Biological and Clinical Relevance of Associated Genomic Alterations in MYD88 L265P and non-L265P–Mutated Diffuse Large B-Cell Lymphoma: Analysis of 361 Cases Free

This article is featured in Highlights of This Issue, p. 2129

Diffuse large B-cell lymphoma (DLBCL) is currently divided into three main molecular subtypes, defined by gene expression profiling (GEP): germinal center B-cell like (GCB), activated B-cell like (ABC), and primary mediastinal B-cell lymphoma (PMBL; refs. 1, 2). One of the characteristics of ABC DLBCL cells is their addiction to constitutive NFκB signaling pathway activation, leading to tumor cell survival (3, 4). Certain alterations, such as activating MYD88 and CARD11 mutations as well as inactivating TNFAIP3 mutations, have been shown to be key to NFκB pathway activation. The importance of MYD88 was first discovered through RNAi screening, identifying the gene as essential for ABC, but not GCB, survival (5). Overall, MYD88 was mutated in 39% of ABC cases, with 29% harboring the highly recurrent MYD88 L265P variant, while it was rare or absent in GCB and PMBL cases. High-throughput sequencing methods have since confirmed the high prevalence of MYD88 mutations in ABC DLBCL (6, 7), but also in primary central nervous system lymphoma (PCNSL; refs. 8, 9) and leg-type (10, 11) DLBCL. More recently, MYD88 L265P has been reported in approximately 90% of Waldenstrom macroglobulinemia (12) and 50% of IgM MGUS (13, 14), as well as less frequently in chronic lymphocytic leukemia (CLL; ref. 15) and marginal zone leukemia (MZL; ref. 16).

MYD88 is an essential adapter protein, which activates the NFκB pathway by interacting with the cytoplasmic portion of Toll-like receptors (TLR) and IL1R and activating downstream signaling components such as IRAK1 and IRAK4 (17). The vast majority of MYD88 mutations, including the L265P variant, target the TIR domain, which is necessary for the interaction with the homologous TIR domains of TLRs and IL1R and ultimate transduction of intracellular signals (18, 19). MYD88 L265P has been shown to be a gain-of-function driver mutation, which confers a selective survival advantage to ABC DLBCL cells, through increased IRAK1 and IRAK4 kinase activity and thus constitutively active NFκB pathway signaling (5). So far, the functional effects of other MYD88 mutations have not been extensively studied due to their lower frequencies in DLBCL cases.

MYD88 locus amplification has been frequently identified in MYD88 L265P–mutated ABC DLBCL cases (5), and less frequently in Waldenstrom macroglobulinemia (12). However, as of now, only in vitro cell line assays and IHC on patient samples have been performed to assess MYD88 expression levels, showing no modification in the presence of MYD88 L265P (5, 20).

The identification of MYD88 mutations is increasingly important in terms of diagnostic relevance, notably with the advent of cell-free DNA (cfDNA) sequencing, which can aid in the diagnosis of difficult-to-biopsy subtypes such as PCNSL and intraocular lymphoma (21, 22). The prognostic relevance of MYD88 mutations, however, is still a matter of debate. Indeed, certain studies have shown MYD88 L265P to be an independent negative prognostic factor in DLBCL, although a large proportion were extranodal cases (23), and in leg-type DLBCL specifically (10), but rather a positive prognostic factor in CLL (24); however, these associations have not been confirmed in all studies and, in the case of CLL, might simply be linked to the mutation's predominance in better-outcome M-CLL (25). The theranostic importance of MYD88 and associated mutations has recently been in the spotlight, with data from a phase I/II clinical trial suggesting that MYD88/CD79B–mutated ABC DLBCL patients are sensitive to ibrutinib (26). Other studies have also pinpointed the importance of studying the associated mutational and GEP characteristics of each case, showing for instance that the proliferation and NFκB activation induced by MYD88 L265P are countered by the induction of TNFAIP3 and require the overexpression of BCL2 (27). A recent study confirmed that the MYD88 L265P mutation cooperates with BCL2 amplifications to drive ABC DLBCL development (28).

The frequency of MYD88 mutations in DLBCL and other hematologic malignancies is well described, and a recent study also analyzed the clinical impact of MYD88 L265P versus other MYD88 mutations in a cohort of 213 DLBCL patients (47 MYD88 mutated and 166 MYD88 WT) with specific targeted MYD88 sequencing (29). However, there has not yet been a large-scale study of a MYD88-mutated patient cohort with additional next-generation sequencing (NGS), copy number variation (CNV), and gene expression data, to thoroughly characterize the associated genomic profiles of these patients. The aims of our study were to compare the L265P and non-L265P mutations in terms of pathologic and genetic features, to better detail the genomic background associated with MYD88 mutations to delineate patients potentially sensitive to targeted therapies, and to define the prognostic value of MYD88 mutations according to different genomic contexts.

A cohort of 361 DLBCL patients was selected among the prospective, multicenter, and randomized LNH-03B (as previously published; ref. 30) and LNH09-7B (NCT01195714) LYSA trials, as well as among patients sequenced at our institution as part of routine procedure. Inclusion criteria included available frozen or FFPE tumor samples, adequate DNA/RNA quality, and sequencing performed at our institute with the NGS Lymphopanel, described hereafter. DNA from the LNH03-B LYSA series was extracted from fresh frozen tumor biopsies (n = 214), DNA from the LNH09-7B trial was extracted from FFPE tumor biopsies (n = 46) and DNA from the patients of our institution was extracted from frozen or FFPE tumor biopsies, according to availability (n = 101, 86% frozen and 14% FFPE). In 4 cases, DNA was obtained at the time of relapse; these cases were not included in clinical or survival analyses. Main clinical features of the patients at diagnosis are indicated in Table 1. All clinical features are recapitulated in Supplementary Table S1. Cell-of-origin (COO) classification was obtained with HGU133+2.0 Affymetrix GeneChip arrays for 214 patients (detailed in Supplementary Methods), with RT-MLPA for 77 patients (31), and with Hans immunohistochemical method for 49 patients. COO information was not available for 21 patients. COO details are listed in Supplementary Table S2. For our analyses, we considered ABC cases to include ABC as defined by Affymetrix, ABC as defined by RTMLPA, and non-GCB as defined by the Hans algorithm. Data for HGU133+2.0 Affymetrix GeneChip microarrays have been deposited in NCBI's Gene Expression Omnibus and is accessible through GEO series accession number GSE87371. For survival analyses, we considered only patients treated with rituximab (R) and additional chemotherapy (R-chemotherapy). Written informed consent was obtained from all participants at the time of enrollment.

Ion Torrent Personal Genome Machine (PGM) Sequencing and PGM data analysis was performed as described previously (30, 32, 33) and detailed in Supplementary Methods. Variant analysis was performed using an in-house–generated bioinformatics pipeline, as described previously (30, 32, 33) and detailed in Supplementary Methods.

All patients were sequenced using the Lymphopanel, which was designed to identify mutations in 34 genes important for lymphomagenesis, based on literature data (Supplementary Table S3; ref. 34) and WES of relapsed/refractory DLBCL sequencing (35). The design covers 87,703 bases using 872 amplicons. Median sequencing depth was 232×. Genes were grouped into 8 specific pathways according to literature (Supplementary Table S4). All identified variants are indicated in Supplementary Table S5.

CNV detection was performed using ONCOCNV software version 5.9 (36). Eight control samples, which were paired normal DNA from leg-type patients, were used to construct the baseline. Amplicons with abnormal depth variability among the baseline were excluded. All identified CNVs are indicated in Supplementary Table S6.

Immunohistochemistry was performed as described previously (37, 38). Staining was performed for CD10, BCL6, MUM1, MYC, BCL2, FOXP1, and IgM. BCL2 and MYC overexpression thresholds were respectively set at 50% and 40%, in accordance with previous publications (39). FISH analysis of MYC, BCL2, and BCL6 rearrangement was performed as described previously (40).

All statistical analyses were performed using R software version 3.1.2 (41). Progression-free survival (PFS) was evaluated from the date of enrolment to the date of disease progression, relapse or death from any cause. Overall survival (OS) was evaluated from the date of enrolment to the date of death from any cause. Multivariate Cox and log-rank tests (“survival” R package version 2.37.7) were used to assess differences in OS and PFS rates calculated by Kaplan–Meier estimates. Survival curves were stopped at a follow-up of 5 years. Statistical differences between all other parameters were determined using χ², Mann—Whitney, or Fisher exact tests when appropriate. P values < 0.05 were considered statistically significant.

Among the 361 patients, 94 (26%) presented MYD88 mutations and 267 (74%) were MYD88 wild-type (WT). Among the MYD88-mutated cohort, 60 (64%) patients presented MYD88 L265P and 34 (29%) presented other MYD88 variants (Fig. 1). Median uncorrected variant allele frequency (VAF) in our cohort was 43.4% (8.7–97.1) for MYD88 L265P and 33.4% for non-L265P variants (7.1–95.3; P = 0.09, Supplementary Fig. S1). Tumor load did not seem to be a confounding variable as this value was uniformly high: among the 214 patients for whom tumor cell percentage was available, the median was 90% and the SD was 8% (data not shown).

MYD88 L265P was the most frequent variant, as expected, representing 62.5% of MYD88 variants (Fig. 2A). The most frequent non-L265P variant was a G→A transition at position 38182292 leading to MYD88 S243N (13.5%), followed by a C→G transversion at position 38182032 leading to MYD88 S219C (8.3%). Others included a G→T transversion at position 38182025, leading to MYD88 V217F (3.1%) and a T→C transition at position 38182259, leading to MYD88 M232T (3.1%). Two patients presented two separate MYD88 mutations, leading to a total of 96 MYD88 variants (Figs. 1 and 2). In both cases, a MYD88 L265P mutation was associated with a G→T transversion at position 38182638 leading to MYD88 R264L: interestingly, these were the only R264L variant cases. A difference in VAF between the L265P and the R264L variants was observed in both cases (43.5% vs. 12.6% for one patient, and 48.2% vs. 20.9% for the other patient, respectively; Supplementary Table S5). Tumor load was estimated at 90% for the first patient, and Integrative Genomics Viewer analysis showed that both variants were on the same allele, suggesting a heterozygous MYD88 L265P mutation and the existence of the R264L variant within a separate subclone (data not shown). For the second patient, both variants were also on the same allele. However, as tumor load was not estimated, we cannot ascertain whether the variants belong to two separate subclones or are simply the expression of a homozygous variant and a heterozygous variant within a DLBCL of 50% tumor load.

Eighty-nine percent of MYD88 L265P—mutant patients with COO information were of ABC subtype, compared with only 37% of WT MYD88 patients and 42% of MYD88 non-L265P–mutant patients (Table 1; Fig. 2B; Supplementary Table S2). Furthermore, MYD88 L265P was detected in 50 of 158 ABC patients (32%). Among patients with S243N and S219C variants, 6 of 13 (46%) and 6 of 8 (75%), respectively, were of GCB subtype (Fig. 2B; Supplementary Table S2).

MYD88 mutations were significantly correlated with age among ABC patients (P = 0.05) but not when restricting data to MYD88 L265P variants only (P = 0.07, Table 1; Supplementary Fig. S2). However, no correlation was found between MYD88 mutations and other clinical parameters, including Ann Arbor stage, IPI, LDH levels, and performance status (Table 1).

Immunohistochemical analysis was available for a large subset of patients (151–224 depending on the antibody). Among ABC patients, the presence of MYD88 mutations did not significantly correlate with the expression of Hans markers, FoxP1 or IgM (Supplementary Table S7). However, BCL2 overexpression was significantly more frequent among ABC MYD88 L265P–mutant patients compared with the remainder (P = 0.025, Supplementary Table S7). This result was not significant when looking at ABC MYD88–mutant DLBCL as a whole, suggesting an immunohistochemical trait specific to MYD88 L265P DLBCL cases. Neither MYC overexpression nor MYC/BCL2 double-expressor profile were significantly different among ABC patients according to MYD88 mutation status (Supplementary Table S7). With a more stringent MYC overexpression threshold (at least 70% positive cells), ABC MYD88–mutant patients tended toward more frequent MYC overexpression (P = 0.095), and this result was statistically significant when comparing ABC MYD88 L265P–mutant patients versus WT and non-L265P–mutant ABC patients (P = 0.04, Supplementary Table S7). GCB patients with MYD88 mutations also tended toward BCL2 overexpression (P = 0.07), and expressed CD10, FoxP1, and IgM more frequently (P = 0.03, P = 0.04, and P = 0.055 respectively) (Supplementary Table S7).

The FISH analysis of MYC, BCL6, and BCL2 rearrangements did not highlight a significant difference according to MYD88 mutation status, whether among ABC or GCB patients (Supplementary Table S7).

In accordance with the prevalence of the L265P variant in the ABC subtype, DLBCL patients with MYD88 L265P presented an ABC mutational profile, with frequently associated PIM1 (52%), CD79B (52%), KMT2D (42%), and PRDM1 (32%) mutations (Fig. 3A; Supplementary Fig. S3A). Thirty-two percent of MYD88 L265P–mutated patients presented both PIM1- and CD79B-associated mutations. Of note, when focusing on the 15 most frequently mutated genes among the Lymphopanel, unsupervised hierarchical clustering identified two distinct groups of MYD88 L265P patients, separating those with and without at least one of the top 4 associated gene mutations (Fig. 3A). Interestingly, DLBCL patients with other MYD88 mutations presented an intermediate mutational profile between that of ABC and GCB patients, reflecting the prevalence of the two most common non-L265P mutations (S243N and S219C) in GCB patients (Fig. 2B). However, one group identified by clustering seemed to present mostly neutral or GCB-enriched mutations (KMT2D 26%, CREBBP 18%, EZH2 15%, BCL2 15%, TNFRSF14 15%, and EP300 12%) except for PIM1 (21%), a common AICDA target; the second group identified by clustering presented both ABC-enriched mutations (TNFAIP3 21% and CD79B 15%) as well as neutral or GCB-enriched mutations (B2M 26%, GNA13 21%, CARD11 18%, ITPKB 15%, and MEF2B 15%; Fig. 3B).

Pathway analysis further showcased the difference in genetic background between these two groups (Supplementary Fig. S3B and S3C). Indeed, MYD88 L265P–mutated patients presented a significantly higher proportion of NFκB pathway mutations (56% vs. 38% of variants, P = 0.001), and of BCR pathway mutations (11% vs. 6%, P = 0.002). Higher proportions of mutations in the epigenetic pathway (21.5% vs. 13%) and in the immunity pathway (14% vs. 5%) were observed in patients with other MYD88 mutations, but these differences were not significant.

There was no significant difference among the most frequent CNVs detected between ABC MYD88 mutant and WT patients (data not shown). Interestingly, the single most frequent gain among MYD88-mutated patients in general was a MYD88 gain (33%), followed by KMT2D (28%), ITPKB (23%), and IRF4 (20%) gains (Supplementary Fig. S3D). Deletions were more frequent, with a high occurrence of CDKN2A/B deletions (56% and 40.5% respectively), as well as TNFAIP3 (47%), PRDM1 (46%), and TP53 (31%) deletions (Supplementary Fig. S3D).

The top five most frequent deletions (CDKN2A/B, PRDM1, TNFAIP3 and TP53) were the same in both MYD88 L265P and non-L265P–mutated cases, although CDKN2A/B deletions occurred significantly more often in MYD88 L265P–mutated cases (P = 0.02 for CDKN2B and P < 10⁻³ for CDKN2A, Supplementary Fig. S3D). In fact, associated CDKN2A/B, PRDM1, and TNFAIP3 chromosomal alterations segregated a cluster of MYD88 L265P patients (Fig. 3C), as did TNFAIP3 and PRDM1 chromosomal alterations only among MYD88 non-L265P patients (Fig. 3D). The association of PRDM1 and TNFAIP3 CNVs is likely due to the same chromosomal alteration, given their localization on chromosome 6q. Deletions of MFHAS1, a potential oncogene shown to be tumorigenic in nude mice (42), were significantly more frequent in MYD88 L265P patients (21.5% vs. 3%, P = 0.004) and MYC deletions tended toward being more frequent in MYD88 L265P patients as well (8.6% vs. 0%, P = 0.06; Supplementary Fig. S3D). On the other hand, MYD88 non-L265P–mutated patients harbored more deletions in B2M and FOXO1 (13% vs. 4% and 8% vs. 2% respectively), although these differences were not statistically significant (Supplementary Fig. S3D).

Gene amplification profiles were quite similar between MYD88 L265P and MYD88 non-L265P–mutated patients: MYD88 gain was the most common in both groups, at nearly identical frequencies (31%–33%, Supplementary Fig. S3D). Interestingly, the MYD88 gain mostly affected the mutated allele, but not exclusively: of cases with MYD88 gain, 33% harbored the L265P variant, 32% harbored other MYD88 variants, and 21% were MYD88 WT. We performed a pyrosequencing assay on cDNA of 15 MYD88 L265P–mutant cases: 2 with no MYD88 CNV, and 13 with MYD88 gain (detailed in Supplementary Methods). This assay highlighted a highly significant positive correlation between MYD88 L265P VAF at the DNA level and relative MYD88 L265P allele expression at the RNA level (Supplementary Fig. S4). KMT2D, ITPKB, IRF4, and STAT6 were also among the most frequently amplified genes in both groups. One notable exception was CARD11 gain, which was significantly more frequent in patients with non-L265P variants (29% vs. 12%, P = 0.02). In fact, among MYD88 non-L265P–mutant cases, associated MYD88 gains clustered with CARD11 chromosomal alterations, as well as with ITPKB and IRF4 alterations (Fig. 3D), whereas among MYD88 L265P–mutant cases, associated MYD88 gains clustered with BCL2 and ITPKB aberrations (Fig. 3C).

Figure 1 recapitulates the distinct mutational and CNV profiles of MYD88 L265P and MYD88 non-L265P DLBCL cases.

Using Affymetrix GeneChip expression data available for 214 patients, we showed that MYD88 expression levels were not significantly modified by the presence of MYD88 L265P in ABC patients (Supplementary Fig. S5A). However, presence of MYD88 gains in ABC patients did increase the expression of MYD88 (P < 10⁻³ Supplementary Fig. S5B).

Previous studies had identified STAT3-high expression as being associated with the MYD88 L265P mutation (5, 43), and we confirmed this in our cohort (P = 0.02, Supplementary Fig. S5C). Furthermore, significant overexpression of PIM1, activated by the JAK–STAT pathway, was also observed in MYD88 L265P–mutant ABC patients (P = 0.02, Supplementary Fig. S5D). This would suggest that positive feedback from the constitutive NFκB pathway activation observed in ABC MYD88–altered patients leads to significant activation of JAK–STAT3–PIM1 signaling.

We sought to better understand the activation of the NFκB pathway by the MYD88 L265P mutation, using gene expression data.

We first analyzed the expression levels of all five NFκB transcription factor family members NfKB1, NFkB2, RelA, RelB, and Rel in ABC patients. Only NFκB1 expression levels were significantly increased in MYD88 L265P patients (P = 0.003, Supplementary Figs. S5E and S6). Furthermore, we observed TNFAIP3 underexpression (P = 0.003, Supplementary Fig. S5F) and a tendency toward CARD11 overexpression in these patients (P = 0.09, Supplementary Fig. S6), also leading to NFκB pathway activation.

We then used a previously described NFκB pathway gene expression signature (44) to create a Linear Predictor Score (LPS), based on the methodology published by Wright and colleagues (2), which reflects NFκB pathway activation (detailed in Supplementary Methods). We retained only genes that significantly separated ABC and GCB subtypes (Supplementary Table S8) and the LPS score significantly separated ABC and GCB patients (Supplementary Fig. S7). Interestingly, ABC patients with CD79B alterations (mutations or gains) had significantly higher LPS scores than those with WT CD79B (P = 0.04, Supplementary Fig. S5G), whereas no significant difference was observed according to MYD88 L265P (P = 0.15, Supplementary Fig. S5H). However, associated CD79B mutations and MYD88 L265P in ABC patients increased the LPS score even more significantly than CD79B alterations alone (P = 0.03, Supplementary Fig. S5I). LPS scores were not significantly increased in ABC patients with NFκB-activating TNFAIP3 or CARD11 alterations, although a tendency was observed in the case of CARD11 alterations (P = 0.08, Supplementary Fig. S7). These results suggest that, according to our LPS score, the presence of CD79B alterations alone activates the NFκB pathway, and that the alteration of both CD79B and MYD88 synergistically increases NFκB pathway activation.

We sought to further validate our GEP data by identifying whether TLR pathway activation was affected by the MYD88 L265P mutation among our cohort. We used a list of genes differentially expressed by primary bone marrow–derived mouse dendritic cells following lipopolysaccharide (LPS) stimulation, based on a previous study (45). Thus, 267 corresponding human genes with Affymetrix microarray probes were analyzed. Of these, 27 (10.1%) were significantly differentially expressed among MYD88 WT and MYD88 L265P patients (Supplementary Fig. S8; Supplementary Table S9). Interestingly, ASAP1 and PTPRS, known to negatively regulate the production of proinflammatory mediators such as interferon in response to LPS, were significantly less expressed among patients with the MYD88 L265P mutation (46, 47). Equally of note, BLNK, whose expression induces higher levels of MYD88-dependent TLRs, was significantly more expressed among patients with the MYD88 L265P mutation (48). These results suggest that the presence of MYD88 L265P variant tends to stimulate TLR pathway activation.

The role of MYD88 L265P has recently been studied in ibrutinib sensitivity, with evidence suggesting that MYD88 L265P alone conveys resistance, whereas the adjunction of CD79A/B mutations conveys sensitivity to ibrutinib (26). However, the adjunction of downstream TNFAIP3 inactivation and/or CARD11 activation in any scenario leads to ibrutinib resistance (26).

Interestingly, TNFAIP3 is mutated in 8% of ABC MYD88 L265P patients, of whom half present truncating stopgain mutations (Figs. 1 and 4). Furthermore, TNFAIP3 is frequently deleted in ABC MYD88 L265P cases (40%), and leads to significantly lesser expression levels, suggesting reduced activity (Figs. 1; Supplementary Fig. S8, P = 0.004). In addition, CARD11 is mutated in 8% and amplified in 14% of ABC MYD88 L265P patients (Figs. 1 and 4). CARD11 gains lead to significantly higher expression levels, suggesting increased activity (Supplementary Fig. S9, P < 10⁻⁴). As a result, 27 of 50 (54%) of ABC MYD88 L265P patients present downstream TNFAIP3 and/or CARD11 alterations, which have been shown to hinder ibrutinib sensitivity. Of the remaining 23 ABC MYD88 L265P patients, 12 (24%) present CD79B mutations and 11 (22%) do not (Figs. 1 and 4).

We performed survival analysis on DLBCL patients explored at the time of initial diagnosis, treated with R-chemotherapy (Table 1, details of treatment in Supplementary Table S1). As expected, OS and PFS were shorter in ABC patients versus GCB patients and in patients with higher IPI (Supplementary Fig. S10). No prognostic difference was identified in ABC MYD88 L265P DLBCL patients compared with either ABC MYD88 WT DLBCL patients (Fig. 5A and B) or ABC MYD88 non-L265P DLBCL patients (Supplementary Fig. S10). Interestingly however, the presence of associated CD79B mutations significantly improved the prognosis of ABC MYD88 L265P–mutant patients, both in OS (P = 0.02) and PFS (P = 0.01; Fig. 5C and D). Moreover, the presence of CD79B mutations alone in ABC DLBCL patients led to significantly better prognosis in PFS (P = 0.02) and tended toward the same result in OS (P = 0.06; Fig. 5E and F). On the other hand, the association of CARD11 or TNFAIP3 alterations did not impact survival (data not shown). Survival analyses were also performed within the same molecular subgroups within patients treated by R-CHOP like therapy and patients treated by R-ACVBP and showed similar results, although with fewer patients (Supplementary Fig. S10).

Given the high prevalence of MYD88 mutations in PCNSL, we compared the prevalence of CNS (meningeal and/or cerebral) relapse in MYD88-mutant and WT DLBCL patients. Of 346 patients for whom data was available, 17 cases experienced CNS relapse and 329 did not (Supplementary Table S1). Clinical characteristics of these 17 cases are available in Supplementary Table S10. There was no statistically significant difference in first-line treatment between cases with CNS relapse and the total cohort, with 94% and 88%, respectively, treated with R-chemotherapy, versus 6% and 10%, respectively, not treated with rituximab (Table 1; Supplementary Table S10). Furthermore, systemic response after first-line treatment was compared with that of 184 evaluable patients with no CNS relapse, and showed no statistically significant difference in the proportion of patients with complete response or unconfirmed complete response (71% of patients with CNS relapse and 70% of patients without) or with partial response (24% of each subgroup). Patients with progressive disease were a minority among both subgroups (6% and 2%, respectively; Supplementary Table S10 and data not shown). Of the 17 cases with CNS relapse, 9 presented MYD88 mutations (53%) and 6 (35%) harbored the L265P variant, whereas only 79 of the 329 (24%) nonrelapsing cases were MYD88 mutant and only 48 (15%) were MYD88 L265P mutant (Supplementary Table S1). This suggests that MYD88-mutant DLBCL cases are significantly more likely to experience CNS relapse than MYD88 WT cases (P = 0.02), as are MYD88 L265P–mutant cases specifically (P = 0.03). When considering only ABC patients, 7 of 11 CNS-relapsing cases were MYD88 mutant (64%), while 52 of 141 non-CNS–relapsing cases were MYD88 mutant (37%). This still tended toward statistical significance (P = 0.1), but would need to be confirmed within a larger cohort.

In this study, we thoroughly examined the associated genomic and transcriptomic landscape of MYD88-mutant DLBCL patients in a large cohort. Furthermore, we strove to analyze subgroups of MYD88-mutant patients, according to the variant presented (L265P or not), and take into account COO information. Our data supports distinct genetic backgrounds for these subgroups and highlights the theranostic and prognostic relevance of examining MYD88 and associated genomic alterations.

Despite the heterogeneity of the methods used to determine COO, we believe that our study ultimately benefits from a subgroup-specific analysis. The gold standard for COO, Affymetrix microarrays, was available for 214 cases and the RTMLPA method, performed in 77 cases, has been favorably compared with Affymetrix microarrays (31). Indeed, the RTMLPA method is proposed as an appropriate alternative to determine COO according to the revised 2016 WHO classification (49). Finally, the Hans immunohistochemical method was used for 49 patients, only when neither Affymetrix nor RTMLPA was available, as recommended by the WHO.

As expected, we highlighted ABC DLBCL mutational profiles among MYD88 L265P–mutated patients, with a high incidence of NFκB pathway–activating mutations, whereas non-L265P mutants harbor a mutational profile more similar to GCB DLBCL. However, unsupervised hierarchical clustering analysis was able to separate distinct mutational profiles within each large MYD88 mutation patient group, highlighting hidden heterogeneity. Interestingly, CNV profiles did not drastically differ according to the MYD88 variant present, although poor prognosis CDKN2A/B deletions were significantly more frequent in MYD88 L265P patients, and clustering identified a group of MYD88 L265P patients with associated CDKN2A/B, TNFAIP3, and PRDM1 deletions, suggesting a synergistic effect of these ABC-related alterations. In contrast, CARD11 gains were more likely to occur in cases with non-L265P mutations, suggesting that CARD11 amplification might act as a bypass to lead to NFκB pathway activation. Unfortunately, MYD88 variants other than L265P remain difficult to analyze specifically despite our cohort size, as few are recurrent and all are far less frequent. However, our study does point toward distinct genetic backgrounds for the two MYD88 mutation subgroups, mostly via a germinal center (GC) versus a non-GC evolution.

GEP analysis was performed, and confirmed previous immunohistochemical reports that the L265P mutation does not affect MYD88 expression levels (20). We also confirmed previous reports of JAK–STAT3–PIM1 signaling activation pathway in the presence of MYD88 alterations (5, 43). MYD88 L265P seemed to induce differential gene expression leading to TLR pathway activation, further confirming previous studies identifying intact TLR activity as crucial for the variant's oncogenic potential (27). Furthermore, we showed that CD79B alteration, but not MYD88 L265P, seems to be sufficient for NFκB pathway activation, but that the associated gene mutations can increase NFκB pathway activation in a synergistic fashion. The original paper by Ngo and colleagues did indeed indicate that MYD88 L265P and CD79B mutations provide nonredundant survival signals to ABC DLCBL cells (5). However, it also highlighted a significant decrease in NFκB pathway signaling when knocking down MYD88 L265P compared with WT MYD88 in an ABC cell line with unknown CD79B mutation status, as well as a significant activation of NFκB reporters when overexpressing MYD88 L265P in GCB cell lines (5). While our GEP data warrants further exploration to confirm functional consequences, the discrepancies between our studies might be related to the existence of additional alterations in patient-derived tumor cells. Furthermore, as the gene signature was established on cell lines with associated CD79B or CARD11 mutations, it is difficult to distinguish whether the measured effect is due to BCR-related NFκB signaling or NFκB signaling as a whole.

The prognostic relevance of MYD88 mutations in DLBCL is still a matter of debate. In this large cohort, we conclude that the MYD88 L265P variant alone is not an independent prognostic factor in ABC DLBCL, as have previous studies (29), but suggest that the MYD88 mutation–related genetic background may have prognostic impact. A previous study came to the opposite conclusion, but included a high percentage of primary extranodal DLBCL cases (43%), which could explain this discrepancy (23). We did, however, demonstrate that MYD88-mutant DLBCL cases were significantly more likely to experience CNS relapse, with a tendency among ABC cases as well, suggesting greater clinical aggressiveness of the disease in some cases. This result did not seem linked to differences in treatment or in response to first-line therapy. This finding supports the data from a recent study on PCNSL, which showed that patients with MYD88-mutant tumors also harbored MYD88 mutations in peripheral blood mononuclear cells, suggesting that MYD88-mutant cells can originate outside of the CNS and develop into PCNSL only after undergoing additional alterations, which confer adaptation to the CNS (50).

We showed that the association of CD79B mutations was sufficient to significantly improve ABC MYD88 L265P patient survival in our cohort, suggesting that these lymphoma cells' genetic addiction to BCR and NFκB pathway activation might be a weakness in the rituximab era. Indeed, this novel finding could potentially be linked to the increased surface BCR expression and BCR signaling induced by CD79B mutations (26, 51) and inhibited by rituximab (52), which might therefore preferentially target CD79B-mutant B cells, explaining their improved prognosis. This data highlights the importance of taking into account patients' genomic profiles as a whole, rather than single mutations, as has been the case in previous studies on MYD88 (29).

A recent phase I/II study reported that ABC DLBCL cases with MYD88 L265P but no CD79 mutations were not sensitive to ibrutinib (26). However, this study did not explore the presence of associated CARD11 or TNFAIP3 alterations to potentially explain ibrutinib resistance. Indeed, according to our study, CD79B WT/MYD88 L265P cases have associated downstream CARD11 and/or TNFAIP3 alterations more often than not. This would suggest that regardless of CD79A/B mutation status, MYD88-mutant patients with no downstream CARD11 or TNFAIP3 alterations might also be potential ibrutinib treatment targets. This hypothesis warrants further investigation of ibrutinib sensitivity in a cohort with targeted analysis of downstream alterations, as the resulting data could lead to doubling the proportion of ABC MYD88 L265P–mutant DLBCL cases likely to respond to ibrutinib treatment, according to our cohort data. In addition, the study performed by Wilson and colleagues warrants validation in a larger cohort, as only 80 R/R DLBCL patients were included. In our study, we have shown that, according to the current literature on the impact of associated alterations, 24% of MYD88 L265P–mutated ABC DLBCL are susceptible to respond to ibrutinib, warranting the widespread use of NGS in clinical settings to best stratify patients according to response potential.

Altogether, our study highlights the relative heterogeneity hidden behind the MYD88 mutant DLBCBL label. Distinct genomic profiles highlight differences in background according to the presence of the MYD88 L265P variant or other MYD88 variants. Furthermore, clinical and prognostic differences were showcased, most notably highlighting the importance of associated CD79B mutations in this DLBCL subpopulation. Finally, this study adds to the field's current knowledge of the theranostic importance of MYD88 mutation status, but also of additional associated alterations, emphasizing the usefulness of genomic profiling to best stratify patients for targeted therapy.

C. Haioun is a consultant/advisory board member for Celgene, Gielad, Jenssen, and Roche. F. Peyrade is a consultant/advisory board member for Jenssen. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Dubois, P.-J. Viailly, E. Bohers, P. Ruminy, F. Jardin

Development of methodology: S. Dubois, P.-J. Viailly, E. Bohers, P. Bertrand, P. Ruminy, C. Maingonnat, F. Jardin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E. Bohers, J.-M. Picquenot, D. Penther, P. Peyrouze, M. Figeac, T. Fest, C. Haioun, T. Lamy, C. Copie-Bergman, R. Delarue, M. André, N. Ketterer, K. Leroy, G. Salles, T.J. Molina, H. Tilly

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Dubois, P.-J. Viailly, E. Bohers, P. Ruminy, S. Mareschal, J.-P. Jais, B. Tesson, B. Fabiani, T.J. Molina, H. Tilly, F. Jardin

Writing, review, and/or revision of the manuscript: S. Dubois, P.-J. Viailly, E. Bohers, P. Bertrand, P. Ruminy, S. Mareschal, J.-M. Picquenot, F. Desmots, T. Lamym, F. Peyrade, M. André, N. Ketterer, G. Salles, T.J. Molina, H. Tilly, F. Jardin

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): V. Marchand, C. Maingonnat, J.-M. Picquenot, M. Figeac, C. Copie-Bergman, G. Salles

Study supervision: P. Ruminy, F. Jardin

We thank Hervé Ghesquières and Jean-Philippe Jais for their critical review.

This study was funded in part by grants from the Institut National du Cancer (INCA), the Direction Gé;né;rale de l'Offre de Soins (DGOS), and the Ligue Contre le Cancer.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.