Transcriptional Subtypes Research Articles

YAP1 and WWTR1 expression inversely correlates with neuroendocrine markers in Merkel cell carcinoma.

BackgroundMerkel cell carcinoma (MCC) is an aggressive neuroendocrine (NE) skin cancer caused by severe UV-induced mutations or expression of Merkel cell polyomavirus (MCPyV) large and small T antigens (LT and ST). Despite deep genetic differences between MCPyV-positive and -negative subtypes, current clinical diagnostic markers are indistinguishable, and the expression profile of MCC tumors is, to our knowledge, unexplored.MethodsHere, we leveraged bulk and single-cell RNA-Seq of patient-derived tumor biopsies and cell lines to explore the underlying transcriptional environment of MCC.ResultsStrikingly, MCC samples could be separated into transcriptional subtypes that were independent of MCPyV status. Instead, we observed an inverse correlation between a NE gene signature and the Hippo pathway transcription factors Yes1-associated transcriptional regulator (YAP1) and WW domain-containing transcriptional regulator 1 (WWTR1). This inverse correlation was broadly present at the transcript and protein levels in the tumor biopsies as well as in established and patient-derived cell lines. Mechanistically, expression of YAP1 or WWTR1 in a MCPyV-positive MCC cell line induced cell-cycle arrest at least in part through TEA domain-dependent (TEAD-dependent) transcriptional repression of MCPyV LT.ConclusionThese findings identify what we believe to be a previously unrecognized heterogeneity in NE gene expression within MCC and support a model of YAP1/WWTR1 silencing as essential for the development of MCPyV-positive MCC.FundingUS Public Health Service grants R35CA232128, P01CA203655, and P30CA06516.

Tumor-suppressive disruption of cancer subtype-associated super enhancer circuits by small molecule treatment.

Transcriptional cancer subtypes which correlate with traits such as tumor growth, drug sensitivity or the chances of relapse and metastasis, have been described for several malignancies. The core regulatory circuits (CRCs) defining these subtypes are established by chromatin super enhancers (SEs) driving key transcription factors (TFs) specific for the particular cell state. In neuroblastoma (NB), one of the most frequent solid pediatric cancer entities, two major SE-directed molecular subtypes have been described: A more lineage-committed adrenergic (ADRN) and a mesenchymal (MES) subtype. Here, we found that a small isoxazole molecule (ISX), a frequently used pro-neural drug, reprogrammed SE activity and switched NB cells from an ADRN subtype towards a growth-retarded MES-like state. The MES-like state shared strong transcriptional overlap with ganglioneuroma (GN), a benign and highly differentiated tumor of the neural crest. Mechanistically, ISX suppressed chromatin binding of N-MYC, a CRC-amplifying transcription factor, resulting in loss of key ADRN subtype-enriched components such as N-MYC itself, PHOX2B and ALK, while concomitently, MES subtype markers were induced. Globally, ISX treatment installed a chromatin accessibility landscape typically associated with low risk NB. In summary, we provide evidence that CRCs and cancer subtype reprogramming might be amenable to future therapeutic targeting.

Analysis of complement system gene expression and outcome across the subtypes of glioma

Glioma is a malignant and poorly understood cancer that occurs in the glial cells of the brain. Gliomas consist of two clinical subcategories: low-grade glioma (LGG) and glioblastoma (GBM). Based on gene expression profile analysis, gliomas are also divided into four transcriptional subtypes and two mutational subclasses based on IDH expression. These subtypes are associated with differing prognoses, histology, and gene profiles. The complement system, a branch of the innate immune system traditionally associated with inflammation and opsonization, has various pro-tumor effects, including immunosuppression, maintenance of glial stem cells, and hypoxic signaling. We analyzed the expression of complement system genes across the transcriptional and IDH-mutational subtypes of LGG and GBM to determine whether these genes are differentially expressed concerning mutation status and transcriptional subclasses. We performed differential gene expression analysis and analyzed the results for gene set enrichments and correlations with outcome status. The results showed that complement system genes are differentially expressed with varying outcomes across different glioma subtypes. Within the transcriptional subtypes, the complement system genes tended to be overexpressed in subtypes with poor response to treatment or increased tumor malignancy. These results suggest that dysregulation of the complement system may significantly contribute to the categorization of transcriptional subtypes and further, may play a role in treatment response and/or overall patient outcome, although more research is needed to confirm this. These findings could help elucidate the interplay between the immune system, gene expression, and glioma pathogenesis.

Abstract A012: Transcription factor expression repertoire basis for epigenetic and transcriptional subtypes of colorectal cancers

Abstract Colorectal cancers (CRCs) form a heterogenous group classified into epigenetic and transcriptional subtypes. The basis for the epigenetic subtypes, exemplified by varying degrees of promoter DNA hypermethylation, and its relation to the transcriptional subtypes is not well understood. We now link cancer-specific transcription factor (TF) expression dysregulation to methylation alterations near TF-binding sites at promoter and enhancer regions in CRCs and their pre-malignant precursor lesions to provide mechanistic insights into the origins and evolution of the CRC molecular subtypes. A gradient of TF-expression changes forms a basis for the subtypes of abnormal DNA methylation, termed CpG-island promoter DNA methylation phenotypes (CIMP), in CRCs and other cancers. CIMP is tightly correlated with cancer-specific hypermethylation at enhancers, which we term CpG-enhancer methylation phenotype (CEMP). CIMP/CEMP coordination appears to be driven by augmented downregulation of TFs with common binding sites at the hypermethylated enhancers and promoters. The dysregulated expression of TFs related to CIMP/CEMP subtypes occurs early during CRC development, detectable in pre-malignant adenomas. TF-based profiling further identifies patients with worse overall survival. Importantly, altered expression of these TFs discriminates the transcriptome-based consensus molecular subtypes (CMS), thus providing a common basis for CIMP and CMS subtypes. Citation Format: Yuba R. Bhandari, Vinod Krishna, Rachael Powers, Sehej Parmar, Sara-Jayne Thursby, Ekta Gupta, Ozlem Kulak, Prashanth Gokare, Joke Reumers, Liesbeth Van Wesenbeeck, Kurtis E. Bachman, Stephen B. Baylin, Hariharan Easwaran. Transcription factor expression repertoire basis for epigenetic and transcriptional subtypes of colorectal cancers. [abstract]. In: Proceedings of the AACR Special Conference: Cancer Epigenomics; 2022 Oct 6-8; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2022;82(23 Suppl_2):Abstract nr A012.

Abstract A069: Hi-C analysis in patient-derived organoids reveals the impact of three-dimensional chromatin structures on pancreatic cancer transcriptional subtypes

Abstract Three-dimensional (3D) chromatin structures connect genes to regulatory elements, thereby contributing to cell-type-specific gene regulation. However, its role in pancreatic cancer remains unclear. Inspired by our previous finding that open chromatin profiles analyzed by ATAC-seq well characterize the biological features and transcriptional status of pancreatic neoplasms, we hereby demonstrate the previously overlooked impact of 3D epigenetic regulation in pancreatic cancers. Through integrative analysis of Hi-C, ChIP-seq, and RNA-seq using patient-derived organoids, 3D genome modules, including compartments, topologically associating domains, and enhancer-promoter loops were found to play important roles in characterizing pancreatic tumors. Interestingly, intraductal papillary mucinous neoplasm (IPMN) lineages tended to have unique 3D chromatin structure profiles. We further found that transcriptional subtypes in pancreatic cancer, either classical/progenitor-like or basal/squamous-like, were largely underpinned by 3D structures. Compartment was responsible for supporting many subtype-specific genes, by synchronously activating genes that were in the vicinity, as represented by CYP2C family genes. Meanwhile, enhancer-promoter loop fine-tuned some key subtype genes, such as GATA6. By exploring the driving forces behind these 3D structures, we found that HNF1B, a transcription factor implicated in the progenitor subtype, was sufficient to induce subtype switch, together with their underlying 3D genome structure alterations. We demonstrate that HNF1B overexpression in squamous-type pancreatic cancer organoids induced the upregulation and downregulation of genes associated with the progenitor and squamous subtype, respectively. Mechanistically, these dynamic transcriptional and 3D structural modulation were accompanied by HNF1B-induced long-range genomic interactions and H3K27ac modification at HNF1B bound sites, implying HNF1B as a driving force. We also provide evidence suggesting that this HNF1B-induced subtype switch was mediated through the phase-separating property of HNF1B, which was enabled by its intrinsically disordered region (IDR). Induction of mutant HNF1B, whose IDR was either deleted (HNF1B-dIDR) or substituted by transcriptional phase-separation forming IDR from another peptide AKAP95 (HNF1B-AKAP95), showed that subtype switching effects of HNF1B were diminished by HNF1B-dIDR or recapitulated by HNF1B-AKAP95. Thus, HNF1B contributes to the transcriptional subtype formation in pancreatic cancers through its phase-separating structural ability. Collectively, the mapping of 3D structural changes induced by transcription factors, such as HNF1B, is useful for further understanding the features of pancreatic cancers. Our study may imply the potential of 3D structures to effectively suppress or unwind the malignant transformation process and to discover subtype-specific therapeutic targets. Citation Format: Hiroyuki Kato, Dosuke Iwadate, Keisuke Yamamoto, Hiroaki Fujiwara, Hideaki Ijichi, Mitsuhiro Fujishiro, Keisuke Tateishi. Hi-C analysis in patient-derived organoids reveals the impact of three-dimensional chromatin structures on pancreatic cancer transcriptional subtypes [abstract]. In: Proceedings of the AACR Special Conference on Pancreatic Cancer; 2022 Sep 13-16; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2022;82(22 Suppl):Abstract nr A069.

Abstract A003: Developmental parallels in pancreatic cancer subtypes revealed by single cell transcriptomics

Abstract Recent research into the transcriptional subtypes in pancreatic ductal adenocarcinoma (PDA) has demonstrated their importance in predicting survival and therapeutic response. The Basal-like subtype is enriched in metastases and generally predicts poorer prognosis, while the Classical subtype is defined by gene expression reflective of pancreatic progenitors. The two-subtype scheme can be further subdivided into five subtypes that help better reflect the vast transcriptional heterogeneity found in PDA tumors. Basal-like A tumors fare worse than Basal-like B tumors. KRAS wildtype tumors are enriched in Classical B, which have the best prognosis, while Classical A is more reflective of the two-subtype Classical program. Hybrid tumors, generally a mix of Basal-like B and Classical A, are intermediate in prognosis between the Classical and Basal-like subtypes. Despite the importance of the subtypes, we have yet to fully understand how subtype programs within single cells of a tumor are intrinsically regulated, and how they contribute to tumor-level transcriptional observations. To investigate, we generated and leveraged a single cell RNA sequencing dataset enriched for malignant tumor cells spanning both primary and metastatic disease. Along with the richly annotated COMPASS dataset featuring bulk RNA sequencing of over 300 laser captured samples, we developed a single sample classifier utilizing gene pairs for assigning Basal-like A, Basal-like B, Hybrid, Classical A or Classical B to bulk tumors as well as individual cells, allowing interrogation of the subtypes across modalities. We find that the bulk subtype of the tumor is generally reflected in the dominant cell subtype. However, there are frequently cells of other subtypes within individual tumors, and substantial differences in cell subtype composition across tumors of the same subtype. In addition, we find that the proportion of Basal-like cells is a significant risk prognosticator. Investigating deeper into the transcriptional drivers of the subtypes reveals differences in expression of programs linked to endodermal development. Transcription factors such as FOXC1 and IRX3 generally related to foregut development are enriched in Basal-like A, forming a gradient of high expression to lower expression from Basal-like A, Basal-like B, Hybrid, Classical A to Classical B. Interestingly, midgut transcription factors such as CDX2 and ELF3 are expressed in the reverse order. Integrating data from single cell endoderm RNA sequencing further supports this finding, suggesting that the subtypes may arise from tumor cells exploiting pathways normally expressed early in development. This work provides a stepping stone to better understanding the mechanisms behind transcriptional subtypes in PDA. Citation Format: Sabrina Ge, Gun Ho Jang, Michelle Chan-Seng-Yue, Eugenia Flores Figueroa, Karen Ng, Stephanie Ramotar, Anna Dodd, Grainne M. O'Kane, Julie M. Wilson, Jennifer J. Knox, Sandra E. Fischer, Steven Gallinger, Faiyaz Notta. Developmental parallels in pancreatic cancer subtypes revealed by single cell transcriptomics [abstract]. In: Proceedings of the AACR Special Conference on Pancreatic Cancer; 2022 Sep 13-16; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2022;82(22 Suppl):Abstract nr A003.

A Multimodal Single Cell Atlas of Human Tonsils Reveals New Insights into T and B Cell Differentiation

Palatine tonsils are secondary lymphoid organs under constant exposure to antigens via the upper respiratory tract which makes them a compelling organ for adaptive immunity development. They represent an important niche for lymphocyte maturation, where they differentiate into a wide variety of functional states, tailoring immune responses to different insults. However, although some efforts have been made to characterise the lymphocyte maturation process in human tonsils at single cell resolution, these studies lacked cell numbers and multimodal, i.e multiomics, information to comprehensively capture the cellular heterogeneity of the process. Here we generated a comprehensive human tonsil atlas of >357,000 cells profiled across five different data modalities, such as transcriptome, epigenome, proteome, adaptive immune repertoire and spatial location. We identified 121 cell subtypes and states across lineages and characterised the maturation process of both B and T lymphocytes. The high number of profiled cells allowed the detection of rare cell subtypes, such as precursor B and T cells, the latter representing <0.01% of detected cells, supporting an ongoing maturation of lymphoid cells outside the bone marrow and thymus, respectively. Beyond these rare precursor lymphocytes, the tonsil is the niche for multiple steps of mature T and B cell differentiation. With regard to mature T cells, we described 37 different T cell (TC) subpopulations being CD4 TC the more representative population with 58,783 cells, which illustrated the maturation process of this lineage. We identified an early bifurcation following antigen recognition by naïve CD4 TC, resulting in two distinct precursor populations of T follicular helper (fh) and non-Tfh cells. Tfh precursor cells continued their differentiation trajectory towards terminally differentiated populations, with functions in the production of mature B cells (BC) and in the enhancement of antibody secretion. On the other branch of the CD4 TC differentiation trajectory, our data revealed three transcriptional Treg subtypes, with functions in the suppression of antitumor responses, COVID-19 and the control of BC antibody production, respectively. The multimodal study design further enabled the interrogation of regulatory circuits driving TC specialization, describing a potential BCL6 superenhancer specifically active in follicular TC. Intriguingly, the same region has been previously linked to BCL6 gene activation in Germinal Center (GC) BC, indicating its function as follicle-specific rather than cell type-specific cis-regulatory element. Mature B cells corresponded to the greatest population of the dataset, with >70% cells assigned to the BC lineage allowing the fine-grained description of the BC maturation process. We identified 8 distinct naïve BC (NBC) subpopulations and comprehensively characterized the activation pathway upon first antigen recognition, further expanding previous findings focused on MYC. Additionally, by the combination of gene expression and chromatin accessibility profiles we disentangled the transcription factor (TF) hierarchy associated with the Light zone (LZ) to Dark Zone (DZ) reentry. We identified the activity NF-kB family genes as one of the first determinants of the TF hierarchy that drives this specialization, which seems to be shared in LZ cells reentering the DZ as well as activated NBCs that enter the DZ for the first time. However, this regulation was not observed in LZ GCBCs that further differentiate into plasma cells (PCs). This final maturation step towards an antibody secreting phenotype was also characterized, identifying the first stages of PC precursor cells derived from either memory or LZ-GCBC and their transcriptional dynamics was also spatially mapped in the tonsillar tissue. Charting the regulatory landscape in PCs we validated the presence of described PC TFs, but we also discovered SIX5 as a novel potential TF associated with the maturation, but not the commitment, of PCs. Moreover, we identified a higher activity of the regulatory network of SIX5 in multiple myeloma. This study presents a multimodal single cell view of the TC and BC maturation process taking place in the tonsil. We not only present our analysis as a resource but we discover a yet unappreciated cellular heterogeneity as well as novel gene regulatory mechanisms driving naive towards highly specialized lymphoid cells.

High ORM1 Expression Marks a Subset of Genetically Adverse-Risk B-ALL Characterized By MDSC Enrichment, T-Cell Dysfunction, and Inferior Overall Survival

Alterations in the transcriptional activity of certain bone marrow microenvironment populations have been shown to contribute to the pathogenesis of B-lymphoblastic leukemia (B-ALL); however, the precise mechanisms by which neoplastic lymphoblasts shape the gene expression programs of reactive immune and stromal cell populations to support a leukemogenic niche remain incompletely characterized. ORM1 is an acute phase protein with immunomodulatory properties that is overexpressed in a subset of high-risk B-ALL. In an initial exploratory analysis of 1,104 adult and pediatric B-ALL RNA-seq transcriptomes annotated with mature survival outcomes (PMID 30643249), high ORM1 expression was significantly associated with inferior event-free and overall survival and predominantly associated with Ph+, Ph-like, and KMT2A-r B-ALL subtypes. We hypothesized that B-ALL with high ORM1 expression is associated with transcriptionally distinct bone marrow microenvironment immune cell subsets that contribute to its aggressive clinical behavior and sought to delineate these alterations through digital cytometry and analysis of single cell proteogenomic sequencing (CITE-seq) data. To delineate the tumor microenvironment (TME) features of ORM1 high B-ALL through digital cytometry, we constructed a novel bone marrow signature matrix that recapitulates the native BM cellular landscape, including hematopoietic progenitors and mesenchymal stem cells, and performed de novo transcriptional subtype discovery agnostic to clinical outcomes using the Ecotyper platform on a cohort of 577 B-ALL microarray transcriptomes from the MILE study (GSE13159). Discovery analysis identified 51 transcriptionally defined TME cell states across 17 cell types. ORM1 overexpression significantly correlated with transcriptionally-distinct neutrophil, promyelocyte, monocyte, and macrophage subsets enriched for immunosuppressive gene expression programs as well as CD8+ and γδ T-cell subsets with hallmark transcriptional features of effector dysfunction and exhaustion. To validate the imputed cell states at single-cell resolution and define their immunophenotypic features, we drew upon several B-ALL CITE-seq datasets. Recovery and assignment of discovery cohort transcriptional states to single-cell transcriptomes in KMT2A-r and Ph+ cases confirmed enrichment for multiple myeloid-derived suppressor cell (MDSC) and dysfunctional T-cell transcriptional subsets. MDSCs were predominantly monocytic but also included neutrophilic promyelocytes with surface expression of CD123 and CD155/PVR, which interacts with TIGIT to suppress immune cell activation. Indeed, enrichment for CD8+ cytotoxic lymphocytes with surface expression of PD-1 and TIGIT was also present. Neutrophilic MDSCs were present at diagnosis, in pre-relapse remission samples and at the time of relapse, suggesting that their persistence following frontline therapy may facilitate the immune evasion and expansion of treatment-resistance leukemic clones that drive eventual relapse. Taken together, high ORM1 expression marks a subset of high-risk B-ALL characterized by enrichment for transcriptionally-distinct immunosuppressive TME cell states which may be targetable with emerging therapeutic strategies aimed at restoring host anti-tumor immunity.

Abstract B075: Sirtuin 6 histone deacetylase is required for the integrated stress response and resistance to inhibition of transcriptional cyclin dependent kinases in Pancreatic Ductal Adenocarcinoma

Abstract Pancreatic Ductal Adenocarcinoma (PDA) can be characterized by two distinct transcriptional subtypes: Classical and Basal-like. The basal PDA subtype is more aggressive and has the worst overall survival. Previous work has shown that Sirtuin 6 histone deacetylase (SIRT6) acts as a tumor suppressor and cooperates with oncogenic KRAS in GEMM models of PDA. Our study identifies SIRT6 as a biomarker for basal PDA and investigates the underlying sensitivity of basal PDA to inhibition of transcriptional cyclin dependent kinases (CDKs). Through analysis of data from hundreds of human PDA tumors, we identified that low SIRT6 expression correlates with basal PDA while the converse is true for classical PDA. Using many in vitro drug assays, genetic manipulation experiments, and in vivo drug studies in human patient-derived xenografts of PDA, we also discovered that basal PDA is uniquely sensitive to the covalent CDK7/12/13 inhibitor THZ1 as well as other inhibitors of transcriptional CDKs. Importantly, removal of SIRT6 can sensitize SIRT6 high/classical PDA to THZ1. Further investigation identified that SIRT6 low/basal PDA has an inactivated Integrated Stress Response (ISR), which is primarily driven by ATF4. To determine how loss of SIRT6 leads to an inability to activate the ISR, we have integrated use of human PDA cell lines, and organoids as well as murine PDA GEMMs where SIRT6 has been deleted or downregulated. We uncovered that SIRT6 regulates the ISR through control of ATF4 translation, specifically through a 3’ UTR open reading frame (uORF) dependent mechanism. To determine whether ISR activation could protect basal PDA cells from transcriptional CDK inhibition, we manipulated expression of ISR pathway components. We found that activation of the ISR in SIRT6 low/basal PDA reduces sensitivity to THZ1, while inactivation of the ISR sensitizes SIRT6 high/classical PDA to this inhibitor. Interestingly, both basal and classical PDA exhibit increased phosphorylation of eIF2a upon inhibition of transcriptional CDKs, which typically indicates activation of the ISR in response to stress. However, this activation of the ISR remains incomplete since ATF4 translation is not upregulated in basal PDA in the absence of SIRT6. Therefore, we discovered that loss of SIRT6 sensitizes basal PDA to inhibition of transcriptional CDKs through an inability to launch a functional ISR response. Furthermore, we have uncovered an important biomarker which controls a stress-induced transcriptional program that may be exploited with targeted therapies in a particularly aggressive PDA subtype. Citation Format: Jessica Gianopulos, Nithya Kartha, Zachary Schrank, Stephanie Dobersch, Sarah Cavender, Bryan Kynnap, Adrianne Wallace-Povirk, Cynthia Wladyka, Juan Santana, Jaeseung C. Kim, Angela Yu, Caroline Bridgwater, Kathrin Fuchs, Sarah Dysinger, Aaron Lampano, Faiyaz Notta, David Price, Andrew Hsieh, Sunil Hingorani, Sita Kugel. Sirtuin 6 histone deacetylase is required for the integrated stress response and resistance to inhibition of transcriptional cyclin dependent kinases in Pancreatic Ductal Adenocarcinoma [abstract]. In: Proceedings of the AACR Special Conference on Pancreatic Cancer; 2022 Sep 13-16; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2022;82(22 Suppl):Abstract nr B075.

Abstract B010: Investigating endoplasmic reticulum stress responses in pancreatic ductal adenocarcinoma under lipid imbalance

Abstract Pancreatic ductal adenocarcinoma (PDAC) is the 4th leading cause of cancer-related death with a five-year survival rate of 10%. The complex genetic makeup and heterogeneity of its microenvironment make PDAC difficult to treat. Desmoplasia, a major feature of PDAC’s TME, are dense fibrotic and inflammatory stromal regions. Desmoplastic reaction results in poor intratumoral perfusion and elevated interstitial fluid pressure, creating regions with severe oxygen and nutrient deprivation. One result is a deficiency in unsaturated fatty acids (uFAs) because the de novo synthesis of uFAs requires oxygen. Our results indicate that PDAC cells experience endoplasmic reticulum (ER) stress and trigger subsequent ER stress responses under uFA limited conditions in vitro. The ER stress responses are governed by three major stress sensors: Inositol-requiring enzyme 1𝛼 (IRE1𝛼), PKR-like ER kinase (PERK) and activating transcription factor 6𝛼 (ATF6𝛼). We hypothesize that ER stress responses are tumor promoting in PDAC and used as a survival strategy under uFA deprivation. In addition, previous studies showed that MYC-driven cancer cells are highly reliant on the ER stress response pathways. Therefore, we hypothesize that the two major PDAC transcriptional subtypes “Classical” and “Basal-like” have distinct dependencies on ER stress response pathways, with the basal-like subtype being more dependent due to an enrichment of MYC gene signatures. Our results suggest that opposite to cancer cell promoting, ER stress responses mediated by IRE1𝛼, PERK and ATF6𝛼 are cytotoxic in vitro under uFA deprived conditions. By analyzing the published bulk RNA-seq from patient samples, we found that the basal-like PDAC subtype is enriched with an ER stress gene signature. To avoid contamination from other cell populations, we are currently analyzing a published scRNA-seq dataset from 15 independent patient samples. We will investigate whether the ER stress signature is also enhanced in basal-like cells. Basal-like PDAC cells are usually more mesenchymal; we therefore created isogenic “epithelial” and “mesenchymal” lines to mimic PDAC subtypes. However, we did not see signaling preferences between the lines. One caveat is that the basal-like subtype have many more features than being mesenchymal, and our isogenic system might not be a faithful modeling of the subtypes. We are studying whether hypoxic PDAC cells experience ER stress in vivo due to a deprivation in uFAs using multiplex fluorescent imaging, stimulated Raman microscopy and/or mass spectrometry. We are evaluating the role of ER stress responses in vivo and see whether inhibiting ER stress response alone or in combination with other reagents gives rise to potential therapeutic opportunities. This project will provide a better understanding of the role ER stress responses play under uFA deprivation and will also enable us to further understand PDAC subtypes in terms of how they utilize ER stress responses. Citation Format: Yanqing (Christine) Jiang, Xu Han, Mai Wang, Raymond Ng, Pamela Burgess-Jones, John Tobias, Nathan Coffey, Andrew Hu, Brian Keith, Celeste Simon. Investigating endoplasmic reticulum stress responses in pancreatic ductal adenocarcinoma under lipid imbalance [abstract]. In: Proceedings of the AACR Special Conference on Pancreatic Cancer; 2022 Sep 13-16; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2022;82(22 Suppl):Abstract nr B010.

Abstract B031: Loss of GATA6 leads to reduced cell fitness and PDA molecular subtype changes

Abstract Background: Pancreatic ductal adenocarcinoma (PDA) is defined by two main transcriptional subtypes, Basal-like &amp; Classical. Classical tumors are more frequent and tend to respond better to chemotherapy with higher expression of genes related to pancreatic lineage development, the epithelial compartment, and adhesion. Conversely, Basal-like tumors are more aggressive, therapy resistant and have higher expression of laminins and keratins. The Classical subtype correlates with GATA6 expression, and ~16% pancreatic tumors show amplification in the transcription factor GATA6 suggesting its importance in the Classical subtype and in tumor survival (Chan-Seng-Yue et al., 2020). The goal of this project is to elucidate the role of GATA6 expression in PDA tumors showing that GATA6 is a key molecular regulator of the Classical subtype influencing tumor growth, maintenance, and phenotype. Methods: Patient derived organoids (PDOs) spanning the spectrum of GATA6 expression were used with CRISPR/Cas9 to establish GATA6 KO lines (n=8). These were then subject to a growth competition assay between cells with and without GATA6 KO and the amount of knockout over time was monitored by Sanger sequencing. Tumor growth curves were generated in a subcutaneous xenograft mouse model over 1.5 months to assess the in vivo growth potential of GATA6 KO cells in a sample with high GATA6 expression. 3’RNA-seq was used to assess gene expression changes in subtype, and GSEA was used to assess pathway enrichment. This was done alongside IHC, capillary Westerns, and flow cytometry to assess changes that GATA6 KO incurs on a functional level. Results: GATA6 KO tended to result in reduced fitness (dropout) in cells when ablated, although the rate and level of dropout varied, suggesting different levels of dependence on GATA6 expression for survival/growth. This was confirmed in vivo when the GATA6 KO failed to grow as robustly as control groups in a sample with high native GATA6 expression. Molecular analysis of all PDO samples showed shifts in expression from Classical to Basal-like. This shift was denoted in changes in subtype score, changes in marker gene expression (i.e., ANXA10 &amp; TP63), and in changes in key pathways associated with each subtype. This shows that GATA6 not only correlates with the Classical subtype but plays a causal role in maintaining it. Even though there was a shift towards Basal-like expression several key characteristics of the subtype were notably absent including enrichment for epithelial to mesenchymal transition, down regulation of MYC and its targets, and the lack of aggressive growth in in vitro and in vivo models (dropout of GATA6 KO cells, low Ki67 expression, and upregulation of apoptosis). This suggests that KO of GATA6 is not sufficient in of itself to cause a true Basal-like phenotype and that instead GATA6 KO might lead to an in-between state that does not quite offer the same growth advantages and may open GATA6 and its pathways to be a potential therapeutic target. Citation Format: Tristan S.D. Woo, Kena F. Figueroa, Sabrina Ge, Irene Xie, Karen Ng, Pinjiang Cao, Julie M. Wilson, Stephanie Ramotar, Jennifer J. Knox, Grainne M. O'Kane, Steven Gallinger, Faiyaz Notta. Loss of GATA6 leads to reduced cell fitness and PDA molecular subtype changes [abstract]. In: Proceedings of the AACR Special Conference on Pancreatic Cancer; 2022 Sep 13-16; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2022;82(22 Suppl):Abstract nr B031.

High-Plex Imaging and Cellular Neighborhood Spatial Analysis Reveals Multiple Immune Escape and Suppression Patterns in DLBCL

Background: DLBCL is a heterogeneous disease of malignant B cells and tumor microenvironment (TME). As a malignancy arising from the immune system, DLBCL tumor cells are known to deploy immune escape and/or suppression strategies to evade immunosurveillance. A deep understanding of the interactions between tumor and its TME is critical to any enduring treatment outcome. Here, we systematically characterize the spatial patterns of major tumor infiltrating immune cells in DLBCL using high-plex imaging data, and characterize them in known high-risk patient groups such as EFS24 failures, ABC subtype, double-hit, and a high risk transcriptional subtype A7 (Stokes submitted). Methods: We generated multiplexed ion beam imaging (MIBI) data for two cohorts (cohort 1 and cohort 2) of newly-diagnosed DLBCL cases (N=55 and N=30) with 17 and 33 markers, respectively. Both cohorts had associated RNAseq and FISH data, enabling determination of molecular high-risk patients. The first cohort also contained MIBI imaging for N=5 normal lymph node (NLN) samples, while the second cohort had associated progression-free survival (PFS) data with R-CHOP treatment. The imaging analysis identified major immune cell phenotypes and their relative abundances. Largely following the methods of Bhate 2022, we calculated cellular neighborhoods (CNs) consisting of each cell and its 20 nearest neighbors for all 532,755 and 643,114 cells in the two cohorts, respectively. We clustered the cellular composition of the CNs to identify characteristic CN clusters for each cohort, and investigated the spatial arrangement, or context (SC) of these CNs, and the CN cluster-specific expression of certain markers and their association with ABC subtype, DHIT, and A7. For cohort 2, we also performed univariate Cox proportional hazard modeling with PFS data. Results: Seven unique CN clusters were identified, each enriched with different types of infiltrating immune cells, which we named with ascending tumor content as: CD4-rich (6.3% tumor), CD8-rich (21%), M1/DC-rich (33%), Tumor/DC mix (46%), M2-rich (56%), Tumor/T mix (59%), and Tumor-rich (87%). These clusters, initially identified in cohort 1, were largely replicated in cohort 2. When contrasting ABC vs. GCB subtypes, we find that ABC has fewer Tumor-rich (p = 0.01) but higher abundance of M2-rich CNs (p=0.043). These M2-rich CNs, which are enriched with immune suppressive CD163+ macrophages, are present only in DLBCL and not in NLN samples, indicating a role of suppressive macrophages in ABC that is previously under-appreciated. CD163+ macrophages are associated with poor outcome in the 2ndcohort (HR=4.4 +/- 0.61, p = 0.015), confirming similar findings in literature (Bohn 2012, Nam 2014, Reiss 2019). As expected, tumor-rich CNs are associated with PFS24 failures (p = 0.026), and in DHIT (p = 0.0023), and A7 (p < 0.001) samples, in both cohorts. A7 samples lacked the two T-cell rich CNs, consistent with its low immune infiltration characteristic (Stokes submitted). Interestingly, the abundance of tumor/DC-mix CNs is also associated with A7 high-risk patients in both cohorts (p < 0.01). The two T cell-rich CNs were least enriched for tumor cells, indicating a structure of T cell exclusion from tumor-rich regions in DLBCL. We identified a SC motif of "T-cell islands" where CD4-rich CNs are enclosed by CD8-rich CNs, which are adjacent to a layer of tumor/T mix CNs and/or M2-rich CNs, followed by tumor-rich CNs (example in Figure 1B). We believe that this layered structure serves as an immune escape strategy, whereby CD4+ T helper cells are excluded from direct contact with tumor cells, which may reduce their antigen presentation capacity; while CD8-rich CN adjacency with M2-rich CNs could result in immunosuppression of cytotoxic CD8+ T cells. Among our other findings in cohort 2, infiltrating CXCR5+ CD8+ T cells are prognostic (PFS HR=0.23; p=0.021), and are highly enriched among the three highest tumor-rich CNs (p < 10-6). Conclusions: Using high-plex imaging and cellular neighborhood spatial analysis we have uncovered multiple spatial patterns between DLBCL tumor cells and their immune cell neighbors, which indicate that tumors exploit multiple mechanisms of immune escape or suppression. Detailed understanding of these mechanisms may provide new insights into tumor-TME interactions that will be useful for improving immune-oncology and combination strategies in DLBCL. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal

EPCO-15. LINKING HISTOLOGICAL GLIOBLASTOMA PHENOTYPES TO TRANSCRIPTIONAL SUBTYPES AND PROGNOSIS USING DEEP LEARNING

Abstract Deep learning is increasingly utilized in medical image analysis and shows promise to substantially inform treatment decisions in clinical practice. While it finds broad applications in radiology or genomics, there are still relatively few applications of deep learning to the analysis of histological images. A disease in which a better understanding of underlying phenotype-genotype correlations seems crucial is glioblastoma, the most common malignant brain tumor in adults that features a dismal prognosis and limited therapeutic options. Using a large cohort of 276 glioblastoma patients with available histological, immunohistochemical, clinical and molecular data, we trained an Xception convolutional neural network to predict patient survival and transcriptional subtypes on histological images alone. We used 5-fold cross validation to evaluate the performance of the trained network. We achieve a high accuracy for predicting overall survival on univariable (p &amp;lt; 0.001) and multivariable (p = 0.01) analysis. Moreover, we observe a high accuracy for predicting the predominant transcriptional subtype with an AUC of 0.84, 0.81, and 0.71 for classical, mesenchymal and proneural subtypes, respectively (all p &amp;lt; 0.001). Finally, we plot heatmaps of the predicted transcriptional subtype and risk scores to directly link them to the histological phenotypes and the tumor microenvironment. Reassuringly, we find an association of the mesenchymal subtype with the perinecrotic niche and of the proneural subtype with the infiltrating edge. In conclusion, we introduce an easily adaptable convolutional neural network model for working with histological data, which can accurately predict outcomes and transcriptional subtypes in glioblastoma.

TMET-19. DEFINING A NOVEL ROLE FOR CD97 IN REGULATING GBM GLYCOLYTIC METABOLISM

Abstract Glioblastoma (GBM) is the most common and aggressive primary brain malignancy. Despite multimodal therapy, disease recurrence is inevitable. To identify novel vulnerabilities of GBM, we performed an arrayed CRISPR/Cas9 screen against select adhesion G protein-coupled receptors (aGPCRs), many of which we found to be de novo expressed in GBM. Knockout of CD97, previously implicated in GBM cell migration, produced the most striking proliferative disadvantage in patient-derived GBM cultures (PDGC) among aGPCRs tested. We found high CD97 surface expression in all our PDGCs, while levels remained nearly undetectable in non-neoplastic brain cells, confirming that CD97 is de novo expressed in GBM. Upon shRNA-mediated knockdown of CD97 in PDGCs from all three TCGA transcriptional subtypes, we observed reduced proliferation, as measured by cell cycle analysis. Notably, CD97 knockdown also significantly reduced tumorsphere formation capacity as measured by limiting dilution assays; an effect that was partially rescued upon CD97 overexpression. To elucidate mechanisms of action of CD97, we performed RNA-sequencing and GO pathway enrichment analysis from PDGCs following CD97 knockdown. The top downregulated pathways involved glycolytic metabolism, specifically involving many genes relevant for glucose-6-phosphate (G6P) and fructose-6-phosphate (F6P) processing. Indeed, when we measured metabolite levels under both steady-state and flux conditions using mass spectrometry, we observed an accumulation of G6P and a depletion of most downstream glycolytic and Krebs cycle metabolites upon CD97 knockdown. Furthermore, Seahorse metabolic assays revealed deficits in both glycolytic metabolism and oxygen consumption. We aim to interrogate the activity of specific glycolytic enzymes involved in processing G6P and F6P, pinpointing how these are influenced by CD97 signaling pathways (MAPK or Akt). Overall; our studies suggest a novel role of CD97 in regulating GBM metabolism (Warburg effect), and provide a strong scientific rationale for developing biologics to target CD97 which appears to be universally and de novo expressed in GBM.

Niche deconvolution of the glioblastoma proteome reveals a distinct infiltrative phenotype within the proneural transcriptomic subgroup

Glioblastoma is often subdivided into three transcriptional subtypes (classical, proneural, mesenchymal) based on bulk RNA signatures that correlate with distinct genetic and clinical features. Potential cellular-level differences of these subgroups, such as the relative proportions of glioblastoma’s hallmark histopathologic features (e.g. brain infiltration, microvascular proliferation), may provide insight into their distinct phenotypes but are, however, not well understood. Here we leverage machine learning and reference proteomic profiles derived from micro-dissected samples of these major histomorphologic glioblastoma features to deconvolute and estimate niche proportions in an independent proteogenomically-characterized cohort. This approach revealed a strong association of the proneural transcriptional subtype with a diffusely infiltrating phenotype. Similarly, enrichment of a microvascular proliferation proteomic signature was seen within the mesenchymal subtype. This study is the first to link differences in the cellular pathology signatures and transcriptional profiles of glioblastoma, providing potential new insights into the genetic drivers and poor treatment response of specific subsets of glioblastomas.

Abstract B027: The development of a multiscale transcriptional atlas of sarcoma

Abstract Objectives: Sarcomas are mesodermal cancers of bone and soft tissue of which there are &amp;gt;60 malignant varieties, many of which can be difficult to diagnose or subtype using traditional histopathology. A universal molecular definition of sarcoma types would therefore be an invaluable tool to the diagnostic pathologist. RNA has the potential to offer a complimentary perspective to cytogenetic- and methylation-based diagnostics, as it represents the active state of the disease at sampling and better reveals its phenotype. Recognizing the potential for RNA-based classification, we set out to create a first-generation transcriptional atlas of sarcoma. Methods: To develop transcriptional definitions of cancers with the potential to further subclassify tumor types, we designed a self-optimizing and scale-adaptive unsupervised method (RACCOON), which groups samples into hierarchically organized clusters. We used this approach on the UCSC Treehouse Childhood Cancer Compendium, a set of 2,178 pediatric and 9,400 adult tumors, 1,130 of which are sarcomas, as well as 1,735 non-neoplastic samples. We then trained an ensemble of convolutional neural networks to classify tumors to these transcriptional clusters. We have now added 624 more sarcoma samples from Toronto centers and international collaborators to better represent the breadth of sarcoma. We are actively sequencing 500 additional samples in partnership with the Gabriella Miller Kids First Research Program to yield an expanded cohort of &amp;gt;2,200 uniformly processed and analyzed sarcomas. Results: Sarcomas organize into two clusters at the highest hierarchical level: one characterized by entities which occur primarily in adults and resemble mature tissue, the other by primarily pediatric entities which exhibit high stemness and resemble embryonic tissue. Several included entities are not bona fide sarcomas but originate from the mesoderm (e.g., Wilms Tumor) signifying a common transcriptional identity for mesodermal neoplasms. Additionally, we demonstrate the first transcriptional subtypes of central osteosarcoma reflecting its major histotypes and representing divergent clinical courses. We also determine Ewing Sarcoma (ES) to be a distinct entity which clusters separately from all other cancers, raising questions of its origin and affinity to sarcoma. When classifying ongoing patients to the atlas, we correctly classified &amp;gt;85% of tumors and corrected the diagnosis of 7%. We find 14% of ES in our dataset were likely misdiagnosed CIC- or BCOR-driven sarcomas. Critically, assigned subtypes are consistent between primary and relapse pairs. Conclusion: RNA-seq is a promising tool for both subtype discovery and classifying sarcoma in ongoing patients. We have already included this tool in tumor boards to help inform patient care. Our method reveals the overarching organization of sarcoma for the first time and specifies its underlying biology. This atlas is ever-growing and is open to the community to contribute. Citation Format: Joshua O. Nash, Federico Comitani, Rose Chami, Sarah Cohen-Gogo, Astra Chang-Schwertschkow, Yael Babichev, Jodi Lees, Noa Alon, Nalan Gokgoz, Stephen Man Yu, Kyoko Yuki, Miranda Lorenti, Zhanqin Liu, Alaina McGoey, Famida Spatare, Bernarld Castro, Kim Tsoi, Hagit Peretz Soroka, Jack Brzezinski, Anita Villani, Albiruni Razak, Abha Gupta, Elizabeth Demicco, Gino Somers, Brendan C. Dickson, Jay S. Wunder, Irene L. Andrulis, David Malkin, Rebecca A. Gladdy, Adam Shlien. The development of a multiscale transcriptional atlas of sarcoma [abstract]. In: Proceedings of the AACR Special Conference: Sarcomas; 2022 May 9-12; Montreal, QC, Canada. Philadelphia (PA): AACR; Clin Cancer Res 2022;28(18_Suppl):Abstract nr B027.

KS01.4.A Transcriptional evolution of glioblastoma point towards changes in bulk composition, mesenchymal sub-type as end-state, and a prognostic association with increased extracellular matrix gene expression

Abstract Background Glioblastoma is the most prevalent and severe type of malignant brain tumor in adults. Although the genetic make-up initiating glioblastoma is increasingly better understood, a better understanding in the mechanisms that drive its evolution, heterogeneity and therapy resistance may reveal new directions for therapy development. To get better insights into glioblastoma evolution, we analyzed and deconvoluted transcriptomes of primary and recurrent glioblastoma resections. Material and Methods Matching primary and secondary resections from n=185 uniformly treated glioblastoma patients were collected as part of EORTC Study 1542 and RNA-sequenced. Data was extended with pairs from n=51 patients from the GLASS study. The datasets were subjected to differential and deconvolution analysis using in-house algorithms. Results When projecting the tumor samples into a reduced Glioblastoma Intrinsic Transcriptional Subtype space, visualization of transitions indicated that the CL subtype switches most often. As we found no more transitions from MES to other subtypes than to be expected by chance, we concluded that MES is an end-state. On average, tumor purity percentages decreased from ~67% to ~50%, mostly due to an increase in macrophages/microglia. Differential expression analysis was performed with correction for the fraction of non-malignant cells. While expression of glioblastoma associated oncogenes did not change significantly over time, marker genes for macrophages/microglia, neurons and oligodendrocytes were up-regulated whereas endothelial cell markers were down-regulated. A cluster of ~30 extracellular matrix associated (ECM) genes increased significantly over time. Single cell RNA-seq, IF-staining and RNA-ish indicated the signature is most strongly expressed near intra-tumoral vessels. Since endothelial marker genes were down-regulated over time, this suggests a form of prognostic vessel progression with a representative transcriptome signature. Conclusion Using a large cohort and validation set of uniformly treated patients, we demonstrate how the glioblastoma transcriptome changes over time with in particular changes the composition of the tumor and its environment. The tumor purity decrease over times suggests a more invasive phenotype or recruitment of non-malignant cells or a combination of both. A post-progression increase in expression of ECM-associated genes expressed near blood vessels was associated with poor survival. Concluding, while no consistent path for transcriptional evolution of tumor cells was observed other than transitions to the MES subtype, glioblastoma becomes more aggressive in time by (re-)organizing its environment.

Archetype tasks link intratumoral heterogeneity to plasticity and cancer hallmarks in small cell lung cancer.

Small cell lung cancer (SCLC) tumors comprise heterogeneous mixtures of cell states, categorized into neuroendocrine (NE) and non-neuroendocrine (non-NE) transcriptional subtypes. NE to non-NE state transitions, fueled by plasticity, likely underlie adaptability to treatment and dismal survival rates. Here, we apply an archetypal analysis to model plasticity by recasting SCLC phenotypic heterogeneity through multi-task evolutionary theory. Cell line and tumor transcriptomics data fit well in a five-dimensional convex polytope whose vertices optimize tasks reminiscent of pulmonary NE cells, the SCLC normal counterparts. These tasks, supported by knowledge and experimental data, include proliferation, slithering, metabolism, secretion, and injury repair, reflecting cancer hallmarks. SCLC subtypes, either at the population or single-cell level, can be positioned in archetypal space by bulk or single-cell transcriptomics, respectively, and characterized as task specialists or multi-task generalists by the distance from archetype vertex signatures. In the archetype space, modeling single-cell plasticity as a Markovian process along an underlying state manifold indicates that task trade-offs, in response to microenvironmental perturbations or treatment, may drive cell plasticity. Stifling phenotypic transitions and plasticity may provide new targets for much-needed translational advances in SCLC. A record of this paper's Transparent Peer Review process is included in the supplemental information.

Correlation of transcriptional subtypes with a validated CT radiomics score in resectable pancreatic ductal adenocarcinoma.

Transcriptional classifiers (Bailey, Moffitt and Collison) are key prognostic factors of pancreatic ductal adenocarcinoma (PDAC). Among these classifiers, the squamous, basal-like, and quasimesenchymal subtypes overlap and have inferior survival. Currently, only an invasive biopsy can determine these subtypes, possibly resulting in treatment delay. This study aimed to investigate the association between transcriptional subtypes and an externally validated preoperative CT-based radiomic prognostic score (Rad-score). We retrospectively evaluated 122 patients who underwent resection for PDAC. All treatment decisions were determined at multidisciplinary tumor boards. Tumor Rad-score values from preoperative CT were dichotomized into high or llow categories. The primary endpoint was the correlation between the transcriptional subtypes and the Rad-score using multivariable linear regression, adjusting for clinical and histopathological variables (i.e., tumor size). Prediction of overall survival (OS) was secondary endpoint. The Bailey transcriptional classifier significantly associated with the Rad-score (coefficient = 0.31, 95% confidence interval [CI]: 0.13-0.44, p = 0.001). Squamous subtype was associated with high Rad-scores while non-squamous subtype was associated with low Rad-scores (adjusted p = 0.03). Squamous subtype and high Rad-score were both prognostic for OS at multivariable analysis with hazard ratios (HR) of 2.79 (95% CI: 1.12-6.92, p = 0.03) and 4.03 (95% CI: 1.42-11.39, p = 0.01), respectively. In patients with resectable PDAC, an externally validated prognostic radiomic model derived from preoperative CT is associated with the Bailey transcriptional classifier. Higher Rad-scores were correlated with the squamous subtype, while lower Rad-scores were associated with the less lethal subtypes (immunogenic, ADEX, pancreatic progenitor). • The transcriptional subtypes of PDAC have been shown to have prognostic importance but they require invasive biopsy to be assessed. • The Rad-score radiomic biomarker, which is obtained non-invasively from preoperative CT, correlates with the Bailey squamous transcriptional subtype and both are negative prognostic biomarkers. • The Rad-score is a promising non-invasive imaging biomarker for personalizing neoadjuvant approaches in patients undergoing resection for PDAC, although additional validation studies are required.

Identification of Crucial Gene Modules Related to the Efficiency of Anti-PD-1/PD-L1 Therapy and Comprehensive Analyses of a Novel Signature Based on These Modules.

Biomarker development for clinical checkpoint inhibition is still in its early stages. It is critical to determine the cause of the lack of a long-term response in patients after immune checkpoint blockade (ICB) treatment and to develop composite biomarkers or signatures to improve personalized approaches. Three modules that were significantly correlated with the immunotherapeutic response were identified. Stimulatory pathways of cellular immunity, extracellular matrix formation-related pathways, and ATP metabolism-related pathways were enriched. Two distinct transcriptional subtypes were determined. Tumor microenvironment (TME) characteristics were highly correlated with “hot” and “cold” tumors. The ICB score was significantly correlated with clinical characteristics including age, Breslow depth, Clerk level, AJCC stage, and T stage. Meanwhile, a low ICB score is characterized by increased activation of immunity, a higher level of immune infiltration, and immune molecule expression. The ICB score showed a robust ability to predict melanoma prognosis in the discovery, internal validation, and external validation cohorts. In addition, a low ICB score was linked to a higher CR/PR rate in the immunotherapeutic cohort. The ICB score could reflect the pre-existing immune features and the expression pattern of “Cold” versus “Hot” tumors in melanoma patients. Thus, it has the potential to serve as a reliable predictor of melanoma prognosis and response to ICB therapy.