MYD88 L265P Mutations, But No Other Variants, Identify a Subpopulation of DLBCL Patients of Activated B-cell Origin, Extranodal Involvement, and Poor Outcome

Diffuse large B-cell lymphoma (DLBCL) is a heterogeneous disease in terms of morphology, immunophenotyping, gene expression profile (GEP), oncogenic aberrations, clinical presentation, and outcome (1–3). Two major subtypes related to the different cell of origin (COO) of the tumor have been distinguished according to GEP: germinal center B-cell like (GCB) and activated B-cell like (ABC). Each lymphoma subtype bears a phenotypic resemblance to B cells at a particular stage of differentiation (known as COO). The germinal center B cell is the normal counterpart of GCB subtype, whereas ABC subtype resembles postgerminal center plasmablasts (1, 4, 5). In addition, they are associated with distinct genetic lesions and oncogenic pathways in tumor development (2, 6, 7). GEP studies have shown that the unfavorable prognosis of ABC DLBCL subtype most likely relies on the constitutive activation of NF-κB transcription complex-blocking apoptosis (7–11). A variety of signaling pathways are capable of inducing NF-κB transcription complex, including B-cell receptor (BCR), CD40, and Toll-like receptor (TLR; refs. 8–14). Oncogenic CARD11, CD79A/B, and TNFAIP3 (also known as A20) mutations activate NF-κB by BCR signaling. Inactivation of TNFAIP3 can activate NF-κB downstream of other pathways besides BCR signaling. Moreover, myeloid differentiation primary response gene 88 (MYD88) mutations may activate this pathway through TLR signaling (8–11, 15).

MYD88 is an adaptor protein of the Toll-like and IL1 receptors that participates in the innate immune response and plays a crucial role in the homeostasis of human B cells (16). After TLR's activation, MYD88 phosphorylates and subsequently recruits IL1R-associated kinases (IRAK) and other downstream proteins, such as TRAF6, resulting in NF-κB, JAK kinase/STAT3 activation, and secretion of IL6, IL10, and IFNβ. Ngo and colleagues described that about 39% of ABC DLBCLs had MYD88 mutations, whereas they were rare in GCB cases (11). MYD88 has also been recurrently found mutated at different frequencies in several lymphoid neoplasms, including the majority of Waldenström macroglobulinemias (90%), in 69% of primary cutaneous leg-type DLBCL, in 38% to 50% of central nervous system lymphomas, in 9% of MALT lymphomas, and in approximately 3% of patients with chronic lymphocytic leukemia (CLL; refs. 11, 17–35). MYD88 L265P mutation is the most frequent and most oncogenic form in DLBCL, although a gain of function triggering an increased production of chemokines has also been demonstrated for other variants (11, 34, 36).

The clinical impact and prognostic value of MYD88 mutations are variable among different lymphoproliferative disorders. Most likely, this is due to the fact that the different cellular pathways affected by MYD88 mutation have different relevance for cell survival according to the stage of B-cell maturation, the genetic/epigenetic mechanisms involved in each type of tumor, and the tumor cell interaction with the microenvironment. Thus, CLL patients with mutated MYD88 show a favorable prognosis (35). In DLBCL, MYD88 mutations have been related to specific extranodal sites (particularly the so-called immune-privileged territories) and might be associated with unfavorable outcome (37). However, very few data are available in patients treated in the rituximab era with information on gene expression-COO and the different mutational variants of MYD88. The aim of the present study was to assess the MYD88 mutational status in a series of patients with de novo DLBCL homogeneously treated with immunochemotherapy in a single institution in order to determine its relationship with initial clinical and biologic features, including COO, as well as its impact on response to therapy and survival.

Two-hundred thirteen patients (115 M/98 F; median age, 65 years), consecutively diagnosed with DLBCL according to the World Health Organization (WHO) classification (3) in a single institution between 2002 and 2012, were selected for the current study. Availability of the histologic material was the only criterion of inclusion. Patients with a recognized disease phase of follicular lymphoma or another type of indolent lymphoma with subsequent transformation into a DLBCL, as well as those with immunodeficiency-associated tumors, posttransplant lymphoproliferative disorders, and with intravascular, primary central nervous system, primary effusion, or primary mediastinum lymphomas were excluded. Staging procedures included patient history and physical examination; performance status according to the Eastern Cooperative Oncology Group (ECOG) scale; presence of B symptoms (fever, night sweats, weight loss) and bulky disease (defined as a tumor diameter >7 cm); blood cell counts; serum biochemistry, including lactate dehydrogenase (LDH) and β2-microglobulin (β2m) levels; chest, abdomen, and pelvis computerized tomography scans; PET/CT and bone marrow biopsy. Posttherapy restaging consisted of a repetition of the abnormal previous tests and/or biopsies. Response was assessed according to conventional criteria (38).

Main clinicobiological variables were recorded and analyzed according to MYD88 mutational status. In Table 1, the main characteristics of patients are listed. All patients were treated in the rituximab era, although in 16 cases, rituximab was discontinued mostly due to drug intolerance. Rituximab associated with cyclophosphamide, adriamycin, vincristine, and prednisone (R-CHOP) was administered to most patients (N = 183, 87%). Response to treatment was as follows: complete response (CR), 151 cases (72%); partial response (PR), 19 (9%); and stable disease or progression, 40 (19%). The median follow-up for surviving patients was 6.0 years (range, 0.78–12). Five-year progression-free survival (PFS) of the series was 53% [95% confidence interval (CI), 46%–60%]. Ninety-one patients eventually died during the follow-up, with a 5-year overall survival (OS) of 61% (95% CI, 54%–68%). PFS and OS curves are depicted in Fig. 1. The study was performed according to the guidelines of the Ethic committee of Hospital Clínic of Barcelona. Informed consent to use both clinical data and histologic material was obtained in accordance with the Declaration of Helsinki.

The diagnosis of DLBCL was based on the WHO classification criteria (3). We assessed the proportion of tumor cells in all cases in order to guarantee a minimum of 40% of tumor presence. Tissue microarrays were constructed using a tissue arrayer (MTA I; Beecher Instruments) and included two 1-mm representative cores of each case. Standard methods for tissue fixation (10% buffered formalin) and processing were used. The panel of antibodies included common B- and T-cell markers, as well as CD10 (clone 56C6), MUM1/IRF4 (clone MUM1p), BCL2 (clone 124), and Ki-67 (clone MIB-1) all from Dako, BCL6, kindly provided by Dr. Roncador (Centro Nacional de Investigaciones Oncologicas, Madrid, Spain), and MYC (clone Y69; Epitomics). The conditions for all these antibodies and their evaluation were as previously described and followed the guidelines recommended by the Lunenburg Lymphoma Biomarker Consortium (39, 40). The thresholds herein used for dichotomizing protein expression were based on these references and were as follows: MYC ≥40%; BCL2 ≥50%; BCL6 ≥50%; and Ki67 index ≥80%. Algorithm of Hans was used to assess COO by immunohistochemistry (IHC) in 186 cases (41).

Split FISH for BCL2, BCL6, MALT1, and MYC was performed using probes and a FISH accessory kit following the manufacturer's recommendations (DAKO). The cut-off values for the interphase FISH analyses were established following Ventura's criteria (42). Tonsil sections were used as controls, and for each sample, 100 to 200 evaluable nuclei with complete FISH signals were scored. For the break-apart probes, the cut-off values were 3% for the detection of the rearrangements and 6% to detect gains. For the dual fusion probe, a cutoff of 11% for both structural and numerical abnormalities was considered (43).

DNA was isolated from frozen and paraffin tissue sections with the QIAamp DNA mini Kit (Qiagen) according to the manufacturer's instructions. Screening for the most frequent MYD88 mutations (L265P, M232T, S219C, V217F, and S222R) was performed using an allele-specific PCR assay achieved by Amplification Refractory Mutation System (ARMS) technology (qBiomarker Somatic PCR Assay; Qiagen). This technique is based on the discrimination by Taqpolymerase between a match and a mismatch at the 3′ end of a PCR primer. The method allows reproducible detection of MYD88 mutations in up to 1% of tumor DNA diluted in wild-type DNA. Cutoff for considering a case as mutated was Ct> 37. In MYD88 L265P cases, a relative quantitation analysis was done using the DLBCL ABC OCI Ly3 cell line as a calibrator.

COO by gene expression (GE-COO) could be determined in 129 samples. In 60 cases, gene expression was assessed by using Affymetrix HG U133 plus 2.0 gene expression arrays as previously described (5, 10, 44). All samples predicted as ABC DLBCL at higher than 90% were called ABC DLBCL. The samples that showed less than 10% of probability of being called ABC DLBCL were classified as GCB DLBCL. All the other cases were considered unclassified DLBCL. In addition, other 69 tumors were classified by a NanoString-based technology (45).

Differences in the clinicobiological features between patients with mutated or unmutated MYD88 were analyzed by using χ² test, Student t test, or nonparametric tests when necessary. The actuarial survival analysis was performed by the Kaplan and Meier method and differences assessed by the log-rank test. To evaluate the prognostic impact of different variables, including MYD88 mutational status in PFS and OS, a multivariate analysis was performed with the stepwise proportional hazards model (Cox model; ref. 46).The replicability of results of the Cox analyses was assessed by bootstrap resampling. One thousand samples were built through random extractions with reposition, so that in each sample, a given patient may either not be represented at all or represented once, twice, or more times. The parameters assessed by resampling were the P value of HR in the Cox regression. Bootstrap resampling allows verifying that the predictive value of MYD88 was not critically dependent on the particular composition of the present series. All tests were two-tailed, and P values <0.05 were considered statistically significant.

MYD88 mutations were found in 47 (22%) cases. The most frequent mutation was L265P detected in 39 cases (83% of mutated cases; 18.3% of the whole series; Table 2). The other eight mutated cases corresponded to S219C and M232F mutations (four cases each). Two tumors showed a double mutation status: L265P/S219C and L265P/M232F and were considered part of the L265P group. No V217F or S222R mutations were observed in the present series.

The clinical and biologic characteristics of the DLBCL cases according to the mutational status of MYD88 are shown in Table 2. The distribution according to GE-COO was as follows: 61 cases (47%) were assigned to GCB type and 54 (42%) to ABC type, whereas 14 (11%) remained unclassified. Among these 129 cases with available information on GE-COO, 32 tumors (25%) carried MYD88 mutations. As expected, MYD88 mutations were more frequently found in ABC DLBCL subtype (N = 22, 69%) but were also detected in eight GCB subtype (25%) and two (6%) unclassified. Interestingly, L265P mutation was significantly more frequent in ABC than in GCB-DLBCL (82% vs. 18%, respectively; P = 0.0003). No significant relationship between the other mutations and GCB or ABC DLBCL subtypes was observed. S219C/M232F mutations (N = 8) were detected in 4 ABC DLBCL and 4 GCB DLBC patients with no differences in the type of mutation (two S219C and two M232F cases in both subtypes).

We analyzed the allelic burden according to the COO only in the MYD88 L265P cases due to the lack of calibrator in the other mutations. The frequency for the mutant allele burden varied between 0.05% and 100% (median, 21%). The distribution by quartiles was as follows: Q1: 0.05–0.28; Q2: 0.29–20.9; Q3: 21.0–54.4; Q4: 54.5–100. The proportion of MYD88-mutated samples with allelic burden higher than the median (21%) was 89% (16/18 cases) for ABC subtype and 0% (0/4 cases) for GCB DLBCL (P = 0.002). In addition, allelic burden <1% was found in three of four GCB tumors showing MYD88 L265P mutation.

By using Hans IHC algorithm, 96 tumors were GCB (52%) and 90 non-GCB (48%). The relationship between MYD88 mutations and the DLBCL subtypes identified by Hans IHC algorithm was also analyzed (Table 2). As expected, there were discrepancies in COO between GEP and IHC, 29% and 23% for GCB and ABC, respectively (Supplementary Table S1). L265P-mutated cases (N = 33) were distributed as follows: 10 GCB DLBCL (30%) and 23 non–GCB DLBCL (70%).

The relationship between MYD88 mutations and rearrangements of MYC, BCL2, and BCL6 is detailed in Table 2. BCL2 rearrangements were less frequently found in MYD88 mutated than unmutated DLBCL (8% vs. 21%, respectively; P = 0.05). We observed seven cases with double hit and one triple hit, being all of them MYD88 wild-type. No other statistically significant correlation was observed.

The association between MYD88 mutations and the protein expression of MYC, BCL2, BCL6, and MUM1 is listed in Table 2. Moreover, we analyzed the coexpression of MYC and BCL2 according to MYD88 mutational status. This particular pattern was observed in 41% of MYD88 wild-type versus 18% of MYD88-mutated cases (P = 0.046). Sixty percent of cases had high Ki67 proliferative index (≥80%). Among MYD88 wild-type cases, high Ki67 was observed in 61%, whereas in MYD88 L265P and other MYD88 mutations, such proportion was 73% and 12%, respectively (P = 0.008).

NF-κB signaling activation due to MYD88 mutation ends up with IL6, IL10, and IFNβ secretion. Thus, MYD88 mutational status was correlated with tumor expression of IL6, IL10, TNFα, and IL1β with no significant associations (Supplementary Table S2). There were no differences according to MYD88 in IL6, IL2-R, and TNF serum levels (Supplementary Table S3).

Main clinical features of all DLBCL patients according to the MYD88 mutational status are listed in Table 3. Patients with MYD88 L265P were older, had higher serum LDH levels, and had more frequently extranodal involvement, particularly testis and breast. MYD88 mutation was observed in only 1 of 23 patients with gastrointestinal involvement. No significant differences were observed in terms of specific location of disease or survival when only non-MYC/BCL2 coexpression cases were considered. On the contrary, patients with MYD88 mutations other than L265P showed a trend for exclusive nodal disease (exclusive nodal involvement: 31%, 49%, and 63% for MYD88 L265P, MYD88 wild-type, and MDY88 other than L265P, respectively; P = 0.06).

Treatment administration did not differ according to MYD88 status. CR rates are listed in Table 3 and were not significantly different based on MYD88. One hundred and ten patients eventually experienced failure, including 85 of 166 cases MYD88 wild type (51%), 23 of 39 MYD88 L265P (59%), and 2 of 8 MYD88 other (25%). Five-year PFS was 54% (95% CI, 46%–62%), 44% (95% CI, 28%–60%), and 75% (95% CI, 45%–100%), respectively (P = 0.09; Fig. 2A). Variables predicting poor PFS were older age, presence of B symptoms, poor performance status, bulky disease, extranodal involvement >1 site, advanced Ann Arbor stage, high serum LDH, high serum β2m, high-risk IPI, high tumor BCL2 expression, and ABC subtype (P < 0.04 in all cases). PFS according to MYD88 mutational status for GCB and ABC subtypes in 115 patients with information available is shown in Fig. 2B and C. The most important variables predicting PFS in the Cox multivariate analysis model with 110 patients were IPI (low vs. intermediate vs. high-risk; HR, 2.915; P, 0.002) and COO (GCB vs. ABC; HR, 2.481; P, 0.002).

Ninety-one patients died after a median follow-up of 6.0 years, including 67 of 166 (40%) with MYD88 wild-type, 22 of 39 (56%) MYD88 L265P (56%), and 2 of 8 (25%) MYD88 other than L265P. Five-year OS was 62% (95% CI, 55%–70%), 52% (95% CI, 36%–68%), and 75% (95% CI, 45%–100%), respectively (P = 0.05; Fig. 3A). In the 39 cases carrying the L265P mutation, the presence of mutation at low tumor burden maintained its unfavorable prognostic significance. Other variables associated with poor OS were older age, presence of B symptoms, poor performance status, bulky disease, extranodal involvement >1 site, advanced Ann Arbor stage, high serum LDH, high serum β2m, high-risk IPI, high tumor BCL2 expression, and ABC subtype (P < 0.03 in all cases). OS according to MYD88 mutational status for GCB and ABC subtypes, as assessed by gene expression, is depicted in Fig. 3B and C, respectively. A multivariate analysis was performed including IPI (low vs. intermediate vs. high-risk), MYD88 L265P (unmutated vs. mutated), and MYD88 other than L265P (unmutated vs. mutated). In the final model with 203 patients, IPI (HR, 2.71; P < 0.001) and MYD88 L265P (HR, 1.786; P = 0.023) were independent variables predicting OS. At the resampling test of replicability for validating the results, MYD88 L265P was selected as an independent predictor of OS in 79% of the 1,000 bootstrap samples, whereas IPI was selected in 95%. The same analysis was performed including COO (GCB vs. ABC). In the model with 110 patients, only IPI (HR, 3.401; P, 0.001) and COO (HR, 2.841; P, 0.002) maintained the prognostic value for OS.

DLBCL is a clinical and biologically heterogeneous disease. Although many patients are currently cured with immunochemotherapy, a considerably proportion is still refractory to treatment or eventually relapse. Nowadays, to elucidate the biologic background of patients with unfavorable outcome is a key objective. Gene expression methods can stratify DLBCL according to the COO into GCB or ABC subtypes. In addition, next-generation sequencing of whole-exome sequences has provided the genetic landscape of this lymphoma. In the ABC subtype, the major signaling alterations are related to the constitutive activation of the NF-κB pathway through chronic stimulation of the BCR, CD40, and TLR pathways. However, the GCB subtype is definitely less dependent upon the regulation of any particular pathway. MYD88 mutation that promotes both NF-κB and JAK/STAT signaling conferring a selective advantage in cell survival (11) is the most frequent genetic alteration found in ABC DLBCLs and is associated with a poor outcome (27,37). The identification and characterization of the molecular basis of the dismal prognosis of patients with MYD88-mutant DLBCL could provide the rationale for new treatment targets in the involved pathways. Thus, MYD88 could emerge as a key biomarker in ABC DLBCL (23, 25, 47–49).

We observed MYD88 mutations in 22% of the whole series, being L265P the most frequent one (39 of 47 cases, 83%). These figures are similar to previously reported studies ranging from 14% to 39% (11, 23, 50), but higher than others (around 10%; refs. 24, 25, 27, 37), which may be due to the technique used. Next-generation sequencing techniques at low coverage are probably less sensitive than allele-specific PCR. Whether or not the presence of mutations at lower burden has prognostic impact is of high clinical interest. It is of note that in the present series, the clinical impact of MYD88 mutation was maintained even at low allelic burden, whereas with ARMS technology, we can detect less than 1% of tumor cells.

While in most of the previous reports, the different MYD88 mutations are considered as a whole to assess their clinical impact, we herein have shown that the type of mutation could be relevant. As previously described, patients with a MYD88 L265P mutation are older, mostly of ABC COO, have extranodal involvement, high serum LDH and high Ki67 index, and a poorer OS (Table 3 and Fig. 3; refs. 27, 37, 50). On the other hand, DLBCLs with MYD88 mutations other than L265P have no COO preference, are typically nodal, and the OS is even better than that of MYD88 wild type. However, this information is based on a small number of patients and therefore should be confirmed in further studies. Even though the favorable prognosis of MYD88 mutations different from L265P could be hampered by the low number of patients, our data clearly show that such patients had a considerable better outcome than patients with MYD88 L265P. Moreover, some biologic data can support this different behavior: in a GCB DLBCL cell line with little endogenous NF-κB, Ngo and colleagues demonstrated that wild-type MYD88 only modestly activated NF-κB, whereas L265P had the strongest NF-κB activation. Other isoforms, including S219C, M232T, V217F, and S222R, showed much less capacity to activate NF-κB than L265P (11). This circumstance could explain the favorable evolution of some MYD88-mutated GCB-DLBCL cases. Patients with MYD88 L265P showed poor PFS and OS. This may be due to other related variables such as older age or ABC subtype rather than to the mutation itself. Similarly to previous data, in our series, MYD88 L265P and IPI were independent variables to predict OS (37). However, when including GE-COO in the model, MYD88 L265P lost its prognostic significance at the multivariate level most likely because the vast majority of MYD88-mutated cases were of ABC origin. Interestingly, on the other hand, the coexpression of MYC and BCL2, typically observed in ABC DLBCLs and considered one of the factors of the adverse outcome in this group (51,52), was less frequently seen in our patients with MYD88 L265P than in wild-type ones (18% vs. 41%, respectively). Moreover, patients with MYD88 mutation were significantly older than those without the mutation. This is consistent with the higher rate of these mutations in ABC DLBCL and the high frequency of this subtype in older patients (53). As described in other series, patients with MYD88 L265P frequently had extranodal involvement, including bone marrow, testes, and breast. Primary central nervous system was not evaluable in the present series because those patients were excluded from the study. It is of note that we did not find especial predilection for gastrointestinal system as previously suggested by others (37, 49, 50).

Different gene alterations in DLBCL, including MYC, BCL2, and BCL6, were analyzed. The only significant correlation was the expected lower frequency of BCL2 rearrangement in MYD88 L265P–mutated tumors. Only one of 39 MYD88 L265P cases showed MYC rearrangement, being in line with a recent publication showing that only one of 37 double-hit DLBCL cases had MYD88 mutation (49).

Despite notable improvements in survival after the introduction of an anti-CD20 monoclonal antibody (rituximab) to front-line therapy, a proportion of patients still remains refractory or eventually relapses. New treatments targeting specific tumor alterations are currently under development. NF-κB activation and resistance to chemotherapy are well-known molecular pathways that could be overcome by these novel therapies; thus, MYD88 pathway could represent a key hub to block the viability of tumor cells. Several studies have shown that bortezomib, a proteasome inhibitor, can act synergistically with R-CHOP inhibiting NF-κB and improve the outcome of ABC DLBCL patients (54–56). DLBCLs carrying MYD88 mutations could be particularly sensitive to those drugs inhibiting either NF-κB or JAK/STAT signaling pathways (57, 58). There are several drugs currently being studied that could target different steps of the pathway, including BTK and PI3K inhibitors. Wilson and colleagues have recently showed 5% of responses to ibrutinib in GCB DLBCL versus 37% in ABC DLBCL due to chronic active BCR signaling in the latter. The authors have hypothesized that ABC DLBCL could arise by two distinct pathogenetic routes, one BCR dependent and other independent. Interestingly, only BCR-dependent cases responded to ibrutinib particularly when MYD88 L265P mutation was present (59). In this setting, MYD88 mutational status along with other mutations may be good biomarkers of response to the targeted therapies in the near future.

In summary, MYD88 mutational status analysis by PCR could help to recognize subgroups of patients with particular features and outcome. The current data highly suggest that MYD88 L265P mutation highlights a subpopulation of older patients with an ABC origin, specific extranodal involvement, and unfavorable prognosis. On the contrary, MYD88 mutations other than L265P, also present in the elderly population, may have a favorable outcome. Due to the low incidence of these mutations, current data should be validated in further studies.

No potential conflicts of interest were disclosed.

Conception and design: J. Rovira, A. López-Guillermo

Development of methodology: J. Rovira, K. Karube, D. Colomer, A. López-Guillermo

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Rovira, L. Colomo, A. Martínez-Trillos, E. Giné, I. Dlouhy, J. Delgado, A. Martínez, E. Campo, A. López-Guillermo

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Rovira, K. Karube, A. Enjuanes, L. Colomo, A. Martínez, E. Campo, A. López-Guillermo

Writing, review, and/or revision of the manuscript: J. Rovira, L. Colomo, E. Giné, L. Magnano, J. Delgado, N. Villamor, E. Campo, A. López-Guillermo

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Rovira, A. Valera, A. Martínez-Trillos, I. Dlouhy

Study supervision: A. López-Guillermo

Other (review clinical data and approve the final version of the manuscript): J. Delgado

This work was partially developed at the Centro Esther Koplowitz (CEK), Barcelona, Spain. The authors thank the HCB-IDIBAPS Biobank-Tumor Bank and Hematopathology Collection for sample procurement and Laura Jiménez for her technical assistance.

This work was funded in part by PI12/01536 (to A. López-Guillermo), RD12/0036/0023(to A. López-Guillermo), RD12/0036/0004 (to D. Colomer), RD12/0036/0036 (to E. Campo), SAF12-38432 (to E. Campo), AGAUR 2013-SGR-0795 (to E. Campo), AGAUR 2014SGR967 (to D. Colomer), and PIE13/00033 (to E. Campo and A. López-Guillermo) from Instituto de Salud Carlos III, Spanish Ministry of Health. E. Campo is an Academia Researcher of the "Institució Catalana de Recerca I Estudis Avançats" (ICREA) of the Generalitat de Catalunya. J. Rovira was supported by an “Emili Letang” grant (Hospital Clínic).

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.