Christopher Sevinsky

Highly multiplexed single-cell analysis of formalin-fixed, paraffin-embedded cancer tissue

Limitations on the number of unique protein and DNA molecules that can be characterized microscopically in a single tissue specimen impede advances in understanding the biological basis of health and disease. Here we present a multiplexed fluorescence microscopy method (MxIF) for quantitative, single-cell, and subcellular characterization of multiple analytes in formalin-fixed paraffin-embedded tissue. Chemical inactivation of fluorescent dyes after each image acquisition round allows reuse of common dyes in iterative staining and imaging cycles. The mild inactivation chemistry is compatible with total and phosphoprotein detection, as well as DNA FISH. Accurate computational registration of sequential images is achieved by aligning nuclear counterstain-derived fiducial points. Individual cells, plasma membrane, cytoplasm, nucleus, tumor, and stromal regions are segmented to achieve cellular and subcellular quantification of multiplexed targets. In a comparison of pathologist scoring of diaminobenzidine staining of serial sections and automated MxIF scoring of a single section, human epidermal growth factor receptor 2, estrogen receptor, p53, and androgen receptor staining by diaminobenzidine and MxIF methods yielded similar results. Single-cell staining patterns of 61 protein antigens by MxIF in 747 colorectal cancer subjects reveals extensive tumor heterogeneity, and cluster analysis of divergent signaling through ERK1/2, S6 kinase 1, and 4E binding protein 1 provides insights into the spatial organization of mechanistic target of rapamycin and MAPK signal transduction. Our results suggest MxIF should be broadly applicable to problems in the fields of basic biological research, drug discovery and development, and clinical diagnostics.

Emerging Understanding of Multiscale Tumor Heterogeneity

Cancer is a multifaceted disease characterized by heterogeneous genetic alterations and cellular metabolism, at the organ, tissue, and cellular level. Key features of cancer heterogeneity are summarized by 10 acquired capabilities, which govern malignant transformation and progression of invasive tumors. The relative contribution of these hallmark features to the disease process varies between cancers. At the DNA and cellular level, germ-line and somatic gene mutations are found across all cancer types, causing abnormal protein production, cell behavior, and growth. The tumor microenvironment and its individual components (immune cells, fibroblasts, collagen, and blood vessels) can also facilitate or restrict tumor growth and metastasis. Oncology research is currently in the midst of a tremendous surge of comprehension of these disease mechanisms. This will lead not only to novel drug targets but also to new challenges in drug discovery. Integrated, multi-omic, multiplexed technologies are essential tools in the quest to understand all of the various cellular changes involved in tumorigenesis. This review examines features of cancer heterogeneity and discusses how multiplexed technologies can facilitate a more comprehensive understanding of these features.

Multiplexed Analysis of Proteins in Tissue Using Multispectral Fluorescence Imaging

We present a new application of multispectral analysis for subcellular measurement of multiple proteins in formalin-fixed paraffin embedded tissue and cells. Typically, the targets of interest are present in the same or spatially overlapping cellular compartments. Such co-localization can complicate analysis and interpretation of the images obtained using traditional fluorescence, especially when spectrally overlapping labels are present. The spectral properties of currently available fluorescent dyes set an upper limit to the number of molecules that can be detected simultaneously with traditional fluorescence. By exciting a set of fluorophores at the same wavelength and unmixing their emission signals from background autofluorescence, we were able to image three targets in a single channel. This parallel imaging approach provides significant advantages for multiplexed analysis of tissues and cells.

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.

Autofluorescence removal using a customized filter set

Quantitative fluorescence microscopy is severely hindered by intrinsic autofluorescence (AF). Endogenous fluorescent molecules in tissue and cell samples emit fluorescence that often dominates signals from specific dyes. This makes AF removal critical to the development and practice of quantitative fluorescence microscopy. In this study, we showed that AF signal could be separated from specific signal using a customized filter set. The filter set used the same excitation and beam splitter as the standard filter set, but the emission filter was red‐shifted 40–60 nm from the peak of the specific dye. This filter set configuration collected mostly AF with minimum contribution from the specific dye. A linear transformation of AF images was required to correct for the difference in exposure and filter configuration. The constants (slope and intercept) in linear transformation were obtained through a pixel to pixel comparison between AF images (no staining) obtained by the standard filter set and the customized AF filter set. After staining of specific dye, the standard filter collecting target dye spectra was used to capture both target signal and AF, whereas customized filter was used to capture only AF. AF removal was accomplished by subtracting the linear transformed AF image from the image obtained from the standard filter. To validate our approach, we examined weak staining of androgen receptor in an AF abundant prostate tissue sample. Our method revealed a similar but cleaner nuclear staining of androgen receptor in a specimen, when compared to a traditional autofluorescence removal method. Microsc. Res. Tech., 76:1007–1015, 2013.

Single-Cell Mass Spectrometry of Subpopulations Selected by Fluorescence Microscopy

Specific subpopulations in a heterogeneous collection of cells, for example, cancer stem cells in a tumor, are often associated with biological or medical conditions. Fluorescence microscopy, based on biomarkers labeled with fluorescent probes, is a widely used technique for the visualization and selection of such cells. Phenotypic differences for these subpopulations at the molecular level can be identified by their untargeted analysis by single-cell mass spectrometry (MS). Here, we combine capillary microsampling MS with fluorescence microscopy for the analysis of metabolite and lipid levels in single cells to discern the heterogeneity of subpopulations corresponding to mitotic stages. The distributions of ATP, reduced glutathione (GSH), and UDP- N-acetylhexosamine (UDP-HexNAc) levels in mitosis reveal the presence of 2-3 underlying subpopulations. Cellular energy is found to be higher in metaphase compared to prometaphase and slightly declines in anaphase, telophase, and cytokinesis. The [GTP]/[GDP] ratio in cytokinesis is significantly higher than in prometaphase and anaphase. Pairwise correlations between metabolite levels show that some molecules within a group, including certain amino acids and nucleotide sugars, are strongly correlated throughout mitosis, but this is not related to their pathway distances. Correlations are observed between monophosphates (AMP and GMP), diphosphates (ADP and GDP), and triphosphates (ATP and GTP) of different nucleosides. In contrast, there is low correlation between diphosphates and triphosphates of the same nucleoside (ADP and ATP).

A computational study on convolutional feature combination strategies for grade classification in colon cancer using fluorescence microscopy data

The cancer diagnostic workflow is typically performed by highly specialized and trained pathologists, for which analysis is expensive both in terms of time and money. This work focuses on grade classification in colon cancer. The analysis is performed over 3 protein markers; namely E-cadherin, beta actin and colagenIV. In addition, we also use a virtual Hematoxylin and Eosin (HE) stain. This study involves a comparison of various ways in which we can manipulate the information over the 4 different images of the tissue samples and come up with a coherent and unified response based on the data at our disposal. Pre- trained convolutional neural networks (CNNs) is the method of choice for feature extraction. The AlexNet architecture trained on the ImageNet database is used for this purpose. We extract a 4096 dimensional feature vector corresponding to the 6th layer in the network. Linear SVM is used to classify the data. The information from the 4 different images pertaining to a particular tissue sample; are combined using the following techniques: soft voting, hard voting, multiplication, addition, linear combination, concatenation and multi-channel feature extraction. We observe that we obtain better results in general than when we use a linear combination of the feature representations. We use 5-fold cross validation to perform the experiments. The best results are obtained when the various features are linearly combined together resulting in a mean accuracy of 91.27%.

A machine learning approach to quantifying noise in medical images

As advances in medical imaging technology are resulting in significant growth of biomedical image data, new techniques are needed to automate the process of identifying images of low quality. Automation is needed because it is very time consuming for a domain expert such as a medical practitioner or a biologist to manually separate good images from bad ones. While there are plenty of de-noising algorithms in the literature, their focus is on designing filters which are necessary but not sufficient for determining how useful an image is to a domain expert. Thus a computational tool is needed to assign a score to each image based on its perceived quality. In this paper, we introduce a machine learning-based score and call it the Quality of Image (QoI) score. The QoI score is computed by combining the confidence values of two popular classification techniques—support vector machines (SVMs) and Naïve Bayes classifiers. We test our technique on clinical image data obtained from cancerous tissue samples. We used 747 tissue samples that are stained by four different markers (abbreviated as CK15, pck26, E_cad and Vimentin) leading to a total of 2,988 images. The results show that images can be classified as good (high QoI), bad (low QoI) or ugly (intermediate QoI) based on their QoI scores. Our automated labeling is in agreement with the domain experts with a bi-modal classification accuracy of 94%, on average. Furthermore, ugly images can be recovered and forwarded for further post-processing.

Inferring Mechanism of Action of an Unknown Compound from Time Series Omics Data

Identifying the mechanism of action (MoA) of an unknown, possibly novel, substance (chemical, protein, or pathogen) is a significant challenge. Biologists typically spend years working out the MoA for known compounds. MoA determination is especially challenging if there is no prior knowledge and if there is an urgent need to understand the mechanism for rapid treatment and/or prevention of global health emergencies. In this paper, we describe a data analysis approach using Gaussian processes and machine learning techniques to infer components of the MoA of an unknown agent from time series transcriptomics, proteomics, and metabolomics data.

Robust single cell quantification of immune cell subtypes in histological samples

Due to the rapid increase in immunotherapies there is an urgent need to develop new tools for robust in situ immune cell-typing and quantification to understand disease mechanisms and therapeutic responses. In this paper, we present a new machine-learning based method for classifying immune cell types in human tissue from highly multiplexed data. The proposed method is based on: i) identifying the most representative cell clusters across multiple slides by performing hierarchical multi-channel and multi-slide clustering; ii) from the clusters of interest, we then learn a biological-phenotypical-taxonomical cell model by solving a multi-class and multi-label classification problem. We have applied this methodology for the simultaneous classification of T and B cells using CD3 and CD20 markers and further sub-classification of T cells (CD3+) into CD4+ and CD8+, and FoxP3+ cells (within CD3+ and CD4+ cells). The method allows estimating statistical measurements used for correlation analysis with clinical data. Our method is generic and can be applied for any cell type classification problem. We obtain an average accuracy of ∼95% across six immune cell types/subtypes following simultaneous classification with this approach.