Next-Generation Integrated Sequencing Identifies Poor Prognostic Factors in Patients with MYD88-Mutated Chronic Lymphocytic Leukemia in Taiwan

Chronic lymphocytic leukemia (CLL) is the most common type of leukemia in the western countries, but the incidence is low in Asia [1]. In our previous study, we detected somatic hypermutation status of the immunoglobulin heavy-chain variable (IGHV) in 71.2% of Taiwanese CLL, a higher ethnicity-dependent usages of IGHV3-23 and stereotyped subset 8, and a lower frequency of 11q deletion [2].

In addition, higher frequencies of MYD88 mutations (2.4%–13% vs. 3%) and TP53 mutations (11%–20% vs. 5%–10%) and lower frequencies of SF3B1 mutations (5%–10% vs. 9%–21%) and 11q deletions (6.9% vs. 10%–25%) were observed in Asian patients as compared to Caucasian patients in several studies [2‒5]. Recently, KMT2D, which encodes for histone lysine methyltransferase, was also reported to be frequently mutated and associated with worse prognosis in Chinese CLL as compared to Western CLL (8% vs. <1%) [5]. The prognostic relevance of MYD88 mutations was also conflicting for patients with CLL reported from different studies that could be attributed to ethnic differences [6‒8].

In this study, we sought to find genetic lesions that could be more enriched within Taiwanese CLL than in non-Taiwanese CLL and their prognostic values by using whole-genome sequencing (WGS) and whole-exome sequencing (WES). We also performed validation of the ethnicity-dependent mutated genes in a larger cohort of CLL patients (n = 215).

Between 1991 and 2020, 215 patients with diagnostic CLL were consecutively diagnosed and followed up in a single tertiary referral center in Taiwan. Diagnosis of CLL was based on the International Workshop of CLL-National Cancer Institute (IWCLL-NCI) criteria [9]. All patients met the criteria of ≥5 × 10⁹/L monoclonal B cells in the peripheral blood with expression of CD5, CD20, and CD23 by flow cytometry. The patients’ clinical features, including age, gender, and Binet stage, are listed in online supplementary Table S1 (for all online suppl. material, see https://doi.org/10.1159/000541709). Peripheral blood or bone marrow mononuclear cells from time of diagnosis from each patient were stored frozen at −80°C. Matched buccal samples were collected from patients as control samples. The study was approved by the Institutional Review Board of Chang Gung Memorial Hospital (201405283B0).

WES was performed using the SureSelect Clinical Research Exome Kit V2 or SureSelect Human All Exon Kit V4/V5 (Agilent Technology, Santa Clara, CA, USA). WGS was performed using the KAPA HyperPrep PCR-free Kit (Roche Kapa Biosciences, Santa Clara, CA, USA). The DNA libraries were sequenced using the Illumina instrument with a standard 100- or 150-bp paired-end read protocol.

The reads were aligned using BWA-MEM (v0.7.17) [10] to the hs37d5 reference genome. PCR duplicates were subsequently marked by Sambamba (v0.6.5) [11]. The alignment coverage statistics were computed by Qualimap (v2.2.1) [12]. The sequencing coverage and quality statistics were summarized in online supplementary Table S2. Short variants were called using Strelka2 (v2.9.4), and only variants with the “PASS” were curated for subsequent analyses. For WES data, additional parameter “-exome” was used to skip the depth filter and positional short variants have to be supported by at least 20× reads, variant depth of at least 5× and a variant-allele frequency of at least 3%. wANNOVAR (7th July 2021) [13] was used to annotate the short variants. The somatic single-nucleotide variants (SNVs) were further analyzed for the composition of mutational signatures within each sample using SigProfiler [14] based on mutational signatures from COSMIC v3.3. Somatic copy number variation (CNV) was called using CNV kit (v0.9.6) and the Log₂ thresholds of 0.5 and −0.7 were used to denote copy gain and copy loss for the WGS data, respectively. Somatic structural variation (SVs) were called using Manta (v0.29.6) [15] and annotated by AnnotSV (v1.2) [16]. Each candidate SV was subjected to the following filtering criteria: (1) SV is supported by at least 3 discordant read-pairs and (2) breakpoints must be supported by at least 3 soft-clipped alignments. Oncoplot and co-mutational diagrams were generated by Maftools (v2.18.0) [17].

Ion AmpliSeq primer pools were used to amplify MYD88 (exons 2–5), IGLL5 (coding region), and KMT2D (coding region) and subsequently sequenced on the Ion Torrent PGM (Life Technologies, Carlsbad, CA, USA) machine with 1000X depth of coverage. Mutations were then analyzed and annotated by Ion Torrent’s propriety software. Sanger sequencing or pyrosequencing were used to validate these mutations.

Overall survival (OS) was defined as the time from diagnosis to death or the latest follow-up as of September 14, 2022. All statistical analyses were carried out using the IBM SPSS Statistics version 17. Categorical variables and tumor mutational burden (TMB) were compared using Fisher exact test and t test, respectively. Multivariate analysis was performed by Cox proportional hazard regression. Survival curves were constructed by Kaplan-Meier estimate and differences were evaluated by log-rank test. Two-tailed p values less than 0.05 were considered as statistically significant.

WGS of 19 tumor/normal pairs and WES of 11 tumor/normal pairs were analyzed. The mean coverage of tumor WES, normal WES, tumor WGS, and normal WGS was 113.9X, 66.8X, 64.9X, and 30.2X, respectively. Somatic non-silent protein-changing short variants with a median of 47 variants and an average of 46.4 (range: 22–149) per sample were detected in our study cohort (online suppl. Fig. S1a). The TMB of CLL was compared with other cancers in TCGA (online suppl. Fig. S1b). TMB of 5 published CLL studies [18‒22] including ours were further compared and showed a significant difference among them (online suppl. Table S3). The activities of 14 mutational signatures (SBS1, 3, 4, 5, 7d, 8, 9, 12, 18, 26, 37, 40, 42, and 89) from COSMIC version 3.3 (https://cancer.sanger.ac.uk/signatures/sbs/) [23] were detection in our study cohort (Fig. 1a). Notably, signature SBS9 was observed in 13 of our samples and 12 of them, except CLL-010, were validated to carry mutated IGHV by Sanger sequencing. Signature SBS5 (28/30, 93.3%) and SBS1 (27/30, 90.0%) were ubiquitous. The activity of SBS5 was negligible in CLL-021 and CLL-027, and the activity of SBS1 was also negligible in CLL-020, CLL-023, and CLL-022. In addition, 5 patients (CLL-006, CLL-017, CLL-018, CLL-020, and CLL-016) carried SBS18 that was associated with damage by reactive oxygen species [23, 24].

In our cohort, 75.9% (22/29) of patients carried mutated IGHV and the most frequently mutated gene was IGLL5 (47%), followed by ATM, MYD88, CHD2, DNAH17, DSCAM, and TCHH with a frequency of 13%–20% and 10% each for AHNAK2, AXDND1, BICRA, COQ2, CSMD3, FLG, KMT2C, KMT2D, MYT1L, OXNAD1, PRAMEF2, RBM43, SSPO, and USH2A (Fig. 1b). The mutations were associated with 5 signaling pathways (online suppl. Fig. S1c): TP53 pathway (11/30, 36.7%), RTK-RAS pathway (9/30, 30.0%), Hippo pathway (8/30, 26.7%), WNT pathway (8/30, 26.7%), and cell cycle control (5/30, 16.7%). None of co-occurring mutations revealing significant association (p < 0.01) was found (Fig. 1c). After filtering mutations with a frequency of more than 1% in ExAC EAS and/or 1000G EAS, 13 of the 21 frequently mutated genes presented in more than 3 patients were defined as recurrently mutated genes (online suppl. Table S4). As shown in Table 1, there were significant differences in the frequencies of 9 of 13 recurrently mutated genes between Taiwanese CLL and Western cohorts from 2 studies (p < 0.05) [20, 22]. Two genes (MYD88 and KMT2D), which were reported to exist high incidence of mutations in Chinese [5] and the most recurrently mutated gene, IGLL5, in 30 WGS/WES were resequenced in another 185 CLL patients for confirmatory purposes.

As shown in Figure 1d, 40.0% (n = 12) and 20.0% (n = 6) of our cohort had at least one mutation of the reported drivers in CLL [19] and CLL-related genes previously listed in COSMIC or PanCancer [25], respectively. In total, 50.0% of patients harbored mutations in CLL-associated genes (group A, n = 15). Seven of the remaining 15 patients (group B, n = 15) had at least one mutation in MYD88, KMT2D, and IGLL5. MYD88 mutations were significantly enriched within group B (5/15 vs. 0/15, p = 0.0421).

Next, we investigated genome complexity (GC) of our samples using CNV larger than 1 Mb from our data. We found that 38.7% of our cancer genomes were inflicted with such CNVs (median: 2 and range: 1–32). As shown in Figure 1e, the most common chromosomal loss and gain were 13q/11q and 12q/7q/12p, respectively. FISH analysis also confirmed that the detected 12q/12p gains from sequencing data were resulted from trisomy 12. We further classified the 30 patients into 3 groups [26] according to their respective CNV loads: low GC (0–2 CNV; n = 26), intermediate GC (3–4 CNV; n = 1), and high GC (≥5 CNV; n = 3). As shown in online supplementary Table S5, all genetic alterations were frequently found in low GC CLLs. Somatic SVs analysis in the 19 WGS samples detected a total of 184 (average: 9.7, range: 3–21) SVs, comprising 67 deletion-type, 25 duplication-type, 33 inversion-type, and 59 interchromosomal-type SVs. As expected, a high frequency of the SVs (40/184, 21.7%) was located within the locus of the IGH gene as a consequence of the somatic expanded cell populations that underwent V(D)J recombination.

Combing the data of 185 targeted sequencing and 30 WGS/WES, the mutational frequencies of IGLL5, MYD88, and KMT2D were 29.3% (63/215), 20.9% (45/215), and 15.0% (32/214), respectively. The frequency of MYD88 mutational spectrum included V217F (26/45, 57.8%), followed by L265P (9/45, 20.0%), S219C (7/45, 15.6%; in which 1 patient also carried V217F), and other mutational types (4/45, 8.9%). The mutation patterns of IGLL5 included non-synonymous mutations (56/63, 88.9%), 5′UTR mutations (5/63, 7.9%), and splice site mutations (2/63, 3.2%). Of the 214 patients with available samples for these 3 gene mutational analysis, 36 patients had 2 coexisting mutations: 22 for MYD88 and IGLL5, 9 for KMT2D and IGLL5, 5 for MYD88 and KMT2D, in which 2 patients carried 3 mutations. Sixty-two of the remaining 176 patients had only one mutation of the 3 genes. Totally, near half (46.7%) of patients had at least one mutation of the 3 genes.

We further analyzed the association of the mutations of these 3 genes with the previously reported genetic factors in 173 patients [2]. As shown in Table 2, MYD88 mutations were significantly associated with IGLL5 mutations (p = 0.0004), mutated IGHV (p < 0.0001), deletion of 13q (del 13q) (p = 0.0164), the absence of del 11q or ATM mutations (p = 0.0109), and wild-type SF3B1 (p = 0.0451). IGLL5 mutations were significantly associated with MYD88 mutations (p = 0.0011), mutated IGHV (p < 0.0001), and del 13q (p = 0.0308). We also observed that 47.2% (25/53) of 53 IGLL5-mutated patients carried additional genetic alterations: 6 with trisomy 12, 3 with TP53 alteration, 3 with ATM alteration, 4 with KMT2D mutation, and 9 with more than one alterations. In contrast, no significant correlation was found between KMT2D mutations and other genetic abnormalities.

Totally, the median OS was 124.7 months (95% CI: 86.8–162.6 months) with a median follow-up of 67.4 months. There was a significant difference in OS between patients with and without MYD88 mutations (p = 0.012, Fig. 2a). However, similar OS was found among patients with different positional mutations of MYD88 (p = 0.494, online suppl. Fig. S2a). Likewise, patients with either KMT2D mutations (p = 0.841, Fig. 2b) or IGLL5 mutations (p = 0.153, Fig. 2c) had comparable OS compared with patients without the corresponding gene mutations and also in mutated IGHV cohort (online suppl. Fig. S2b, S2c). Regarding the prognostic impact of co-occurring gene mutations on OS, we observed a more favorable OS in patients with MYD88 mutation alone when compared to patients with co-occurring KMT2D (mean survival: 185.3 months vs. 137.8 months, p = 0.153, online suppl. Fig. S2d) or IGLL5 mutations (mean survival: 196.9 months vs. 130.0 months, p = 0.078, online suppl. Fig. S2e) but did not reach statistical significance. Interestingly, a significantly superior OS was found in patients with MYD88 mutation alone (median not reached) when compared with MYD88-mutated patients with co-occurring KMT2D and/or IGLL5 mutations (median survival: 131.8 months, 95% CI: 72.0–191.6) (p = 0.007, Fig. 2d).

With Cox proportional hazards model for univariate regression, the significant risk factors consisted of age, Rai stage, Binet stage, IGHV status, TP53 alterations, and MYD88 mutations. We further integrated KMT2D and IGLL5 mutations with other risk factors in the multivariate analysis (Table 3). Age ≥60 (p = 0.001), Rai stage III+IV (p = 0.014), Binet stage B+C (p < 0.0001), unmutated IGHV (p = 0.021), and TP53 alterations (p = 0.0007) were independent risk factors, whereas MYD88 mutations with wild-type KMT2D and wild-type IGLL5 were independent predictor for favorable OS (p = 0.028).

We analyzed the distribution of mutational signatures for CLL in a Taiwanese cohort. The most prevalent was SBS5 (clock-like) followed by SBS1 (clock-like), SBS9 (polymerase η activity), and SBS18 (reactive oxygen species). Mutational signatures, SBS1, SBS5, and SBS9, were reported to be frequent in CLL [27], which is consistent with our finding. SBS8 and SBS18 were likely generated by DNA damage caused by oxidative stress. SBS8 was increased in samples with mutated genes from the double-strand repair pathway [28]. These highlighted the implication of DNA repair activities in the tumorigenesis of CLL and the relevance of DNA damaging agents as therapeutics options too. SBS9 could be induced during DNA replication by polymerase η as part of somatic hypermutation in lymphoid cells, and this process was associated with B-cell-derived hematologic malignancies [23], which was enriched within our mutated IGHV samples [27]. This explains why SBS9 was prevalent in our cohort of Taiwan CLL too.

Comparison of WGS results between 485 European CLL [27] and 30 WGS/WES samples in this study showed that Hippo pathway and gain of 7q were unique to our cohort, supporting that there was difference in the genetic landscapes of CLLs between the West and Asia. Hippo pathway has recently been discussed in leukemia including chronic myeloid leukemia, acute myeloid leukemia, acute lymphoblastic leukemia, and CLL [29]. None of our WGS/WES samples harboring the 3 ethnicity-associated gene mutations had complex karyotypes, which was reported to be an adverse prognostic marker for CLL [30, 31]. In this study, survival analysis of 215 CLL patients further confirmed a favorable OS in patients with MYD88 mutations as well as a comparable OS for patients with or without IGLL5 or KMT2D mutations.

The mutational frequencies of IGLL5, MYD88, and KMT2D were different between Taiwanese and Western CLL cohorts. The high prevalence of MYD88 and KMT2D mutations were also observed in other Asian countries compared to at most 4% and 1% of MYD88 and KMT2D mutations in CLL, respectively, in the West [20, 21]; 13% for MYD88 mutations and 8% for KMT2D mutation in China [5]; 28% and 6.2%, respectively, in Korea [32]. Of the 45 patients with MYD88 mutations, results of immunofixation electrophoresis were available in 21 patients, of whom none had monoclonal IgM paraprotein whereas 19 patients had monoclonal IgG paraprotein (16 IgG kappa and 3 IgG lambda), all had low or normal serum IgM levels. In lymphoplasmacytic lymphoma, the great majority of MYD88 mutations are L265P [33, 34]. The immunohistochemistry (IHC) study of the bone marrow biopsy specimens and CXCR4 mutational analysis were performed in the 9 patients carrying MYD88 L265P. Only 1 patient had a nonsense CXCR4 mutation but without IgM with light chain restriction by IHC. Only 1 patient had a tiny population (1%) monotypic plasma cells (IgM kappa); of note, this case had a CLL-specific 13q deletion by FISH. Thus, the possibility of lymphoplasmacytic lymphoma in our CLL cohort could be ruled out. All these results supported that MYD88 and KMT2D mutational frequencies were different in CLL patients of different ethnics. Recently, IGLL5 was also reported to be the most recurrently mutated gene in a cohort of 52 Chinese CLL (29%) [35], which was concordant with 29.3% of our 215 patients. The frequency (60/215, 27.9%) of IGLL5 mutations still was the highest one in our cohort after removing 3 patients with IGLL5 mutations located in the coding region overlap with IGJ1 gene. On the other hand, variable mutational frequency of IGLL5 was found in the West, 30% of 30 patients from USA analyzed with WGS data [36], 15% of 233 CLL from France analyzed by using 54 genes targeted sequencing [37], and 2.8% (15/538) from a study reported by Landau et al. [20]. Thus, whether IGLL5 mutation was rarely found in the West remains to be determined. Furthermore, the range of variant frequency (VF) of MYD88, IGLL5, and KMT2D were 16.1%–56.4%, 6.7%–90.5%, and 14.0%–64.4%, respectively. Mutations with ≤10% of VF were defined as subclonal mutations [38]. Only one mutation, IGLL5 (VF: 6.7%), was present as a subclonal mutation in this study. Thus, subclonal mutations are not an issue for the higher frequency and prognostic impact of the 3 genetic mutations.

The MYD88 V217F and MYD88 L265P mutations occurred commonly in patients with CLL [5]. The frequency of V217F mutation (58%) was higher than that of L265P (20.0%) in Taiwan compared to 18% and 68%, respectively, in the West [39]; 47% and 41% in Korea [32]; 32% and 45% in China [5]. There was no significant difference in survival between patients with MYD88 L265P and MYD88 V217F in our cohort as well as in other studies [32, 39]. These results indicated that the different positional mutations in MYD88 did not affect the survival of patients with CLL.

The histone H3 lysine 4 (H3K4) methyltransferase KMT2D can act as coactivators of tumor suppressors, such as p53, or affect the expression of other tumor suppressor genes [40, 41]. KMT2D mutations could result in the loss of H3K4 mono- and di-methylation at enhancer regions or promoters, driving germinal center B-cell expansion [41]. The higher frequency of KMT2D mutations in Asia including our cohort, compared to the West, suggested that the regulation of cell cycle and chromatin modification, which interact with KMT2D, should be considered in the leukemogenesis of CLL in Asia. In other words, comparable outcomes between patients with or without KMT2D mutations in our cohort, as well as in the studies from Korea [32] and China [5] might be due to the other complex interactions involving KMT2D.

The immunoglobulin lambda-like polypeptide IGLL5 is involved in B-cell activation through the BCR signaling pathway [42]. In this study, IGLL5 mutations were highly co-existed with other alterations and occurred more frequently in the Taiwanese CLL patients compared to those from the Western countries [2]. However, patients with IGLL5 mutations did not have inferior OS than patients without IGLL5 mutations. In contrast, patients with MYD88 mutations together with IGLL5 mutations had adverse OS compared to patients with MYD88 mutations alone, supporting that IGLL5 mutation had a negative prognostic impact in MYD88-mutated patients.

We showed all MYD88-mutated patients in this study carried mutated IGHV and either IGLL5 mutations or KMT2D mutations had no impact on outcome of patients with mutated IGHV. Thus, effect of IGHV status in outcome analysis of MYD88-mutated patients co-occurring with KMT2D or IGLL5 mutations can be ruled out. In this study, we found that MYD88-mutated patients without KMT2D and IGLL5 mutations had longer OS as compared to patients with co-occurrence of MYD88 mutations with KMT2D and/or IGLL5 mutations. We further analyzed outcome of 24 MYD88-mutated patients, who carried either IGLL5 and/or KMT2D mutations and also had available data of other CLL-related genetic alterations, including NOTCH1 mutation, SF3B1 mutation, BIRC3 mutation, IGHV mutational status, ATM alteration (ATM mutation and 11q deletion), and TP53 alteration (TP53 mutation and 17p deletion). Only 3 patients had co-occurring alterations: 2 with 17p deletion and 1 with ATM alteration; we then excluded these 3 patients for survival analysis and found a comparable median OS (p = 0.999). Thus, the influence of co-occurring mutations with known prognostic impact on the poor outcome of MYD88-mutated patients with either IGLL5 and/or KMT2D mutations can be excluded. Only 1 of our patients received ibrutinib as front-line therapy, and none of the 45 MYD88-mutated patients received novel agents, including BTK inhibitors, BCL-2 inhibitors, or PI3K inhibitors, as first-line therapy. Thus, the favorable outcome of MYD88-mutated patients without KMT2D and/or IGLL5 mutations was not caused by treatment with new agents. It had been reported that NF-kB signaling pathway in B cells can be regulated through BCR in addition to toll-like receptors/MYD88 pathway [43]. Phelan et al. [44] identified a supercomplex combining MYD88 and BCR-pathway in diffuse large B-cell lymphoma. Fowler et al. [45] also showed that BCR activation reduced trimethylation of H3K27 [45], which could accumulate due to the loss of KMT2D function [46]. Thus, there could exist cross-talk mechanisms between MYD88 and KMT2D or IGLL5 in the disease progression of CLL.

In sum, through WGS/WES analysis on 30 CLLs, we identified 9 mutated genes: IGLL5, MYD88, TCHH, DSCAM, AXDND1, BICRA, KMT2D, MYT1L, and RBM43. Targeted sequencing confirmed higher frequencies of IGLL5, MYD88, and KMT2D mutations in additional 185 Taiwanese CLL patients compared to those in Western cohorts. We also discovered that the frequency of V217F was higher than L265P in the MYD88 gene of our cohort. The co-occurring of MYD88 mutations with IGLL5 and/or KMT2D mutations resulted in unfavorable OS for MYD88-mutated patients.

We thank Mr. Tung-Huei Lin for statistical analysis, Ms. Ting-Yu Huang for secretarial assistance, and Ms. Shih-Yu Wang for bioinformatics analysis.

The study was approved by the Institutional Review Board of Chang Gung Memorial Hospital (201405283B0). Written informed consent was obtained from all participants to participate in the study.

The authors have no conflicts of interest.

This work was supported by Chang Gung Memorial Hospital (CMRPG3E0271-3) to L.-Y.S., National Medical Research Council (MOH-OFYIRG22jul-0017) to J.Q.L., and Flagship Program of Precision Medicine for AsiaPacific Biomedical Silicon Valley (109-0324-01-19-06) to S.-F.T.

L.-Y.S. designed and supervised the study; M.-C.K., P.-N.W., H.-W.K., J.-H.W., and L.-Y.S. provided patients’ samples and their clinical data; L.-Y.S., Y.-J.H., and J.Q.L. developed the methodology; C.-C.C. performed experiments; L.-Y.S., Y.-J.H., J.Q.L., J.S.H., S.-F.T., and C.K.O. analyzed and interpreted the data; and Y.-J.H. and L.-Y.S. wrote the manuscript.

Y.-J.H. and J.Q.L. contributed equally as co-first authors.

All data generated or analyzed during this study are included in this article and its supplementary material files. Further inquiries can be directed to the corresponding author.