Proinflammatory GSDMD activation in live macrophages and DLBCL cells marks cognate interactions and better prognosis

Key Points

• •Cytoplasmic cleaved GSDMD in live activated macrophages and DLBCL cells marks cognate interactions and predicts better patient survival.
• •High-level death-resistance of DLBCL cells, T-cell pyroptosis, and endothelial GSDME expression predict poorer patient survival in DLBCL.

Visual Abstract

The Visual Abstract was created by Xiaoya Xu-Monette using Adobe Photoshop.

Abstract

Pyroptosis is a form of programmed cell death characterized by the cleavage of the gasdermin (GSDM) family proteins that form pores in the plasma membrane, cell rupture, and the release of proinflammatory cytokines. In this study, we performed immunohistochemistry for cleaved gasdermin D (GSDMD), gasdermin E (GSDME) N-terminus, and gasdermin C (GSDMC) in 2 different cohorts of diffuse large B-cell lymphoma (DLBCL) and analyzed the impact of GSDM expression on prognosis and immunity. The results showed frequent cleaved GSDMD (N-terminal) expression. Only cytoplasmic GSDMD N-terminal expression correlated with significantly better patient survival in the 2 cohorts. In contrast, GSDME was mainly expressed in the vascular endothelium, and correlated with significantly adverse prognostic effect. Correlating with the multiplex fluorescent immunohistochemistry results, we found that cytoplasmic GSDMD N-terminal expression was associated with increased CD38⁺ (activated) M1 macrophages in both cohorts, cognate interactions between live DLBCL cells and activated M1 macrophages (and T cells), and lower PD-1/PD-L1 expression in the analyzed cases. In contrast, T-cell pyroptosis, lymphoma cell-resistance to cell death, and phagocytosis by M2 macrophages were observed in tissues with predominantly nuclear GSDMD N-terminal expression. Bulk gene expression profiling and deconvolution analysis revealed associations of cytoplasmic GSDMD N-terminal expression with downregulation of “Don’t eat me”–signaling genes, upregulation of many RNA genes, decreased frequency of “inflammatory” lymphoma microenvironment subtype, increased frequencies of prognostically favorable cell states and ecotypes, and decreased frequency of T-cell exhaustion state. In summary, this study showed distinct cellular and subcellular patterns of 3 GSDM proteins and their associated immune response phenotypes and prognostic effects, with implications for novel therapeutic strategies for B-cell lymphoma.

Introduction

Diffuse large B-cell lymphoma (DLBCL), the most prevalent type of aggressive lymphoid neoplasm, is characterized by the diffuse proliferation of malignant large B cells with large nuclei derived from germinal center B cells, which normally are selected through apoptosis.^1,2 DLBCL has a high complete response rate to standard chemoimmunotherapy with R-CHOP (rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone). Despite this, 10% to 15% of cases are refractory, and 20% to 25% of patients relapse after achieving an initial response. Standard salvage therapy with high-dose chemotherapy followed by autologous stem cell transplantation or with anti-CD19 chimeric antigen receptor (CAR) T cells can rescue only ∼40% of refractory/relapsed patients.^3,4

Pyroptosis, a novel type of cell death different from apoptosis, has garnered attention in cancer research. It is a lytic and inflammatory form of programmed cell death that was initially discovered as an antimicrobial mechanism in macrophages.^5,6 The pyroptosis pathway in macrophages was delineated 10 years ago, that upon inflammasome activation,⁷ gasdermin D (GSDMD) protein is cleaved by caspases to release the N-terminal fragment from its autoinhibitory C-terminal domain,8, 9, 10 enabling subsequent oligomerization and pore formation on cell membranes, which leads to the release of cytosolic contents and proinflammatory cytokines such as interleukin 1β (IL-1β) and IL-18 and pyroptosis of macrophages.^8,11, 12, 13, 14 However, a previous study¹⁵ showed that GSDMD pore formation is not sufficient for macrophage cell death, and when membrane rupture is blocked, it leads to the hyperactivation state of living macrophages secreting IL-1 for an extended time. Another prior study¹⁶ showed that the hyperactive cell state of IL-1–releasing living dendritic cells potentiates T-cell activation and adaptive immune responses; however, GSDMD was not analyzed in that study. Importantly, a new-generation CAR T-cell product, huCART19-IL18, which produces IL-18 (which can be released via pyroptosis), has shown an overall response rate of 80% in patients with B-cell lymphoma refractory to or relapsed after anti-CD19 CAR T-cell therapy.¹⁷

Moreover, gasdermin E (GSDME) N-terminal activities were discovered to induce pyroptosis in various solid cancers and enhance the antitumor effects of chemotherapies and killer lymphocytes (including CAR T cells).^18,19 In contrast, gasdermin C (GSDMC) was shown to have oncogenic activities associated with chemoresistance,²⁰ including activating the AKT/mTOR pathway.²¹

DLBCL has the highest bulk GSDMD expression across 33 cancer types (http://gepia2.cancer-pku.cn/#general); all high-grade non-Hodgkin lymphoma samples in the Human Protein Atlas (proteinatlas.org)^22,23 showed cytoplasmic expression of GSDMD and GSDME proteins upon testing via immunohistochemistry (IHC). However, whether lymphoma cells undergo pyroptosis and confer clinical significance in patients with DLBCL has not been studied.

In this study, we performed IHC for 3 gasdermins (GSDMs) in a large number of DLBCL samples and determined their prognostic impact. To gain insights into the cell types expressing GSDM proteins and the cell states of lymphoma cells, macrophages, and T cells in the tumor microenvironment (TME), we examined single-cell biomarkers on multiplex fluorescent IHC (mfIHC) images and correlated different types of GSDM expression with quantified biomarker expression and bulk gene expression profiling (GEP) data.

Patients and methods

Patients

A total of 523 patients with de novo DLBCL in a large DLBCL cohort from the International DLBCL Consortium Program were evaluated for GSDMD, GSDME, and GSDMC expression using diagnostic samples, including 369 patients treated with R-CHOP and 154 patients treated with CHOP after diagnosis. Serial formalin-fixed, paraffin-embedded tissue microarrays (TMAs) were constructed using leftover pathology specimens.²⁴

IHC for GSDMD, GSDME, and GSDMC

IHC for GSDMD, GSDME, and GSDMC was performed on 3 sets of 14 TMA slides using a Ventana Discovery Ultra automated platform (Roche Diagnostics, Indianapolis, IN). The antibodies used were cleaved GSDMD (Asp275) (E7H9G) rabbit monoclonal antibody 36425 (1:200) from Cell Signaling Technology, Inc (Danvers, MA; Research Resource Identifier: AB_2799099); anti–GSDME-N-terminal rabbit monoclonal antibody (1:100; catalog no. GTX81693) from GeneTex, Inc (Irvine, CA; Research Resource Identifier: AB_11179150); and GSDMC polyclonal antibody (1:100; NBP1-91926) from Novus Biologicals (Centennial, CO). Antigen retrieval was performed using Ventana Cell Conditioning 1 for 36 minutes. The primary antibody was incubated for 60 minutes, and the signal was developed with anti-rabbit horseradish peroxidase. Finally, the ChromoMap DAB Detection Kit (Roche Diagnostics, Indianapolis, IN) and hematoxylin were used for counterstaining. GSDMD, GSDME, and GSDMC staining was evaluated for 357, 334, and 334 patients in the R-CHOP cohort and 154, 141, and 149 patients in the CHOP cohort, respectively.

mfIHC

As previously described, mfIHC using 13 antibodies conjugated to either Cy3 or Cy5 was performed on 1 set of serial TMAs to analyze immune checkpoint expression.²⁵ Programmed cell death 1 (PD-1), PD-1 ligand 1 (PD-L1), PD-1 ligand 2 (PD-L2), and Cytoxic T lymphocyte associated protein 4 (CTLA-4) expression in B cells, T cells, macrophages, and natural killer cells were digitally quantified for pathologist-marked tissue regions of interest (ROIs) in tissues from patients who were treated with R-CHOP after biopsy, including 285 patients in the GSDM discovery cohort in this study.

New mfIHC was performed²⁶ on another set of serial TMAs of DLBCL samples (including 12 TMAs in this GSDM study) using the same MultiOmyx immunofluorescence platform (NeoGenomics Laboratories, Aliso Viejo, CA) and methods as previously described for the immune checkpoint study²⁵ and a new primary antibody panel: rabbit anti–c-Myc (EP121/Y69; Epitomics), mouse anti–Bcl-2 (100/D5; Leica Biosystems), anti-CD38 (SPC32; Leica Biosystems), anti-PAX5 (D7H5X; Cell Signaling Technology, Inc), anti-CD8 (C8/144B; Dako), anti-CD31 (89C2; Cell Signaling Technology, Inc), anti-CD4 (EPR6855; Abcam), anti-CD11c (D3V1E; Cell Signaling Technology, Inc), anti-CD3 (F7.2.38; Dako), anti-CD20 (EP459Y; Abcam), anti-CD19 (LE-CD19; Bio-Rad), anti-CD68 (KP1; BioLegend), and anti-CD163 (EDHu-1; Bio-Rad). Fluorescently labeled secondary antibodies were obtained from Jackson ImmunoResearch Laboratories, Inc. Sequential staining images of ROIs were acquired using a 20× camera on an IN Cell Analyzer 2200 microscope (GE HealthCare Life Sciences) equipped with high-efficiency fluorochrome-specific filter sets for DAPI (4′,6-diamidino-2-phenylindole), Cy3, and Cy5. A series of algorithms (https://www.neogenomics.com/Paletrratm/)²⁵ were applied, and single cells were classified as positive or negative for 13 markers based on the immunofluorescence intensity detected in the membrane, cytoplasmic, or nuclear regions of single cells.

GEP

GEP Affymetrix chip data (GSE31312²⁷) available for 293 cases were normalized using the Robust Multichip Average algorithm implemented in the R package affy (release 3.16). Two-class unpaired significance analysis of microarrays was performed using the R package samr (version 3.0). Heat maps were generated using the Cluster 3.0 software and Java TreeView.

Tools used to deconvolute the bulk GEP data to infer TME characteristics included the Lymphoma EcoTyper framework (https://github.com/digitalcytometry/ecotyper), which estimates the proportion of cell states of 13 cell types and classifies lymphoma ecotypes (LE),²⁸ and the lymphoma microenvironment (LME) classifier (https://github.com/BostonGene/LME) through gene set enrichement analysis of 22 gene signatures of microenvironment components.^29,30

Statistical analysis

Overall survival (OS) was defined as the time from diagnosis to the last follow-up or death due to any cause. Progression-free survival (PFS) was defined as the time from diagnosis to the last follow-up or disease progression. The Kaplan-Meier method with the log-rank test was used to compare the OS and PFS rates. Comparisons of characteristics between the 2 groups were performed using the χ² test and unpaired t test or U test.

This study was conducted in accordance with the principles of the Declaration of Helsinki. Material transfer agreements and data collection protocols were approved as minimal to no risk or were exempted by the institutional review board of each participating institution.

Results

Cytoplasmic cleaved GSDMD expression is associated with significantly better OS in the 2 DLBCL cohorts

Serial dilutions of antibodies were tested in 2 normal tonsil samples, and 3 GSDM antibodies at optimal dilutions were selected to stain 3 sets of 14 TMA slides with variable numbers of DLBCL and control tissues. The clinical features of the patients are presented in supplemental Table 1. IHC showed that cleaved GSDMD N-terminal (GSDMD-n) expression was the most prevalent, showing >1% expression in 78.5% of DLBCL tissues, yet the abundance, staining intensity, and cellular and subcellular location of GSDMD-n were significantly heterogeneous in the DLBCL cohort. We evaluated GSDM expression by H scores, the positivity percentage multiplied by staining intensity (1+, weak; 2+, moderate; 3+, strong), and 2 subcellular types of expression : nuclear and cytoplasmic (including membranous). Figure 1A and supplemental Figure 1A-B show histograms and scatterplots for the percentages and H scores of GSDMD-n⁺ cells in the DLBCL cohort. In the R-CHOP cohort, 26 samples showed only cytoplasmic expression, 131 samples showed predominantly nuclear expression, 178 samples showed both nuclear and cytoplasmic expression, and 22 samples were GSDMD⁻. In the CHOP cohort, the number of cases with the abovementioned expression patterns was 12, 41, 93, and 8, respectively.

Figure 1.

Prevalence and prognostic effects of GSDMD-n expression in patients with DLBCL. (A) Top: a representative IHC image of a DLBCL pretreatment sample using an antibody against cleaved GSDMD (Asp275) with 1:200 dilution. Original magnification ×40. Bottom: a histogram for the patient numbers at cleaved GSDMD expression levels regardless of cytoplasmic or nuclear location and a histogram for the patient numbers at cytoplasmic cleaved GSDMD expression levels in the DLBCL study cohort. Patients with 3% and <3% expression levels were plotted in the bars at 5% and 0% expression levels (bin centers), respectively. (B) Cytoplasmic GSDMD-n expression, with a cutoff of H score >6, was associated with significantly better OS of patients with DLBCL in both the R-CHOP and CHOP cohorts. Patients were treated with R-CHOP or CHOP after biopsy and diagnosis. (C) Very high nuclear expression of GSDMD-n, with a cutoff of H score > 90, was associated with significantly poorer OS among patients treated with R-CHOP and a strong trend of poorer OS in the CHOP cohort. HR, hazard ratio.

To determine the clinical relevance, we first correlated GSDMD-n expression with patient survival in the discovery cohort (patients treated with R-CHOP after diagnosis) and then tested the prognostic significance in the validation cohort (those with CHOP treatment) using the same cutoffs. The CHOP cohort compared with the R-CHOP cohort had significantly lower proportions of patients with B symptoms at presentation and tumors with Ki-67 ≥70%, RELA/p65 expression, p63 expression, and BCL6 rearrangement and significantly higher proportions of patients with Eastern Cooperative Oncology Group scores of >1; bulky tumors; REL proto-oncogene c-Rel, Germinal center B-cell expressed transcript 1 protein (GCET1), CD10, and MDM2 expression; and radiotherapy (supplemental Table 1), as well as higher mean/median H scores of GSDMD-n expression (supplemental Figure 1B).

We found that cytoplasmic GSDMD-n expression, with an optimal expression cutoff of H score >6, correlated with significantly better OS in both the discovery and validation cohorts, stratifying 38% and 56% of the discovery and validation cohorts, respectively (Figure 1B). The effects on PFS were also favorable in both cohorts and reached significance in the CHOP cohort (supplemental Figure 1C). In contrast, very high nuclear GSDMD-n expression, with an H score cutoff of 90, was observed in only 2% and 9% of the R-CHOP and CHOP cohorts, respectively, and was associated with significantly poorer OS and PFS in the discovery cohort and a trend of poorer OS in the validation cohort (Figure 1C; supplemental Figure 1D). If cytoplasmic and nuclear expression were not differentiated, GSDMD-n positivity (any type) did not show a significant prognostic effect in the DLBCL cohort.

The DLBCL cohort has been molecularly characterized in our previous studies. Comparisons between cytoplasmic GSDMD-n⁺ and GSDMD-n⁻ patients in the R-CHOP cohort, as shown in supplemental Table 2, revealed that cytoplasmic GSDMD-n⁺ patients had significantly lower frequencies of NF-κB p50 and p52 protein expression; higher frequencies of AKT^high and IL-6 expression^24,31, 32, 33; and no significant differences in frequencies of TP53 mutations³⁴ or overexpression of MDM2,³⁵ survivin,³⁶ p16, p21, MYC, and BCL2³⁷ via IHC. Despite the low number of cases with high nuclear GSDMD-n expression, we compared these cases with others and found that these cases in the R-CHOP cohort had significantly higher proportions of patients presenting with B symptoms, p65 expression, and loss of CD37 and p63 expression^24,38 (supplemental Table 3). However, these significant associations were not validated in the CHOP cohort (without IL-6 data available), although loss of CD37 was also more frequent (75% vs 63%) in the nuclear GSDMD-n^high cases.

Cytoplasmic GSDMD-n expression is associated with increased CD38⁺ M1 macrophages and lower PD-1/PD-L1 expression in DLBCL

Both the discovery and validation cohorts were analyzed via 13-plex fluorescent IHC for 3 myeloid lineage markers (CD68, CD11c, and CD163), 6 B-cell–related markers (CD20, CD19, PAX5, MYC, BCL2, and CD38),³⁹ 3 T-cell markers (CD3, CD4, and CD8), and an endothelial marker (CD31). Immunofluorescence counts of single and multiple markers in the ROIs were digitally quantified for tissues that passed both tissue and staining quality control in all staining rounds. Quantification data were obtained for 314 R-CHOP and 92 CHOP cases in this study, including absolute cell counts, cell densities, and cell type–specific biomarker expression.

Correlation analysis revealed that, in both the R-CHOP and CHOP cohorts, cytoplasmic GSDMD-n⁺ expression was significantly associated with higher CD68⁺CD163⁻CD38⁺ cell densities and CD38⁺ percentage in CD68⁺CD163⁻ cells (Figure 2A; supplemental Figure 2). In the R-CHOP cohort with PD-1/PD-L1 expression analyzed,²⁵ cytoplasmic GSDMD-n⁺ cases had significantly lower mean and median PD-L1⁺ percentages in macrophages (CD68⁺) and lower PD-1⁺ percentages in T cells (Figure 2A; supplemental Figure 3A).

Figure 2.

Correlative analysis of GSDMD-n expression in DLBCL. (A) Scatterplots showing significant associations with cytoplasmic GSDMD-n expression. In both the discovery and validation cohorts, DLBCL samples with cytoplasmic GSDMD-n expression had significantly higher median CD38⁺CD68⁺CD163⁻ cell densities and higher mean and median CD38⁺ percentage in CD68⁺CD163⁻ cells. In the discovery cohort (R-CHOP cohort; PD-1/PD-L1 were not imaged or quantified in the validation cohort), cytoplasmic GSDMD-n expression was significantly associated with lower median and mean PD-L1 percentages in macrophages and lower PD-1 percentages in T cells. (B) Scatterplots showing associations of lower T cells with high GSDMD-n nuclear expression. Each dot represents 1 patient. Bars and error bars in 3 cell density plot represent median cell density ±95% confidence interval. Bars and error bars in 2 percentage plots represent mean percentage ± standard deviation (SD) values.

In contrast, the small number of DLBCL cases with high nuclear GSDMD-n expression showed negative associations with T-cell abundance in the 2 cohorts: significantly lower median CD8⁺ T-cell density in the R-CHOP cohort and significantly lower median T-cell density and CD4⁺ T-cell density in the CHOP cohort (Figure 2B).

GSDMD-n is expressed in lymphoma cells, macrophages, and T cells, exhibiting morphology of pyroptosis, cognate interaction with Ki-67 expression, or phagocytosis

To decipher the cell types of GSDMD-n⁺ and surrounding cells, we overlaid GSDMD-n⁺ IHC images with mfIHC images, Ki-67 IHC, and hematoxylin and eosin (H&E) images and examined details of single cells for all cases with a cytoplasmic-only or nuclear-only expression pattern and 10 cases with >90% of cells being GSDMD-n⁺. GSDMD-n⁺ cells were found to be tumor or immune cells and varied in Ki-67, BCL2, and MYC expressions.

Figure 3 shows the mfIHC images of a GSDMD-n^high (mixed cytoplasmic and nuclear expression pattern) case with a superb prognosis (alive at the last follow-up, >19 years). Pores in DLBCL cell membranes were distinctive via CD20⁺ and PAX5⁺ immunofluorescence (Figure 3) and spread remarkably in the tissue. Moreover, DAPI⁺ staining diffused the tissue, suggesting loss of plasma membrane integrity and leakage of nuclear content. Other morphological characteristics of pyroptosis, such as cell swelling and membrane blebbing, were also observed in mfIHC images. This case showed Ki-67⁻ (Figure 3), p53⁻,³⁴ MYC⁻, and BCL2⁻ IHC³⁷; high PD-1 expression in T cells (60%); high PD-L1 expression in macrophages (54%) via mfIHC²⁵; and necrosis regions via H&E staining in our previous studies.

Figure 3.

Representative mfIHC and IHC images for a DLBCL tissue sample with widely spread GSDMD-n expression (both nuclear and cytoplasmic expression). Three representative GSDMD-n⁺ lymphoma cells with PAX5⁺ nuclei were marked as ①-③. Enlarged mfIHC images showed pores in the CD20⁺ plasma membrane of PAX5⁺ cells and a region with CD20⁺ and CD11c⁺ colocalization (immunofluorescence color mixing). This patient had long survival (>19 years). IF: immunofluorescence.

However, such widely spread CD20⁺ pores and pyroptosis morphology of tumor cells on mfIHC were rare among GSDMD-n⁺ cases, suggesting that most DLBCL tumors were pyroptosis resistant. Instead of apparent CD20⁺ pores, cognate interactions between lymphoma and immune cells, readily revealed via mfIHC, were found to be a common feature of the cytoplasmic GSDMD-n⁺ cases examined. As shown in Figure 4, the GSDMD-n⁺ cytoplasmic regions of the representative M1 macrophage (CD11c⁺CD163⁻CD68⁺) and lymphoma (CD20/PAX5⁺) single cell displayed color mixing of CD20⁺ and CD11c⁺ immunofluorescence, and T (mostly CD4⁺ T) cells neighboring or proximal to GSDMD-n⁺ cells demonstrated color mixing of CD20⁺ and CD3⁺ immunofluorescence, suggesting antigen presentation (by CD11c and CD4). The CD11c⁺ membrane of the representative GSDMD-n⁺ macrophage was continuously bounding (Figure 4A), but the CD20⁺ membrane of the representative lymphoma cell (Figure 4B) was porous where it interacted with a macrophage and T cells. In contrast to the cells in Figure 3, both GSDMD-n⁺ single cells in Figure 4 were Ki-67⁺. In addition, the macrophage was weakly CD38⁺, and the lymphoma cell was BCL2⁺ (also weakly MYC⁺) via mfIHC. The prognosis of this patient was favorable (follow-up of 5.5 years, alive) despite high Ki-67 (90%),³⁴ MYC (60%), and BCL2 (80%) IHC scores³⁷ and 41% of macrophages expressing PD-L1 via mfIHC.²⁵

Figure 4.

Characterization via mfIHC and IHC of cells showing positive cytoplasmic GSDMD-n staining. (A) Images of different biomarkers for a representative macrophage. (B) Images of different biomarkers for a representative lymphoma cell. Both cells showed CD20⁺ and CD11c⁺ colocalization (immunofluorescence color mixing) and were in vicinity of T cells showing color mixing of CD20⁺ and CD3⁺ immunofluorescence. The macrophage was M1 type, CD38⁺, Ki-67⁺, close to T (especially CD4 T) cells interacting with B cells (via CD20⁺ and CD3⁺ immunofluorescence color mixing), and surrounded by BCL2⁺ lymphoma cells. The PAX5⁺ lymphoma cell showed discontinuity of CD20⁺ plasma membrane (where the lymphoma cell was overlapped with a T cell and a macrophage) and was Ki-67⁺ and BCL2⁺. The patient had good prognosis (follow-up of >5.4 years). Original magnification for IHC images ×40; original magnification for mfIHC images ×20.

In contrast, none of the nuclear-only GSDMD-n⁺ cases we examined displayed color mixing of CD20⁺/CD11c⁺ immunofluorescence. Figure 5 shows a representative case with high nuclear GSDMD-n⁺ expression, which was in B cells, macrophages, and T cells on mfIHC images. The representative lymphoma cells 2 and 3 showed some CD20⁺ discontinuity, and both were BCL2⁺ (via mfIHC) and Ki-67⁺ (via IHC). Interestingly, cell 3, which showed CD20⁺ porous membrane and very weak PAX5⁺ immunofluorescence, colocalized with CD163⁺ and CD68⁺ immunofluorescence with a low level of CD20⁺/CD163⁺ color mixing, suggesting possible phagocytosis. In contrast, the representative cell 1 in Figure 5 was a CD8⁺, Ki-67⁻, probably dead (via H&E staining) T cell. This DLBCL case showed high Ki-67 (95%),³⁴ high MYC (80%), and low BCL2 (30%) expression via IHC³⁷; high PD-1 expression in T cells (71%); high PD-1 expression in B cells (74%), and high PD-L1 expression in macrophages (69%) and B cells (36%) via mfIHC.²⁵ Regarding prognosis, this patient died 12.5 months after the DLBCL diagnosis.

Figure 5.

IHC, mfIHC, and H&E images for a representative DLBCL tissue sample with high GSDMD-n nuclear expression and poor prognosis. Three representative cells were highlighted. Cell 1 was a CD8 T cell without Ki-67 expression. Cells 2 and 3 were both BCL2⁺ and Ki-67⁺ B cells. Cell 3 showed CD20⁺ immunofluorescence discontinuity (membrane pores) and very weak PAX5 expression and colocalized with M2 macrophage markers.

GEP analysis revealed associations of cytoplasmic GSDMD-n expression with a lower frequency of inflammatory LME subtype, increased cell states in LE6 and LE8, and upregulation of RNA genes

To gain genome-wide molecular insights, we compared the GEP data²⁷ of patients with and without cytoplasmic GSDMD-n expression. A large number of significantly differentially expressed genes (DEGs) with false discovery rate <5 were identified (supplemental Figure 3B; supplemental Table 4), including upregulated IL13 (related to M2a macrophages) and IL23R (related to M1 and M2a macrophages)^40,41 and downregulated pyroptosis regulators DPP8 and MAPK8,^42,43 LILRB1 and LILRB3 (macrophage receptors for “Don’t eat me” signal and M2 skewing),^44,45 FCGR2C, IL10RA (related to M2b, M2c, and M2d macrophages), and IL16.⁴⁶ Notably, 228 of the 858 (27%) genes upregulated in the cytoplasmic GSDMD-n⁺ group were RNA genes (noncoding RNAs, antisense RNAs, microRNAs, etc) or pseudogenes. Previous studies have indicated that noncoding RNA genes are regulators of and are often upregulated during apoptosis, necrosis, autophagy,^47,48 and pyroptosis.⁴⁹ Moreover, of our interest, median and mean levels of CD47 (which signals “Don’t eat me” to macrophages) and FCGR2A (more expressed on M2 macrophages), relevant for macrophage functional state, were lower in cases with cytoplasmic GSDMD-n, although the false discovery rate was nonsignificant (0.12 and 0.08, respectively).

However, prominent heterogeneity was observed in DEG expression, particularly in the cytoplasmic GSDMD-n⁻ group. Median-centered unsupervised clustering based on the expression of DEGs resulted in a DEG^high cluster with only ∼50% of patients being cytoplasmic GSDMD-n⁺, mostly in subcluster 2.2 in Figure 6A. Only subcluster 2.2, but not 2.1, had significantly better OS and PFS than other cases (Figure 6A; supplemental Figure 3C).

Figure 6.

GEP comparison and deconvolution analysis. (A) Heat map for 2 distinct clusters via unsupervised clustering based on expression of significantly DEGs (Affymetrix, GSE31312) between cytoplasmic GSDMD-n⁺ and GSDMD-n⁻ patients. The case plot below shows the GSDMD-n status and GEP-based LME classification.²⁹ Prognostic analysis showed superior OS of cluster 2.2 compared with cluster 2.1 and cluster 1. Violin plots show that CD47 and FCGR2A were significantly downregulated in DLBCLs with cytoplasmic GSDMD-n expression. (B) Comparisons of EcoTyper estimates²⁸ between cytoplasmic cleaved GSDMD⁺ and GSDMD⁻ patients. The top plot shows differences in proportions of 5 lymphoma ecotypes (LE); the bottom plot summarizes significant differences in fractions of prognostic cell states. The most significant associations with T_FH cell state S1 in LE6 and plasma cell state S3 in LE8 (P = .0002 and P < .0001, respectively) in line with the favorable prognostic effect of cytoplasmic cleaved GSDMD expression are highlighted. (C) Cytoplasmic GSDMD-n⁺ cases had a lower proportion of “inflammatory” LME subtype in the total numbers of prognostically unfavorable “depleted” and “inflammatory” LME subtypes. ∗∗∗∗P < .0001; ∗∗∗P < .001; ∗∗P < .01; ∗P < .05. Cyto, cytoplasmic; LME, Lymphoma MicroEnvironment.

To understand the TME part in GEP,²⁷ the Lymphoma EcoTyper software²⁸ and LME classification algorithms²⁹ were applied to deconvolute the bulk GEP data. Cytoplasmic GSDMD-n⁺ DLBCL cases more frequently had prognostically favorable LE6 and LE8 ecotypes according to the Lymphoma EcoTyper software²⁸ and less frequently had LE4 (prognostically unfavorable) and LE7 (favorable) ecotypes. The EcoTyper cell states significantly associated with cytoplasmic GSDMD-n included prognostically favorable plasma cell state S3 in the LE8 ecotypes, T_FH cell state S1 in the LE6 ecotype, and prognostically unfavorable dendritic cell state S4 in the LE1 ecotype (Figure 6B), whereas CD8 T-cell state S4 (immune exhaustion) in the LE4 ecotype was significantly less frequent (supplemental Figure 4).

LME classification²⁹ revealed TME heterogeneity and its prognostic influence among patients with cytoplasmic GSDMD-n⁺ expression; on the other hand, the prognostic effects of the LME subtypes, including the prognostically unfavorable “inflammatory” LME subtype, showed dependence on cytoplasmic GSDMD-n expression (supplemental Figure 3C). Contrary to the assumption that pyroptosis is proinflammatory, cytoplasmic GSDMD-n⁺ compared with GSDMD-n⁻ cases less frequently had the “inflammatory” LME subtype (Figure 6C). Comparing 3 clusters in Figure 6A, both subclusters 2.1 and 2.2 (especially 2.1, which had poor prognosis) had a higher proportion of the “depleted” LME subtype than unsupervised subcluster 1.

Endothelial GSDME expression shows significantly adverse prognostic effects in DLBCL, whereas tumor GSDME and GSDMC expressions are rare

Figure 7A shows GSDME-N-terminal staining results in the representative DLBCL case in Figure 1A. Expression was mainly in the vascular endothelium; GSDME⁺ leukocytes included B cells (representative cell 2), T cells (cells 3, 8, and 9), and macrophages (cells 1, 4, 5, 6, and 7) with a cytoplasmic expression pattern. In total, 63.3% of the discovery cohort and 70.2% of the validation cohort had GSDME expression in >1% of endothelial cells with an intensity score of +1, and the median level was 5% in both cohorts (supplemental Figures 1B and 5A). Consistently in the 2 study cohorts, endothelial GSDME⁺ vs GSDME⁻ tissues showed significantly higher median and mean CD31⁺ and CD3⁺ T-cell densities via mfIHC (Figure 7A; supplemental Figure 5B). Leukocyte GSDME expression was rare, with a median of 0% in both cohorts, and significantly associated with GSDMD-n⁺ expression in both the discovery and validation cohorts (Figure 7B; supplemental Figure 5C).

Figure 7.

Analysis of GSDME and GSDMC expression in the 2 DLBCL cohorts. (A) Left: IHC images for GSDME expression in the same DLBCL case from Figure 1A showing a representative endothelial expression region and a leukocyte expression region. Right: The leukocyte GSDME⁺ region was characterized via examining its mfIHC images. Nine cells showing GSDME staining positivity in the cytoplasm (or possibly in the membrane or cytoplasm of neighboring cells as some cells were too dense to distinguish) are marked in the GSDME IHC image and mfIHC images for this DLBCL case. GSDME expression in the vascular endothelium was associated with significantly poorer survival in the 2 DLBCL cohorts (optimal cutoffs: ≥5% in the R-CHOP cohort and >10% in the CHOP cohort), and the adverse effect was independent of the favorable effect of cytoplasmic GSDMD-n in the discovery cohort. The scatterplot shows that endothelial GSDME expression was significantly associated with CD31⁺ cell abundance via independent mfIHC in the 2 study cohorts. Each dot represents 1 patient; lines and error bars represent mean CD31⁺ cell densities ± SD. (B) GSDME expression in leukocytes showed a significantly favorable prognostic effect in the CHOP cohort and a nonsignificant trend of better OS in the R-CHOP cohort. The scatterplot shows that leukocyte GSDME expression was significantly associated with GSDMD-n expression. (C) Image of GSDMC IHC in the same DLBCL case from Figures 1A and 7A. (D) High GSDMC expression was significantly associated with endothelial GSDME expression in the R-CHOP cohort only and was associated with the opposite prognostic effects in the 2 DLBCL cohorts. ∗∗∗∗P < .0001; ∗∗∗P < .001. cyto-GSDMD, cytoplasmic cleaved GSDMD expression; EC, endothelial cell; HR, hazard ratio; (t), t test (unpaired, 2-tailed); (U), U test (unpaired, 2-tailed).

Prognostic analysis revealed that significantly in both R-CHOP and CHOP cohorts endothelial GSDME expression was associated with worse OS and PFS (Figure 7A; supplemental Figure 5D; optimal cutoffs: median of ≥5% in the R-CHOP cohort and >10% in the CHOP cohort), as well as higher frequencies of Eastern Cooperative Oncology Group score of >1. In addition, significantly in the R-CHOP but not the CHOP cohort, bone marrow involvement (19% vs 4.8% and P = .0002) and p53 expression (39.9% vs 26.5% and P = .022) were more frequent in endothelial GSDME⁺ patients than in GSDME⁻ patients. The adverse effect of endothelial GSDME expression on OS was independent of the favorable prognostic effect of cytoplasmic GSDMD-n expression (Figure 7A; supplemental Figure 5E). In contrast, rare leukocyte GSDME expression (cutoff: >0% in only 17 CHOP-treated and 32 R-CHOP–treated patients) showed a significantly favorable impact on OS in the CHOP cohort and a trend toward better OS in the R-CHOP cohort (Figure 7B).

GSDMC was expressed in 22.9% of the R-CHOP cohort and 34.2% of the CHOP cohort (supplemental Figures 1B and 6A). Figure 7C shows a GSDMC⁺ region in the representative DLBCL case in Figures 1A and 7A. High GSDMC expression, with a cutoff of >20%, showed opposite effects in the discovery and validation cohorts (Figure 7D; supplemental Figure 6B). GSDMC expression showed a significant association with endothelial GSDME (but not GSDMD-n⁺) expression in the discovery (but not the validation) cohort (Figure 7D; supplemental Figure 6C). Correlation analysis of GSDMC expression found correlations similar to those for endothelial GSDME⁺ (supplemental Figure 6D).

Discussion

Integrating the IHC, multiplex IHC,^25,39 and H&E staining and bulk GEP results, we demonstrated the cellular and subcellular expression patterns of 3 GSDMs and revealed their differential effects on patient survival and immune cell abundance, cell state, and immune checkpoint expression in 2 DLBCL cohorts.

The detailed subcellular and spatial analysis of cleaved GSDMD adds a novel understanding of the TME and prognosis of DLBCL. Cytoplasmic GSDMD-n expression was associated with significantly better OS and increased CD38⁺ M1 macrophages in both the R-CHOP and CHOP cohorts, although the 2 cohorts had significant differences in clinical and molecular features and treatment. A novel finding of this study is that cytoplasmic (including membranous) GSDMD-n expression is a hallmark of cognate interactions between macrophages and lymphoma cells, as indicated by the color mixing of CD20 and CD11c revealed via mfIHC with single-cell subcellular resolution. The representative GSDMD-n⁺ macrophage in Figure 4 was membrane bound, Ki-67⁺, and CD38⁺, and “Don’t eat me” signal genes were downregulated, consistent with a hyperactivation state defined in previous studies.⁵⁰ CD38 is traditionally viewed as an “activation marker”51, 52, 53 and is more expressed in M1 than in M2 macrophages.⁵⁴ The cognate interaction of macrophages with lymphoma cells might enhance antigen presentation and helper T cell response, indicated by CD20 and CD3/CD4 immunofluorescence color mixing in Figure 4. Thus, enhancing the pyroptosis pathway would be valuable for therapeutics that modulate macrophage-mediated tumor clearance (such as anti-CD47) in DLBCL. However, our GEP analysis did not identify a commonly known upstream inflammasome signaling pathway enriched in DLBCL with cytoplasmic or nuclear GSDMD-n expression.

Unexpectedly, nuclear subcellular location of GSDMD-n was often observed. A previous study showed that, under hypoxia in the TME, nuclear GSDMD can promote chemotherapy-induced apoptosis rather than pyroptosis⁵⁵; however, that study only analyzed full-length GSDMD but not cleaved GSDMD. In our study, a few cases had high nuclear expression levels without cytoplasmic GSDMD-n and poor clinical outcomes, which were potentially attributable to T-cell pyroptosis (cell 1 in Figure 5) and lymphoma cell resistance to cell death (lymphoma cells 2 and 3 in Figure 5), in contrast to the superior prognosis of the patient in Figure 3 with both cytoplasmic and nuclear GSDMD-n⁺ cells showing typical pyroptotic morphology.⁵⁶ However, future functional studies are warranted to understand the biological underpinnings of the prognostic effects and the cause of GSDMD activation (such as pyroptosis spread⁵⁷) and pyroptosis resistance in DLBCL.

In summary, this study is, to our knowledge, the first to reveal the patterns of cleaved GSDMD, GSDME-N-terminal, and GSDMC expression in DLBCL and their effects on the TME and survival of patients with DLBCL. Cytoplasmic expression of GSDMD-n was associated with cognate interactions between activated macrophages and DLBCL cells and predicted better patient survival. Nuclear GSDMD-n expression was observed in pyroptotic cells (including T cells) and live DLBCL cells resistant to cell death without cognate interactions. GSDME expression was mostly observed in endothelial (rather than in lymphoma) cells and adversely affected patient survival. These results have implications for understanding immune responses and discovering novel therapies to overcome chemoresistance in DLBCLs harboring antiapoptotic mechanisms.

Conflict-of-interest disclosure: E.D.H. received research support from Eli Lilly and Company and stock options from Abcon Therapeutics. H.N. and Q.A. are employees of NeoGenomics Laboratories, Inc. The remaining authors declare no competing financial interests.