Tee-ing up a New Follicular Lymphoma Classification System

Follicular lymphomas (FL) are indolent yet largely incurable tumors that arise from B cells undergoing the naturally mutagenic germinal center (GC) response, during which B cells undergo immunoglobulin affinity maturation. Through this process, B cells undergo repeated, rapid, and dramatic phenotypic transitions prior to exiting the GC reaction as memory or plasma cells that confer long-term immunity (1). GCs themselves are composed of transcriptionally heterogeneous subpopulations of B cells with distinct potential to proliferate, mutate, survive, and exit the GC. The instructions that drive these many cell fate decisions derive from signals received from specialized CD4⁺ T follicular helper cells (TFH), follicular dendritic cells (FDC), and others (1). All these cell types can be present within the rich lymphoma microenvironment (LME) of FLs, where they presumably support malignant B cells. However, FLs may also contain other CD4⁺, CD8⁺, and lymphoid stromal populations with potential pro- or antilymphoma functions (2). Underlining the clinical relevance of these complex immune scenarios is the poor response of FL patients to immune-checkpoint therapies, but more robust responses to CART or bispecific antibodies (3). However promising, there are still no definitive biomarkers to predict which FL patients are likely to best respond to such treatments. These considerations have emphasized a need to better characterize the composition of FL B cells, and how they intersect with and determine the composition and immune functionality of their respective LMEs.

The advent of powerful single-cell genomic and multiparameter imaging technologies affords a path forward to capture such critical information. Along these lines, in this issue of Blood Cancer Discovery, Han and colleagues (4) provide the most comprehensive and potentially clinically actionable single-cell dissection of FLs to date. They performed single-cell RNA-seq in a set of 20 FLs as well as BCR and TCR sequencing in patients for which they also had available exome sequencing and flow cytometry. As noted previously by smaller studies (5–7), they observed considerable heterogeneity in B-cell transcriptomes between and within each patient. The interpretation of such findings has varied but likely reflects a combination of the natural phenotypic plasticity of GC B cells, as well as potential genetic clonal diversification. Intriguingly, both of these scenarios are suggested by Han and colleagues identifying clonally related plasma cells in certain patients, as well as BCR evolution of plasma cell clones beyond their FL origin (4). This is notable as FLs are usually considered to be blocked in differentiation at the GC stage. Indeed, the three FL hallmark chromatin modifier mutations in KMT2D, CREBBP, and EZH2 result in the downregulation of genes involved in GC exit and plasma cell differentiation (1). However, in animal models reflecting all three mutations, plasma cell formation is attenuated but not fully blocked (1), consistent with observations in FL patients from Han and colleagues, all of whom have at least one of these three mutations.

FLs are described to contain variable proportions of TFH (usually CXCR5⁺PD1⁺) and FDCs, reflecting their origin in the GC reaction. Exactly how these cell populations play into disease pathogenesis is unclear, although recent studies provide some insight. For example, Béguelin and colleagues showed that EZH2 gain-of-function mutations impair interactions between GC B cells and TFH cells, while at the same time enhancing interactions with FDCs (8). This leads to the expansion of aberrant GCs with increased abundance and/or arborization of FDCs and relative depletion of TFH cells. EZH2-mutant GC B cells were relatively resistant to the blockade of T cells but were highly dependent on FDCs (8). Active bidirectional cross-talk between at least a subset of FLs and lymphoid stromal cells can skew their differentiation toward particular phenotypes, for example, through TGFβ signaling (2). CREBBP loss-of-function mutations result in diminished MHCII expression and facilitate immune evasion of incipient lymphomas (1). Hence, the interplay between somatic mutations, immune signaling, and the lymph node microenvironment is multifactorial, heterogeneous, and potentially clinically significant.

Prior scRNA-seq studies in smaller numbers of patients attempted to characterize their immune microenvironment and complement of T-cell subsets. For example, Haebe and colleagues identified regulatory T cells (Treg), TFH, and T effector subpopulations (memory, activated, and exhausted) in FLs, as well as two myeloid and three dendritic cell populations (6). Comparing these immune populations at distinct tumor sites within the same patient showed a relatively similar microenvironment composition, but some heterogeneity in T-cell subsets and their expression of checkpoint genes (6). Herein, Han and colleagues performed detailed analyses of T-cell populations within their FL patients, which accounted for approximately 87% of all nonmalignant cells and included CD4, CD8, and NKT subsets, with a fairly heterogeneous distribution of these various T cells across patients (4). It is possible that T cells could be overrepresented in these specimens because stromal populations could be technically depleted given their fragility during cell suspension preparation. Among CD8 cells, the authors identified variable proportions of naïve, effector, and exhausted subsets, consistent with prior reports. As expected, the CD4 cells included naïve, TFH, and Tregs, but also identified a subset of CD4 T cells (CD4_CTL) expressing cytotoxic markers such as GZMA/K, NKG7, and EOMES. CD4_CTL cells were not previously known to associate with FL, and in this case manifested phenotypic overlap with TFH by expressing CXCL13 and PDCD1 (Fig. 1, middle). The presence of CD4_CTL was confirmed by multispectral immunofluorescence of an independent cohort of FL specimens. This is interesting as it is not clear to what extent TFH cells are able to physiologically or pathologically manifest phenotypic plasticity. For example, it is debated whether and to what extent TFH cells are able to transition to the T follicular regulatory cell phenotype, typically found at low abundance in GCs. It is intriguing to speculate whether aberrant signaling from FL cells could induce TFH diversification into CD4_CTL, and whether such effects could contribute to FL pathogenesis.

The rich and heterogeneous nature of the LME has made it difficult to distinguish whether there are specific and discrete patterns of cells that could be used to classify these tumors into clinically relevant entities. Related to this question, recent studies using mRNA expression deconvolution to dissect the gene-expression profiles of large cohorts of diffuse large B-cell lymphomas (DLBCL) were able to generate new LME classification schemes (9, 10). These studies are based on the principle that the RNA profiling of lymphoma biopsies reflects the transcriptional signatures of all the various cell types that accompany malignant B cells, which can be detected and even quantified using computational approaches that can reconstruct the LME composition. For example, Kotlov and colleagues examined a cohort of over 4,600 DLBCL gene-expression profiles, which allowed them to classify DLBCLs (which are closely related to FLs) into four subtypes based on whether they manifested an immune depleted, inflammatory, GC-like, or mesenchymal LME (9). The depleted subtype featured the worst clinical outcome independent of somatic mutation profiles, and the authors used the various LME profiles to discover novel therapeutic vulnerabilities. Such LME deconvolution approaches have not been performed at a large scale in FLs. Taking into account the information afforded by their scRNA-seq dissection of the FL LME, Han and colleagues used the Kotlov RNA-deconvolution approach focusing specifically on T cells, from the gene-expression profiles of 1,269 FL patients obtained from public databases. Unbiased clustering analysis identified four T cell–defined classes of FL—“immune depleted,” “naïve,” “warm,” and “intermediate”—based on FL cell and T-cell composition. “Warm” FLs featured exhausted CD8 cells, TFH, Tregs, and CD4_CTL, the “naïve” class contained CD4-naïve, CD-naïve, and CD8 effector cells, whereas intermediate FLs had an abundance of B cells and depletion of naïve CD4 and CD8 cells (Fig. 1, top). The “immune-depleted” FLs contained abundant lymphoma cells and few T cells and were associated with inferior failure-free survival, as was observed in DLBCL (4, 9). The other three categories showed more favorable and similar clinical outcomes, although their distinct T-cell composition may reflect specific vulnerabilities to particular immunotherapy approaches.

LME patterns might be linked to particular somatic mutations, given that mutations of CREBBP, KMT2D, and EZH2 all disrupt the expression of different sets of immune sig­naling genes (1). Prior studies showed that EZH2 gain-of-function mutations result in aberrant silencing of MHCII and T-cell adhesion/interaction genes, CREBBP loss-of-function results in repression of MHCII and B-cell receptor signaling genes, and KMT2D in repression of CD40 and germinal center exit genes (1). However, all of those reports focused only on the impact of each gene individually, without taking into account that these three hallmark mutations often cooccur together, making it harder to dissect their contributions in many cases. In general accordance with these data, scRNA-seq studies reported herein showed similar perturbations in expression profiles and shed light on how combinations of mutations could further impact immune signaling genes (4). The authors focused in particular on the silencing of MHCII expression, given its strong link to immune evasion and high incidence (58%) in their FLs (4). Importantly, MHCII repression was not limited to patients with CREBBP and/or EZH2, or even KMT2D mutation, suggesting that other somatic mutations beyond those affecting chromatin modifiers could yield similar effects. Importantly, low MHCII expression was strongly associated with reduced CD8_exh and CD4_CTL based on scRNA-seq, mRNA deconvolution in their larger data sets, and independent IHC studies in relation to CD4_CTLs. These data suggest a novel role for the CD4_CTL MHCII-dependent immune surveillance of FLs. Low MHCII, on the other hand, was associated with a higher abundance of naïve CD4 and CD8 T cells. The implication is that MHCII staining intensity alone may predict important T-cell LME characteristics in FL patients.

To further explore the diagnostic potential of MHCII expression to direct T-cell immunotherapies, the single-cell gene-expression profiles of CD4 and CD8 cells were used to identify transcriptionally defined clusters. Strikingly, these clusters segregated patients based on MHCII expression by FL B cells. Examining the signatures of these various clusters revealed several findings with significant diagnostic and therapeutic implications. Cluster-defining genes included those encoding specific T-cell activation signaling and transcriptional factor, as well as checkpoint genes such as TIGIT, CTLA4, and LAG3. Expression of these genes was anticorrelated with MHCII expression in FL B cells (4). Hence, immunologic exhaustion in the FL LME is linked to persistent MHCII expression in these tumors. Taking into account the expression of checkpoint inhibitor therapeutic target pairs allowed the authors to derive a “combination immunotherapy index,” whereby exhausted CD8 T cells expressing high TIGIT and LAG3 and CD4 cells expressing OX40 and CTLA4 were characteristically present in MHCII high FLs and low in MHCII low FLs (4).

Collectively, these important discoveries strongly expand on the theme of MHCII expression being a critical determinant of immune surveillance, linked, on the one hand, to T-cell exhaustion and CD4_CTL infiltration when significantly expressed, and potential immune evasion from CD4_CTL and lack of checkpoint therapy potential when repressed (Fig. 1, bottom). Importantly, the genetic composition of FLs would not have necessarily predicted these distinct immune states pointing to the need for additional histologic features especially MHCII to be considered together with mutation profiles in the design of precision immunotherapies. The primary human FL confirmation of the association of reduced MHCII expression in EZH2, CREBBP, and KMT2D-mutant lymphomas by Han and colleagues provides the basis for using immune augmentation therapies with EZH2 inhibitors and HDAC3 inhibitors to restore antigen presentation profiles of FLs and enhance the activity of host T cells (perhaps CD4_CTL) and adoptive T cell–directed immunotherapies.

A.M. Melnick reports grants and personal fees from Epizyme and personal fees from Daiichi Sankyo during the conduct of the study; grants and personal fees from Jansen, personal fees from AstraZeneca, Treeline Biosciences, BMS, and Exo Therapeutics outside the submitted work.

A.M. Melnick is funded by NCI/NIH R35 CA220499, NCI/NIH P01 CA229086-01A1, LLS-SCOR 7012-16, LLS-TRP 6572–19, the Samuel Waxman Cancer Research Foundation, the Follicular Lymphoma Consortium, and the Chemotherapy Foundation.