Search

Recent advancements in molecular profiling have revealed distinct breast cancer subtypes, but many clinical NGS assays rely on gene panels, such as PAM50, limiting their clinical utility. Basal-like breast cancer (BLBC), one of the most aggressive subtypes, has highly variable molecular and clinical characteristics. Now, the tumor microenvironment (TME) is recognized as a vital participant in tumor progression and therapeutic response. The development of more refined classifications based on the TME, capable of accounting for tissue heterogeneity, may improve NGS clinical utility for BLBC. Here, we apply our transcriptomic-based approach, recently described by Bagaev et. al., to classify the BLBC TME into discrete immune portraits, to potentially improve clinical outcomes and facilitate therapeutic decisions. We collected a cohort of 1,708 BLBC samples based on the expression levels of 50 genes (PAM50) from 10 publicly available datasets, with clinical outcomes available (n = 819). Using methodology described by Bagaev et. al., 31 functional gene expression signatures (Fges) were selected, and unsupervised dense Louvain clustering was performed to identify TME subtypes. A novel RNA-seq deconvolution algorithm was used to determine the cell types within the TME. Validation of the histological features, including stroma, tumor infiltrating lymphocytes (TILs), and tertiary lymphoid structures (TLS), relative to gene expression patterns in the TME subtypes was performed by automated and manual annotation of BLBC H&E slides (n = 146) from an independent TCGA cohort. Overall survival (OS) analysis was performed using Kaplan-Meier and Cox regression methods. We revealed 5 BLBC subtypes with distinct expression patterns: immune-enriched, non-fibrotic (IE, 19%), B-cell–enriched, TLS-like (TLS, 25.5%), granulocyte-enriched (G, 12.8%), fibrotic (F, 28%), and immune desert (D, 17.7%) (Table). IE tumors featured an active immune TME, with high immune checkpoint expression and T cell activity. The TLS subtype also had an immune-rich TME, presenting high levels of B cells, T helper cells, and TLS (p < 0.001). The TLS subtype exhibited the highest number of stromal TILs on TCGA H&E slides. The G subtype was characterized by high expression of granulocytes and granulocyte traffic molecules. In accordance with our findings, deconvolution predicted the highest percent of neutrophils in the G subtype (p < 0.001). The F subtype demonstrated the highest levels of angiogenesis, stromal Fges, and VEGFR1-3, FGFR1, and EGFR expression. By histological evaluation, 84% of F subtype samples demonstrated a medium or high level of fibrosis. The D subtype showed a high proliferation rate and low stromal and immune Fges. Indicative of proliferation rate, CCNB1 and cyclin B1 were highest in G, D, and IE subtypes. OS analysis revealed a significant association between TME subtypes (TLS = baseline; log HR G = 0.87, p < 0.05; IE = 0.39, p = 0.18; F = 0.99, p < 0.05; D = 1.21, p < 0.05), and survival outcomes. The immune-enriched subtypes, IE and TLS, demonstrated good prognosis and higher expression of immune checkpoint genes, while immune desert D and granulocyte-enriched G subtypes exhibited the worst OS. Using our transcriptomic-based approach, BLBC was classified into 5 distinct subtypes, each with unique therapeutic vulnerabilities. Further investigation of these TME subtypes may lead to potential clinical utility as a prognostic tool to improve clinical decision making. Table. Characteristics of Basal-like breast cancer (BLBC) tumor microenvironment (TME) subtypes. TLS - Tertiary lymphoid structures; CAF - cancer-associated fibroblasts Citation Format: Svetlana Khorkova, Diana Shamsutdinova, Vladimir Kushnarev, Lev Popyvanov, Daniil Dymov, Anastasia Zotova, Ivan Valiev, Zoya Antysheva, Anna Love, Jessica H. Brown, Alexander Bagaev, Nikita Kotlov, Nathan Fowler. A molecular classification system for basal-like breast cancer based on the tumor microenvironment is prognostic for survival [abstract]. In: Proceedings of the 2022 San Antonio Breast Cancer Symposium; 2022 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2023;83(5 Suppl):Abstract nr P4-09-02.

Follicular lymphoma (FL) is a B-cell malignancy with a complex tumor microenvironment that is rich in nonmalignant immune cells. We applied single-cell RNA sequencing to characterize the diverse tumor and immune cell populations of FL and identified major phenotypic subsets of FL T cells, including a cytotoxic CD4 T-cell population. We characterized four major FL subtypes with differential representation or relative depletion of distinct T-cell subsets. By integrating exome sequencing, we observed that somatic mutations are associated with, but not definitive for, reduced MHC expression on FL cells. In turn, expression of MHCII genes by FL cells was associated with significant differences in the proportions and targetable immunophenotypic characteristics of T cells. This provides a classification framework of the FL microenvironment in association with FL genotypes and MHC expression, and informs different potential immunotherapeutic strategies based upon tumor cell MHCII expression. Significance: We have characterized the FL-infiltrating T cells, identified cytotoxic CD4 T cells as an important component that is associated with tumor cell–intrinsic characteristics, and identified sets of targetable immune checkpoints on T cells that differed from FLs with normal versus low MHC expression. See related commentary by Melnick, p. 374. This article is highlighted in the In This Issue feature, p. 369

The spatial anatomy of hematopoiesis in bone marrow has been extensively studied in mice and other preclinical models, but technical challenges have precluded a commensurate exploration in humans. Institutional pathology archives contain thousands of paraffinized bone marrow core biopsy tissue specimens, providing a rich resource for studying the intact human bone marrow topography in a variety of physiologic states. Thus, we developed an end-to-end pipeline involving multiparameter whole tissue staining, in situ imaging at single-cell resolution, and artificial intelligence (AI)-based digital image analysis, and then applied it to a cohort of disease-free samples to survey alterations in the hematopoietic topography associated with aging. Our data indicate heterogeneity in marrow adipose tissue (MAT) content within each age group, and an inverse correlation between MAT content and proportions of early myeloid and erythroid precursors, irrespective of age. We identify consistent endosteal and perivascular positioning of hematopoietic stem and progenitor cells (HSPCs) with medullary localization of more differentiated elements and, importantly, uncover new evidence of aging-associated changes in cellular and vascular morphologies, microarchitectural alterations suggestive of inflammaging, and diminution of a potentially active megakaryocytic niche. Overall, our findings suggest that there is topographic remodeling of human hematopoiesis associated with aging. More generally, we demonstrate the potential to deeply unravel the spatial biology of normal and pathologic human bone marrow states using intact archival tissue specimens.

Supplementary Figure from Follicular Lymphoma Microenvironment Characteristics Associated with Tumor Cell Mutations and MHC Class II Expression

Supplementary Data from Follicular Lymphoma Microenvironment Characteristics Associated with Tumor Cell Mutations and MHC Class II Expression

Follicular lymphoma (FL) is a B-cell malignancy with a complex tumor microenvironment that is rich in nonmalignant immune cells. We applied single-cell RNA sequencing to characterize the diverse tumor and immune cell populations of FL and identified major phenotypic subsets of FL T cells, including a cytotoxic CD4 T-cell population. We characterized four major FL subtypes with differential representation or relative depletion of distinct T-cell subsets. By integrating exome sequencing, we observed that somatic mutations are associated with, but not definitive for, reduced MHC expression on FL cells. In turn, expression of MHCII genes by FL cells was associated with significant differences in the proportions and targetable immunophenotypic characteristics of T cells. This provides a classification framework of the FL microenvironment in association with FL genotypes and MHC expression, and informs different potential immunotherapeutic strategies based upon tumor cell MHCII expression.Significance:We have characterized the FL-infiltrating T cells, identified cytotoxic CD4 T cells as an important component that is associated with tumor cell–intrinsic characteristics, and identified sets of targetable immune checkpoints on T cells that differed from FLs with normal versus low MHC expression.See related commentary by Melnick, p. 374.This article is highlighted in the In This Issue feature, p. 369

Diffuse large B-cell lymphoma (DLBCL) is a biologically and clinically heterogeneous disease. Transcriptomic and genetic characterization of DLBCL has increased the understanding of its intrinsic pathogenesis and provided potential therapeutic targets. However, the role of the microenvironment in DLBCL biology remains less understood. Here, we performed a transcriptomic analysis of the microenvironment of 4,655 DLBCLs from multiple independent cohorts and described four major lymphoma microenvironment categories that associate with distinct biological aberrations and clinical behavior. We also found evidence of genetic and epigenetic mechanisms deployed by cancer cells to evade microenvironmental constraints of lymphoma growth, supporting the rationale for implementing DNA hypomethylating agents in selected patients with DLBCL. In addition, our work uncovered new therapeutic vulnerabilities in the biochemical composition of the extracellular matrix that were exploited to decrease DLBCL proliferation in preclinical models. This novel classification provides a road map for the biological characterization and therapeutic exploitation of the DLBCL microenvironment.Significance:In a translational relevant transcriptomic-based classification, we characterized the microenvironment as a critical component of the B-cell lymphoma biology and associated it with the DLBCL clinical behavior establishing a novel opportunity for targeting therapies.This article is highlighted in the In This Issue feature, p. 1307

Supplementary Figures

Supplementary Tables

Supplementary Figure S2. The mpMRI-visible PCa tumors show marked differences between the malignant and non-malignant compartments. A, Box plots showing WD per mpMRI-invisible and -visible PCa tumors. P-values were calculated with the Mann-Whitney U test. B, Combination of WD plotted against the malignant cell density per mpMRI-visible and -invisible PCa tumors and patient relapse.

Supplementary Table S1. Clinicopathologic characteristics of each mpMRI-invisible and -visible PCa tumor.

Purpose:Multiparametric MRI (mpMRI) has become an indispensable radiographic tool in diagnosing prostate cancer. However, mpMRI fails to visualize approximately 15% of clinically significant prostate cancer (csPCa). The molecular, cellular, and spatial underpinnings of such radiographic heterogeneity in csPCa are unclear.Experimental Design:We examined tumor tissues from clinically matched patients with mpMRI-invisible and mpMRI-visible csPCa who underwent radical prostatectomy. Multiplex immunofluorescence single-cell spatial imaging and gene expression profiling were performed. Artificial intelligence–based analytic algorithms were developed to examine the tumor ecosystem and integrate with corresponding transcriptomics.Results:More complex and compact epithelial tumor architectures were found in mpMRI-visible than in mpMRI-invisible prostate cancer tumors. In contrast, similar stromal patterns were detected between mpMRI-invisible prostate cancer and normal prostate tissues. Furthermore, quantification of immune cell composition and tumor-immune interactions demonstrated a lack of immune cell infiltration in the malignant but not in the adjacent nonmalignant tissue compartments, irrespective of mpMRI visibility. No significant difference in immune profiles was detected between mpMRI-visible and mpMRI-invisible prostate cancer within our patient cohort, whereas expression profiling identified a 24-gene stromal signature enriched in mpMRI-invisible prostate cancer. Prostate cancer with strong stromal signature exhibited a favorable survival outcome within The Cancer Genome Atlas prostate cancer cohort. Notably, five recurrences in the 8 mpMRI-visible patients with csPCa and no recurrence in the 8 clinically matched patients with mpMRI-invisible csPCa occurred during the 5-year follow-up post-prostatectomy.Conclusions:Our study identified distinct molecular, cellular, and structural characteristics associated with mpMRI-visible csPCa, whereas mpMRI-invisible tumors were similar to normal prostate tissue, likely contributing to mpMRI invisibility.

Supplementary Data file with Methods and each Supplementary figure/table linked to legend

Although there are immune checkpoint inhibitors (ICIs) available for the treatment of renal cell carcinoma (RCC), the utility of PD-L1 detection by immunohistochemistry (IHC) as a predictive biomarker in clear cell RCC (ccRCC) remains controversial. Nevertheless, alternative methods for PD-L1 detection, such as RNA sequencing (RNA-Seq), may be clinically useful in ccRCC; therefore, we sought to determine the ability of RNA-Seq to accurately and sensitively detect PD-L1 expression across different ccRCC clinical samples in comparison with IHC.Patients with ccRCC (n=127) who received treatment from Washington University in St. Louis between 2018 and 2020 were identified. Tumors from these patients were analyzed using RNA-Seq and IHC.PD-L1 detection by RNA-Seq strongly correlated with IHC (P.001), which was further validated using two independent datasets. Furthermore, RNA-Seq analysis identified an immune-enriched (higher PD-L1 positivity) and an immune-desert (lower PD-L1 positivity) microenvironment of ccRCC, which also correlated with IHC (P.00001).The results demonstrate the ability of RNA-Seq to detect PD-L1 in various ccRCC clinical samples compared to IHC. Ultimately, these findings suggest that PD-L1 detection by RNA-Seq can be further developed to determine the clinical utility of this methodology in ccRCC.

SUMMARYReference atlases, molecular and spatial maps of mammalian tissues, are critical resources for discovery efforts and translational research. Their utility is dependent on operationalizing the resulting data by identifying cell types, histological patterns, and predictive biomarkers underlying health and disease. The human lymph node (LN) offers a compelling use case because of its importance in immunity, structural and cellular diversity, and neoplastic involvement. One hematological malignancy, follicular lymphoma (FL), evolves from developmentally blocked germinal center B cells residing in and trafficking through these tissues. To promote survival and immune escape, tumor B cells undergo significant genetic changes and extensively remodel the lymphoid microenvironment. Here, we present an integrated portrait of healthy and FL LNs using multiple genomic and advanced imaging technologies. By leveraging the strengths of each platform, we identified several tumor-specific features and microenvironmental patterns enriched in individuals who experience early relapse, the most high-risk of FL patients.

The BostonGene Tumor PortraitTM Test (TPT) is a comprehensive molecular profiling assay with both integrated genomic and transcriptomic analysis. The combined analysis offers additional information that is potentially critical for personalized clinical decision making. We retrospectively reviewed 2 sarcoma cohorts that received the TPT, MG (n=21) and MD (n=40) (Table 1), and evaluated reported events that were not previously identified by institute respective targeted panels. A total of 19 (MG) and 23 (MD) additional clinically-relevant SNV/indels, CNAs, and fusions were identified by the TPT, including an amplification of EZH2, 2 novel fusions with MDM2 as a binding partner, and a known diagnostic fusion, NAB2-STAT6. A high TMB was observed in 5 patients. MSI was stable for all patients. HLA class I loss of heterozygosity was found in 7 patients. In both cohorts, we observed significant changes in the expression of potentially targetable genes, with 81% of patients presenting with increased expression of such targets. Transcriptomic-based tumor microenvironment (TME) classification found that 51% and 49% of patients had an Immune-Enriched (IE) subtype (predicted favorable IO treatment outcomes) and Fibrotic/Desert (F/D) subtypes (predicted poor response to IO treatment), respectively. Predicted response to IO therapy through TME subtyping agreed with previous reports, as angiosarcomas were enriched in the IE subtype (83%), and bone sarcomas were enriched in the F (30%) compared to the IE (6%) subtype. Further, these results supported matching patients with biomarker-driven clinical trials at their respective centers. In summary, the TPT identified more clinically-relevant biomarkers compared to respective institute targeted gene panels, highlighting the potential value of comprehensive molecular profiling for patients with highly heterogeneous diseases, such as sarcoma, to support personalized treatment decisions. Table 1. Genetic and expression alterations identified by the BostonGene Tumor PortraitTM Test Event type analyzed in targeted panels Event type Cohort (N patients)* % cohort with event or # biomarkers Events, status, or subtype Yes SNV/Indel MG (21) 10% FAT4 Y558_V590delinsCAfs*5 LOF mutation COL2A1 W1348Nfs*13 LOF mutation Yes SNV/Indel MD (40) 6% KMT2B R1779* LOF mutation KMT2C S143Vfs*3 LOF mutation PRKDC Y4046* LOF mutation Yes Germline MG (21) 5% MUTYH rs36053993, germline, pathogenic Yes Germline MD (40) 3% BTD rs13078881, germline, pathogenic Yes CNA MG (21) 43% IGF1R amplification +8 copies MGMT loss FANCA loss MCL1 amplification +10 copies MYOCD amplification +3 copies AKT2 amplification +8 copies AMACR amplification +6 copies YAP1 amplification +7 copies TRAF2 loss EZH2 amplification +2 copies HMGA2 amplification +100 copies FRS2 amplification +61 copies DDIT3 amplification +39 copies Yes CNA MD (40) 27.5% ERK1 amplification +2 copies FOXO1 amplification +5 copies HMGA2 amplification +2, +8 copies KMT2C amplification +3 copies NCOR1 amplification +7 copies TERT amplification +6 copies TSPAN31 amplification +30 copies HSF1 loss Yes Fusions MG (20) 10% MDM2-CR1 MDM2-TXNDC12 AC090825.1-IGF1R Yes Fusions MD (39) 15% FUS-KIAA1549 KIAA1549-CREB3L2 HMGA2-LRRC37A3 NAB2-STAT6 PPP1R12A-PAWR RB1-ZAR1L No TMB** MG+MD (61) 8% 35.67 mut/Mb (Desmoid fibromatosis) 8.48 mut/Mb (Cutaneous angiosarcoma) 8.9 mut/Mb (Sarcoma, NOS) 16.1 mut/Mb (Skin Angiosarcoma) 18.8 mut/Mb (Undifferentiated pleomorphic sarcoma) No MSI status MG+MD (61) 100% Stable No HLA loss of heterozygosity MG+MD (61) 11% HLA-I LOH (Leiomyosarcoma (N=1), Ewing Sarcoma (N=1)) HLA-A (Dedifferentiated liposarcoma (N=1)) HLA-I (Chondrosarcoma (N=3), High-grade sarcoma (N=1)) No Targetable surface molecule overexpression MG+MD (59) 81% CTLA4, EGFR, ERBB2, MAGEA3, PD1, PDL1, TIM3, TROP, ERBB2, FOLR1, NECTIN4, ROR1, TROP2 No Molecular Functional PortraitTM type MG+MD (59) 32% Fibrotic No Molecular Functional PortraitTM type MG+MD (59) 17% Immune Desert No Molecular Functional PortraitTM type MG+MD (59) 24% Immune-Enriched, Fibrotic No Molecular Functional PortraitTM type MG+MD (59) 27% Immune-Enriched, Non-fibrotic No MHC deficiency MG+MD (59) 3% MHC class I/II deficiency No MHC deficiency MG+MD (59) 3% MHC class II deficiency No FDA label biomarkers MG+MD (59) 7 biomarkers - Desmoid fibromatosis, TMB 35.67 mut/Mb (Pembrolizumab) - Angiosarcoma, TMB 16.1 mut/Mb (Pembrolizumab); - Undifferentiated pleomorphic sarcoma, MSI, TMB 18.8 mut/Mb (Pembrolizumab); - Cutaneous angiosarcoma, PALB2 pathogenic germline variant (Olaparib); - Chondrosarcoma IDH1 R132L (Ivosidenib) No Transcriptomic biomarkers MG+MD (59) 29 biomarkers CD8+ T cell number; PDL1 expression level; SLFN11 expression level; SMARCB1 expression; CD8+ T cell number No Diagnostic biomarkers MG+MD (59) 9 biomarkers NAB2-STAT6; MXI1-NUTM1; EWSR1-NR4A3; EWSR1-FLI1; EWSR1-CREB3L1; NAB2-STAT6 - - - - *RNA analysis failed for 2 patients. Therefore, any expression based analysis was performed on n=20 and n=39 for the MG and MD cohort, respectively. **Tumor mutational burden is considered high for FFPE samples when TMB>8mut/Mb. Citation Format: Gregory M. Cote, Anna Novokreshchenova, Irina Zhuk, Aida Aukhadieva, Sandrine Degryse, Aleksei Shevkoplias, Anna Belozerova, Krystle Nomie, Nikita Kotlov, Georgy Sagardze, Lev Bedniagin, Jessica Brown, Ekaterina Postovalova, Alexander Bagaev, Nathan Fowler, J. Andrew Livingston, Vinod Ravi. Use of comprehensive molecular profiling to identify additional clinically-relevant alterations compared to targeted gene panels [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 244.

Minimally invasive next-generation sequencing-based (NGS) liquid biopsy (LB) tests using cell-free DNA (cfDNA) are increasingly becoming an important tool for the clinical management of cancer patients. The ability to accurately detect molecular alterations from a single plasma sample is valuable for biomarker discovery, diagnostics, and disease monitoring. However, challenges for this assay include low concentrations of circulating tumor DNA (ctDNA) within the cfDNA plasma fraction and PCR and sequencing errors, which can lead to decreased sensitivity in variant calling. Further, somatic mosaicism due to clonal hematopoiesis of indeterminate potential (CHIP) in plasma makes accurate interpretation of LB results a challenge. Incorrect classification of false positives can lead to inappropriate therapeutic management. Here, we present an LB assay composed of a proprietary error reduction algorithm designed to eliminate false positives due to PCR and NGS errors and a white blood cell sequencing (WBC-seq) component to detect and bioinformatically filter out CHIP-related mutations. Consisting of a custom panel of 216 FDA/NCCN pan-cancer genes, our cfDNA LB assay was analytically validated with commercially available reference standards and human cell lines. At a clinically relevant sequencing depth of 4000x, we observed a limit of detection (LOD) of 0.2% variant allele frequency for single nucleotide variants (SNVs) and insertions and deletions (indels) with overall accuracies of 98% and 99% and sensitivities of 91% and 83%, respectively. To assess the LB assay performance on clinical samples, orthogonal whole-exome sequencing (WES) with subsequent standard somatic variant calling was performed on paired peripheral blood lymphocytes (PBLs) and plasma samples from 12 patients. The WES assay results were concordant with those obtained using our LB assay for SNVs (CCC = 0.75) and indels (CCC = 0.90). Further, our assay detected an additional 53% and 50% SNVs and indels, respectively, compared to WES. Finally, 8 out of 11 clinically actionable mutations were detected by both WES and the LB assay, while 3 additional SNVs were detected in the LB assay. Taken together, the use of WBC-seq to filter CHIP-derived mutations combined with our error reduction algorithm resulted in a substantial reduction in false positive variant calls without a compromise in the sensitivity or classification accuracy of SNVs and indels. Concordance was observed between traditional WES analysis and the LB assay developed here. These findings highlight the potential for our LB cfDNA assay to be used to improve personalized therapeutic options and guide clinical decision making. Citation Format: Anastasiya Yudina, Alexey Efremov, Danielle Sookiasian, Ekaterina Nuzhdina, Svetlana Podsvirova, Madison Chasse, Tori Conroy, Noel English, Sergey Starikov, Olesia Klimchuk, Dmitry Tabakov, Anna Love, Kushal Suryamohan, Artur Baisangurov, Cagdas Tazearslan, Nathan Fowler, Alexander Bagaev. Analytical validation of a liquid biopsy test using cell-free circulating tumor DNA for mutational profiling [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 1042.

The ability to locate a tumor primary site for patients with a cancer of unknown primary (CUP) is a major obstacle in providing personalized therapeutic options and access to clinical trials. Despite the recent use of molecular-based tools to identify the tumor tissue-of-origin (TOO), overall survival for CUP patients remains low. Here, we present an AI-based tool that predicts the TOO by using genomic and transcriptomic data to classify CUP tumors into hierarchically-organized molecular subgroups. The TOO predictor was composed of DNA, RNA, and Consensus classifiers that were hierarchically organized with respect to molecular diagnosis with upper level clusters based on common molecular features reflecting similar cell of origin, and lower levels containing specific diagnoses for further classification. The ML-based DNA classifier was trained on a dataset of publicly available genomic data generated from 8,000 samples, and independently validated using more than 5,500 samples. The ML-based RNA classifier was trained on a dataset of publicly available transcriptomic data created from more than 10,100 samples with tumor- and normal-specific features for each cancer type, and independently validated using 20,000 samples. The Consensus classifier, combining outputs from both DNA and RNA algorithms, was trained on a dataset of genomic and transcriptomic data from 1,000 samples, and validated on an independent dataset of 2,000 samples. Each classifier contained features selected based on data analysis according to the weighted F1-score, and the best hyperparameters for the final model. The 3-classifier algorithm predicts TOO for 33 cancer types and subtypes belonging to solid neoplasms, independently of sample source, sequencing methods, and cohort. Validation of the Consensus classifier showed a higher accuracy (95% f1-score) compared to the DNA and RNA classifiers (79% and 93% f1-scores, respectively), along with a high sensitivity (95%), specificity (99%), and precision (96%), as it takes into account both genomic events and expression patterns. The TOO predictor was prospectively validated on approximately 298 clinical samples with a known diagnosis using all classifiers. The diagnosis was identified in 90% of clinical cases (295/297) by the Consensus classifier with 90% sensitivity and 94% precision. The call rate for the DNA and RNA classifiers was above 95%. Of note, sensitivity of the top 4 predicted diagnoses was > 90% for all 3 classifiers, and the calculated rule-out accuracy of the Consensus classifier was 97%. In conclusion, an ML-based algorithm was developed that utilizes genomic and transcriptomic data to accurately predict the TOO for CUP tumors. Utilizing the Consensus classifier after DNA and RNA classifiers helps to identify the TOO of the tumor with high specificity, which can guide precision oncology therapeutic options. Citation Format: Zoia Antysheva, Daria Kiriy, Anton Sivkov, Alexander Sarachackov, Alexandra Boyko, Naira Samarina, Nara Shin, Jessica H. Brown, Ivan Kozlov, Viktor Svekolkin, Alexander Bagaev, Nathan Fowler, Nikita Kotlov. An ML-based tool for predicting tissue of origin for cancer of unknown primary (CUP) based on genomic and transcriptomic data. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5405.

Recent advances in immunotherapy demonstrate the need to further understand the characteristics of an individual cancer patient’s immune system and how it influences responses to cancer treatment. Here, we developed an immunoprofiling platform to evaluate the features in the blood of cancer patients to test the hypothesis that peripheral immune cell heterogeneity could be used to stratify these patients into different categories or immunotypes to monitor disease progression and treatment response. To that end, we established a unique diagnostic immunoprofiling assay and analytical framework based on the analysis of leukocytes in the peripheral blood using multiparameter flow cytometry. Supervised manual gating of flow cytometry data from a cohort of 50 healthy donors identified 415 cell types and immune activation states that were used to train and later independently validate machine learning models to automatically identify immune cell subsets from raw cytometry data. By applying this tool to peripheral blood samples from a mixed cohort of 299 healthy donors and 323 cancer patients, we developed a machine-learning classification model that can differentiate between these two groups with 93% accuracy. This model was further refined using spectral clustering with bootstrapping, revealing 5 clusters, or immunotypes, characterized by specific physiological immune profiles: (1) Myeloid-derived suppressor/NK cell, (2) Terminally-differentiated CD8+ T cells, (3) Mixed CD4+ T helper cells, (4) CD4+ Th1 & CD8+ T cell memory, and (5) Naive T and B lymphocytes. Interestingly, very few healthy donors could be found in clusters 1 and 2 but were assigned most frequently to cluster 5. Matched RNA-seq was used to further validate these profiles using the cellular deconvolution algorithm, Kassandra, and differential gene expression analysis revealed immunotype-specific signatures that are consistent with immune response potential. Patients in the terminally-differentiated CD8+ T cell cluster had a narrower range of HLA-types than the other clusters, and TCR repertoire analysis indicated significantly increased clonality and reduced clonotype diversity. Within this cluster there was a high degree of overlap between TCR sequences in the peripheral blood and the tumor, indicating a relationship between peripheral blood immunotype and tumor infiltration. Altogether, the establishment of these immunotypes using peripheral blood immunoprofiling represents a promising signature that can be used to identify and stratify cancer patients that will benefit from immune-based therapies. Citation Format: Daniiar Dyikanov, Iris Wang, Tatiana Vasileva, Polina Shpudeiko, Polina Turova, Arseniy A. Sokolov, Olga Golubeva, Evgenii Tikhonov, Anna Kamysheva, Ilya Krauz, Mary Abdou, Madison Chasse, Tori Conroy, Nicholas R. Merriam, Boris Shpak, Anastasia Radko, Anastasiia Kilina, Lira Nigmatullina, Linda Balabanian, Christopher J. Davitt, Alexander A. Ryabykh, Olga Kudryashova, Cagdas Tazearslan, Ravshan Ataullakhanov, Alexander Bagaev, Aleksandr Zaitsev, Nathan Fowler, Michael F. Goldberg. Comprehensive immunoprofiling of peripheral blood reveals five conserved immunotypes with implications for immunotherapy in cancer patients [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 6664.

Homologous recombination deficiency (HRD) is characterized by a defective double-stranded DNA repair mechanism due to alterations in the homologous recombination (HR) pathway. Deleterious mutations in HR pathway genes can help identify potential responders to platinum-based chemotherapies and PARP inhibitors (PARPi), which promote apoptosis in HRD cells. Until recently, the use of PARPi in breast cancer (BC) was limited to advanced or metastatic disease with pathogenic or likely pathogenic germline BRCA (gBRCA) mutations. However, the use of PARPi has expanded to treat early-stage HER2-negative BC with a high risk of recurrence and gBRCA mutations, HER2-negative BC with somatic BRCA mutations, and triple-negative BC. To date, no universal method for HRD scoring is accepted; therefore, biomarkers are needed to stratify patients into PARPi responders and nonresponders more effectively. Because PARPi are used to treat HRD-driven cancers, we aimed to identify genomic consequences of HRD. For HRD-specific features, we calculated the proportion of long-focal total copy number alterations (LF-tCNA), which estimates amplification events for long-focal segments. Aneuploidy is especially prevalent in medullary, metaplastic, and invasive micropapillary BCs. These copy number variations are not necessarily consequences of HRD but can influence HRD scores. Genome-wide loss of heterozygosity (gwLOH) can be a result of HRD and used as a biomarker for HRD. We developed several scoring methods based on LF-tCNA, gwLOH, and aneuploidy scores. These scores were tested as predictors of HRD, defined as known loss-of-function germline mutations in BRCA1, BRCA2, PALB2, or BARD1 (TCGA-BRCA cohort, n = 1,032). To calculate the optimal HRD score, we used multivariate logistic regression analysis with HRD as an outcome and LF-tCNA, LOH, and ploidy as predictors. Based on the results of logistic regression, LF-tCNA and ploidy were selected to calculate HRD status. The positive and negative predictive values (PPV, NPV) were used to set the upper and lower thresholds, respectively. Samples with HRD scores above 7 (PPV) have the greatest potential for PARPi and platinum-based therapy response, while scores below -1 (NPV) were considered HRD wild type (WT). To validate the developed HRD score, we tested the score by combining BC cohorts (MSK_NCI and MET500 cohorts, n = 164) and used gBRCA1/2 pathogenic mutations as an outcome. Our HRD scoring system distinguished HRD WT from HRD-positive samples (AUC = 0.81) more effectively than a previously reported score based on LOH, telomeric-allelic imbalance (TAI), and large-scale state transitions (LST) (AUC = 0.72). We tested our HRD scoring system across tumor microenvironment molecular subtypes. The median HRD score in the basal-like subtype was increased compared to other subtypes, supporting the prevalence of gBRCA1/2 pathogenic mutations in this subtype. Moreover, the number of HRD-positive patients and the percent positive agreement, defined as the proportion of positive test results from our HRD also positive for HR mutations, were calculated for each BC subtype. Only 22% of basal-like HRD-positive samples also carried germline HR pathway mutations, which indicates 78% of HRD-positive patients who might benefit from PARPi or platinum treatment would have been missed by germline HR gene panels. This BC HRD scoring method is a promising tool for identifying HRD patients who may respond to PAPRi and platinum-based therapies, but additional studies are required for clinical validation. Citation Format: Maria Gurileva, Nikita Kotlov, Anna Love, Alexandra Melnikova, Krystle Nomie, Nathan Fowler, Alexander Bagaev. An HRD scoring system based on long-focal copy number alterations predictive of PARP inhibitor response [abstract]. In: Proceedings of the 2022 San Antonio Breast Cancer Symposium; 2022 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2023;83(5 Suppl):Abstract nr P5-02-45.

Introduction Tertiary lymphoid structures (TLSs) and tumor-infiltrating lymphocytes (TILs) in breast carcinomas are prognostic for survival and predictive of certain therapy responses. The presence of TLSs and TILs are identified by manual pathological examination; however, this method often lacks reproducibility, limiting its use in routine clinical practice. Here, we demonstrate that morphological evaluation of whole slide images (WSIs) using an artificial intelligence (AI)-based analytic workflow comprised of convolutional neural network (CNN) deep learning models that accurately and reproducibly characterizes TILs, measured as the lymphocyte immune-infiltrated area (LIIA), and TLSs in the tumor microenvironment (TME) of breast carcinomas. Methods We collected a cohort of 445 TCGA breast cancer H&E WSIs, including clinical and sequencing data, and divided this cohort into luminal invasive lobular carcinoma (ILC) (n = 192), HER2-enriched (n = 110), and basal-like (n = 143) molecular subtypes. After 55 samples were excluded due to artifacts or incomplete clinical annotation, a total of 390 samples were analyzed. A combination of CNN-based deep learning models was used to detect and classify the tumor area, TLSs present in the TME, TLS density (number of TLS per mm2 of tumor), and lymphocyte-rich regions. The LIIA was calculated as the area of the stromal and TIL components of the TME. Validation was performed by manually annotating 10 random WSIs from the dataset. Spatial model predictions of the tumor and TLSs were combined to identify TLS locations. Each model’s predictions were verified by univariate (Kaplan-Meier) and multivariate (Cox regression) survival analyses, and the log-rank test was used to calculate overall survival. Additionally, the relationship between TLSs and LIIAs with CD274 expression (PD-L1) and a high tumor mutational burden (TMB > 10) was analyzed. Statistical analyses included Spearman’s rank correlation and Mann-Whitney tests. Results TLS were detected in 53% (n = 207) of the samples, with a mean density of 26.02 TLS/mm2 (Q3 = 5.53 TLS/mm2). TLS density was higher in basal-like subtype samples compared to luminal and HER2-enriched subtypes. While LIIA and TMB-high samples exhibited a significant relationship (p = 0.00001), no significant association was found between TME and TLS quantities or density. PD-L1 gene expression exhibited weak to moderate correlations with predicted LIIA in basal-like (r = 0.38, p = 0.00001) and HER2-enriched subtypes (r = 0.38, p = 0.0001). The luminal subtype had no significant correlation between PD-L1 expression and predicted LIIA. As a result, LIIA and TLS were characterized as positive prognostic factors for the basal-like subtype. After adjusting for age, stage, and grade, the LIIA and TLS density were found to be significant independent positive prognostic overall survival factors for the basal-like subtype (LIIA HR: 0.02, p = 0.003; TLS-high group HR: 0.09, p = 0.002). For the HER2-enriched subtype, TLS density was also a significant predictor (HR: 0.05, p = 0.035), while LIIA was not a statistically significant prognostic factor (HR: 0.0002, p = 0.08). Associations were not observed between the TLSs and LIIA between the ILC subtypes and survival outcomes. The same result was observed for univariate analyses. Conclusion The developed analytic pipeline accurately identified the presence of LIIA and TLS on H&E slides, demonstrating the potential of CNN for automated characterization of the breast cancer TME. AI-based TLS and LIIA quantification can be a robust tool for pathology processes, offering additional information to help in clinical decision-making. This approach can be used to detect features of immune morphology biomarkers in other cancer types. Citation Format: Vladimir Kushnarev, Daniil Dymov, Nadezhda Lukashevich, Lev Popyvanov, Anna Belozerova, Diana Shamsutdinova, Aida Akaeva, Yury Popov, Svetlana Khorkova, Ivan Valiev, Anastasia Zotova, Jessica H. Brown, Anna Love, Alexander Bagaev, Ekaterina Postovalova, Nathan Fowler. AI-based prediction of tertiary lymphoid structures and lymphocyte immune infiltration in breast carcinomas [abstract]. In: Proceedings of the 2022 San Antonio Breast Cancer Symposium; 2022 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2023;83(5 Suppl):Abstract nr P6-04-15.

Personalized cancer vaccines designed to target neoantigens represent a promising new treatment paradigm in oncology. In contrast to classical idiotype vaccines, we hypothesized that ‘polyvalent’ vaccines could be engineered for the personalized treatment of follicular lymphoma (FL) using neoantigen discovery by combined whole exome sequencing (WES) and RNA sequencing (RNA-Seq). Fifty-eight tumor samples from 57 patients with FL underwent WES and RNA-Seq. Somatic and B-cell clonotype neoantigens were predicted and filtered to identify high-quality neoantigens. B-cell clonality was determined by alignment of B-cell receptor (BCR) CDR3 regions from RNA-Seq data, grouping at the protein level, and comparison to the BCR repertoire of RNA-Seq data from healthy individuals. An average of 52 somatic mutations per patient (range: 2-172) were identified, and two or more (median: 15) high-quality neoantigens were predicted for 56 of 58 samples. The predicted neoantigen peptides were composed of missense mutations (76%), indels (9%), gene fusions (3%), and BCR sequences (11%). Building off of these preclinical analyses, we initiated a pilot clinical trial using personalized neoantigen vaccination combined with PD-1 blockade in patients with relapsed or refractory FL (#NCT03121677). Synthetic long peptide (SLP) vaccines were successfully synthesized for and administered to all four patients enrolled to date. Initial results demonstrate feasibility, safety, and potential immunologic and clinical responses. Our study suggests that a genomics-driven personalized cancer vaccine strategy is feasible for patients with FL, and this may overcome prior challenges in the field.

Intratumor heterogeneity (ITH) represents a major challenge for anticancer therapies. An integrated, multidimensional, multiregional approach dissecting ITH of the clear cell renal cell carcinoma (ccRCC) tumor microenvironment (TME) is employed at the single-cell level with mass cytometry (CyTOF), multiplex immunofluorescence (MxIF), and single-nucleus RNA sequencing (snRNA-seq) and at the bulk level with whole-exome sequencing (WES), RNA-seq, and methylation profiling. Multiregional analyses reveal unexpected conservation of immune composition within each individual patient, with profound differences among patients, presenting patient-specific tumor immune microenvironment signatures despite underlying genetic heterogeneity from clonal evolution. Spatial proteogenomic TME analysis using MxIF identifies 14 distinct cellular neighborhoods and, conversely, demonstrated architectural heterogeneity among different tumor regions. Tumor-expressed cytokines are identified as key determinants of the TME and correlate with clinical outcome. Overall, this work signifies that spatial ITH occurs in ccRCC, which may drive clinical heterogeneity and warrants further interrogation to improve patient outcomes.

Cellular deconvolution algorithms virtually reconstruct tissue composition by analyzing the gene expression of complex tissues. We present the decision tree machine learning algorithm, Kassandra, trained on a broad collection of9,400 tissue and blood sorted cell RNA profiles incorporated into millions of artificial transcriptomes to accurately reconstruct the tumor microenvironment (TME). Bioinformatics correction for technical and biological variability, aberrant cancer cell expression inclusion, and accurate quantification and normalization of transcript expression increased Kassandra stability and robustness. Performance was validated on 4,000 HE slides and 1,000 tissues by comparison with cytometric, immunohistochemical, or single-cell RNA-seq measurements. Kassandra accurately deconvolved TME elements, showing the role of these populations in tumor pathogenesis and other biological processes. Digital TME reconstruction revealed that the presence of PD-1-positive CD8

Gene expression profiling is widely used in oncology research and in clinical settings for decision making. Despite the cross-platform correlation of gene expression values, ideally, each measurement should be evaluated against a cohort of samples sequenced using the same methodology. Clinical samples, preserved as FFPEs, often undergo exome capture-based RNA-seq; research samples, stored as fresh/frozen (FF), undergo poly-A RNA-seq, producing high quality expression data. Thus, development of sequencing protocols and data processing algorithms are necessary to provide the same quality gene expression measurements from FFPE samples. Further, while several batch effect correction algorithms exist to neutralize the batch effect between samples across large cohorts, the majority cannot be applied to an individual sample, raising the need to develop an algorithm for single sample projection to improve gene expression-based personalized clinical decision-making. To improve the quality of RNA reads from FFPE tissues, exome capture enrichment of RNA transcripts was optimized and the concordance with poly-A RNA-seq was increased by adding non coding 3’ and 5’ UTR region to the probes. After testing the performance of multiple different extraction methods, a 0.88 correlation was achieved between exome-capture-based and poly-A RNA-seq protocols. To further align the sequencing methodologies, we designed a batch-correction ML-based algorithm by performing a series of paired RNA-seq experiments from the same sample using exome-capture-based and poly-A RNA-seq; we applied linear modeling on the training subset (N = 64) and verified the performance on the validation subset (N = 24). For each gene, 5-20 correlated genes belonging to the TCGA combined pan-cancer datasets were selected and trained using the Lasso model. Over 82% of genes (total N = 20,062) correlated across the two RNA-seq methodologies for each sample after correction (ccc value > 0.5), and approximately 94% of cancer-specific and microenvironment-related genes correlated (ccc value > 0.5). The algorithm significantly outperformed other batch correction methods, with ccc values > 0.8 for 51.37% of the 20,062 genes compared with ~3% for PCA, 26% for MNN, and 28% for ComBat. Our algorithm showed improved performance by correction of 77% of the 1,890 clinically-relevant genes (ccc values > 0.8) compared with 15% for PCA, 39% for MNN, and 40% for ComBat. Here, we developed combinatory technology with a batch correction algorithm trained and developed on FFPE or FF tumor samples, using exome capture-based sequencing or poly-A RNA-seq, that enables the projection of a single sample onto a larger cohort. Future application of this correction tool will enable direct analysis of gene expression of single tumor samples to support potential gene expression-based treatment decisions. Citation Format: Nikita Kotlov, Kirill Shaposhnikov, Cagdas Tazearslan, Ilya Cheremushkin, Madison Chasse, Artur Baisangurov, Svetlana Podsvirova, Svetlana Korkova, Yaroslav Lozinsky, Katerina Nuzhdina, Elena Vasileva, Dmitry Kravchenko, Krystle Nomie, John Curran, Nathan Fowler, Alexander Bagaev. Combinatory technologies for single sample gene expression projection onto a cohort sequenced with a different technology for personalized clinical decision-making [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1216.

Despite the breast cancer microenvironment being recognized as a critical participant in tumor progression and therapeutic response, current technologies for tumor cell-to-cell interaction analyses are limited to qualitative and descriptive histological methods. Here, we utilized multiplex immunofluorescence (MxIF), RNA sequencing (RNA-seq), and microarray gene expression to quantitatively assess, using the BostonGene automated pipeline, the immune microenvironment composition of clinical stage 2 or 3 primary estrogen receptor-positive (ER+) HER2-negative (HER-) breast tumors (n = 29) from patients treated with neoadjuvant palbociclib in combination with anastrozole. Microenvironment characteristics of pretreatment tumor biopsies were associated with molecular subtype, cytokine signatures, and Ki67 response. MxIF analysis (21 markers, 2-13 ROIs per slide) of the tumor immune composition revealed that while the cellular composition was similar for each tumor region of the same patient, significant cellular heterogeneity was observed in tumor biopsies from different patients. The spatial heterogeneity was found in the distribution of both immune infiltrate cells and stromal components. T cells, CD68+ and CD163+ macrophages, and B cells were the predominant cell populations in the stromal component. Further MxIF analysis showed an increase in Ki67+ tumor cells in nonresponders (NR) versus responders (R). The number of immune cells was similar between Luminal A and B breast cancer molecular subtypes. A lower number of Ki67+ tumor cells was found in Luminal A compared to Luminal B tumors, as expected. MxIF analysis of cell-to-cell interactions revealed that distinct groups of cellular neighborhoods were significantly increased in NR compared with R: the immune inflammation cluster with a high density of CD4 T cells, Tregs, and CD8 T cells in the stromal component (p = 0.03); CD163+ macrophage-enriched cluster (p < 0.001); and tertiary lymphoid structures cluster (p < 0.001). Transcriptomic analysis showed that while CXCL1, CCL17, CCL13, CXCL9, and CCL5 expression correlated with the high density of CD4 and CD8 T cells in the stromal component cluster, VEGFA, IL1B, CXCL16, CCL3, and CCL7 were increased in the CD163+ and CD68+ macrophage-enriched clusters, indicating that differential cytokine expression is associated with the tumor architecture.Comprehensively characterizing, via MxIF, the immune microenvironment of ER+ HER2- breast cancer provided resolution of cellular architecture and elucidated spatial relationships among the immune and stromal cells of the tumor. Quantitatively assessing breast tumor cell-to-cell interactions has the potential to identify imaging markers of response, ultimately leading to the development of effective therapy options. Citation Format: Maria Bruttan, Souzan Sanati, Anna Belozerova, Ilia Galkin, Akshaya Ramachandran, Pavel Ovcharov, Grigory Perelman, Viktor Svekolkin, Ekaterina Postovalova, Alexander Bagaev, Krystle Nomie, Shana Thomas, Jeremy Hoog, Ravshan Ataullakhanov, James Hsieh, Cynthia Ma. Tumor microenvironment heterogeneity identifies potential biomarkers of response in ER-positive breast cancers treated with palbociclib [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 6151.

Classifying tumors based on histological grade is used for many cancers to determine a patient’s prognosis and select the appropriate treatment. Although low grade (G1) and high grade (G3/G4) tumors have been shown to predict patient survival in many cancers, an intermediate histological grade (G2) does not have clear prognostic significance. Therefore, to allow a more accurate assessment of a patient’s risk, recent studies have focused on utilizing gene expression data to improve prognostic stratification of tumors into either a high-risk or low-risk group. Here, we present a molecular-based classifier that can use gene expression data from both RNA sequencing (RNA-seq) and microarray profiling to achieve accurate prediction of tumor grade for breast cancer (BRCA), lung adenocarcinoma (LUAD), and clear cell renal cell carcinoma (ccRCC). The tumor grade predictor was developed by training a machine-learning algorithm to classify tumors into categories of low (mG1) and high (mG3/mG4) molecular grades for each cancer type using rank transformation of gene expression data from multiple datasets. This resulted in the revealing of features of the mG3/mG4 groups not shown by histological methods (i.e., the Warburg effect, proliferation signaling pathways, inactivation of tumor suppressors, fibrotic and immunosuppressive microenvironment, and the loss of normal cell morphology), allowing for a more accurate stratification between the high- and low-grade groups. Importantly, by using rank transformation to analyze data from both RNA-seq and microarray profiling, the predictor functioned independent of various sources of batch effect, resulting in the ability to work with single samples and not only datasets. Validation of the tumor grade classifier with pathological classification of G1 and G3/G4 samples from RNA-seq and microarray expression data into mG1 and mG3/mG4 groups showed a more accurate performance than other gene expression-based methods, which was demonstrated by it’s sensitivity-specificity value (AUC scores: BRCA - 0.95, LUAD - 0.89, ccRCC - 0.83). Further, cox-regression analysis showed that prognostic value of mG1 and mG3 groups were comparable to histological grades (BRCA - 0.71 vs. 0.71; LUAD - 0.66 vs. 0.62; ccRCC - 0.74 vs. 0.71), demonstrating the ability of the classifiers to successfully split G2 histological samples into molecular grades. In conclusion, we developed a tumor grade predictor that utilizes expression data from multiple datasets to accurately classify tumors for three cancer types. The resulting prognostic ability of the molecular grades demonstrates their capability of assessing the levels of risk for G2 samples, suggesting the potential for molecular-based tumor grade classifiers to be used in future clinical application. Citation Format: Zoya Antysheva, Maria V. Guryleva, Ivan Valiev, Nikita Kotlov, Viktor Svekolkin, Olga Gancharova, Vladimir Kushnarev, Yuriy Popov, Anna Belozerova, Dmitry Tabakov, Jessica H. Brown, Krystle Nomie, Alexander Bagaev, Nathan Fowler. Molecular-based tumor grade predictor for breast cancer, clear cell renal cell carcinoma, and lung adenocarcinoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1227.

Checkpoint immunotherapy (CPI) and its combination with tyrosine kinase inhibitor (TKI) therapy are widely used in the treatment of many cancers, including clear cell renal cell carcinoma (ccRCC). Although recent advances in these therapies demonstrate effectiveness, a subset of patients are unresponsive. In this study, using deep immune profiling by mass cytometry (CyTOF), we analyzed peripheral blood mononuclear cell (PBMC) samples from 86 ccRCC patients pre- and post-treatment. The cohort included patients treated with first-line CPI (n = 43), TKI (n = 15) and combinatorial CPI+TKI (n = 28), with PBMC samples collected pre- (n = 41) and post- (n = 69) therapy (n = 24 patients total with paired samples). Samples were characterized by CyTOF analysis of 30 different cell surface markers. Using t-distributed stochastic neighbor embedding (tSNE) and Phenograph clustering, the presence of immune populations and subpopulations were characterized and associated with therapeutic response as well as longitudinally tracked. While most cell types did not differ between pre- and post-treatment states, a decrease in CD8 and CD4 PD1+ T cell subsets was observed in patients treated with CPI and CPI+TKI (p < 0.001). Also, increased CD8 effector memory T cells were observed in post-treatment samples in those treated with CPI or CPI+TKI (p < 0.001).Subsequent analysis of pre-treatment samples in the context of clinical outcome revealed specific features associated with favorable response to immunotherapy. Responders (R) in comparison with non-responders (NR) were characterized by an increased number of naive CD4 T cells (p = 0.03), CD8 T cells (p = 0.038), immature NK cells (p = 0.005) and lower percentage of granulocytes (p = 0.012). Using unsupervised hierarchical clustering, 8 patient groups were identified with a unique immune profile enriched by a certain PBMC population. Three groups (G1-G3), which included a high ratio of R, were enriched by naive T cells, immature NK cells and non-classical monocytes, respectively. The G4 group, enriched with B cells, consisted of R and NR in equal numbers. The G5-G8 groups, enriched with granulocytes, classical monocytes, mature NK-cells and effector CD8 T cells, respectively, were primarily comprised of NR. Patients with favorable pre-treatment immune profiles also belonged to favorable groups post-treatment, with a similar trend observed with unfavorable profiles. In conclusion, this analysis revealed a remarkable association between the initial immune status or “portrait” of the patient before therapy and subsequent therapeutic outcome as well as the development of therapeutic resistance. Citation Format: Svetlana Shishkova, Dmitry Tabakov, Evgeny Egorov, Akshaya Ramachandran, Yang Lyu, Krystle Nomie, Jessica Brown, Vladimir Zyrin, Alexander Zaytcev, Natalia Miheecheva, Ekaterina Postovalova, Alexander Bagaev, Ravshan Ataullakhanov, James Hsieh. Deep immune profiling by mass cytometry revealed an association between the state of immune system before treatment and response to checkpoint inhibitor therapy in clear cell renal cell carcinoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 2061.

Human papillomavirus (HPV)-associated Head and Neck Squamous Cell Carcinoma (HPV+HNSCC) is now the most common HPV-associated malignancy in the United States. Current treatments can be associated with severe side-effects or lack of efficacy yet prognostic biomarkers are limited, slowing efforts to personalize treatment in HPV+HNSCC. Here, we describe the use of a transcriptomic-based analytical platform to analyze expression patterns of viral transcripts, the tumor microenvironment (TME), and viral genome integration, and associate these features with overall survival. Functional gene expression signatures were analyzed on publicly available HPV+HNSCC expression data (n=266). Unsupervised clustering analysis revealed 5 distinct and novel TME types across patients (immune-enriched non-fibrotic, immune-enriched fibrotic, fibrotic, immune-desert, immune-enriched luminal). These microenvironment subtypes were highly correlated with both overall survival and patient prognosis. Tumors with an immune-enriched microenvironment showed the highest survival rates, whereas fibrotic TME types were associated with poor survival (p < 0.05). Unsupervised clustering of a HPV+HNSCC cohort from The Cancer Genome Atlas (TCGA) (n=53), based on HPV transcript expression, revealed 4 HPV-related subtypes. Each subtype was enriched for distinct viral transcripts: E2/E5, E6/E7, E1/E4 and L1/L2. We then validated TME and HPV transcript-related classifications on an independent HPV+HNSCC cohort (n=132). Utilizing both viral transcript and TME subtypes, we found that the E2/E5 HPV subtype was associated with an immune-enriched TME and had a higher overall survival compared to the other subtypes. The E2/E5 subtype was also enriched for samples without HPV-genome integration, suggesting that HPV episomal DNA status and E2/E5 expression pattern may drive an inflamed microenvironment and improved prognosis. In contrast, E6/E7 subtype samples were associated with the fibrotic and depleted TME types, with lower values of T-cell and B-cell gene expression signatures and a lower survival rate. Both E1/E4 and L1/L2 subtypes were associated with the immune-enriched luminal TME types. These findings suggest that HPV-transcript expression patterns may drive modulation of the TME, and hence impact prognosis. Further validation of the relationships between viral gene expression, TME, and prognosis is warranted to understand if such subtypes could aid in the development of prognostic biomarkers for treatment selection. Citation Format: Daria Kiriy, Dmitry Tychinin, Nikita Kotlov, Olga Kudryashova, Anastasia Nikitina, Andrey Tyshevich, Naira Samarina, Ksenia Demina, Sandrine Degryse, Susan Raju Paul, Mark Poznansky, Krystle Lang Kuhs, James S. Lewis, Robert L. Ferris, Xiaowei Wang, Alexander Bagaev, Nathan Fowler, Lori Wirth, Daniel Faden. Viral transcript and tumor immune microenvironment-based transcriptomic profiling of HPV-associated head and neck cancer identifies subtypes associated with prognosis [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 3823.

3142 Background: Previous studies of non-small cell lung cancer (NSCLC) have shown that TLS can be predictive of therapy response and a positive prognostic factor for survival. Currently, TLS identification is manually performed by pathologists with limited morphological criteria. Standardizing TLS detection with an automated DIA workflow could guide clinical trials in precision medicine by improving patient stratification. Here, we investigate the reproducibility and sensitivity of our DIA platform for evaluating TLS in LUAD using digital histopathology and machine learning. Methods: TLS were assessed by 3 pathologists on whole slide images (WSI) in a validation cohort of 22 LUAD samples using current TLS characterization criteria of dense lymphoid structures, the presence/absence of a germinal center, and high endothelial venules (HEVs). The intraclass correlation coefficient (ICC) was used to measure reproducibility between pathologists. The BostonGene DIA platform was used to train models for automated TLS detection. Quantitative measurements of area, lymphocyte number, and density of each TLS were obtained. A prospective cohort of 8 samples was used to compare pathologist and DIA identification of TLS. Normalized numbers of TLS in the tumor area were used for cohort stratification for overall survival (OS) analysis using the Kaplan-Meier method in an independent clinical cohort of 104 TCGA-LUAD patients. Results: A panel of 3 pathologists identified 326 unique TLS from 22 samples. Between-pathologist detection of TLS, independent of germinal center or HEV criteria, resulted in good reproducibility with an ICC of 0.77. Our DIA platform exhibited excellent reproducibility with an ICC of 0.94 when compared to validated prospective cohort annotation. In total, 155 and 189 TLS were identified by pathologists and our DIA platform, respectively. The DIA platform demonstrated a markedly improved sensitivity of 0.91 for TLS identification. Furthermore, OS analysis revealed that a TLS density greater than 0.94 TLS per mm2 of tumor assessed by DIA is a statistically significant independent biomarker of better OS in the LUAD cohort from TCGA. Conclusions: These results demonstrate the BostonGene DIA platform detects TLS in LUAD, with improved reproducibility and sensitivity over previous methods. Additionally, the DIA platform showed a TLS density greater than 0.94 TLS per mm2 of tumor is a positive prognostic marker for OS in LUAD. Standardized TLS DIA identification can be exploited in digital pathology applications for future clinical trials, informing clinicians of predictive and prognostic information during the decision-making process.

Despite a characteristic indolent course, a substantial subset of follicular lymphoma (FL) patients has an early relapse with a poor outcome. Cells in the microenvironment may be a key contributor to treatment failure. We used a discovery and validation study design to identify microenvironmental determinants of early failure and then integrated these results into the FLIPI. In total, 496 newly diagnosed FL grade 1–3 A patients who were prospectively enrolled into the MER cohort from 2002 to 2012 were evaluated. Tissue microarrays were stained for CD4, CD8, FOXP3, CD32b, CD14, CD68, CD70, SIRP-α, TIM3, PD-1, and PD-L1. Early failure was defined as failing to achieve event-free survival at 24 months (EFS24) in immunochemotherapy-treated patients and EFS12 in all others. CyTOF and CODEX analysis were performed to characterize intratumoral immunophenotypes. Lack of intrafollicular CD4 expression was the only predictor of early failure that replicated with a pooled OR 2.37 (95%CI 1.48–3.79). We next developed a bio-clinical risk model (BioFLIPI), where lack of CD4 intrafollicular expression moved patients up one FLIPI risk group, adding a new fourth high-risk group. Compared with BioFLIPI score of 1, patients with a score of 2 (OR 2.17; 95% CI 1.08–4.69), 3 (OR 3.53; 95% CI 1.78–7.54), and 4 (OR 8.92; 95% CI 4.00–21.1) had increasing risk of early failure. The favorable intrafollicular CD4 T cells were identified as activated central memory T cells, whose prognostic value was independent from genetic features. In conclusion, lack of intrafollicular CD4 expression predicts early failure in FL and combined with FLIPI improves identification of high-risk patients; however, independent validation is warranted.

Purpose: Multiparametric MRI (mpMRI) has become an indispensable radiographic tool in diagnosing prostate cancer. However, mpMRI fails to visualize approximately 15% of clinically significant prostate cancer (csPCa). The molecular, cellular, and spatial underpinnings of such radiographic heterogeneity in csPCa are unclear. Experimental Design: We examined tumor tissues from clinically matched patients with mpMRI-invisible and mpMRI-visible csPCa who underwent radical prostatectomy. Multiplex immunofluorescence single-cell spatial imaging and gene expression profiling were performed. Artificial intelligence–based analytic algorithms were developed to examine the tumor ecosystem and integrate with corresponding transcriptomics. Results: More complex and compact epithelial tumor architectures were found in mpMRI-visible than in mpMRI-invisible prostate cancer tumors. In contrast, similar stromal patterns were detected between mpMRI-invisible prostate cancer and normal prostate tissues. Furthermore, quantification of immune cell composition and tumor-immune interactions demonstrated a lack of immune cell infiltration in the malignant but not in the adjacent nonmalignant tissue compartments, irrespective of mpMRI visibility. No significant difference in immune profiles was detected between mpMRI-visible and mpMRI-invisible prostate cancer within our patient cohort, whereas expression profiling identified a 24-gene stromal signature enriched in mpMRI-invisible prostate cancer. Prostate cancer with strong stromal signature exhibited a favorable survival outcome within The Cancer Genome Atlas prostate cancer cohort. Notably, five recurrences in the 8 mpMRI-visible patients with csPCa and no recurrence in the 8 clinically matched patients with mpMRI-invisible csPCa occurred during the 5-year follow-up post-prostatectomy. Conclusions: Our study identified distinct molecular, cellular, and structural characteristics associated with mpMRI-visible csPCa, whereas mpMRI-invisible tumors were similar to normal prostate tissue, likely contributing to mpMRI invisibility.

Diffuse large B-cell lymphoma (DLBCL) is a biologically and clinically heterogeneous disease. Transcriptomic and genetic characterization of DLBCL has increased the understanding of its intrinsic pathogenesis and provided potential therapeutic targets. However, the role of the microenvironment in DLBCL biology remains less understood. Here, we performed a transcriptomic analysis of the microenvironment of 4,655 DLBCLs from multiple independent cohorts and described four major lymphoma microenvironment categories that associate with distinct biological aberrations and clinical behavior. We also found evidence of genetic and epigenetic mechanisms deployed by cancer cells to evade microenvironmental constraints of lymphoma growth, supporting the rationale for implementing DNA hypomethylating agents in selected patients with DLBCL. In addition, our work uncovered new therapeutic vulnerabilities in the biochemical composition of the extracellular matrix that were exploited to decrease DLBCL proliferation in preclinical models. This novel classification provides a road map for the biological characterization and therapeutic exploitation of the DLBCL microenvironment.Significance:In a translational relevant transcriptomic-based classification, we characterized the microenvironment as a critical component of the B-cell lymphoma biology and associated it with the DLBCL clinical behavior establishing a novel opportunity for targeting therapies.This article is highlighted in the In This Issue feature, p. 1307

The clinical use of molecular targeted therapy is rapidly evolving but has primarily focused on genomic alterations. Transcriptomic analysis offers an opportunity to dissect the complexity of tumors, including the tumor microenvironment (TME), a crucial mediator of cancer progression and therapeutic outcome. TME classification by transcriptomic analysis of >10,000 cancer patients identifies four distinct TME subtypes conserved across 20 different cancers. The TME subtypes correlate with patient response to immunotherapy in multiple cancers, with patients possessing immune-favorable TME subtypes benefiting the most from immunotherapy. Thus, the TME subtypes act as a generalized immunotherapy biomarker across many cancer types due to the inclusion of malignant and microenvironment components. A visual tool integrating transcriptomic and genomic data provides a global tumor portrait, describing the tumor framework, mutational load, immune composition, anti-tumor immunity, and immunosuppressive escape mechanisms. Integrative analyses plus visualization may aid in biomarker discovery and the personalization of therapeutic regimens.

Background Agonists of TLR3 and TLR4 are effective immunoadjuvants for different types of vaccines. The mechanisms of their immunostimulatory action differ significantly; these differences are particularly critical for immunization with non-replicating adenovirus vectors (rAds) based vaccines. Unlike traditional vaccines, rAd based vaccines are not designed to capture vaccine antigens from the external environment by antigen presenting cells (APCs), but rather they are targeted to the de novo synthesis of vaccine antigens in APCs transfected with rAd. To date, there is no clear understanding about approaches to improve the efficacy of rAd vaccinations with immunoadjuvants. In this study, we investigated the immunoadjuvant effect of TLR3 and TLR4 agonists on the level of activation of APCs during vaccination with rAds. Results We demonstrated that TLR3 and TLR4 agonists confer different effects on the molecular processes in APCs that determine the efficacy of antigen delivery and activation of antigen-specific CD4+ and CD8+ T cells. APCs activated with agonists of TLR4 were characterized by up-regulated production of target antigen mRNA and protein encoded in rAd, as well as enhanced expression of the co-activation receptors CD80, CD86 and CD40, and pro-inflammatory cytokines TNF-α, IL6 and IL12. These effects of TLR4 agonists have provided a significant increase in the number of antigen-specific CD4+ and CD8+ T cells. TLR3 agonist, on the contrary, inhibited transcription and synthesis of rAd-encoded antigens, but improved expression of CD40 and IFN-β in APCs. The cumulative effect of TLR3 agonist have resulted in only a slight improvement in the activation of antigen-specific T cells. Also, we demonstrated that IFN-β and TNF-α, secreted by APCs in response to TLR3 and TLR4 agonists, respectively, have an opposite effect on the transcription of the targeted gene encoded in rAd. Specifically, IFN-β inhibited, and TNF-α stimulated the expression of target vaccine antigens in APCs. Conclusions Our data demonstrate that agonists of TLR4 but not TLR3 merit further study as adjuvants for development of vaccines based on recombinant adenoviral vectors. Electronic supplementary material The online version of this article (10.1186/s12865-018-0264-x) contains supplementary material, which is available to authorized users.