Consensus Non-negative Matrix Factorization Research Articles

Spatial transcriptomics reveals influence of microenvironment on intrinsic fates in melanoma therapy resistance.

Resistance to cancer therapy is driven by both cell-intrinsic and microenvironmental factors. Previous work has revealed that multiple resistant cell fates emerge in melanoma following treatment with targeted therapy and that, in vitro, these resistant fates are determined by the transcriptional state of individual cells prior to exposure to treatment. What remains unclear is whether these resistant fates are shared across different genetic backgrounds and how, if at all, these resistant fates interact with the tumor microenvironment. Through spatial transcriptomics and single-cell RNA sequencing, we uncovered distinct resistance programs in melanoma cells shaped by both intrinsic cellular states and the tumor microenvironment. Consensus non-negative matrix factorization revealed shared intrinsic resistance programs across different cell lines, highlighting the presence of universal and unique resistance pathways. In patient samples, we demonstrated that these resistance programs coexist within individual tumors and associate with diverse immune signatures, suggesting that the tumor microenvironment and distribution of resistant fates are closely connected. Single-cell resolution spatial transcriptomics in xenograft models revealed both intrinsically determined and extrinsically influenced resistant fates. Overall, this work demonstrates that each therapy resistant fate coexists with a distinct immune microenvironment in tumors and that, in vivo, tissue features, such as regions of necrosis, can influence which resistant fate is adopted.

An integrated single-cell RNA-seq map of human neuroblastoma tumors and preclinical models uncovers divergent mesenchymal-like gene expression programs

BackgroundNeuroblastoma is a common pediatric cancer, where preclinical studies suggest that a mesenchymal-like gene expression program contributes to chemotherapy resistance. However, clinical outcomes remain poor, implying we need a better understanding of the relationship between patient tumor heterogeneity and preclinical models.ResultsHere, we generate single-cell RNA-seq maps of neuroblastoma cell lines, patient-derived xenograft models (PDX), and a genetically engineered mouse model (GEMM). We develop an unsupervised machine learning approach (“automatic consensus nonnegative matrix factorization” (acNMF)) to compare the gene expression programs found in preclinical models to a large cohort of patient tumors. We confirm a weakly expressed, mesenchymal-like program in otherwise adrenergic cancer cells in some pre-treated high-risk patient tumors, but this appears distinct from the presumptive drug-resistance mesenchymal programs evident in cell lines. Surprisingly, however, this weak-mesenchymal-like program is maintained in PDX and could be chemotherapy-induced in our GEMM after only 24 h, suggesting an uncharacterized therapy-escape mechanism.ConclusionsCollectively, our findings improve the understanding of how neuroblastoma patient tumor heterogeneity is reflected in preclinical models, provides a comprehensive integrated resource, and a generalizable set of computational methodologies for the joint analysis of clinical and pre-clinical single-cell RNA-seq datasets.

Abstract 881: Cell-type-specific gene programs suggest immune and stromal drivers of relapse in colorectal cancer

Abstract Background: Molecular profiling of colorectal cancer (CRC) has demonstrated marked heterogeneity between tumors that appear histologically similar. Although existing classification systems such as the Consensus Molecular Subtypes (CMS) have shown clinical utility, these bulk sequencing-based methods fail to adequately address the intra-tumoral heterogeneity of CRC. To disentangle the complex landscape and identify clinically-relevant properties of CRC we generated a single cell sequencing (scRNASeq) database from 225 tumor samples derived from resection of primary colon or liver metastases. Our analysis defines cell-type-specific gene programs significantly associated with relapse and survival and suggests interventions which may abrogate their effect. Methods: Utilizing a bio-repository of CRC specimens with an average of 5 years of clinical follow-up, we performed scRNASeq on 225 patient samples comprising ~ 1,000,000 total cells and 2.4 billion gene expression measurements. Following traditional clustering approaches, we then inferred gene expression programs through consensus non-negative matrix factorization (cNMF) to characterize cell activities. Gene program activity was explored on a cell-type-specific level and associated with clinical outcomes such as relapse. Treatment effects were additionally explored. Results: We identified 56 cell-type specific gene programs significantly associated with relapse in adult CRC patients. Interestingly these programs spanned all cellular compartments with the majority found in stromal or immune, as opposed to tumor cells. Program activity such as the non-canonical Granulocyte Chemotaxis program in B Cells was protective of relapse (relapse median expression 0% activity vs non-relapse activity 40% pAdj <.05 × 1050) whereas the Epithelial Mesenchymal Transition (EMT) program in Cancer Associated Fibroblasts (CAF) was strongly associated with relapse (relapse median expression 16% activity vs non-relapse activity 0% pAdj <.05 × 1010). Cumulative program usage enabled stratification on a patient level to provide an estimate of comparative disease free survival (DFS) and overall survival (OS); gene-programs were significantly associated with outcomes such as the CAF-EMT program which was associated with a DFS HR of 3 (pAdj <.0005). Responses to chemotherapy and treatment effect were also explored; prior treatment with Bevacizumab was associated with a significant reduction in the CAF-EMT program (11% activity in non-exposed patients versus 0% activity in exposed patients pAdj <.05 × 1010). Conclusions: Analysis of the largest CRC scRNASeq cohort to date identifies a number of previously undescribed cell-type specific gene programs significantly associate with relapse after surgery. These programs provide insights into the underlying mechanisms and suggest that the TME plays an important role in relapse. Citation Format: Nicholas J. Hornstein, Mingshuang Wang, Abdelrahman Yousef, Saikat Chowdhury, Fadl Zeineddine, Mohammad Zeineddine, Mahmoud Yousef, Shuangjie You, Betul Gunes, David Menter, Scott Kopetz, John Paul Shen. Cell-type-specific gene programs suggest immune and stromal drivers of relapse in colorectal cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 881.

Abstract 6846: Tumor cell metaplastic programs and failure of T cell remodeling drive resistance and progression to frontline chemoimmunotherapy in advanced gastric cancer

Abstract Adding aPD1 to 5-FU/platinum in advanced gastroesophageal adenocarcinomas (GEA) has yielded modest and heterogeneous results. Understanding cooperativity between these two treatment modalities will inform novel combination treatments for GEA. Towards this end, we conducted a trial (n = 47) sequentially adding pembrolizumab to 5-FU/platinum in previously untreated advanced GEA. Using serial biopsy of the primary tumor at baseline, after one cycle of 5-FU/platinum, and after the addition of pembrolizumab we transcriptionally profiled 358,067 single cells and obtained TCR-sequences from single tumor-resident T cells. Stratifying patients into slow- (> 6 months) and fast-progressing (< 6 months) groups based on progression-free survival allowed us to quantify differences in immune, stromal and epithelial subpopulations across treatment time points. Using topic modeling with consensus non-negative matrix factorization (cNMF), we identified gene expression programs in each population at each time point and developed a methodology to verify the stability of those programs across the cohort using cross-validation. Faster clinical progression featured decreased M1-like macrophage repolarization and increased MUC5A and MSLN containing metaplasia programs in tumor cells. We next selected features from cell state gene programs and identified trajectories of tumor cells across a spectrum of treatment sensitive and resistant states. We identified several factors that correlated with these trajectories and regulons that may be influencing them. After pembrolizumab we observed slower progression was associated with further increased CD8 T cell infiltration with increased CXCL13+ T cells. We next calculated covarying programs between patient samples at each time point and performed permutation testing to identify which programs (and respective cell types they were expressed in) were significantly correlated to infer the presence of multicellular tumor communities. We found that the tumor cell interferon program co-varied with the CXCL13 T cell program, and the emergence of a covarying module with tumor-reactive CXCL13 program expression happens as early as after the first dose of chemotherapy. To better understand the behavior of tumor reactive T cells within the tumor microenvironment, we leveraged TCR-sequencing data to study T cell clonality and diversity across treatment time points and between fast and slow progressors. Our data provide an invaluable resource to dissect chemoimmunotherapy response in tumor cells, understand T cell behavior and propose strategies to drive anti-tumor immune hub formation to expand the portion of patients benefiting from immunotherapy approaches. Citation Format: Samuel J. Wright, Minae An, Byung Hoon Min, You Jeong Heo, Milan Parikh, Lynn Bi, Hyuk Lee, Taejun Kim, Song-Yi Lee, Jeonghyeon Moon, Ryan J. Park, Matthew R. Strickland, Woong Yang Park, Won Ki Kang, Kyoung-Mee Kim, Seung Tae Kim, Arnav Mehta, Samuel J. Klempner, Jeeyun Lee. Tumor cell metaplastic programs and failure of T cell remodeling drive resistance and progression to frontline chemoimmunotherapy in advanced gastric cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 6846.

Abstract 868: A novel unsupervised machine learning model applied to neuroblastoma single-cell RNA-seq data reveals a drug-induced mesenchymal-like gene expression program

Abstract High-risk neuroblastoma is one of the most difficult to treat pediatric cancers, with a survival rate of only about 50% and significant long-term consequences from current chemotherapy. Preclinical models such as cell lines and mice are the backbone of drug development and experimental-mechanistic oncology, but the development of new treatments is hampered in part by a lack of understanding of the direct clinical relevance of such data collected in preclinical models. Despite this, few formal methods have been developed to determine how these various models represent/resemble primary patient tumors. Here, we present the first comprehensive single-cell RNA-seq analysis of neuroblastoma across an extensive cohort of patient tumors and a variety of preclinical model systems (n = 126 total samples assembled - the largest cohort of its kind). By building an innovative unsupervised machine learning method, which we term “automatic consensus nonnegative matrix factorization” (acNMF), we have integrated and contrasted the transcriptional landscapes of patient tumors with those of cell lines, patient-derived xenografts (PDX), and genetic mouse models (GEMM).Using these tools, we discovered the dominant adrenergic gene expression programs found in neuroblastoma patient tumors were preserved across all preclinical models. However, the presumptive chemo-resistant mesenchymal-like programs, while identifiable in cell lines, were primarily restricted to subpopulations of cancer-associated fibroblasts and Schwann-like cells in vivo. Surprisingly, a mesenchymal-like program could be acutely chemotherapy-induced in GEMM and was evident in pre-treated patient and PDX samples, suggesting a previously uncharacterized mechanism of therapy escape. In addition to these core findings, our computational tools were able to further delineate the classical neuroblastoma adrenergic and mesenchymal gene expression programs, discovering for example, novel subpopulations of cancer associated fibroblasts. These behaviors were conserved across tumors and most preclinical models, which we validated by RNA in situ hybridization, which is a high resolution ultra sensitive spatial transcriptomics technology. Our work cautions against overreliance on traditional preclinical models without recognizing their limitations and we offer a nuanced, high-resolution view of neuroblastoma pre-clinical systems for advancing therapeutic development. We have launched an open-source web resource, featuring this integrated map to aid the scientific community in further exploration of these data and hypothesis generation (available at http://pscb.stjude.org). Citation Format: Rich Chapple, Xueying Liu, Sivaraman Natarajan, Margaret I. Alexander, Yuna Kim, Anand Patel, Christy W. LaFlamme, Min Pan, William C. Wright, Hyeong-Min Lee, Yinwen Zhang, Meifen Lu, Selene C. Koo, Courtney Long, John Harper, Chandra Savage, Melissa D. Johnson, Thomas Confer, Walter J. Akers, Michael A. Dyer, Heather Sheppard, John Easton, Paul Geeleher. A novel unsupervised machine learning model applied to neuroblastoma single-cell RNA-seq data reveals a drug-induced mesenchymal-like gene expression program [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 868.

Unraveling the underlying pathogenic factors driving nonalcoholic steatohepatitis and hepatocellular carcinoma: an in-depth analysis of prognostically relevant gene signatures in hepatocellular carcinoma

BackgroundNonalcoholic steatohepatitis (NASH) is a progressive manifestation of nonalcoholic fatty liver disease (NAFLD) that can lead to fibrosis, cirrhosis, and hepatocellular carcinoma (HCC). Despite the growing knowledge of NASH and HCC, the association between the two conditions remains to be fully explored. Bioinformatics has emerged as a valuable approach for identifying disease-specific feature genes, enabling advancements in disease prediction, prevention, and personalized treatment strategies.Materials and methodsIn this study, we utilized CellChat, copy number karyotyping of aneuploid tumors (CopyKAT), consensus Non-negative Matrix factorization (cNMF), Gene set enrichment analysis (GSEA), Gene set variation analysis (GSVA), Monocle, spatial co-localization, single sample gene set enrichment analysis (ssGSEA), Slingshot, and the Scissor algorithm to analyze the cellular and immune landscape of NASH and HCC. Through the Scissor algorithm, we identified three cell types correlating with disease phenotypic features and subsequently developed a novel clinical prediction model using univariate, LASSO, and multifactor Cox regression.ResultsOur results revealed that macrophages are a significant pathological factor in the development of NASH and HCC and that the macrophage migration inhibitory factor (MIF) signaling pathway plays a crucial role in cellular crosstalk at the molecular level. We deduced three prognostic genes (YBX1, MED8, and KPNA2), demonstrating a strong diagnostic capability in both NASH and HCC.ConclusionThese findings shed light on the pathological mechanisms shared between NASH and HCC, providing valuable insights for the development of novel clinical strategies.

DCCNMF: Deep Complementary and Consensus Non-negative Matrix Factorization for multi-view clustering

Deep non-negative matrix factorization-based methods have recently been explored in multi-view clustering due to their ability to deal with complex non-linear data. Although a few methods exist that learn both complementary and consensus information simultaneously by adding new regularizations or adjusting hyperparameters without providing a solid architecture, they do not fully exploit this information. This paper proposes a novel method called Deep Complementary and Consensus Non-negative Matrix Factorization (DCCNMF) that combines the strengths of Non-negative Matrix Factorization and Coupled Non-negative Matrix Factorization using a novel architecture to simultaneously learn both complementary and consensus information present in multi-view data. Two manifold regularization terms, namely complementary manifold and consensus manifold are introduced to preserve the view-specific and view-shared geometric structures during the dimensionality reduction from higher to lower order. Further, smoothness regularization is employed to achieve distinct low-order representations by maximizing the cosine similarity among data points with similar orientations and minimizing it among data points with dissimilar orientations. DCCNMF and benchmark methods are evaluated on several real-world datasets and the results demonstrate that DCCNMF significantly outperforms state-of-the-art multi-view learning approaches.

MODL-50. CLONAL EXPANSION OF SPONTANEOUS AND INJECTED TUMORS IN SYNGENEIC MOUSE MODELS OF GLIOBLASTOMA

Abstract INTRODUCTION Glioblastoma (GBM) tumors are highly heterogenous and plastic. To better understand how tumors respond to treatment, it is important to use immunocompetent mouse models. We've created various mouse GBM models with common genetic mutations found in different subtypes of GBM, including classical, proneural, mesenchymal, and BRAF-driven. METHODS Using a combination of CRISPR and PiggyBac Transposon constructs, we injected plasmid mixtures into the ventricles of the mouse brain during the post-natal period. Electroporation was then used to enable the plasmids to enter nearby neural stem cells, which led to the spontaneous formation of tumors (spontaneous model). We collected tumor cells from these models and injected them into the adult mouse brain to create the injected mouse GBM model. Here we use spatial and single cell transcriptome profiling approaches to characterize these models at endpoint. Spatial transcriptome data was collected using the Visium 10X platform and tumor spots were deconvoluted from non-malignant spots using consensus non-negative matrix factorization. Copy number alterations (CNAs) were inferred using non-malignant spots and single-cell clusters as a reference. RESULTS We found that copy number states are largely homogenous within each sample, supporting an early population expansion in vivo. CNAs found in spontaneous tumors are also found in the matched injected tumors, which in turn harbour additional CNAs, suggesting that further selection occurs in vitro. Some of the CNAs include relevant cancer genes, suggesting the observed clonal expansions are potentially driven by additional somatic alterations that cooperate with the engineered driver mutations. Studying these models using orthogonal methods will allow us to discover alterations that synergize with the original drivers and delineate principles of GBM initiation and progression in an immune-competent environment.

EPCO-13. CNMF-SNS, A FRAMEWORK FOR UNSUPERVISED INTEGRATION OF MULTIOMICS DATA, IDENTIFIES INVASION AND RECURRENCE ASSOCIATED PROGRAMS IN GBM

Abstract Glioblastoma (GBM) is an aggressive cancer of the brain for which novel therapies are urgently needed. Heterogeneity in GBM underlies tumor evolution and resistance to therapy and involves genetic diversity of tumor cells, spatially distinct niches harbouring stem-like cells, and immune cells spanning anti- and pro-tumor states. No single profiling methodology can capture all these facets of tumor biology in the same space and time, necessitating multi-omics data collection and computational methods suited for multi-omic integration across and within cohorts. We therefore set out to develop a generalizable framework for integration that can bridge across cohorts and data types. Using consensus non-negative matrix factorization (cNMF), we discover gene expression programs (GEPs) corresponding to cell types and/or states in an unsupervised fashion. Rather than choosing a single rank (number of programs), we identify both high and low-resolution programs for each dataset, and then integrate programs from all datasets and ranks into a solution network space (SNS) on which graph algorithms identify communities of highly similar GEPs identified in all or some datasets. These communities are characterized by gene set enrichment analyses and association with sample metadata. SNS communities can be used to transfer sample metadata across datasets, enabling multi-omics and multi-cohort integration, even in the absence of shared samples or cells. We showcase the utility of cNMF-SNS in integrating multiomics datasets of GBM from single-cell to bulk RNA-Seq and mass spectrometry proteomics. We further demonstrate the power of cNMF-SNS to define and cross-annotate programs identified in 5 complementary mass spectrometry datasets of GBM, layering on survival, driver gene mutation status, clinical features, and imaging features. We demonstrate robust mapping of biologically and clinically relevant processes, highlighting how programs for tumor cell invasion and therapy resistance can be mined for novel therapeutic targets. cNMF-SNS is available at https://github.com/MorrissyLab/cNMF-SNS.

Stratification of Multiple Myeloma Patients in the Commpass Study By Single Cell Transcriptomics of Plasma Cells Associates with Clinical Outcome

Introduction: There is substantial variation in the prognosis and response to therapy among multiple myeloma (MM) patients. This diversity is influenced by both intrinsic tumor features and the tumor microenvironment. We hypothesize that the clinical outcomes post IMiD and/or PI are linked to molecular subtypes dictated by tumor sample transcriptomics. We previously showed that molecular subtyping based on the CoMMpass bulk RNAseq data has an association with time to second-line therapy (log rank p = 0.0018; Hamidi et al., Blood 140: 7208-7210). Here, we applied the same molecular subtyping schema to single cell RNAseq isolated plasma cell transcriptomic profiles. Methods: Bone marrow biopsies from newly diagnosed MM patients (N = 200) in the MMRF CoMMpass study (NCT01454297) were collected, enriched by CD138 negative bead selection, and scRNAseq was performed by MMRF Immune Profiling Consortium, a collaboration between MMRF and laboratories at Beth Israel Deaconess Medical Center, Icahn School of Medicine at Mt. Sinai, Mayo Clinic, Washington University in St. Louis, and Emory University. After quality control and unsupervised clustering, plasma cells were subset out and pseudo-bulk expression profiles were generated for baseline samples with more than 100 plasma cells (N = 95). Unsupervised machine learning approach based on consensus non-negative matrix factorization (cNMF) was applied to identify the transcriptomics-based molecular subtypes. A random forest model, built from scRNAseq-based molecular subtypes (scNMF), was applied to bulk RNAseq data with a larger sample size (N = 754) to train a predictive molecular subtyping (pNMF). Potential cell-cell communication events based on scRNAseq were also predicted by LIgand-receptor ANalysis frAmework (LIANA). Results: 705,959 single cells passed the quality control and were broadly annotated as T/NK cells, B cells, plasma cells, myeloid cells, erythroid, progenitor, and stromal cells. Consistent with previous reports, CD138 negative bead selection did not completely remove plasma cells, resulting in an average of 18.5% plasma cells (n = 130,213). Subsequent unsupervised clustering of plasma cells exhibited a patient specific clustering pattern, suggestive of the heterogeneity of these tumor-derived cells. Unsupervised NMF on the pseudo-bulk expressions of baseline plasma cells identified five patient subsets. In the same cohort of patients (N = 95), subtypes assigned by the single cell had a better association with time to second-line therapy ( p-value = 0.0027 in scNMF as compared to p = 0.13 in bulk NMF). The median time to 2nd line of 2,443 days in scNMF5.1 versus 585 days in scNMF5.4; and the median time to 2nd line of 1,597 days in NMF5.2 versus 850 days in NMF5.3. A predicted subtyping based on scRNAseq data (pNMF) had a significantly better association with time to second-line therapy (Figure 1A, log rank p = 0.00088), compared to the published molecular subtyping based on bulk RNAseq (log rank p = 0.0018). The median time to 2nd line of 1,342 days in pNMF5.1 versus 597 days in pNMF5.5; compared to the median time to 2nd line of 1,347 days in NMF5.2 versus 850 days in NMF5.3. The best clinical outcome subset pNMF5.1 was characterized with higher levels of IFN pathways (Figure 1B, Hallmark interferon gamma and alpha pathways Wilcoxon p = 0.003 and p = 0.010, respectively) and higher levels of tumor-promoting inflammatory microenvironment (Hallmarks inflammatory response and TNFα signaling via NF-κB p = 0.003 and p = 0.010, respectively). The worst outcome subset pNMF5.5, on the other hand, displayed higher levels of glycolysis, a promoting factor of tumor development ( p < 0.001). Moreover, the total number of cell-cell interactions between the major cell types differed based on scRNAseq subtypes, informing the underlying biology driving the molecular heterogeneity. We identified a positive feedback loop of CXCL12/CXCR4 signaling between stromal and plasma cells, enhancing tumor cell proliferation. Conclusions: The identification of molecular subtypes based on single-cell data through an innovative unsupervised machine learning approach can have significant implications for determining which NDMM patients are likely to respond to established therapies; and the cellular level resolution allows deconvolution of cell-cell signaling, enhancing our understanding of the underlying molecular biology and mechanisms of resistance.

Multi-Omic Insights into the Mutational Landscape and Dysregulated Transcriptional Programs of Autoimmune B-Cells in Systemic Lupus Erythematosus

Systemic Lupus Erythematosus (SLE) is an autoimmune disease with diverse manifestations, from mild and sporadic to severe and life-threatening. SLE heterogeneity arises from various factors, including genetic, epigenetic, and immune niche-related features (Kaul et al., 2016; Jenks et al., 2018). However, the exact interactions and hierarchy between these features remain unclear. Recent studies have found somatic mutations in genes typically mutated in B-cell lymphomas, in subsets of B-cells that are associated with SLE pathogenesis (Piotrowski et al., 2015; Pullabhatla et al., 2018; Singh et al., 2020; Brown et al., 2022). These mutations may support B-cell clonal bursts, but direct information from clinical specimens on the impact of these mutations on B-cell lineages and their influence on SLE progression or malignant transformation is missing. To define the somatic mutational landscape of B-cells in SLE, we first profiled peripheral blood mononuclear cells obtained from a pilot cohort of four SLE patients using whole-exome sequencing (WES; 200X depth). Following GATK best practices (Van der Auwera and O'Connor, 2020), we identified somatic alterations in DDX11 [variant allele frequency (VAF): 10.2%], DNMT3A [VAF: 3.4%]; PMS2 [VAF: 47.9%], and BICD1 [VAF: 47.3%]; these genes have been linked to lymphoid malignancies and SLE pathogenesis (Piotrowski et al., 2015; Pullabhatla et al., 2018; Niroula et al., 2021; Saeed et al., 2021). We validated the presence of the top-identified mutations using droplet digital PCR technology on sorted B-cell subsets (naïve, memory, and double-negative [DN; CD19 hiIgD loCD27 lo]) from these SLE patients. DN cells are a heterogeneous subtype of autoreactive memory B-cells that play a crucial role in SLE pathogenesis being the precursors of autoantibody producing plasma cells (Wei et al., 2007; Jenks et al., 2018). To investigate the transcriptional and epigenetic alterations in the mutation-harboring B-cells, we then conducted single-cell multi-omic sequencing. We profiled sorted activated B-cells (CD19 hiIgD lo) from the four SLE cases, to largely exclude non-autoreactive naïve B-cells. To maintain a comparison with naïve B cells, we included a small proportion of these cells in the input. In total, we analyzed 35,339 cells, and identified different B-cell subsets with distinct gene expression patterns, including DN cells ( n = 1,437 cells; ZEB2 hiTRAF5 lo, Jenks et al., 2018). To uncover, in an unbiased way, transcriptional differences between B-cell subsets, we catalogued variation in transcriptomic signatures derived from the scRNA data using consensus non-negative matrix factorization. We found a unique gene expression program exclusively present in DN cells ( Panel A), which includes key genes implicated in immune regulation and the development of autoimmunity, such as PTPN22 (Chung & Criswell, 2007; Tizaoui et al., 2021) . Pathway analysis of the DN-specific program revealed an enrichment of complement system genes (FDR = 0.074; Hallmark), which are crucial for immune regulation and tolerance, such as CBLB (Tang et al., 2019). Notably, in DN cells from one of the SLE samples carrying lymphoma/SLE-associated mutations in BICD1, PMS2, DDX11, we observed an altered expression of this gene program compared to DN cells from other samples ( P-value = 3.6x10 -11; Panel A, inset), suggesting that a malfunctioning complement system in DN cells may lead to an overly active immune response (Weinstein et al., 2021). In line with this notion, using chromatin accessibility profiles as a measure of transcription factor (TF) regulatory activities, we showed increased activity of TFs involved in interferon-dependent innate immune response and in B-cell activation and differentiation (RFX and OCT-1, respectively; FDR < 0.05), in the DN B-cells of this sample ( Panel B). In summary, we uncovered a unique transcriptional program associated with autoimmune DN B-cells. This program showed enrichment in key genes and pathways involved in immune response regulation. The disruption of this program, accompanied by the somatic SLE-associated mutations, may lead to overactivation of DN B cells, ultimately contributing to autoimmunity. Further analyses involving a larger cohort of clinical specimens are needed to directly link B-cell genotypes with their transcriptional and epigenetic programs and to elucidate the role of the identified mutations in SLE pathobiology.

Characterizing the key genes of COVID-19 that regulate tumor immune microenvironment and prognosis in hepatocellular carcinoma.

Hepatocellular carcinoma (HCC), a highly heterogeneous malignant tumor associated with a poor prognosis, is a common cause of cancer-related deaths worldwide, with a limited survival benefit for patients despite ongoing therapeutic breakthroughs. Coronavirus disease 2019 (COVID-19), a severe infectious disease caused by severe acute respiratory syndrome coronavirus type 2 (SARS-CoV-2), is a global pandemic and a serious threat to human health. The increased susceptibility to SARS-CoV-2 infection and a poor prognosis in patients with cancer necessitate the exploration of the potential link between the two. No studies have investigated the relationship of COVID-19 genes with the prognosis and tumor development in patients with HCC. We screened prognosis-related COVID-19 genes in HCC, performed molecular typing, developed a stable and reliable COVID-19 genes signature for predicting survival, characterized the immune microenvironment in HCC patients, and explored new molecular therapeutic targets. Datasets of HCC patients, including RNA sequencing data and clinical information, were obtained from The Cancer Genome Atlas (TCGA), International Cancer Genome Consortium (ICGC), and Gene Expression Omnibus (GEO) databases. Prognosis-related COVID-19 genes were identified by univariate Cox analysis. Molecular typing of HCC was performed using the consensus non-negative matrix factorization method (cNMF), followed by the analysis of survival, tumor microenvironment, and pathway enrichment for each subtype. Prognostic signatures were constructed using LASSO-Cox regression models, and receiver operating characteristic (ROC) curves were used to validate the predictive performance of the signature. The same approach was used for the test and external validation sets. Seven software packages were applied to determine the abundance of immune infiltration in HCC patients and investigate its relationship with the risk scores. Gene set enrichment analysis (GSEA) was used to explore the potential mechanisms by which the COVID-19 genes affect hepatocarcinogenesis and prognosis. Three types of machine learning methods were combined to identify the most critical genes in the signature and localize their expression at the single cell level. We identified 53 prognosis-related COVID-19 genes and classified HCC into two molecular subtypes (C1, C2) by using the NMF method. The prognosis of C2 was significantly better than that of C1, and the two subtypes differed remarkably in terms of the tumor immune microenvironment and biological functions. The 17 COVID-19 genes were screened using the LASSO regression method to develop a 17 COVID-19 genes signature, which demonstrated a good predictive performance for 1-, 2- and 3-year OS of patients with HCC. The risk score as an independent prognostic factor for HCC has better predictive accuracy than traditional clinical variables. Patients in the TCGA cohort were categorized by risk score into the high- and low-risk groups, with the high-risk group mainly enriched in the immune modulation-related pathways and the low-risk group mainly enriched in the metabolism-related pathways, suggesting that the COVID-19 genes may affect disease progression and prognosis by regulating the tumor immune microenvironment and metabolism in HCC. NOL10 was identified as the most critical gene in the signature and hypothesized to be a potential therapeutic target for HCC. Objectively, the COVID-19 genes signature developed in this study, as an independent prognostic factor in HCC patients, is closely associated with the prognosis and tumor immune microenvironment of HCC patients and indicates that they may regulate the development of HCC in multiple ways, providing us with new perspectives for understanding the molecular mechanisms of HCC and finding effective therapeutic targets.

Identification of a basement membrane-based risk scoring system for prognosis prediction and individualized therapy in clear cell renal cell carcinoma.

Clear cell renal cell carcinoma (ccRCC) belongs to one of the 10 most frequently diagnosed cancers worldwide and has a poor prognosis at the advanced stage. Although multiple therapeutic agents have been proven to be curative in ccRCC, their clinical application was limited due to the lack of reliable biomarkers. Considering the important role of basement membrane (BM) in tumor metastasis and TME regulation, we investigated the expression of BM-related genes in ccRCC and identified prognostic BM genes through differentially expression analysis and univariate cox regression analysis. Then, BM-related ccRCC subtypes were recognized through consensus non-negative matrix factorization based on the prognostic BM genes and evaluated with regard to clinical and TME features. Next, utilizing the differentially expressed genes between the BM-related subtypes, a risk scoring system BMRS was established after serial analysis of univariate cox regression analysis, lasso regression analysis, and multivariate cox regression analysis. Time-dependent ROC curve revealed the satisfactory prognosis predictive capacity of BMRS with internal, and external validation. Multivariate analysis proved the independent predictive ability of BMRS and a BMRS-based nomogram was constructed for clinical application. Some featured mutants were discovered through genomic analysis of the BMRS risk groups. Meanwhile, the BMRS groups were found to have distinct immune scores, immune cell infiltration levels, and immune-related functions. Moreover, with the help of data from The Cancer Immunome Atlas (TCIA) and Genomics of Drug Sensitivity in Cancer (GDSC), the potential of BMRS in predicting therapeutic response was evaluated and some possible therapeutic compounds were proposed through ConnectivityMap (CMap). For the practicability of BMRS, we validated the expression of BMRS-related genes in clinical samples. After all, we identified BM-related ccRCC subtypes with distinct clinical and TME features and constructed a risk scoring system for the prediction of prognosis, therapeutic responses, and potential therapeutic agents of ccRCC. As ccRCC systemic therapy continues to evolve, the risk scoring system BMRS we reported may assist in individualized medication administration.

Association of RNA-modification "writer" genes with prognosis and response to immunotherapy in patients with low-grade glioma.

RNA modification is a key regulatory mechanism involved in tumorigenesis, tumor progression, and the immune response. However, the potential role of RNA modification "writer" genes in the immune microenvironment of gliomas and their effect on the response to immunotherapy remains unclear. The purpose of this study was to evaluate the role of RNA modification "writer" gene in the prognosis and immunotherapy response of low-grade glioma (LGG). The consensus non-negative matrix factorization (CNMF) method was used to identify different RNA modification subtypes. We used a novel eigengene screening method, the variable neighborhood learning Harris Hawks optimizer (VNLHHO), to screen for eigengenes among the RNA modification subtypes. We constructed a principal components analysis score(PCA_score)-based prognostic prediction model and validated it using an independent cohort. We also analyzed the association between PCA_score and the immune and molecular features of LGG. The results suggested that LGG can be divided into two different RNA modification-based subtypes with distinct prognostic and molecular features. High PCA_score was significantly associated with a poor prognosis in LGG and was an independent prognostic factor. A nomogram containing PCA_score and clinical features was constructed, and it showed a significant predictive value. PCA_score was negatively correlated with tumor purity and the abundance of CD4+ T cells in LGG patients. LGG patients with high PCA_score had lower Tumor Immune Dysfunction and Exclusion scores and showed an immunotherapy response. In conclusion, we report a novel RNA modification-based prognostic model for LGG that lays the foundation for evaluating LGG prognosis and developing more effective therapeutic strategies for these tumors.

Novel Transcriptomics Based Stratification Approach Using the Commpass Dataset Reveals Mechanistic Insights into Current and Upcoming MM Therapies

Introduction: Prognosis and response to Multiple Myeloma (MM) therapy can vary markedly even among patients at the same disease stage and with comparable clinical characteristics. This heterogeneity is driven by tumor intrinsic features such as somatic alterations impacting the genetic and epigenetic landscape of a patient as well immune dysfunction in the tumor microenvironment (TME). Tumor gene expression profiles have been correlated with clinical outcomes and are potentially useful for precision medicine approaches in MM. In this study, we hypothesized that clinical outcomes and variable responses to therapeutic modalities in MM may be driven by expression of critical gene sets, and sought to identify clinically actionable molecular subtypes using the available RNAseq data from newly diagnosed MM (NDMM) patients in the COMPASS functional genomics study. Methods: Unsupervised machine learning approach based on consensus non-negative matrix factorization (cNMF) was applied to the most variable genes (n=6019; top 10%) based on median absolute deviation (MAD) of log2(TPM) in the RNAseq data derived from CD138 positive bone marrow patient samples collected at baseline (N=766) to identify transcriptomic-based molecular subtypes. After the optimal number of clusters was determined using their cophenetic coefficient, the subtypes were assessed for their association with clinical outcomes in patients with available data (N=753). Survival analyses included Kaplan-Meier (KM) curves, log-rank tests, univariate and multivariate Cox regression, and Wald-tests. Subtypes were also biologically characterized by gene signature scoring using well-curated pathway signatures (GSVA analysis using Hallmark pathways), cell type enrichment (xCell enrichment). Results: Our cNMF based analysis identified five clusters of patients based on the 6,019 (top 10%) most variable genes. These novel subtypes (NMF5.1-5) had unique biology associated with time-to-second-line of therapy (Figure 1A, log rank p=0.0018). Biological differences between NMF5.3 and NMF5.2 were investigated further to understand mode of action and mechanism of resistance to IMiD/PI treatment in NDMM patients. NMF5.2 was the best prognostic cluster and associated with the longest time to 2nd line (median time to 2nd line 1,444 days) while NMF5.3 is associated with the shortest time to 2nd line (831 days). NMF5.2 was characterized with a more intact immune system including higher immune cell activation and immune microenvironment scores as well as higher levels in the INF pathway (Hallmark Interferon gamma and alpha pathways Wilcoxon p-value<0.001). NMF5.3, on the other hand, displayed lower scores for immune activation and had more of immunosuppressive/ exhausted immune microenvironment (Wilcoxon p-value<0.001). Ongoing analyses to evaluate the baseline expression level of immunotherapeutic targets in MM like BCMA, FcRH5, GPRC5D and additional immune checkpoint antigens and its association with NMF5.2 and 3 will also be discussed. Conclusions: Unique molecular subtypes identified via a novel unsupervised machine learning may have implications on our understanding of how NDMM patients respond to existing therapies and suggest that the addition of immune checkpoints can improve patients’ clinical outcome. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal

Single-Cell Profiling of CD8 Landscape in Treatment Naïve and Relapsed/Refractory AML Patients Reveals Distinct Effector Cellular States Predictive of Outcomes

Background: The success of allogeneic stem cell transplantation mediated by the graft versus leukemia effect of engrafted T-cells strongly supports the premise to investigate T cell biology in acute myeloid leukemia (AML). Of particular interest are CD8 cells, which are key players in eliciting anti-cancer cytotoxic activities but are understudied in AML. In this study, we conducted single cell RNA profiling in bone marrows from healthy donors, treatment naïve (RxNaive) and relapsed/refractory (R/R) AML patients to investigate the CD8 landscape in AML. Methods: We conducted paired single cell gene expression (scRNA) and T-cell repertoire (scTCR) profiling of bone marrow mononuclear cells from 3 healthy donors and 23 RxNaive adult AML patients (median age 71.81 years; range 51-86 years) and leveraged our previously published data of 8 relapsed/refractory AML patients (median age 73 years; range 64-88 years). After doublet removal and quality assessment, a total of 51,219 T cells passed the filtering criteria, of which 23,320 (45.5%) were CD8 cells. A total of 17,782/23,320 (74.1%) CD8 cells had paired scTCR. We applied principal component analysis to identify sources of variability across CD8 cells, then used Harmony-based linear correction integrated with K-nearest neighbors for cluster identification. We then applied non-linear dimensionality reduction (UMAP) to project cells into 2 dimensions. Results: In total, 6 CD8 T cell subtypes were defined including naïve, memory, effector memory, cytotoxic and exhausted T cell subtypes. Panel 1 shows the UMAP projection of CD8 cells overlaid with neighborhoods demonstrating differential abundance of cells by patient type (healthy, RxNaive or R/R). Naïve, memory, MAIT and exhausted clusters were shared across healthy, RxNaive and R/R patients (Panel 1, left). Remarkably, the effector clusters (effector memory and cytotoxic) demonstrated distinct patterns of disease states. Specifically, the Top effector clusters (Panel 1, top right) were largely contributed by R/R AML patients. However, the Bottom effector clusters (Panel 1, bottom right) were enriched by cells from healthy and RxNaive. Only 2.5% of R/R effector cells were part of the bottom effector clusters. This indicated that naïve and memory cells of RxNaive and R/R AML patients are biologically similar, however, effector cells took diverging paths in RxNaive versus R/R AML patients. We next explored the biological programs that govern the Top clusters and thus associate with R/R state as compared to the Bottom clusters, associated with RxNaive AML patients. We conducted consensus non-negative matrix factorization coupled with gene expression profiling and pathway analysis. This revealed positive enrichment for IFNa and IFNg signaling and significant negative enrichment for TNFa and TGFb signaling in the Top clusters, suggesting a cytokine dependent remodeling of CD8 effector function in R/R cells. We next explored whether the top 25 differential genes (herein referred to as CD8 Dysfunctional Signature) in the Top clusters (enriched for R/R CD8 cells) can predict outcomes in two cohorts of newly diagnosed AML patients (Abbas et al 2021 and TCGA AML cohorts). Interestingly, the CD8 Dysfunctional Signature was indeed predictive of outcomes, with patients harboring higher CD8 Dysfunctional Signature scores having worse outcomes (p=0.0054, Panel 2). We next explored whether the TCR repertoires are also different in AML of RxNaive and R/R patients. Indeed, TCR profiling revealed lower clonotype diversity and significant clonotype expansion in R/R patients, with 34% of R/R cells with TCR data exhibiting hyperexpansion (clone size >100). However, these hyperexpanded clonotypes were largely derived from dysfunctional CD8 cells. Conclusion: We characterized the CD8 landscape in RxNaive and R/R AML patients and defined distinct states of effector CD8 cells predictive of outcomes. CD8+ T-cells in R/R patients are markedly different from healthy/RxNaive patients, and characterized by a dysfunctional state, skewed distribution, lower clonal diversity, and significant clonal hyperexpansion. This dysfunctional state may be due to the AML microenvironment actively shaping the T-cell repertoires or the chemotherapy received by the patient, which may narrow the repertoire by eliminating active T-cells. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal

Decoding the colorectal cancer ecosystem emphasizes the cooperative role of cancer cells, TAMs and CAFsin tumor progression

BackgroundSingle-cell transcription data provided unprecedented molecular information, enabling us to directly encode the ecosystem of colorectal cancer (CRC). Characterization of the diversity of epithelial cells and how they cooperate with tumor microenvironment cells (TME) to endow CRC with aggressive characteristics at single-cell resolution is critical for the understanding of tumor progression mechanism.MethodsIn this study, we comprehensively analyzed the single-cell transcription data, bulk-RNA sequencing data and pathological tissue data. In detail, cellular heterogeneity of TME and epithelial cells were analyzed by unsupervised classification and consensus nonnegative matrix factorization analysis, respectively. Functional status of epithelial clusters was annotated by CancerSEA and its crosstalk with TME cells was investigated using CellPhoneDB and correlation analysis. Findings from single-cell transcription data were further validated in bulk-RNA sequencing data and pathological tissue data.ResultsA distinct cellular composition was observed between tumor and normal tissues, and tumors exhibited immunosuppressive phenotypes. Regarding epithelial cells, we identified one highly invasiveQuery cluster, C4, that correlated closely with tumor-associated macrophages (TAMs) and cancer-associated fibroblasts (CAFs). Further analysis emphasized the TAMs subclass TAM1 and CAFs subclass S5 are closely related with C4.ConclusionsIn summary, our study elaborates on the cellular heterogeneity of CRC, revealing that TAMs and CAFs were critical for crosstalk network epithelial cells and TME cells. This in-depth understanding of cancer cell-TME network provided theoretical basis for the development of new drugs targeting this sophisticated network in CRC.

183. IMPACTS OF NEOADJUVANT CHEMORADIOTHERAPY ON THE IMMUNE LANDSCAPE OF ESOPHAGEAL SQUAMOUS CELL CARCINOMA

Abstract Neoadjuvant chemoradiation (neoCRT) followed by surgery is the most common approach for locally advanced resectable esophageal squamous cell carcinoma (ESCC) patients. Tumor microenvironment (TME) has been implicated in tumor response to treatment. Here, we aimed to investigate the impacts of neoCRT on the complex immune landscape in ESCC TME. Single-cell RNA sequencing (scRNA-seq) were conducted to examine the neoCRT-driven cellular and molecular dynamics in ESCC samples obtained before and after neoCRT. Consensus non-negative matrix factorization analysis was applied to decipher coherent gene expression programs of tumor cells. Activities of pathways and transcription factors of individual cells were analyzed by gene set variation and gene regulatory network analyses, respectively. Cell fate tracing was inferred by RNA velocity analysis. Multiplex fluorescent immunohistochemistry was performed to validate the changes of immune cells identified by scRNA-seq in another 25 pairs of pre- and post-neoCRT ESCC samples. Expression programs of cell cycle, adhesion, stress response, and epithelial differentiation in pre-treatment tumor cells related to neoCRT response. Infiltration of CD8+ T cells with exhausted cells dominant increased in both major and minor responders after neoCRT. Treg cell differentiation decreased after neoCRT in major responders but increased in minor ones. NeoCRT promoted the maturation of cDC1s but decreased the number of cDC2s. Expression of M2 macrophage markers increased after neoCRT. Immune-related pathways were enriched at higher levels in CD8+ T cells and macrophages of major responders as compared to minor responders before neoCRT, which was reversed after neoCRT. Our comprehensive picture of the neoCRT-related dynamic immune ecosystem provides deeper insights into immunological mechanisms associated with ESCC response to neoCRT, which may aid in future development of immune-strategies for improving ESCC treatment.

Deciphering the tumor microenvironment cell-infiltrating landscape reveals microenvironment subtypes and therapeutic potentials for nonsquamous NSCLC.

Recent studies highlighted the clinicopathologic importance of the tumor microenvironment (TME) in delineating molecular attributes and therapeutic potentials. However, the overall TME cell infiltration landscape in nonsquamous non-small cell lung cancer (NSCLC) has not been comprehensively characterized. In this study, we used consensus non-negative matrix factorization molecular subtyping to determine TME cell infiltration patterns and identified 3 TME clusters (TME-C1, -C2, -C3) characterized by distinct clinicopathologic features, infiltrating cells, and biological processes. Proteomics analyses revealed that cyclic GMP-AMP-stimulator of interferon genes immune signaling-mediated protein and phosphorylation levels were significantly upregulated in inflammation-related TME-C2 clusters. The score extracted from the TME-related signature (TMEsig-score) divided patients with NSCLC into high- and low-score subgroups, where a high score was associated with favorable prognosis and immune infiltration. The genomic landscape revealed that patients with low TMEsig-score harbored more somatic copy number alterations and higher mutation frequency of driver genes involving STK11, KEAP1, SMARCA4, and others. Drug sensitivity analyses suggested that tumors with high TMEsig-score were responsible for favorable clinical response to immune checkpoint inhibitor treatment. In summary, this study highlights that comprehensive recognizing of the TME cell infiltration landscape will contribute to enhancing our understanding of TME immune regulation and promote effectiveness of precision biotherapy strategies.

Molecular classification of hormone-sensitive and castration-resistant prostate cancer, using nonnegative matrix factorization molecular subtyping of primary and metastatic specimens.

BackgroundDespite the rapidly evolving therapeutic landscape, immunotherapy has demonstrated limited activity in prostate cancer. A greater understanding of the molecular landscape, particularly the expression of immune‐related pathways, will inform future immunotherapeutic strategies. Consensus nonnegative matrix factorization (cNMF) is a novel model of molecular classification analyzing gene expression data, focusing on biological interpretation of metagenes and selecting meaningful clusters.ObjectiveWe aimed to identify molecular subtypes of prostate cancer using cNMF and correlate these with existing biomarkers to inform future immunotherapeutic strategies.MethodsA cohort of archival tumor specimens from hormone‐sensitive and castration‐resistant disease was studied. Whole transcriptomic profiles were generated using TruSeq RNA Access technology and subjected to cNMF. Comprehensive genomic profiling was performed with the FoundationOne assay. NMF subtypes were characterized by gene expression pathways, genomic alterations and correlated with clinical data, then applied to The Cancer Genome Atlas data set.ResultsWe studied 164 specimens, including 52 castration‐resistant and 13 paired primary/metastatic specimens. cNMF identified four distinct subtypes. NMF1 (19%) is enriched for immune‐related and stromal‐related pathways with transforming growth factor β (TGFβ) signature. NMF2 (36%) is associated with FOXO‐mediated transcription signature and AKT signaling, NMF3 (26%) is enriched for ribosomal RNA processing, while NMF4 (19%) is enriched for cell cycle and DNA‐repair pathways. The most common gene alterations included TMPRSS22 (42%), TP53 (23%), and DNA‐repair genes (19%), occurring across all subtypes. NMF4 is significantly enriched for MYC and Wnt‐signaling gene alterations. TMB, CD8 density, and PD‐L1 expression were low overall. NMF1 and NMF4 were NMF2 was associated with superior overall survival.ConclusionsUsing cNMF, we identified four molecularly distinct subtypes which may inform treatment selection. NMF1 demonstrates the most inflammatory signature with asuppressive TGFβ signature, suggesting potential benefit with immunotherapy combination strategies targeting TGFβ and PD‐(L)1. Prospective studies are required to evaluate the use of this novel model to molecularly stratify patients for optimal treatment selection.