Genetic and Phenotypic Attributes of Splenic Marginal Zone Lymphoma

: Splenic marginal zone B-cell lymphoma (SMZL) is a heterogeneous clinico-biological entity. The clinical course is variable, multiple genes are mutated with no unifying mechanism, essential regulatory pathways and surrounding microenvironments are diverse. We sought to clarify the heterogeneity of SMZL by resolving different subgroups and their underlying genomic abnormalities, pathway signatures and microenvironment compositions to uncover biomarkers and therapeutic vulnerabilities. We studied 303 SMZL spleen samples collected through the IELSG46 multicenter, international study (NCT02945319) by using a multiplatform approach. We carried out genetic and phenotypic analyses, defined self-organized signatures, validated the findings in independent primary tumor meta-data and in genetically modified mouse models, and determined correlations with outcome data. We identified two prominent genetic clusters in SMZL, termed NNK (58% of cases, harboring NF-κB, NOTCH and KLF2 modules) and DMT (32% of cases, with DNA-damage response, MAPK and TLR modules). Genetic aberrations in multiple genes as well as cytogenetic and immunogenetic features distinguished NNK- from DMT-SMZLs. These genetic clusters not only have distinct underpinning biology, as judged by differences in gene-expression signatures, but also different outcome, with inferior survival in NNK-SMZLs. Digital cytometry and in situ profiling segregated two basic types of SMZL immune microenvironments ABSTRACT Splenic marginal zone B-cell lymphoma (SMZL) is a heterogeneous clinico-biological entity. The clinical course is variable, multiple genes are mutated with no unifying mechanism, essential regulatory pathways and surrounding microenvironments are diverse. We sought to clarify the heterogeneity of SMZL by resolving different subgroups and their underlying genomic abnormalities, pathway signatures and microenvironment compositions to uncover biomarkers and therapeutic vulnerabilities. We studied 303 SMZL spleen samples collected through the IELSG46 multicenter, international study (NCT02945319) by using a multiplatform approach. We carried out genetic and phenotypic analyses, defined self-organized signatures, validated the findings in independent primary tumor meta-data and in genetically modified mouse models, and determined correlations with outcome data. We identified two prominent genetic clusters in SMZL, termed NNK (58% of cases, harboring NF-κB, NOTCH and KLF2 modules) and DMT (32% of cases, with DNA-damage response, MAPK and TLR modules). Genetic aberrations in multiple genes as well as cytogenetic and immunogenetic features distinguished NNK- from DMT-SMZLs. These genetic clusters not only have distinct underpinning biology, as judged by differences in gene-expression signatures, but also different outcome, with inferior survival in NNK-SMZLs. Digital cytometry and in situ profiling segregated two basic types of SMZL immune microenvironments termed immune-suppressive SMZL (50% of cases, associated with inflammatory cells and immune checkpoint activation) and immune-silent SMZL (50% of cases, associated with an immune-excluded phenotype) with distinct mutational and clinical connotations. In summary, we propose a nosology of SMZL that can implement its classification and also aid in the development of rationally targeted treatments.

subgroups and their underlying genomic abnormalities, pathway signatures and microenvironment compositions to uncover biomarkers and therapeutic vulnerabilities. We studied 303 SMZL spleen samples collected through the IELSG46 multicenter, international study (NCT02945319) by using a multiplatform approach. We carried out genetic and phenotypic analyses, defined self-organized signatures, validated the findings in independent primary tumor meta-data and in genetically modified mouse models, and determined correlations with outcome data.
We identified two prominent genetic clusters in SMZL, termed NNK (58% of cases, harboring NF-κB, NOTCH and KLF2 modules) and DMT (32% of cases, with DNAdamage response, MAPK and TLR modules). Genetic aberrations in multiple genes as well as cytogenetic and immunogenetic features distinguished NNK-from DMT-SMZLs. These genetic clusters not only have distinct underpinning biology, as judged by differences in gene-expression signatures, but also different outcome, with inferior survival in NNK-SMZLs. Digital cytometry and in situ profiling segregated two basic types of SMZL immune microenvironments termed immune-suppressive SMZL (50% of cases, associated with inflammatory cells and immune checkpoint activation) and immune-silent SMZL (50% of cases, associated with an immuneexcluded phenotype) with distinct mutational and clinical connotations. In summary, we propose a nosology of SMZL that can implement its classification and also aid in the development of rationally targeted treatments. The incidence of SMZL is increasing, mainly because of improved diagnostic techniques resulting in more patients being diagnosed every year. However, in parallel, life expectancy of patients with SMZL is not improving. Compared to other indolent B-cell neoplasms, the survival of patients with SMZL is unsatisfactory (5-year relative survival ~79%) [1][2][3] , and no breakthrough treatment advances have been seen 3 .
SMZL is heterogeneous at multiple levels. The clinical course is variable, with some patients having prolonged survival and a proportion (~20%) experiencing rapidly progressive disease and survival less than 5 years 4 . SMZL lacks a unifying genetic lesion. Multiple mutated genes have been identified, which are restricted to a fraction of cases. [5][6][7][8][9][10][11] Inflammatory cells are expanded in a subset of SMZL 12 , suggesting the existence of different microenvironments. Clinical trials evaluating novel agents provide glimpse into signaling pathways that are essential for SMZL, but sensitivity to these agents is not always observed [13][14][15] .
The IELSG46 study (NCT02945319) is a multicenter, international, retrospective, observational study that aims at resolving the heterogeneity of SMZL into subgroups by using a multiplatform approach, with the belief that it might yield a nosology of SMZL that could be implemented for disease classification, and result in the discovery of novel biomarkers and therapeutic vulnerabilities.

Patients
Inclusion criteria of the IELSG46 study were: i) age >18 years; ii) SMZL diagnosis by spleen histopathologic examination; iii) availability of tumor material from spleen collected before initiation of medical therapy; iv) availability of baseline and follow-up annotations. Patients who received any anti-tumor medical therapy before splenectomy were excluded. The study was conducted in accordance with the

LyV3.0 CAncer Personalized Profiling by deep Sequencing Assay
A CAPP-seq protocol was used for mutation and copy number abnormality (CNA) analysis. Libraries derived from tumor genomic DNA of FFPE (n=246) or

Patient characteristics
Patients diagnosed with SMZL on splenic resections were registered in the IELSG46 study by 28 centers in Europe and the US. A total of 373 patients were initially identified. Seventy cases were excluded due to alternative diagnoses on central pathology review or insufficient material (Supplementary Figure 1A). Table 1 lists the clinical characteristics of the 303 patients with confirmed SMZLs. Deletion 7q and use of IGHV1-2*04 allele, which are recurrent in SMZL 18,19 , were detected in 26.4% and 33.9% of cases, respectively. Based on the percentage of IGHV gene identity to the germline, 11.1% of cases were 'truly unmutated' (100% homology) (Supplementary Table 1). Median follow-up after splenectomy was 10.6 years, with 86 deaths. At ten years, overall survival was 68.5% and relative survival compared to the matched general population was 82.1% (Supplementary Figure 1B). After splenectomy, which counted as first line therapy, 10.6% of patients received These data confirmed the representativeness of the study cohort and the lack of biases due to the inclusion of splenectomized patients. 3,20,21

A genetic classifier for SMZL
We investigated mutations using the LyV3.0 CAPP-seq assay, which targeted ~280 kb of genomic space by deep sequencing and that has been specifically designed to cover the majority of coding regions known to be recurrently mutated in mature B-cell neoplasms (Supplementary Table 2 Table 4).
To segregate SMZLs into discrete genetic classes supported by coordinated mutational profiles, we started with a set of pathway-driven seed modules comprising components of several B-cell programs (Supplementary Table 5). Genes were assigned to a module based on published literature and database annotations [23][24][25][26][27] .
Genes that were attributed to multiple modules (eg, KLF2) were not assigned and seeded individually. Unsupervised analysis of mutational co-occurrence between all lesion pairs revealed overall significantly stronger exclusivity between intra-pathway lesions ( Figure 1B). Mutual exclusivity of mutations within a pathway could reflect functional redundancy and supported their aggregation within a seed ( Figure 1C).
We then applied hierarchical clustering on principal components (HCPC) and discovered four groups of tumors (clusters) with discrete genetic signatures overall accounting for 86.4% of cases, and an additional subset without detectable mutations in the interrogated genomic space (13.6% of cases) ( Figure 2A). The algorithm converged on genetic clusters that for simplicity were termed NNK (58.2%  Table 7).

Phenotypic differences among SMZL clusters
We applied gene expression profiling to FFPE tissue samples to explore phenotypic differences among SMZL clusters. The concordance between expression profiling of FFPE tissues and gold standard RNA-seq was validated in 57 cases with available paired frozen and FFPE SMZL specimens (Supplementary Figure 2D Table 8 and 9). For each signature, we calculated an activity score that is directly associated with the magnitude of a particular effect among populations of cells 32 . We then analyzed the correlation between the NNK-and DMT-SMZL clusters and the signatures ( Figure   3A). NNK-SMZLs expressed significantly higher levels of genes belonging to the NOTCH2 pathway and of genes that are activated by non-canonical NF-κB transcription factors 33 . A proliferation signature also characterized NNK-SMZLs.
Conversely, DMT-SMZLs had a signature of impaired TP53 and apoptosis functions.
We used immunohistochemistry to comparatively screen NNK-SMZLs vs DMT-SMZLs for evidence of NOTCH and NF-κB biochemical activation ( Figure 3B and C; Supplementary Table 10). Consistent with the high incidence of NOTCH and NF-κB pathway mutations in NNK-SMZLs, they frequently showed nuclear staining for NOTCH, canonical-and non-canonical NF-κB proteins, while nuclear TP53 expression strongly correlated with DMT-SMZLs ( Figure 3B and C).
Paired bone marrow biopsies of 96 patients were revised by the local pathologists. The bone marrow morphology was typical (i.e. nodular or nodularinterstitial pattern, with or without intrasinusoidal infiltration) in 85 cases (88.5%) and atypical (i.e. paratrabecular or diffuse patterns without intrasinusoidal infiltration) in the remaining 11 cases (11.5%). With the limitations imposed by the sample size, the majority of cases showing an atypical bone marrow morphology belonged to the DMT molecular cluster, while cases belonging to the NNK, CBS and PA molecular clusters preferentially or exclusively showed a typical bone marrow morphology (Supplementary Table 11).

Immune microenvironment of SMZL
The immune-microenvironment of SMZL is poorly understood. We thus profiled the expression of 1402 genes focusing on tumor/immune interactions, and used 16 signatures (Supplementary Table 12   To confirm the different composition of the "immune-suppressive" and 'immune-silent'' SMZL classes, we applied digital cytometry deconvolution 41 , which inferred a heterogeneous and microenvironment-rich composition in the "immunesuppressive' class of SMZL, and a high ratio of neoplastic-to-microenvironmental cells in the 'immune-silent' class ( Figure 5A, Supplementary Table 14).  Table 11).
The immune microenvironment classes of SMZL were equally distributed across the molecular clusters ( Figure 4B). Taken together these data indicated that in SMZL, such as in DLBCL, microenvironment signatures provide additional information that is not fully captured by a multi-gene mutational signature 23 . In DLBCL, individual gene mutations rather than molecular signatures provided mechanistic insights into the interaction between lymphoma cells and the immune microenvironment [42][43][44][45][46][47] . To understand at a more granular level the mechanisms by which SMZL might induce specific microenvironment patterning, we comparatively assessed the spectra of the most recurrent mutations between the "immunesuppressive" vs the "immune-silent'' classes of SMZL. KLF2 mutations were enriched in the "immune-suppressive" class of SMZL and depleted in the 'immune-silent'' class ( Figure 4C). The tumor mutational burden, that is an approximation for neoantigen load and corresponds to the number of non-synonymous mutations per

Clinical course of the SMZL clusters
Baseline clinical features distributed evenly across molecular clusters and immune microenvironment classes of SMZL (Supplementary Table 15). To understand whether any molecular cluster or immune microenvironment class of SMZL was more aggressive than the others, we correlated these variables with relative survival. Lymphoma-specific deaths accounts for less than one-half of total deaths in SMZL 3 . Accordingly, relative survival provides a more accurate measure of excess mortality experienced by patients than overall survival, without requiring  Figure 7). Of note, neither low Hb, nor low albumin associated with any molecular or microenvironmental SMZL subtypes, making it unlikely that they can act as confounders in the inferior relative survival of NNK and immune-suppressive SMZL (Supplementary Table 15).

DISCUSSION
Our study highlights the complexity of SMZL, which comprises four distinct genetically defined molecular clusters, and two distinct phenotypically defined immune-microenvironment classes. The molecular framework for SMZL that we present here provides an evolving understanding of its pathogenesis, and can be  1q, 3q, 6q, 7q, 8p, 12q, 17p, and 18q 54 . Second, contrary to DLBCL, SMZL has low genomic complexity and lacks recurrent translocations 6,55,56 . Third, no novel recurrent fusions were detected by RNA-seq.
Fourth, mutations in MYC, BCL2 and BCL6, which are a surrogate of their translocations with an immunoglobulin gene partner, 23 were extremely rare in the cohort, consistent with the knowledge that such structural variants are virtually absent in SMZL 55,56 .
Our study was entirely based on spleen tissue samples. Currently, SMZL can be diagnosed without the need for splenectomy by integrating bone marrow histology with cell morphology and immunophenotype in the blood and bone marrow 57 . This notion prompts the question on as to how can we identify the molecular clusters and microenvironmental classes of SMZL with a minimally invasive approach (i.e. without splenectomy). The almost universal dissemination of lymphoma cells in the blood allows the use of "liquid biopsy" approaches for characterizing the genetics of SMZL.
The validity of this approach is confirmed by the evidence that the molecular profile of circulating SMZL cells matches that of cells residing in the spleen 6  In summary, our multiplatform genomic analysis elucidates SMZL pathogenesis, and provides a conceptual edifice to advance the classification and development of precision therapies for SMZL. T., C.T., C. Thieblemont, T.T., A.T., G.V., C.V., U.V., R.W., F.Z., T.Z. and P.L.Z provided study material and clinical data and contributed to manuscript revision; L.C. and H.K. performed bioinformatics analysis, contributed to data interpretation and manuscript preparation; E.Z. provided key scientific insights and contributed to data interpretation and manuscript revision; D.R. designed the study, interpreted data, and wrote the manuscript.   signatures. HCPC clustering reveals two major microenvironment classes ("immunesuppressive" and "immune-silent") in the SMZL cohort . The differential expression of the microenvironment signatures is also represented as density plots on the right.