Gene Expression Deconvolution Research Articles

GLDADec: marker-gene guided LDA modeling for bulk gene expression deconvolution.

Inferring cell type proportions from bulk transcriptome data is crucial in immunology and oncology. Here, we introduce guided LDA deconvolution (GLDADec), a bulk deconvolution method that guides topics using cell type-specific marker gene names to estimate topic distributions for each sample. Through benchmarking using blood-derived datasets, we demonstrate its high estimation performance and robustness. Moreover, we apply GLDADec to heterogeneous tissue bulk data and perform comprehensive cell type analysis in a data-driven manner. We show that GLDADec outperforms existing methods in estimation performance and evaluate its biological interpretability by examining enrichment of biological processes for topics. Finally, we apply GLDADec to The Cancer Genome Atlas tumor samples, enabling subtype stratification and survival analysis based on estimated cell type proportions, thus proving its practical utility in clinical settings. This approach, utilizing marker gene names as partial prior information, can be applied to various scenarios for bulk data deconvolution. GLDADec is available as an open-source Python package at https://github.com/mizuno-group/GLDADec.

Distinctive whole-brain cell types predict tissue damage patterns in thirteen neurodegenerative conditions.

For over a century, brain research narrative has mainly centered on neuron cells. Accordingly, most neurodegenerative studies focus on neuronal dysfunction and their selective vulnerability, while we lack comprehensive analyses of other major cell types' contribution. By unifying spatial gene expression, structural MRI, and cell deconvolution, here we describe how the human brain distribution of canonical cell types extensively predicts tissue damage in 13 neurodegenerative conditions, including early- and late-onset Alzheimer's disease, Parkinson's disease, dementia with Lewy bodies, amyotrophic lateral sclerosis, mutations in presenilin-1, and 3 clinical variants of frontotemporal lobar degeneration (behavioral variant, semantic and non-fluent primary progressive aphasia) along with associated three-repeat and four-repeat tauopathies and TDP43 proteinopathies types A and C. We reconstructed comprehensive whole-brain reference maps of cellular abundance for six major cell types and identified characteristic axes of spatial overlapping with atrophy. Our results support the strong mediating role of non-neuronal cells, primarily microglia and astrocytes, in spatial vulnerability to tissue loss in neurodegeneration, with distinct and shared across-disorder pathomechanisms. These observations provide critical insights into the multicellular pathophysiology underlying spatiotemporal advance in neurodegeneration. Notably, they also emphasize the need to exceed the current neuro-centric view of brain diseases, supporting the imperative for cell-specific therapeutic targets in neurodegeneration.

Distinctive whole-brain cell types predict tissue damage patterns in thirteen neurodegenerative conditions

For over a century, brain research narrative has mainly centered on neuron cells. Accordingly, most neurodegenerative studies focus on neuronal dysfunction and their selective vulnerability, while we lack comprehensive analyses of other major cell types’ contribution. By unifying spatial gene expression, structural MRI, and cell deconvolution, here we describe how the human brain distribution of canonical cell types extensively predicts tissue damage in 13 neurodegenerative conditions, including early- and late-onset Alzheimer’s disease, Parkinson’s disease, dementia with Lewy bodies, amyotrophic lateral sclerosis, mutations in presenilin-1, and 3 clinical variants of frontotemporal lobar degeneration (behavioral variant, semantic and non-fluent primary progressive aphasia) along with associated three-repeat and four-repeat tauopathies and TDP43 proteinopathies types A and C. We reconstructed comprehensive whole-brain reference maps of cellular abundance for six major cell types and identified characteristic axes of spatial overlapping with atrophy. Our results support the strong mediating role of non-neuronal cells, primarily microglia and astrocytes, in spatial vulnerability to tissue loss in neurodegeneration, with distinct and shared across-disorder pathomechanisms. These observations provide critical insights into the multicellular pathophysiology underlying spatiotemporal advance in neurodegeneration. Notably, they also emphasize the need to exceed the current neuro-centric view of brain diseases, supporting the imperative for cell-specific therapeutic targets in neurodegeneration.

Characterizing Immune Infiltration in Esthesioneuroblastoma Subtypes Through Gene Expression Deconvolution

BackgroundEsthesioneuroblastoma (ENB) is a rare cancer deriving from the olfactory mucosa. Among the basal or neural genomic subtypes, the basal subtype is associated with poorer survival, and poor differentiation, and higher levels of tumor-infiltrating immune cells (TIICs). The immune microenvironment of these ENB subtypes remains unclear. We used an established machine-learning algorithm on ENB transcriptomic profiles. MethodsThe authors characterized 22 immune cell populations using the CIBERSORTx deconvolutional machine-learning pipeline on RNAseq data from 18 ENB cases. The characterization aimed to elucidate differences in relative proportions and populations of TIICs between basal and neural ENB. ResultsNo differences in age, Hyams, Dulguerov, IDH2 mutation, or PD-L1 expression were seen between basal and neural subtypes of ENB (p>0.05). No difference in median overall survival was also appreciated (52.0±13.1 vs. 50.0±43.2 months, p=0.5). As a cohort, M2 macrophages were the most abundant subpopulation (14%) followed by naïve B-cells (13%) and CD4 memory resting T cells (12%). No gross differences on CD20, CD4, or CD8 cells/mm2 were apparent on gross histology (p>0.05). However, further analysis showed that activated CD4 memory T cells were significantly increased in the basal ENBs, whereas resting dendritic cells were increased in the neural ENB subtype. The TIIC profiles alone could not differentiate between basal and neural ENB but did suggest immunoprofile differences. ConclusionBasal and neural subtypes display distinct TIIC involvement, which may impact their difference in outcome. These findings provide the framework for further investigation in novel immunomodulation strategies for ENB.

FSHD muscle shows perturbation in fibroadipogenic progenitor cells, mitochondrial function and alternative splicing independently of inflammation.

Facioscapulohumeral muscular dystrophy (FSHD) is a prevalent, incurable myopathy. FSHD is highly heterogeneous, with patients following a variety of clinical trajectories, complicating clinical trials. Skeletal muscle in FSHD undergoes fibrosis and fatty replacement that can be accelerated by inflammation, adding to heterogeneity. Well controlled molecular studies are thus essential to both categorize FSHD patients into distinct subtypes and understand pathomechanisms. Here, we further analyzed RNA-sequencing data from 24 FSHD patients, each of whom donated a biopsy from both a non-inflamed (TIRM-) and inflamed (TIRM+) muscle, and 15 FSHD patients who donated peripheral blood mononucleated cells (PBMCs), alongside non-affected control individuals. Differential gene expression analysis identified suppression of mitochondrial biogenesis and up-regulation of fibroadipogenic progenitor (FAP) gene expression in FSHD muscle, which was particularly marked on inflamed samples. PBMCs demonstrated suppression of antigen presentation in FSHD. Gene expression deconvolution revealed FAP expansion as a consistent feature of FSHD muscle, via meta-analysis of 7 independent transcriptomic datasets. Clustering of muscle biopsies separated patients in an unbiased manner into clinically mild and severe subtypes, independently of known disease modifiers (age, sex, D4Z4 repeat length). Lastly, the first genome-wide analysis of alternative splicing in FSHD muscle revealed perturbation of autophagy, BMP2 and HMGB1 signalling. Overall, our findings reveal molecular subtypes of FSHD with clinical relevance and identify novel pathomechanisms for this highly heterogeneous condition.

BEDwARS: a robust Bayesian approach to bulk gene expression deconvolution with noisy reference signatures

Differential gene expression in bulk transcriptomics data can reflect change of transcript abundance within a cell type and/or change in the proportions of cell types. Expression deconvolution methods can help differentiate these scenarios. BEDwARS is a Bayesian deconvolution method designed to address differences between reference signatures of cell types and corresponding true signatures underlying bulk transcriptomic profiles. BEDwARS is more robust to noisy reference signatures and outperforms leading in-class methods for estimating cell type proportions and signatures. Application of BEDwARS to dihydropyridine dehydrogenase deficiency identified the possible involvement of ciliopathy and impaired translational control in the etiology of the disorder.

Transcriptomic Comparison of Human Peripartum and Dilated Cardiomyopathy Identifies Differences in Key Disease Pathways

Peripartum cardiomyopathy (PPCM) is a rare form of acute onset heart failure that presents in otherwise healthy pregnant women around the time of delivery. While most of these women respond to early intervention, about 20% progress to end-stage heart failure that symptomatically resembles dilated cardiomyopathy (DCM). In this study, we examined two independent RNAseq datasets from the left ventricle of end-stage PPCM patients and compared gene expression profiles to female DCM and non-failing donors. Differential gene expression, enrichment analysis and cellular deconvolution were performed to identify key processes in disease pathology. PPCM and DCM display similar enrichment in metabolic pathways and extracellular matrix remodeling suggesting these are similar processes across end-stage systolic heart failure. Genes involved in golgi vesicles biogenesis and budding were enriched in PPCM left ventricles compared to healthy donors but were not found in DCM. Furthermore, changes in immune cell populations are evident in PPCM but to a lesser extent compared to DCM, where the latter is associated with pronounced pro-inflammatory and cytotoxic T cell activity. This study reveals several pathways that are common to end-stage heart failure but also identifies potential targets of disease that may be unique to PPCM and DCM.

Abstract 5401: Computational cancer cell gene expression deconvolution from tumor bulk RNA-seq via the machine learning algorithm Helenus

Abstract Background: Biomarker gene expression is becoming more commonly utilized for clinical decision-making in oncology clinical practice. However, complex tumor tissue comprises a population of cancer cells (CC) and the tumor microenvironment (TME), causing expression signals belonging to the CC and TME calculated from bulk RNA-seq of the tumor tissue to be indistinguishable. To circumvent this, Helenus, a gene expression deconvolution tool, was developed to estimate TME-specific gene expression, consequently, providing precise CC-specific gene expression. Methods: Helenus performs the “subtraction” of TME gene expression from the total expression calculated from bulk RNA-seq of the tumor tissue. To accurately reconstruct the CC expression profile, LightGBM gene models were trained on artificial transcriptomes created from > 1,000 different solid tumor cancer cell lines and > 3,000 samples of various TME cellular proportions. The LighGBM gene models included genes expressed predominantly in the TME (e.g., CD3E), both the TME and the CC (e.g., BCL6), or in the CC (e.g., HER2). The input features included: 1) RNA percentages of TME cell types predicted by the cell deconvolution tool Kassandra (Zaitsev et al., 2022); 2) evaluation of TME target gene expression via the estimation of its weighted average expression profile in TME cell populations; and 3) a set of TME- and CC-specific genes. The resulting predictions were adjusted based on the CC cell fraction. To evaluate Helenus’ performance, CC and TME RNA were mixed at different ratios using various cancer cell lines and peripheral blood-derived TME cell populations and suspensions of tumor cells prepared from cancer tissue across multiple tumor purity dilutions. Results: Helenus deconvolution resulted in an increased concordance correlation value from 0.73 to 0.98 between the real gene expression profile of pure CC and the reconstructed CC expression from bulk RNA-seq. Helenus showed high concordance between the gene expression profile of sorted cancer cell lines and the deconvolved gene expression across a wide range of CC RNA concentrations (20-90%) mixed with imitated TME RNA at varying concentrations. Helenus demonstrated high performance calculating gene expression of multiple clinically relevant biomarkers in the TME:cancer cell line mixes: CD274 (PD-L1) (mean absolute error [MAE] ~3.5-fold reduction); HLA-A (~2.8-fold MAE reduction); MKI67 (Ki-67, ~2.2-fold MAE reduction), ERBB2 (HER2, ~1.7-fold MAE reduction). Helenus deconvolved CC expression and found significant correlations with CC gene amplifications and deletions (e.g., BCL-2, VNN3) independent of tumor purity (p < 0.003). Conclusion: Helenus, the CC gene expression deconvolution tool, was developed with high accuracy to contribute to tumor diagnosis, disease monitoring, treatment decisions, and clinically relevant biomarker identification. Citation Format: Valentina Beliaeva, Ekaterina Ivleva, Boris Shpak, Daniil Litvinov, Anastasia Zotova, Krystle Nomie, Daniiar Dyikanov, Alexander Kuznetsov, Maria Savchenko, Aleksandr Zaitsev, Nathan Fowler, Alexander Bagaev. Computational cancer cell gene expression deconvolution from tumor bulk RNA-seq via the machine learning algorithm Helenus. [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 5401.

Abstract P4-09-10: Spatial protein and RNA expression in tumor, immune and stromal cells in BRCA1/2 mutated metastatic breast cancer

Abstract Background: Spatial transcriptomics and proteomics are cutting-edge techniques that allow for quantification of RNA and protein expression within separate cell populations or regions of interest in a tissue biopsy specimen. We aimed to independently characterize the tumor cells and immune infiltrate in metastatic site biopsy specimens from patients with somatic or germline BRCA1/2 mutations. Methods: We performed spatial transcriptomics and proteomics using the Nanostring GeoMx Digital Spatial Profiler (DSP) on 9 formalin-fixed paraffin-embedded (FFPE) tissue biopsy specimens from 5 patients with metastatic breast cancer and a known somatic or germline BRCA1/2 mutation. Each patient had 1-3 separate metastatic site tissue biopsy specimens. Cell populations of interest were delineated with fluorescently labeled antibodies to pan-cytokeratin on tumor cells, CD45 of immune cells, and a nuclear stain. On each sample, 4-5 regions of interest (ROIs) were selected near the interface of the metastatic tumor deposit and surrounding normal tissue. Each ROI was segmented into 3 areas of interest (AOIs): pan-cytokeratin-positive tumor cells, CD45-positive immune cells, and the remaining stromal cells negative for both pan-cytokeratin and CD45. Data was background corrected and normalized with internal negative and positive internal controls. Relative abundance of immune cell subtypes was derived from CD45+ cell gene expression data using the TIMER algorithm. Results: The expression of 57 proteins and 84 RNAs was quantified in each of the 396 AOIs. Separate analysis of CD45+ cells, pan-cytokeratin cells, and stromal cell populations confirmed expected differential expression of immune, tumor and stromal markers, respectively. Protein expression of the estrogen receptor (ER) and progesterone receptor (PR) on pan-cytokeratin-positive tumor cells using the DSP correlated with pathologist reviewed immunohistochemistry (IHC) for ER and PR. Significant differential expression was found within each cell population between hormone receptor (HR) positive and negative cohorts, including enrichment for S100B in HR negative samples. Comparison will be shown for expression of 14 overlapping genes/proteins. Immune cell compartment gene expression deconvolution showed heterogeneity in the relative proportion of infiltrating immune cell subtypes between samples and also between ROIs within a sample. Changes in gene and protein expression from a patient with 3 serial biopsy specimens can be tracked with substantial heterogeneity across distinct AOIs within individual specimens and across metastatic sites.. Conclusions: Spatial transcriptomics and proteomics is feasible for concurrent, relative quantification of protein and gene expression in tumor, immune, and stromal cells in FFPE tissue biopsies from patients with metastatic breast cancer harboring germline or somatic BRCA1 or BRCA2 mutations. Heterogeneity within individual samples and across serial time points offers opportunity for methods development in the analyses and interpretation of spatial transcriptomic/proteomic data. Citation Format: Katharine A. Collier, Katherine Miller, Zaibo Li, Jesse Westfall, David Tallman, Mark Vater, Gabriel Tinoco, Elaine Mardis, Daniel Stover. Spatial protein and RNA expression in tumor, immune and stromal cells in BRCA1/2 mutated metastatic breast cancer [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 P4-09-10.

Bioinformatics Tools for Bulk Gene Expression Deconvolution in Diabetic Retinopathy.

Retinal neovascularization is one of the leading causes of vision loss and a hallmark of proliferative diabetic retinopathy (PDR). The immune system is observed to be involved in the pathogenesis of diabetic retinopathy (DR). The specific immune cell type that contributes to retinal neovascularization can be identified via a bioinformatics analysis of RNA sequencing (RNA-seq) data, known as deconvolution analysis. Previous study has identified the infiltration of macrophages in the retina of rats with hypoxia-induced retinal neovascularization and patients with PDR through a deconvolution algorithm, known as CIBERSORTx. Here, we describe the protocols of using CIBERSORTx to perform the deconvolution analysis and downstream analysis of RNA-seq data.

Radiosensitivity is associated with antitumor immunity in estrogen receptor-negative breast cancer.

This study evaluated radiosensitivity and the tumor microenvironment (TME) to identify characteristics of breast cancer patients who would benefit most from radiation therapy. We analyzed 1903 records from the Molecular Taxonomy of Breast Cancer International Consortium cohort using the radiosensitivity index and gene expression deconvolution algorithms, CIBERSORT and xCell, that estimates the TME composition of tumor samples. In this study, patients were stratified according to TME and radiosensitivity. We performed integrative analyses of clinical and immuno-genomic data to characterize molecular features associated with radiosensitivity. Radiosensitivity was significantly associated with activation of antitumor immunity. In contrast, radioresistance was associated with a reactive stromal microenvironment. The immuno-genomic analysis revealed that estrogen receptor (ER) pathway activity was correlated with suppression of antitumor immunity. In ER-negative disease, the best prognosis was shown in the immune-high and radiosensitive group patients, and the lowest was in the immune-low and radioresistant group patients. In ER-positive disease, immune signature and radiosensitivity had no prognostic significance. Taken together, these results suggest that tumor radiosensitivity is associated with activation of antitumor immunity and a better prognosis, particularly in patients with ER-negative breast cancer.

Blockade of the Immune Checkpoint CD47 By TTI-621 Potentiates the Response to Anti-PD-L1 in Cutaneous T Cell Lymphoma

Background: Blockade of PD1/PD-L1 (NCT03011814) and CD47/SIRPα (NCT02890368) have shown promising and durable therapeutic outcomes in cutaneous T-cell lymphoma (CTCL), but resistance and/or relapses are seen in a subset of patients. To improve outcomes of immunotherapy regimens for CTCL patients, we investigated the effects of dual CD47 and PD-L1 blockade as a strategy for targeting both the innate and adaptive immune system. Methods: We performed RNA sequencing, gene expression deconvolution (CIBERSORT) and flow cytometry analysis to assess the immune cell composition, and CD47, SIPRα and PD-L1 expression in 45 CTCL patient samples and CTCL cell lines. CTCL specimens at baseline and during treatment with anti-CD47/SIRPα (TTI621) and anti-PD-L1 (durvalumab) were used to analyze immune cell gene expression. We performed functional assays utilizing our macrophage and CTCL cell line culture systems to assess macrophage polarization, phagocytic activity of macrophages against CTCL cells, T cell-mediated cytotoxicity using a chromium release assay, and CTCL cell line proliferation before and after CD47/SIRPα (TTI-621) and anti-PD-L1 antibody (durvalumab) treatment. Results: Our data showed that CTCL cells and CTCL cell lines overexpress CD47 and PD-L1 compared with healthy control cells, and overexpression of CD47 and PD-L1 was associated with high MYC expression in CTCL cell lines. CIBERSORT analysis of CTCL samples compared to healthy control indicated that the fraction of M2 tumor-associated macrophages (TAMs) was increased, while M0/1 were significantly lower. Our NanoString data showed decreased M2 genes, immature DC-related genes, and the NK cell inhibitory receptor related genes, but up-regulation of M1 genes and the NK cell stimulatory receptors associated genes for CTCL patients after treatment with TTI-621 compared to baseline. In vitro, treatment with TTI-621 increased the phagocytic activity of macrophages against CTCL cells and enhanced CD8+ T-cell-mediated killing. Moreover, TTI-621 synergized with anti-PD-L1 (durvalumab) to reprogram M2-like tumor-associated macrophages to M1-like phenotypes by inhibiting MYC expression. Simultaneous exposure to TTI-621 and anti-PD-L1 inhibited growth of CTCL cell lines more potently than either antibody alone. RNA sequencing analysis indicated these effects were mediated by cell‒death-related pathways, such as apoptosis, autophagy, and necroptosis. Conclusions: Collectively, our findings demonstrate that CD47 and PD-L1 are critical regulators of the immune microenvironment in CTCL and that dual targeting of CD47 and PD-L1 may potentiate anti-tumor responses in CTCL.

HiTIMED: hierarchical tumor immune microenvironment epigenetic deconvolution for accurate cell type resolution in the tumor microenvironment using tumor-type-specific DNA methylation data

BackgroundCellular compositions of solid tumor microenvironments are heterogeneous, varying across patients and tumor types. High-resolution profiling of the tumor microenvironment cell composition is crucial to understanding its biological and clinical implications. Previously, tumor microenvironment gene expression and DNA methylation-based deconvolution approaches have been shown to deconvolve major cell types. However, existing methods lack accuracy and specificity to tumor type and include limited identification of individual cell types.ResultsWe employed a novel tumor-type-specific hierarchical model using DNA methylation data to deconvolve the tumor microenvironment with high resolution, accuracy, and specificity. The deconvolution algorithm is named HiTIMED. Seventeen cell types from three major tumor microenvironment components can be profiled (tumor, immune, angiogenic) by HiTIMED, and it provides tumor-type-specific models for twenty carcinoma types. We demonstrate the prognostic significance of cell types that other tumor microenvironment deconvolution methods do not capture.ConclusionWe developed HiTIMED, a DNA methylation-based algorithm, to estimate cell proportions in the tumor microenvironment with high resolution and accuracy. HiTIMED deconvolution is amenable to archival biospecimens providing high-resolution profiles enabling to study of clinical and biological implications of variation and composition of the tumor microenvironment.

A novel molecular signature identifies mixed subtypes in renal cell carcinoma with poor prognosis and independent response to immunotherapy

BackgroundRenal cell carcinoma (RCC) is a heterogeneous disease comprising histologically defined subtypes. For therapy selection, precise subtype identification and individualized prognosis are mandatory, but currently limited. Our aim was to refine subtyping and outcome prediction across main subtypes, assuming that a tumor is composed of molecular features present in distinct pathological subtypes.MethodsIndividual RCC samples were modeled as linear combination of the main subtypes (clear cell (ccRCC), papillary (pRCC), chromophobe (chRCC)) using computational gene expression deconvolution. The new molecular subtyping was compared with histological classification of RCC using the Cancer Genome Atlas (TCGA) cohort (n = 864; ccRCC: 512; pRCC: 287; chRCC: 65) as well as 92 independent histopathologically well-characterized RCC. Predicted continuous subtypes were correlated to cancer-specific survival (CSS) in the TCGA cohort and validated in 242 independent RCC. Association with treatment-related progression-free survival (PFS) was studied in the JAVELIN Renal 101 (n = 726) and IMmotion151 trials (n = 823). CSS and PFS were analyzed using the Kaplan–Meier and Cox regression analysis.ResultsOne hundred seventy-four signature genes enabled reference-free molecular classification of individual RCC. We unambiguously assign tumors to either ccRCC, pRCC, or chRCC and uncover molecularly heterogeneous tumors (e.g., with ccRCC and pRCC features), which are at risk of worse outcome. Assigned proportions of molecular subtype-features significantly correlated with CSS (ccRCC (P = 4.1E − 10), pRCC (P = 6.5E − 10), chRCC (P = 8.6E − 06)) in TCGA. Translation into a numerical RCC-R(isk) score enabled prognosis in TCGA (P = 9.5E − 11). Survival modeling based on the RCC-R score compared to pathological categories was significantly improved (P = 3.6E − 11). The RCC-R score was validated in univariate (P = 3.2E − 05; HR = 3.02, 95% CI: 1.8–5.08) and multivariate analyses including clinicopathological factors (P = 0.018; HR = 2.14, 95% CI: 1.14–4.04). Heterogeneous PD-L1-positive RCC determined by molecular subtyping showed increased PFS with checkpoint inhibition versus sunitinib in the JAVELIN Renal 101 (P = 3.3E − 04; HR = 0.52, 95% CI: 0.36 − 0.75) and IMmotion151 trials (P = 0.047; HR = 0.69, 95% CI: 0.48 − 1). The prediction of PFS significantly benefits from classification into heterogeneous and unambiguous subtypes in both cohorts (P = 0.013 and P = 0.032).ConclusionSwitching from categorical to continuous subtype classification across most frequent RCC subtypes enables outcome prediction and fosters personalized treatment strategies.

Retinal Transcriptome and Cellular Landscape in Relation to the Progression of Diabetic Retinopathy.

PurposePrevious studies that identify putative genes associated with diabetic retinopathy are only focusing on specific clinical stages, thus resulting genes are not necessarily reflective of disease progression.This study identified genes associated with the severity level of diabetic retinopathy using the likelihood-ratio test (LRT) and ordinal logistic regression (OLR) model, as well as to profile immune and retinal cell landscape in progressive diabetic retinopathy using a machine learning deconvolution approach.MethodsThis study used a published transcriptomic dataset (GSE160306) from macular regions of donors with different degrees of diabetic retinopathy (10 healthy controls, 10 cases of diabetes, 9 cases of nonproliferative diabetic retinopathy, and 10 cases of proliferative diabetic retinopathy or combined with diabetic macular edema). LRT and OLR models were applied to identify severity-associated genes. In addition, CIBERSORTx was used to estimate proportional changes of immune and retinal cells in progressive diabetic retinopathy.ResultsBy controlling for gender and age using LRT and OLR, 50 genes were identified to be significantly increased in expression with the severity of diabetic retinopathy. Functional enrichment analyses suggested these severity-associated genes are related to inflammation and immune responses. CCND1 and FCGR2B are further identified as key regulators to interact with many other severity-associated genes and are crucial to inflammation. Deconvolution analyses demonstrated that the proportions of memory B cells, M2 macrophages, and Müller glia were significantly increased with the progression of diabetic retinopathy.ConclusionsThese findings demonstrate that deep analyses of transcriptomic data can advance our understanding of progressive ocular diseases, such as diabetic retinopathy, by applying LRT and OLR models as well as bulk gene expression deconvolution.

Determination of the Stage Composition of Plasmodium Infections from Bulk Gene Expression Data.

ABSTRACTMalaria symptoms are caused by the development of the parasites within the blood of an infected host. Bulk RNA sequencing (RNA-seq) of infected blood can reveal interactions between parasites and the host immune system during an infection, but because multiple developmental stages with distinct transcriptional profiles are concurrently present in infected blood, it is necessary to correct such analyses for differences in cell composition among samples. Gene expression deconvolution is a statistical approach that has been developed for inferring the cell composition of complex tissues characterized by bulk RNA-seq using gene expression profiles from reference cell types. Here, we describe the evaluation of a species-agnostic reference data set that can be used for efficient and accurate gene expression deconvolution of bulk RNA-seq data generated from any Plasmodium species and for correct gene expression analyses for biases caused by differences in stage composition among samples.IMPORTANCE Differences in cell type proportions among samples can introduce artifacts in gene expression analyses and mask genuine differences in gene regulation. Gene expression deconvolution allows estimation of the proportion of each cell type present in one sample directly from bulk RNA sequencing data, but this approach requires a reference data set with the signature profile of each cell type. Here, we evaluate the suitability of a rodent malaria parasite gene expression data set for estimating the proportions of each parasite developmental stage present in bulk RNA sequencing data generated from blood-stage infections with the human parasites Plasmodium falciparum and Plasmodium vivax. These analyses provide a species-agnostic approach for reliably estimating stage proportions in infected human blood and correcting subsequent gene expression analyses for these variations.

Abstract 3355: The Offonome: Methods for characterizing gene expression of rare cell populations from whole tumor profiling data

Abstract Background: Tumors are composed of diverse cell types. While several approaches have been proposed to dissect the cellular heterogeneity using profiling data such as gene expression or methylation, it remains challenging to discover rare cell types from high-throughput sequencing data. To address this need, we apply a novel approach in which we hypothesize that rare cell subpopulations might be detected based on the expression of differential sets of transcripts, generally expressed at low levels proportional to the infiltrating cell percentage. In this case, the challenge is the precise measurement of gene expression at low levels or the detection of genes specifically attributable to those cell populations as being “off” or “on”. Methods: We propose a novel computational approach for measuring the “offonome”, defined as a gene’s dichotomized state of being “on” or “off”. To accurately determine on/off status of a gene, we compare single nucleotide-level RNA-seq data modeled as vectors in high dimensional data space using the angles between samples. In this framework, sample vectors determined to have low angles can be defined as sharing an “on” gene versus sample vectors with high angle defining a state of “off”. We apply this approach to more than 5,000 human cancer samples drawn from the Cancer Genome Atlas (TCGA) to identify minor cell populations. Additionally, we compare this approach to the alternative gene abundance strategy in which on/off is defined by read-counts-based gene expression estimates from the same dataset. Results: Using the overall and sub-populations, we identified minor cell populations across the TCGA dataset including in a sub-cohort of 522 TCGA head and neck squamous cell carcinoma (HNSCC). The Offonome approach applied genome-wide in HNSCC identified 5802 genes that were either always or sometimes “off” defined by angle-based statistics. Turning our attention to genes that were sometime, but not always off, we applied unsupervised learning to characterize clusters of genes within HNSCC tumors. At least five gene clusters were clearly observed, each of which was strongly associated with specific signaling pathways. For example, one cluster identified infiltrating myocytes, defined by infiltrating normal muscle cells, found only in tumors of the oral tongue. In addition to clustering analysis, we utilized the approach to identify high expression of genes found in outliers, or even a single sample such as those associated with salivary origin such as olfactory receptors. Conclusions: We conclude that on/off status of a gene provides new insights into gene expression deconvolution, thereby characterizing interesting gene clusters. These results offer strong evidence that our approach can be useful for picking up rare transcripts from individual samples such as leukocyte transcripts and stromal transcripts. Citation Format: Hyo Young Choi, David N. Hayes. The Offonome: Methods for characterizing gene expression of rare cell populations from whole tumor profiling data [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 3355.

Abstract 1212: ExTIME: Extended tumor immune micro-environment cell mixture deconvolution using DNA methylation and a novel tumor-site-specific hierarchical approach

Abstract Background: The solid tumor microenvironment is heterogeneous and varies in composition by tumor type. Previous gene expression and DNA methylation deconvolution approaches for tumor micro-environment have had some success for major cell types. However, existing methods lack specificity to tumor type and detailed cell types. We developed 21 tumor-specific DNA methylation-based libraries. We employed a novel hierarchical approach in 3 major tumor microenvironment components (tumor, angiogenic, immune) to profile 17 cell types (see methods below). Methods: DNA methylation data on tumor samples (n=6183) and normal control samples (n=689) for 21 tumor sites were downloaded from GEO and TCGA to develop tumor-type-specific libraries. The top 1000 most informative differentially methylated CpG (DMC) sites were identified using InfiniumPurify for 21 tumor types to project tumor cell proportion. Epithelial, endothelial, and stromal cell samples were used to identify DMCs to profile the cells in the angiogenic environment. Basophil, eosinophil, neutrophil, dendritic cell, monocyte, B naïve, B memory, CD4T naïve, CD4T memory, CD8T naïve, CD8T memory, T regulatory, and natural killer cells were used to identify DMCs to deconvolve the immune environment. In conjunction with the constrained projection/quadratic programming approach, a novel hierarchical approach was employed with six layers and 12 libraries per tumor type to project cell proportions in first, tumor, second, angiogenic, and third, immune micro-environments. The method was validated using purified samples and experimental artificial mixtures. Results: 12 libraries were developed per tumor site to deconvolve 17 cell types in 21 tumors. A preliminary application of the method on TCGA data investigating the association between angiogenic cells and survival revealed worse survival outcomes with a higher proportion of angiogenic cell proportions in BLCA (p<0.01) and HNSC (p=0.02), a higher endothelial cell proportion in CESC (p=0.04), a higher epithelial cell proportion in COAD (p=0.02), a lower endothelial proportion in KIRC (p<0.01), and a lower epithelial proportion in LUAD (p=0.04). Further analyses will be done to investigate the angiogenic and immune microenvironments with prognosis across tumor sites. Conclusion: We developed a DNA methylation-based algorithm, ExTIME, to estimate cell proportions in the tumor microenvironments. This novel approach increased the specificity and accuracy of cell projection by employing a tumor-site-specific hierarchical model. Furthermore, the ExTIME profiles the tumor microenvironment to the most granular level compared to the existing methods. ExTIME’s capability of depicting the cellular composition in tumors promises a better understanding of the cell heterogeneity and its relationship with prognosis across cancers. Citation Format: Ze Zhang, Brock C. Christensen, Lucas A. Salas. ExTIME: Extended tumor immune micro-environment cell mixture deconvolution using DNA methylation and a novel tumor-site-specific hierarchical approach [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 1212.

Cell type and gene expression deconvolution with BayesPrism enables Bayesian integrative analysis across bulk and single-cell RNA sequencing in oncology

Inferring single-cell compositions and their contributions to global gene expression changes from bulk RNA sequencing (RNA-seq) datasets is a major challenge in oncology. Here we develop Bayesian cell proportion reconstruction inferred using statistical marginalization (BayesPrism), a Bayesian method to predict cellular composition and gene expression in individual cell types from bulk RNA-seq, using patient-derived, scRNA-seq as prior information. We conduct integrative analyses in primary glioblastoma, head and neck squamous cell carcinoma and skin cutaneous melanoma to correlate cell type composition with clinical outcomes across tumor types, and explore spatial heterogeneity in malignant and nonmalignant cell states. We refine current cancer subtypes using gene expression annotation after exclusion of confounding nonmalignant cells. Finally, we identify genes whose expression in malignant cells correlates with macrophage infiltration, T cells, fibroblasts and endothelial cells across multiple tumor types. Our work introduces a new lens to accurately infer cellular composition and expression in large cohorts of bulk RNA-seq data.

Abstract P5-06-16: Radiosensitivity and immune cell infiltration signature predict clinical outcome of patients in the molecular taxonomy of breast cancer international consortium (METABRIC) study cohort

Abstract Purpose/Objectives There are growing evidences that tumor radiosensitivity is associated with immune activation in solid tumors. However, previous studies have not examined the relationship between patterns of immune infiltration and radiation response according to diverse molecular subtypes of breast cancer. This study evaluated radiosensitivity and immune infiltration signature to define a group of patients would benefit most from radiation therapy. Materials/Methods We performed integrative analyses of the clinical, genomic, transciptomc, and immunogenomic data to characterize the molecular features associated with radiosensitivity. We analyzed 1,903 data sets of METABRIC Study cohort using the radiosensitivity index (RSI), CIBERSORT and xCell, gene expression deconvolution algorithm which estimates the immune composition of tumor samples. According to RSI cut-point (0.3745), the patients were grouped into radiosensitive (RS, RSI-low tumors, n=1,082) versus radioresistant (RR, RSI-high tumors, n=821). Next, patients were divided into two groups accoring to immune and stromal scores, immune infiltrated (IF, n= 650) and immune excluded (IE, n=1253) subgroups. and performed integrative analyses of the clinical, genomic, transciptomc, and immunogenomic data to characterize the molecular features associated with radiosensitivity. Results RS group showed improved prognosis compared to RR group, and IF group had better survival than IE group, however, these tendency was only significant in ER negative group. Radiosensitivity was significantly associated with activation of antitumor immunity. In contrast, radio-resistance was associated with metastatic properties, such as epithelial mesenchymal transition and angiogenesis. Differentially expressed gene analysis revealed that ER signaling pathway is correlated with suppression of antitumor immunity. Higher ER was associated with higher fraction of M2 macrophage and lower PD-1 expression. Integration of immune signature and radiosensitivity index further stratified patients into four subgroups. In ER-negative disease, IF and RS group were associated with the best prognosis, whereas in ER-positive disease, immune signature and radiation response have no prognostic significance. On multivariable cox regression, both integrated immune infiltration and radiosensitivity signature was independent predictors of relapse free survival (HR 1.12, 95% CI 1.06-1.19, p < 0.001) and overall survival (HR 1.06, 95% CI 1.01-1.11, p = 0.028). Conclusion Taken together, this results suggested that tumor radiosensitivity was associated with activation of antitumor immunity and led to better relapse free survival particularly in ER negative group. Citation Format: Byung-Hee Kang. Radiosensitivity and immune cell infiltration signature predict clinical outcome of patients in the molecular taxonomy of breast cancer international consortium (METABRIC) study cohort [abstract]. In: Proceedings of the 2021 San Antonio Breast Cancer Symposium; 2021 Dec 7-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2022;82(4 Suppl):Abstract nr P5-06-16.