatal day 20. Using computational single cell analysis methods, we simultaneously mapped the evolving temporal and spatial location of epithelial cells expressing subsets of differentiation and progenitor markers throughout salivary gland development. We mapped epithelial cell differentiation markers, including aquaporin 5, PSP, SABPA, and mucin 10 (acinar cells); cytokeratin 7 (ductal cells); and smooth muscle α-actin (myoepithelial cells) and epithelial progenitor cell markers, cytokeratin 5 and c-kit. We used pairwise correlation and visual mapping of the cells in multiplexed images to quantify the number of single- and double-positive cells expressing these differentiation and progenitor markers at each developmental stage. We identified smooth muscle α-actin as a putative early myoepithelial progenitor marker that is expressed in cytokeratin 5-negative cells. Additionally, our results reveal dynamic expansion and redistributions of c-kit- and K5-positive progenitor cell populations throughout development and in postnatal glands. The data suggest that there are temporally and spatially discreet progenitor populations that contribute to salivary gland development and homeostasis.

Frames-Based Denoising in 3D Confocal Microscopy Imaging

In this paper, we propose a novel denoising method for 3D confocal microscopy data based on robust edge detection. Our approach relies on the construction of a non-separable frame system in 3D that incorporates the Sobel operator in dual spatial directions. This multidirectional set of digital filters is capable of robustly detecting edge information by ensemble thresholding of the filtered data. We demonstrate the application of our method to both synthetic and real confocal microscopy data by comparing it to denoising methods based on separable 3D wavelets and 3D median filtering, and report very encouraging results.