Gene expression profiling of gray zone lymphoma

Abstract

Gray zone lymphoma (GZL), a B-cell lymphoma with features intermediate between large B-cell lymphoma (LBCL) and classic Hodgkin lymphoma (cHL), is a rare and poorly defined entity. Alongside GZL, a subset of Epstein-Barr virus (EBV)-positive diffuse large B-cell lymphoma (DLBCL) has been described with polymorphic/GZL-like morphology (polymorphic-EBV-L). To fill the important gap in our understanding of the pathogenic process underlying these entities, we performed a gene expression study of a large international cohort of GZL and polymorphic-EBV-L, combined with cHL and primary mediastinal large B-cell lymphoma (PMBCL) cases. In an unsupervised principal component analysis, GZL cases presented with intermediate scores in a spectrum between cHL and PMBCL, whereas polymorphic-EBV-L clustered distinctly. The main biological pathways underlying the GZL spectrum were related to cell cycle, reflecting tumor cell content, and extracellular matrix signatures related to the cellular tumor microenvironment. Differential expression analysis and phenotypic characterization of the tumor microenvironment highlighted the predominance of regulatory macrophages in GZL compared with cHL and PMBCL. Two distinct subtypes of GZL were distinguishable that were phenotypically reminiscent of PMBCL and DLBCL, and we observed an association of PMBCL-type GZL with clinical presentation in the "thymic" anatomic niche. In summary, gene expression profiling (GEP) enabled us to add precision to the GZL spectrum, describe the biological distinction compared with polymorphic-EBV-L, and distinguish cases with and without thymic involvement as 2 subgroups of GZL, namely PMBCL-like and DLBCL-like GZL.

Graphical abstract

Figure 1.

PCA in the combined cHL, GZL, polymorphic-EBV-L, and PMBCL dataset. (A) Schematic summarizing the GZL pathologic spectrum. Group 0 corresponds to cases with morphology more typical of cHL but with strong and diffuse CD20 expression on all tumor cells, whereas group 3 is represented by cases with a more typical morphology of LBCL but intense and diffuse CD30 expression. Group 3 samples can present with either PMBCL or DLBCL morphology. Between these 2 extremes are cases with a more intermediate morphology (labeled as bona fide GZL) with an immunophenotype divergent from the morphology. *Group 0 cases with mixed-cellularity–like morphology have less fibrosis. **Group 3 samples with PMBCL morphology can present with fibrosis. ^$Some group 2 cases can present a partial loss of CD20 expression. (B) The GZL, polymorphic-EBV-L, cHL, and PMBCL cases are shown in PC space. The x-axis represents the PC1 score (explaining 20% of the variance in the dataset), and the y-axis represents the PC2 score (17% of the variance). Each point represents a patient, colored based on its pathologic classification (cHL, PMBCL, polymorphic-EBV-L [“EBV”] or GZL group 0-3). cHL samples have high PC2 and low PC1 scores, whereas PMBCL samples have high PC1 and low PC2 scores. GZL samples lie in between, supporting the spectrum hypothesis of GZL as a continuum between cHL and PMBCL. The majority of the polymorphic-EBV-L samples can be isolated within PC2. The main biological pathways associated with PC1 and PC2 are described in colored boxes, and the direction of the corresponding arrows reflects positive enrichment of the pathway. H, Hodgkin; Ig, immunoglobulin; M/L, medium to large; NK, natural killer; RS, Reed-Sternberg.

Figure 2.

Unsupervised clustering of the cHL, GZL, polymorphic-EBV-L, and PMBCL samples based on the main biological pathways. Pathways were identified by using preranked GSEA of genes ordered by their correlation with PC1 or PC2 scores. The top 10 genes included in the core set of the significant pathways were selected. The pathways were grouped into functional categories including cell cycle (high in PMBCL and group 3), epithelial-mesenchymal transition (EMT)/matrisome/extracellular matrix (ECM), cellular microenvironment (NK cells, T cells/allograft rejection, IFN), and humoral response. Unsupervised clustering of the samples based on expression of these pathway genes revealed a spectrum with PMBCL and group 3 at one extreme, and cHL cases mixed with GZL groups 0, 1, and 2.

Figure 3.

PCA of GZL spectrum samples. (A) The left plot shows group 0, bona fide GZL, and group 3 samples in PC1 vs PC2 space, reflecting 20% and 11% of the variance, respectively. The middle box plot shows the distribution of PC1 scores based on GZL group (each point represents a sample). Group 3 cases have significantly distinct PC1 scores, whereas group 0 do not (pairwise Student t tests). Group 1 and 2 GZL cases (bona fide GZL) also have similar PC1 scores (P = .25). The right plot shows that B-cell receptor (BCR) pathway genes were strongly negatively correlated with PC1 score, suggesting association with group 3 cases. (B) The left box plot shows PC1 scores based on thymic status within this cohort. In preranked GSEA (middle panel), genes negatively correlated with PC1 score were enriched in the class I antigen presentation pathway, suggesting that non-thymic cases may present a stronger MHC-I presentation machinery compared with thymic cases. Validating the GSEA findings, cases with thymic involvement had significantly lower MHC-I expression (assessed via IHC) compared with cases without thymic involvement (97% vs 66% for thymic vs non-thymic, respectively; χ² test, P = .00066).

Figure 4.

Expression of the bona fide GZL signature genes. Unsupervised clustering of the thymic bona fide GZL (groups 1-2), cHL (EBV^neg only), and PMBCL cases shows 2 distinct clusters, with a small number of discordant PMBCL and GZL cases.

Figure 5.

Microenvironment composition as assessed by IHC staining. (A) Box plots represent the percentage of nucleated cells that are positive for each given marker (P values calculated with Student t tests). T-cell subsets were analyzed by using CD3, CD4, CD8, FOXP3, PD1, and LAG-3. Macrophage subsets were identified by using CD68 and CD163. (B) Representative images of cHL, thymic bona fide GZL (ie, groups 1 and 2 with thymic involvement), and PMBCL tissue stains. Original magnification at 400× for all images. **P ≤ 0.01; ***P ≤ .001; ****P ≤ .0001. H&E, hematoxylin and eosin; ns, not significant (P > .05).

Figure 6.

Comparison of thymic and non-thymic bona fide GZL cases. (A) Unsupervised clustering of the bona fide GZL (groups 1-2) samples using the 30 genes defined in the Lymph3Cx signature. The 6 DLBCL genes form a distinct cluster, suggesting that the sample clustering may reflect tumor cell biology. The majority of non-thymic GZL cases form a cluster on the right with high DLBCL gene expression. Conversely, the thymic cases show high expression of the PMBCL genes. (B) Enrichment of the 6 DLBCL and 24 PMBCL Lymph3Cx genes was assessed in the GZL cases using preranked GSEA, with genes ordered by the DE score between thymic and non-thymic GZL. Thymic cases are enriched in the PMBCL signature (adjusted P < .0001), whereas non-thymic cases were enriched in the DLBCL signature (adjusted P = .05). (C) Main clinical and IHC distinctions between thymic and non-thymic GZL. The H score for PDL1 was dichotomized as high (≥150) or low (<150). The H score for PDL2 was dichotomized as positive (>0) or negative (= 0). The MHC-I score was dichotomized as negative, low (cytoplasmic or no expression, or membranous expression in <90% of the tumor cells), or high (membranous expression in >90% of the tumor cells). (D) IHC of representative bona fide GZL thymic and non-thymic tumors for PD1, PD-L1, PD-L2, and MHC-I staining.  Original magnification at 400× for all images.