Wang, Xue

Abstract This study benchmarks the robustness and resilience of computational deconvolution methods for estimating cell-type proportions in bulk tissues, with a focus on comparing reference-based and reference-free methods. Robustness is evaluated by generating in silico pseudo-bulk tissue RNA sequencing data from cell-level gene expression profiles derived from four different tissue types, with simulated cellular composition at varying levels of heterogeneity. To assess resilience, we intentionally alter single-cell RNA profiles to create pseudo-bulk tissue RNA-seq data. Deconvolution estimates are compared with ground truth using Pearson’s correlation coefficient, root mean squared deviation, and mean absolute deviation. The results show that reference-based methods are more robust when reliable reference data are available, whereas reference-free methods excel in scenarios lacking suitable reference data. Furthermore, variations in cell-level transcriptomic profiles and cell composition have emerged as critical factors influencing the performance of deconvolution methods. This study provides significant insights into the factors affecting bulk tissue deconvolution performance, which are essential for guiding users and advancing the development of more powerful and reliable algorithms in the future.

<p style='text-indent:20px;'>Complete deconvolution analysis for bulk RNA-seq data is important and helpful to distinguish whether the differences of disease-associated GEPs (gene expression profiles) in tissues of patients and normal controls are due to changes in cellular composition of tissue samples, or due to GEPs changes in specific cells. One of the major techniques to perform complete deconvolution is nonnegative matrix factorization (NMF), which also has a wide-range of applications in the machine learning community. However, the NMF is a well-known strongly ill-posed problem, so a direct application of NMF to RNA-seq data will suffer severe difficulties in the interpretability of solutions. In this paper, we develop an NMF-based mathematical model and corresponding computational algorithms to improve the solution identifiability of deconvoluting bulk RNA-seq data. In our approach, we combine the biological concept of marker genes with the solvability conditions of the NMF theories, and develop a geometric structures guided optimization model. In this strategy, the geometric structure of bulk tissue data is first explored by the spectral clustering technique. Then, the identified information of marker genes is integrated as solvability constraints, while the overall correlation graph is used as manifold regularization. Both synthetic and biological data are used to validate the proposed model and algorithms, from which solution interpretability and accuracy are significantly improved.</p>

Abstract Introduction: The immune microenvironment of high-risk HR+ breast cancer (BC) is poorly understood. BEAUTY is a prospective neoadjuvant chemotherapy (NAC) study of stage I-III BC patients treated with neoadjuvant weekly taxane followed by anthracycline-based chemotherapy. Among clinical luminal B BC (St. Gallen criteria) from BEAUTY, we used high-plex digital spatial profiling (DSP) to characterize the intra-epithelial tumor and stromal immune landscapes to 1) assess the impact of NAC on these landscapes and 2) identify immune biomarkers predictive of response to NAC. Methods: The tissue set included FFPE sections of resected tumors from 35 patients (median 51y; range: 21-71y) with clinically-defined luminal B BC (ER &gt; 10%/grade 2/Ki67 ≥ 15% or ER &gt; 10%/grade 3), and 16 paired pre-NAC biopsies. Nanostring GeoMX DSP was used to quantitate 58 immune and BC biomarker proteins in intra-epithelial, cytokeratin-positive tumor segments and adjacent stromal (cytokeratin-negative/SYTO13 (nuclear stain)-positive) segments. Data were normalized using the geometric mean of two negative controls (RbIgG and MsIgG1). A general linear model with negative binomial identified differentially-expressed (DE) proteins in pre/post-NAC tumors. Based on DE protein data and biologic function, a targeted protein subset (N=19) was evaluated in pre-NAC biopsies for associations between protein abundance and residual cancer burden (RCB) class (0-II vs. III) (Wilcoxon-rank sum test: p-value &lt; 0.025 considered significant) or RCB ‘breast only’ index (Spearman correlation). Results: Comparing tumor segments in pre-NAC specimens, intra-epithelial segments were predictably enriched in cytokeratin, ER, PR and Ki-67, but also CD127 and NY-ESO-1; whereas stromal segments were enriched in proteins associated with T cell subsets (e.g. CD3, CD4, CD8), macrophages (CD68 and CD163), antigen presentation/dendritic cells (CD11c, HLA-DR) and immune checkpoint proteins (PD-L1, PD-L2, B7-H3, TIM-3, VISTA) among others. While NAC did not alter the spatial distribution of proteins (intra-epithelial vs. stromal segments), it markedly attenuated the immune landscape, with decreased abundance of functionally-diverse immune proteins (e.g. CD45, CD3, CD4, CD127, granzyme B, CTLA4; STING, B7-H3, CD11c, and CD68, log2FC: 0.27-1.52 p &lt; 0.05), including low abundance proteins PD-1, PD-L1, PD-L2, CD20, and OX40L (log 2FC: 0.15-1.16, p &lt; 0.05). Both PR and Ki-67 decreased in post-NAC tumors whereas ER was not significantly altered (p &lt; 0.05). CD8, CD14, CD163, HLA-DR, IDO-1 and TIM-3 were not significantly altered by NAC. In the pre-NAC biopsies (which had subsequent residual tumor burdens of RCB class 0 or I (n=1 each), II (n=6), and III (n=8), and median RCB ‘breast only’ index of 3.39 (range: 0.00-38.02), 19 proteins were used for RCB analysis (CD3, CD4, CD8, CD14, CD34, CD44, CD45, CD68, CD127, CD163, CTLA4, granzyme B, STING, B7-H3, fibronectin, Ki-67, HLA-DR, SMA and TIM3). Among them, stromal CD127 was significantly higher in RCB class III than RCB class 0-II (p=0.021) and RCB ‘breast only’ index was positively correlated with intra-epithelial granzyme B (rho = 0.61; p=0.012), and negatively correlated with intra-epithelial Ki-67 (rho= -0.71; p=0.0022). Conclusion: In this series of clinical luminal B BC, NAC markedly attenuated the tumoral immune landscapes with a small set of “NAC-resistant” immune proteins. Among a targeted set, stromal CD127 was significantly higher in RCB III, and RCB breast-only index was positively correlated with intra-epithelial granzyme B. These data provide insight into the impact of NAC on HR+ BC, and identify potential immune biomarkers to predict response to neoadjuvant chemotherapy. Citation Format: Jodi M Carter, Judy C Boughey, Jun He, Vera J Suman, Xue Wang, Jennifer M Kachergus, Krishna R Kalari, Liewei Wang, Richard Weinshilboum, Ann M Moyer, Sarah A McLaughlin, Alvaro Moreno-Aspitia, Donald W Northfelt, Richard J Gray, James N Ingle, E. Aubrey Thompson, Matthew P Goetz. Neoadjuvant chemotherapy selectively alters spatially-defined immune landscapes in clinical luminal B HR+/HER2- breast cancers: Analysis of the breast cancer genome guided therapy study (BEAUTY) [abstract]. In: Proceedings of the 2020 San Antonio Breast Cancer Virtual Symposium; 2020 Dec 8-11; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2021;81(4 Suppl):Abstract nr PD7-05.

Abstract Background: The intrinsic immune response in triple negative breast cancer (TNBC) has both prognostic and predictive utility, and immune-related biomarkers are now actionable targets in TNBC. Current modalities for immune profiling are limited by lack of spatial context and precise quantitation. We used quantitative digital spatial profiling (DSP) to 1) characterize the immune architecture of tumoral and stromal regions in TNBC and 2) identify prognostic immune signatures. Methods: From the FinXX trial of adjuvant capecitabine therapy, tumor sections (FFPE) matched for patient characteristics, treatment arm and RFS-based outcome (N=44) were assessed with Nanostring GeoMxTM DSP. Briefly, with 3-plex immunofluorescence (pan-cytokeratin, CD45, CD68 with SYTO-13 dye for nuclei), spatially-defined CD45-enriched or CD68-enriched stromal segments and adjacent tumor segments (N=950) were selected for digital quantitation (NanoString nCounter) of 39 immune proteins (e.g. immune cell profiling proteins: CD3, CD4, CD8, CD11c, CD20, CD56, CD68, HLA-DR, PD-1, PD-L1, Granzyme B), immune drug targets (e.g. VISTA, STING, IDO1), and activation status-related proteins (e.g. ICOS, PD-L2, CD40, CD40L, CD44). Normalized, differentially-expressed (DE) proteins were evaluated, and weighted gene co-expression network analysis (WGCNA) was performed to identify modules of highly co-expressing immune proteins, and association with RFS as a categorical variable was determined. Results: 20 DE proteins in tumor segments were associated with RFS. Among them, those with the highest RFS-associated fold changes (FC) included CD20, CD40, CD56, HLA-DR, Granzyme B, and PD-L2 (FC): 2.1-4.4) (p &lt; 0.05). In contrast, only 4 DE proteins were associated with RFS in stromal segments (Granzyme B, IDO-1, PD-L1, and PD-L2 (FC: 1.5-3.3) (p &lt; 0.05). By WGCNA, 2 tumor-based immune modules [IM1 (lymphocyte infiltration) and IM2 (lymphocyte activation)] were identified based on their co-expression. The top three co-expressed proteins in IM1 were CD4, CD40, and CD45 (r &gt;0.9). The three top co-expressed proteins in IM2 were PD-L1, PD-L2, and Granzyme B (r &gt; 0.85). Together with CD56 expression, IM1 and IM2 comprised a set of 3 intratumoral immune features that associated with RFS. For recurring tumors, all 3 immune features clustered with low scores . Among non-recurrent tumors, two distinct, non-overlapping clusters with inverse characteristics were identified: high IM2, low IM1, low CD56 vs. low IM2, high IM1, and high CD56. Overall, across all 3 immune features, PDL2, CD40 and CD56 expression most strongly associated with RFS. Conclusion: With spatially-defined, precise quantification of immune proteins, we identified 3 immune features, including two distinct immune modules, IM1 and IM2 that correlate with durable RFS. These immune modules of highly co-expressing proteins were associated with RFS only when present in the tumor compartment but not tumor-adjacent, immune cell-rich stroma. Citation Format: Jodi M. Carter, Saranya Chumsri, Yaohua Ma, Xue Wang, Jennifer M. Kachergus, Douglas Hinerfeld, Heather Ann Brauer, Sarah Warren, Heikki Joensuu, Edith A. Perez, Aubrey Thompson. Quantitative digital spatial profiling reveals novel prognostic intratumoral immune signatures in triple-negative breast cancer [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 1004.

Abstract Objective: Allogeneic stem cell transplant (ASCT) is a potentially curative treatment for many hematologic malignancies. HLA match is the principal factor in donor selection but the total antigenic landscape as seen by the donor immune cells extends beyond the HLA and minor HLA antigens. Thus, despite "perfect" HLA match graft versus host disease (GVHD) caused by alloreactive donor immune cells reacting to disparate antigens in the recipient, remains a significant cause of morbidity and mortality. Whole exome sequencing (WES) which assays all the protein coding nucleotide sequences provides a powerful means to assess the entire antigenic makeup of the individual. We examined the WES of sibling donor-recipient pairs undergoing BMT to assess the influence of the extent of genetic disparity and specific gene polymorphisms on GVHD. Methods: Sixteen matched sibling donor-recipient pairs receiving allogeneic stem cell transplants between June 2008 and April 2013 had samples available for retrospective WES. Transplant data including conditioning regimen, age of donor and recipient at the time of transplant, development and type of GVHD and cause of death were recorded. The sequencing was performed using Illumina HiSeq2000 and the single nucleotide variants (SNVs) and small insertions and deletions (INDELs) were identified using the haplotype caller within the Genome Analysis Toolkit GATK v2.7. Functional variants including non-synonymous, frame-shift, truncating, and splice site SNVs and INDELs identified were compared between donor and recipient and among those with GVHD and those without GVHD (non-GVHD). One-sided Fisher's exact test was used to determine the statistical significance of the SNPs in GVHD and non-GVHD. Candidate genes with a p value of <0.05 were further evaluated for their biochemical pathway and possible role in inflammatory or immune responses. Results: Nine of 16 patients had a diagnosis of AML or high grade myelodysplastic syndrome, 3 had a diagnosis of non- Hodgkin lymphoma, 1 had Hodgkin lymphoma, 1 had multiple myeloma, and 2 had CML. Six patients were conditioned with a myeloablative regimen, others with a reduced intensity conditioning regimen. Eight of 16 patients were female and 7 of 16 donors were female. Nine of 16 patients developed stage II-IV acute and/or chronic GVHD with 7 of them developing GVHD of the skin. A total of 12,623 genes with 43,670 functional variants were identified using WES. 134 genes had a p value < 0.05 and enrichment analysis was performed on these genes. 12 out of 134 genes were associated with gene ontology (GO) term "immune system development" with an unadjusted enrichment p-value 0.0015. Three (IL31RA, OSMR, SIRPA) out of 134 genes were associated with "inflammation_Jak-STAT pathway" with an adjusted enrichment p-value 0.07. Variants of IL31RA, OSMR, and PTX3 were identified as genes with polymorphisms associated with GVHD (p= 0.003, 0.01, and 0.01, respectively) with 5 variants identified in IL31RA, 7 variants in OSMR, and 2 variants in PTX3. Further investigation of these three genes revealed their role in inflammatory or immune responses. Discussion: GVHD is a cytokine mediated immune-inflammatory process that can occur despite "perfect" HLA matching. WES identified the extensive genetic variability between donors and recipients implying large allo-reactivity potential that can lead to GVHD. Quantification of genetic variation may allow for more precise determination of GVHD risk and better selection of donors. In our cohort, WES identified polymorphisms in IL31RA, OSMR, and PTX3 as statistically significantly associated with development of GVHD. The protein products of these three genes have each been shown to be involved in inflammatory processes and are influenced by other cytokines. IL31RA and OSMR are found in many tissues including the skin and lungs. They function both independently and as a heterodimer receptor capable of binding multiple cytokines including IL-31, IL-6, and TNF-α. PTX3 is a reactive protein that increases in response to cytokine release related to systemic inflammation. Levels of this protein have been used as a prognostic marker in many disease processes with high levels correlating with poor outcomes. The small sample size of our cohort precludes any firm conclusions but provides promising hypotheses for further evaluation in larger studies. Disclosures No relevant conflicts of interest to declare.

Abstract Waldenströms Macroglobulinemia (WM) is a unique form of lymphoplasmacytic lymphoma, in which the tumor compartment is comprised of B-lymphocytes, plasmacytoid lymphocytes and plasma cells. The majority of WM cases are immunophenotypically characterized by expression of surface antigens that are present on both B-cells (CD19/20) and plasma cells (CD38/138). However, in advanced stage disease, which has been treated with various chemoimmunotherapeutics the molecular profile of WM cells can drastically shift resulting in a majority of the WM tumor compartment exhibiting loss of CD20 and concurrent upregulation of atypical surface antigens (Barakat et al, Am J Clin Pathol 2011; Morice et al Mod Pathol 2009). These external immunophenotypic changes are associated with underlying epigenetic and genomic alterations, whose integrated study may provide further insight into the biology of advanced stage WM. Moreover, such an analysis holds clinical potential by uncovering of therapeutic vulnerabilities, which can direct the selection of appropriate therapeutics for patients with CD20- WM. With this in mind, we conducted comprehensive molecular testing coupled with functional studies to understand if underlying genomic/epigenetic changes correlated with drug sensitivity using appropriate preclinical models of CD20+ WM (BCWM.1 cell line) and CD20- WM (RPCI-WM1 cell line). Importantly, the index patient from whom the CD20- cells were developed had received a total of 6 lines of treatment, two of which contained rituximab and was documented as being refractory to rituximab. Immunophenotyping of CD20- cells showed differential downregulation of 19 and upregulation of 7 CD antigens relative to CD20+ WM cells (cutoff at 1.5 fold; >20% gated expression). Upregulated antigens were associated with a phenotype reminiscent of a memory B-cell with plasmacytoid features. Next, we conducted RNA-Seq analysis (Illumina HiSeq), which revealed downregulation of 6,673 and upregulation of 4,594 genes (>1.5 fold cutoff) in CD20- WM cells compared with CD20+ cells. Gene pathway enrichment analysis (NextBio, Illumina Inc.) of mRNA differentially expressed in CD20- cells showed significant enrichment of upregulated genes corresponding to Protein Transport (p<1.8E-98) and Response to Unfolded Protein Gene Ontology groups (p<5.9E-20). Interestingly, pathway analysis of downregulated genes in CD20- WM cells showed enrichment within the B-cell receptor-signaling (BCR) pathway (Broad MSigDB) (p<6.1E-14) relative to CD20+ WM cells. BCR signaling has been linked to dysregulated expression of several micro-RNAs (miRs). Consistent with pathway analysis results, we observed decreased levels of numerous miRs (Exiqon qPCR) in CD20- vs. CD20+ WM cells, which target and regulate components involved in BCR activation. To examine further epigenetic changes that may account for decreased BCR signaling, we examined the global methylation profile (Infinium HumanMethylation 450K array) of CD20- and CD20+ WM cells and noted hypermethylation of the promoter sites (200 base pairs upstream) for a number of transcription factors involved in the BCR pathway. We therefore examined sensitivity of CD20- WM cells to the BTK inhibitor ibrutinib, which abrogates BCR signaling. As anticipated, CD20- WM cells retained >90% viability and demonstrated significantly less apoptosis in presence of ibrutinib (4uM) compared to CD20+ cells, suggesting decreased reliance on canonical BCR signaling for survival. In parallel and guided by pathway analysis results, we exposed both CD20- and CD20+ WM cells to a novel 19S proteasome inhibitor (VLX1570), which caused profound proteotoxicity and ER stress at concentrations ranging between 250 – 500nM. We noted CD20- cells were significantly more sensitive (p<0.01) to 19S proteasome inhibition than CD20+ WM cells indicating the formers critical dependence on tightly regulated protein homeostasis. In summary we describe for the first time, a comprehensive molecular and immunophenotypic investigation of CD20- WM cells compared to CD20+ WM cells, using validated preclinical models of the two disease phenotypes. Overall our data suggests significant architectural differences between CD20+ and CD20- WM, which can potentially impact the behavior of the cells and their sensitivity towards novel therapeutics being developed for treatment of advanced stage WM. Disclosures Martin: Janssen: Honoraria.