Kirsten E. Diggins

Methods for discovery and characterization of cell subsets in high dimensional mass cytometry data

The flood of high-dimensional data resulting from mass cytometry experiments that measure more than 40 features of individual cells has stimulated creation of new single cell computational biology tools. These tools draw on advances in the field of machine learning to capture multi-parametric relationships and reveal cells that are easily overlooked in traditional analysis. Here, we introduce a workflow for high dimensional mass cytometry data that emphasizes unsupervised approaches and visualizes data in both single cell and population level views. This workflow includes three central components that are common across mass cytometry analysis approaches: (1) distinguishing initial populations, (2) revealing cell subsets, and (3) characterizing subset features. In the implementation described here, viSNE, SPADE, and heatmaps were used sequentially to comprehensively characterize and compare healthy and malignant human tissue samples. The use of multiple methods helps provide a comprehensive view of results, and the largely unsupervised workflow facilitates automation and helps researchers avoid missing cell populations with unusual or unexpected phenotypes. Together, these methods develop a framework for future machine learning of cell identity.

Deep phenotyping of Tregs identifies an immune signature for idiopathic aplastic anemia and predicts response to treatment

Idiopathic aplastic anemia (AA) is an immune-mediated and serious form of bone marrow failure. Akin to other autoimmune diseases, we have previously shown that in AA regulatory T cells (Tregs) are reduced in number and function. The aim of this study was to further characterize Treg subpopulations in AA and investigate the potential correlation between specific Treg subsets and response to immunosuppressive therapy (IST) as well as their in vitro expandability for potential clinical use. Using mass cytometry and an unbiased multidimensional analytical approach, we identified 2 specific human Treg subpopulations (Treg A and Treg B) with distinct phenotypes, gene expression, expandability, and function. Treg B predominates in IST responder patients, has a memory/activated phenotype (with higher expression of CD95, CCR4, and CD45RO within FOXP3(hi), CD127(lo) Tregs), expresses the interleukin-2 (IL-2)/STAT5 pathway and cell-cycle commitment genes. Furthermore, in vitro-expanded Tregs become functional and take on the characteristics of Treg B. Collectively, this study identifies human Treg subpopulations that can be used as predictive biomarkers for response to IST in AA and potentially other autoimmune diseases. We also show that Tregs from AA patients are IL-2-sensitive and expandable in vitro, suggesting novel therapeutic approaches such as low-dose IL-2 therapy and/or expanded autologous Tregs and meriting further exploration.

In Vivo Autofluorescence Imaging of Tumor Heterogeneity in Response to Treatment

Subpopulations of cells that escape anti-cancer treatment can cause relapse in cancer patients. Therefore, measurements of cellular-level tumor heterogeneity could enable improved anti-cancer treatment regimens. Cancer exhibits altered cellular metabolism, which affects the autofluorescence of metabolic cofactors NAD(P)H and FAD. The optical redox ratio (fluorescence intensity of NAD(P)H divided by FAD) reflects global cellular metabolism. The fluorescence lifetime (amount of time a fluorophore is in the excited state) is sensitive to microenvironment, particularly protein-binding. High-resolution imaging of the optical redox ratio and fluorescence lifetimes of NAD(P)H and FAD (optical metabolic imaging) enables single-cell analyses. In this study, mice with FaDu tumors were treated with the antibody therapy cetuximab or the chemotherapy cisplatin and imaged in vivo two days after treatment. Results indicate that fluorescence lifetimes of NAD(P)H and FAD are sensitive to early response (two days post-treatment, P < .05), compared with decreases in tumor size (nine days post-treatment, P < .05). Frequency histogram analysis of individual optical metabolic imaging parameters identifies subpopulations of cells, and a new heterogeneity index enables quantitative comparisons of cellular heterogeneity across treatment groups for individual variables. Additionally, a dimensionality reduction technique (viSNE) enables holistic visualization of multivariate optical measures of cellular heterogeneity. These analyses indicate increased heterogeneity in the cetuximab and cisplatin treatment groups compared with the control group. Overall, the combination of optical metabolic imaging and cellular-level analyses provide novel, quantitative insights into tumor heterogeneity.

Cutting Edge: Redox Signaling Hypersensitivity Distinguishes Human Germinal Center B Cells

Differences in the quality of BCR signaling control key steps of B cell maturation and differentiation. Endogenously produced H2O2 is thought to fine tune the level of BCR signaling by reversibly inhibiting phosphatases. However, relatively little is known about how B cells at different stages sense and respond to such redox cues. In this study, we used phospho-specific flow cytometry and high-dimensional mass cytometry (CyTOF) to compare BCR signaling responses in mature human tonsillar B cells undergoing germinal center (GC) reactions. GC B cells, in contrast to mature naive B cells, memory B cells, and plasmablasts, were hypersensitive to a range of H2O2 concentrations and responded by phosphorylating SYK and other membrane-proximal BCR effectors in the absence of BCR engagement. These findings reveal that stage-specific redox responses distinguish human GC B cells.

Characterizing Phenotypes and Signaling Networks of Single Human Cells by Mass Cytometry

Single cell mass cytometry is revolutionizing our ability to quantitatively characterize cellular biomarkers and signaling networks. Mass cytometry experiments routinely measure 25-35 features of each cell in primary human tissue samples. The relative ease with which a novice user can generate a large amount of high quality data and the novelty of the approach have created a need for example protocols, analysis strategies, and datasets. In this chapter, we present detailed protocols for two mass cytometry experiments designed as training tools. The first protocol describes detection of 26 features on the surface of human peripheral blood mononuclear cells. In the second protocol, a mass cytometry signaling network profile measures 25 node states comprised of five key signaling effectors (AKT, ERK1/2, STAT1, STAT5, and p38) quantified under five conditions (Basal, FLT3L, SCF, IL-3, and IFNγ). This chapter compares manual and unsupervised data analysis approaches, including bivariate plots, heatmaps, histogram overlays, SPADE, and viSNE. Data files in this chapter have been shared online using Cytobank ( http://www.cytobank.org/irishlab/ ).

Characterizing cell subsets using marker enrichment modeling

Learning cell identity from high-content single-cell data presently relies on human experts. We present marker enrichment modeling (MEM), an algorithm that objectively describes cells by quantifying contextual feature enrichment and reporting a human- and machine-readable text label. MEM outperforms traditional metrics in describing immune and cancer cell subsets from fluorescence and mass cytometry. MEM provides a quantitative language to communicate characteristics of new and established cytotypes observed in complex tissues.

High-Dimensional Analysis of Acute Myeloid Leukemia Reveals Phenotypic Changes in Persistent Cells during Induction Therapy

The plasticity of AML drives poor clinical outcomes and confounds its longitudinal detection. However, the immediate impact of treatment on the leukemic and non-leukemic cells of the bone marrow and blood remains relatively understudied. Here, we conducted a pilot study of high dimensional longitudinal monitoring of immunophenotype in AML. To characterize changes in cell phenotype before, during, and immediately after induction treatment, we developed a 27-antibody panel for mass cytometry focused on surface diagnostic markers and applied it to 46 samples of blood or bone marrow tissue collected over time from 5 AML patients. Central goals were to determine whether changes in AML phenotype would be captured effectively by cytomic tools and to implement methods for describing the evolving phenotypes of AML cell subsets. Mass cytometry data were analyzed using established computational techniques. Within this pilot study, longitudinal immune monitoring with mass cytometry revealed fundamental changes in leukemia phenotypes that occurred over time during and after induction in the refractory disease setting. Persisting AML blasts became more phenotypically distinct from stem and progenitor cells due to expression of novel marker patterns that differed from pre-treatment AML cells and from all cell types observed in healthy bone marrow. This pilot study of single cell immune monitoring in AML represents a powerful tool for precision characterization and targeting of resistant disease.

Mass cytometry deep phenotyping of human mononuclear phagocytes and myeloid-derived suppressor cells from human blood and bone marrow

The monocyte phagocyte system (MPS) includes numerous monocyte, macrophage, and dendritic cell (DC) populations that are heterogeneous, both phenotypically and functionally. In this study, we sought to characterize those diverse MPS phenotypes with mass cytometry (CyTOF). To identify a deep phenotype of monocytes, macrophages, and DCs, a panel was designed to measure 38 identity, activation, and polarization markers, including CD14, CD16, HLA‐DR, CD163, CD206, CD33, CD36, CD32, CD64, CD13, CD11b, CD11c, CD86, and CD274. MPS diversity was characterized for 1) circulating monocytes from healthy donors, 2) monocyte‐derived macrophages further polarized in vitro (i.e., M‐CSF, GM‐CSF, IL‐4, IL‐10, IFN‐γ, or LPS long‐term stimulations), 3) monocyte‐derived DCs, and 4) myeloid‐derived suppressor cells (MDSCs), generated in vitro from bone marrow and/or peripheral blood. Known monocyte subsets were detected in peripheral blood to validate the panel and analysis pipeline. Then, using various culture conditions and stimuli before CyTOF analysis, we constructed a multidimensional framework for the MPS compartment, which was registered against historical M1 or M2 macrophages, monocyte subsets, and DCs. Notably, MDSCs generated in vitro from bone marrow expressed more S100A9 than when generated from peripheral blood. Finally, to test the approach in vivo, peripheral blood from patients with melanoma (n = 5) was characterized and observed to be enriched for MDSCs with a phenotype of CD14+HLA‐DRlowS100A9high (3% of PBMCs in healthy donors, 15.5% in patients with melanoma, P < 0.02). In summary, mass cytometry comprehensively characterized phenotypes of human monocyte, MDSC, macrophage, and DC subpopulations in both in vitro models and patients.

Abstract B27: Phenotypic plasticity and heterogeneity in small cell lung cancer (SCLC): Novel molecular subtypes and potential for targeted therapy.

Background: SCLC (15% of lung cancers) exhibits: 1) rapid growth and early fatal metastasis; 2) neuroendocrine features; 3) high initial responsiveness to chemotherapy and radiation; 4) aggressive recurrence with 5% 5-year patient survival. Gene expression and mutation profiling efforts to identify oncogenic mutations, gene amplifications or signatures with clinical utility in SCLC have thus far been unfruitful. In addition, prognostic or diagnostic markers for SCLC are scarce. Hence, there is a dire need for investigating molecular subtypes and oncogenic drivers in SCLC. We hypothesize that deregulated networks, rather than single genes, drive SCLC phenotype. Results: We previously identified a SCLC-specific gene co-expression network (Blue module, by Weighted Gene Co-expression Network Analysis - WGCNA) from a lung cancer patient dataset, and derived a SCLC-specific hub network (SSHN) signature that: 1) separated SCLC from other lung cancer types and normal lung in both genomic and proteomic independent datasets; 2) identified 2 SCLC subtypes with high and low SSHN expression in both patient specimens and cultured cell lines. Spleen tyrosine kinase (SYK) was validated as a candidate oncogenic driver of one subtype, as SYK targeted small-interfering RNA significantly decreased viability via increased death in high SYK-expressing SCLC cell lines. Due to the lack of larger SCLC patient datasets, we have now applied the SSHN classifier to the 53 SCLC cell lines from the Cancer Cell Line Encyclopedia (CCLE) and validated the SSHN-high and low subtypes. From this larger dataset, it is evident that the SSHN-defined subtypes are not totally separate. Rather, they are connected by gradual intermediate shades. This gradation became clearer by applying WGCNA to SCLC cell lines from CCLE, which identified 2 gene co-expression modules – Blue and Turquoise, that overlap with modules from patient datasets described above. The Blue module is enriched in neuroendocrine signaling, the Turquoise in mesenchymal adhesion-related pathways. Eigengene expression of the 2 modules (MEblue, MEturquoise) is anti-correlated, and all 53 SCLC cell lines are distributed along this anti-correlation diagonal. Expression of the neuroendocrine marker CD56 is highest in cells at one end of this diagonal (MEblue-high cell lines), and decreases towards the other end (MEturquoise-high cell lines), whereas the mesenchymal marker CD44 has an opposite trend. Multi-dimensional flow cytometry data, visualized with viSNE, indicated that SCLC cell lines are heterogeneous with respect to several additional cell surface and cytoplasmic markers and that, in general, there is a gradient of expression of these markers that tends to correlate with the neuroendocrine (e.g., SYK) to mesenchymal (e.g., TGFbeta receptor II) phenotype gradient. Finally, at the neuroendocrine end of the phenotypic spectrum (MEblue-high) cells grow in suspension, whereas they become increasingly adherent towards the mesenchymal end (MEturquoise-high). Conclusion: Our data provide strong evidence for a heterogeneous phenotypic space in SCLC that may define distinct subtypes. This heterogeneity was previously unsuspected in human SCLC, although evidence for it was reported in genetic mouse models of SCLC {Calbo J, et.al, Cancer Cell, 2011}. Classification of human SCLC cell lines along a neuroendocrine to mesenchymal differentiation gradient should apply to human tumors as well, since the WGCNA network classifiers overlap. However, further studies in patients are warranted to prove the existence of distinct SCLC subtypes, as well as to probe their translational value for biomarkers and targeted treatment. Citation Format: Akshata Ramrao Udyavar, Megan Hoeksema, Kirsten Diggins, Jonathan Irish, Pierre P. Massion, Vito Quaranta. Phenotypic plasticity and heterogeneity in small cell lung cancer (SCLC): Novel molecular subtypes and potential for targeted therapy. [abstract]. In: Proceedings of the AACR-IASLC Joint Conference on Molecular Origins of Lung Cancer; 2014 Jan 6-9; San Diego, CA. Philadelphia (PA): AACR; Clin Cancer Res 2014;20(2Suppl):Abstract nr B27.

Generating Quantitative Cell Identity Labels with Marker Enrichment Modeling (MEM)

Multiplexed single‐cell experimental techniques like mass cytometry measure 40 or more features and enable deep characterization of well‐known and novel cell populations. However, traditional data analysis techniques rely extensively on human experts or prior knowledge, and novel machine learning algorithms may generate unexpected population groupings. Marker enrichment modeling (MEM) creates quantitative identity labels based on features enriched in a population relative to a reference. While developed for cell type analysis, MEM labels can be generated for a wide range of multidimensional data types, and MEM works effectively with output from expert analysis and diverse machine learning algorithms. MEM is implemented as an R package and includes three steps: (1) calculation of MEM values that quantify each features relative enrichment in the population, (2) reporting of MEM labels as a heatmap or as a text label, and (3) quantification of MEM label similarity between populations. The protocols here show MEM analysis using datasets from immunology and oncology. These MEM implementations provide a way to characterize population identity and novelty in the context of computational and expert analyses.