Whole-exon sequencing of human myeloma cell lines shows mutations related to myeloma patients at relapse with major hits in the DNA regulation and repair pathways

HMCLs were previously characterized [1, 6, 7]. HMCLs were cultured in RPMI-5% fetal calf serum with or without 3 ng/ml of IL6 [1, 6, 7]. Gene expression profile of HMCLs has been previously published [1]. The gene expression profile of primary MM cells was assessed from 414 patients (Arkansas) as previously described [1, 8].

DNA sample processing was performed according to Agilent Technologies (Santa Clara, CA, USA) using the sureselect target enrichment system kit (Human all exon v6, library version 1.6). Sequencing was performed on HiSeq 2500, High Output in paired-end 2 × 100 bp. The reads were aligned (BWA-v0.7.10-r789) to the GRCh37 human reference genome. Duplicated reads were marked by the Picard tool (v1.119), indels were realigned around capture (± 500 bp), and base quality recalibration was finally performed (Genome Analysis Toolkit [GATK-v3.2.2]). In the absence of germline DNA, variants were called by the GATK unified genotyper. Variants were processed through vcf2maf-1.6.15 to obtain a final Mutant Annotation File (maf). The variants’ biological effects predictions were carried out using Ensembl’s VEP-annotator-v.86. Variant annotation database versions were as follows: ExAC-r0.3.1 [likely germline variants], dbSNP-v.144 [known variants], COSMIC-71 and ClinVar-v.201507 [clinical significance of known variants].

Variants that were present more than three times were removed, as well as variants with Global Allele Frequency in ExAC databases over 1% (with respect to ethnicity frequencies when known). Finally, clinically benign mutants, as annotated by ClinVar, were removed (“benign” or “likely benign”). Only protein-coding variants were used for subsequent analyses, and structural protein coding genes (actin, myosin, collagen, fibronectin, vitronectin, tenascin, laminin, titin, obscurin, plectin, aggrecan, and mucins) were removed.

Exon loss was estimated from the read depth using ExomeCOPY and CANOES. The results were validated by visual inspection of the BAM read depth in Integrative Genomics Viewer (IGV; Broad Institute). Genes with frequent variants were selected and were assessed by direct Sanger sequencing on cDNA.

The cell count and viability were measured using the MTT assay. The cell cycle distribution was assessed by propidium iodide incorporation. Rb phosphorylation was assessed by western blotting (Cell Signaling; 4H1 and S807-811). The area under the curve (AUC) was estimated using Graphpad Prism v7.0 for palbociclib (0–1 μM), CX5461 (0–1 μM), and trametinib (0–25 nM). The responses to melphalan, bendamustine, FAS and TRAIL-R agonist antibodies, PRIMA-1^Met, dexamethasone, RITA, ABT-737, and ABT-199 were previously reported [6, 9‐14]. Results were scaled (mean-centered and standardized) to provide a z-score.

Analyses were performed under R 3.4.4. Fisher’s test was carried out with the resampling of parameters for robustness. The somatic interaction plot code was adapted from Gerstung et al. [15]. Enrichment analyses were carried out by ReactomePA and clusterProfiler [16, 17], p values were adjusted for multiple testing by the false discovery rate (q = 0.05). For the Reactome determination, KEGG and GO annotations were used. MAF manipulation was performed using the maftools packages [18]. Oncoprints, heatmaps, and Chord-Diagrams were performed with ComplexHeatmap R-package. Considering the number of samples, the linear regressions between scores and drug responses were calculated by robust a linear regression using a M-estimator (rlm, MASS package) in order to discard outliers. Coefficients were further bootstrapped by Boot function (car package), with 5000 replicates (seed = 22,062,016) and considered significant if the 95% confidence interval (95% CI) did not overlap with zero, only β1 coefficients are presented in the text.

WES was performed in 33 HMCLs of European, American or Asian origin, 19 having been derived in the presence of exogenous IL6 (Additional file 1: Table S1). After global SNP enrichment analysis on 609,585 bi-allelic SNPs (SNPRelate package [19]), three groups of HMCLs were identified: a group gathering HMCLs of Pacific/Japanese origin (AMO-1, KMM1, KMS12PE, KMS11, NAN8, OPM2) and a cluster encompassing all other HMCLs except MM1S, which was individualized as African ethnicity (Additional file 1: Figure S1). To remove ethnic-related SNPs, HMCLs were filtered with Global Allele Frequencies, plus East Asian frequencies for the Pacific/Japanese cluster and African frequencies for MM1S. Because of the lack of normal DNA from patients from whom the HMCLs were derived, we could not easily discriminate the constitutive SNPs from the tumor-associated mutations. Thus, we excluded variants shared by more than 3 HMCLs of the 33: indeed, the most mutated genes in HMCLs and myeloma patients [20], i.e., RAS and TP53, never displayed more than three identical variants across the HMCL collection. For NRAS, the most frequent variant was c.38G-A (Gly12Asp) in JJN-3, Karpas620, and Nan7, while the only TP53 shared variant was 406G-A (Karpas620, XG11). Variant effect predictions were carried out as described in the “Methods” section. We further removed variants of genes uniformly low expressed across the collection (maximum of the considered gene inferior to the first quartile mean expression of the microarray). After filtering, we retained 15,602 variants, spanning over 7641 genes (Maf file, Additional file 2). Most mutated samples were KMM1 and KMS12PE with 916 and 755 variants, respectively. The most frequent variant was missense (n = 14,309; 92%), while frameshifts occurred in 273 variants (1.7%), insertions or deletions without frameshifts occurred in 226 cases (1.4%), and 482 variants (3.1%) were nonsense mutations (Additional file 1: Figure S2). Single mutations were mainly C > T transitions (63%, Additional file 1: Figure S3), corresponding to spontaneous deamination of 5-methyl cytosine. HMCLs age was not associated with a particular mutation (Fisher test, FDR > 0.05), but younger cell lines displayed a lower mutation load (β = 4.29, 95% CI = [1.07; 9.47]). Mutations were confirmed in 18 genes by direct sequencing of RT–PCR products as previously reported for RAS and TP53 [1, 9] (Additional file 1: Table S1). Although amplification of genes was not assessed because of the high number of chromosome abnormalities across the HMCL collection, exon losses were reported as described in the “Methods” section. Main variants are presented in Lollipop Plots (Additional file 3).

Figure 1 shows the most frequently altered genes across the collection and recurrent in MM [20‐22]. Residues modified by mutants are provided in Additional file 1: Figure S4. Five genes, i.e., TP53, KRAS, NRAS, CDKN2C and PRKD2 were altered in at least 21% and up to 67% of HMCLs.

HMCLs shared a similar mutations rate with MM patients, either at diagnosis (DMM) or relapse (RMM) for DIS3 (12% in HMCLs, 10% in DMM and 13% in RMM), PRDM1 (3%, 2%, and 5%, respectively), BIRC3 (3%, 2%, and 3%, respectively), and EGR1 (3%, 4%, and 4%, respectively) (Fig. 2). Of note, these very similar rates among HMCLs, DMM and RMM were in favor of early pathogenic mutations, poorly affected by subsequent treatment selection or cell culture. KRAS mutation rates were roughly shared between HMCLs, DMM, and RMM (21%, 24.7%, and 27%, respectively). By contrast, the NRAS mutation rate increased from DMM (19%) to RMM (24%) and HMCLs (30%). Similarly, TP53 (67%), CDKN2C (33%), PRKD2 (18%), FAM46C (15%), and BRAF (15%) mutation rates displayed a dramatically increased frequency in HMCLs compared with those in primary myeloma cells, either in DMM or RMM [2, 20‐24] (Fig. 2). These high frequencies in the FAM46C, TP53, BRAF, and NRAS rates might be in line with either successive relapses and/or secondary plasma cell leukemia (PCL), from which HMCLs are mostly derived [25‐29].

WES revealed that HMCLs displayed frequent mutations in Fanconi anemia genes (PALB2 [12%], FANCI [12%], FANCA [9%], FANCD2 [9%], BRCA2 [9%]) as well as in helicases (such as RECQL4, 15%, and BLM, 15%) and epigenetic modifiers (e.g., TET2, 15% and SETD2, 6%). FANC family genes were recently reported to be mutated mostly in patients at relapse [24, 30, 31], suggesting that these mutations did not occur in vitro in continuously replicating cells but in vivo. Mutations in epigenetic modifiers were recently described as being more frequent at relapse [20, 22, 24, 32], such as histone methyl-transferases (6.9% vs 17%, in DMM and RMM, respectively) and DNA methylation modifiers (1.9% vs 8.3%, respectively). On the other hand, genes involved in apoptotic pathways (extrinsic, intrinsic, execution) displayed few mutations or deletions, showing that cell death resistance was not associated with major defects in the apoptotic machinery.

These data collectively showed that HMCLs displayed mutations/deletions related to myeloma cells from patients at diagnosis (KRAS, DIS3, EGR1, PRDM1 and BIRC3) and relapse/progression to PCL (TP53, NRAS, BRAF, FANC genes).

We further analyzed mutually exclusive and co-occurring mutations/deletions, as well as their associations with IgH translocation (Additional file 1: Figure S5). Abnormalities in the Ras/MAPK pathway were mutually exclusive: KRAS, NRAS, HRAS, FGFR3 (and to a lesser extent BRAF) displayed mutually exclusive mutations (p < 0.05), and FGFR3 mutations were exclusively found in t(4;14) HMCLs overexpressing FGFR3 (p = 0.003, Additional file 1: Figure S5). TP53 hits were mutually exclusive to mutations in ATM (p = 0.03), as previously reported in all B cell malignancies [33]. By contrast, several co-occurring mutations were found in DNA damage/repair/epigenetic modifiers for instance, in BLM and FANCD2, RECQL5, and ATM, HDAC7, and DOT1L (p < 0.05). CDKN2A deletion was found in HMCLs with CDKN2C mutations/deletions. RECQL4 and BLM mutations were significantly associated with t(11;14), p = 0.03 and p = 0.002, respectively.

To provide a comprehensive landscape of mutations/deletions, we next performed global analysis of altered pathways based on the whole data of the mutated genes.

Gene Ontology (GO) enrichment analysis showed that most of the dysregulated biological processes were related to Rho GTPase signal transduction, Cell cycle/DNA replication and DNA damage (check-points before replication, DNA repair, DNA unwinding) (Additional file 1: Figure S6A). GO molecular functions such as helicase activity, nuclease activity, and Rho GTPase activity were also highly enriched (Additional file 1: Figure S6B-C).

Reactome Pathway Enrichment analysis revealed oncogenic MAPK signaling. After relaxing the q value at 0.1, pathways involved in DNA repair, p53 regulation of activity, and DNA double helix-modifying pathways were highlighted, as well as defects in the SUMOylation of DNA replication proteins, DNA damage response and repair proteins, cell cycle regulation by p53, resolution of D-loop structures through Holliday junction intermediates, and DNA repair (Additional file 1: Figure S6D).

Among cancer-associated pathways, the main KEGG-enriched pathways were the Fanconi anemia pathway (q value = 0.006), cell cycle (q = 0.01), prostate cancer (q = 0.02), chronic myeloid leukemia (q = 0.03), non-small cell lung cancer (q = 0.02), bladder cancer (q value = 0.02), and hepatocellular carcinoma (q = 0.02) (Fig. 6e). Non-homologous end joining (q = 0.04), platinum resistance (q = 0.04), base-excision-repair (q = 0.04), and mismatch-repair (q = 0.04) were also enriched in KEGG pathways. The prostate cancer, bladder cancer, non-small lung cancer, and non-cancer-related pathways revealed by KEGG enrichment analysis were mostly due to the high RAS/BRAF mutation rates. The hepatocellular carcinoma signature was also enriched by Wnt signaling mutations, while breast cancer signature displayed Notch, Wnt, and PI3K altered signaling.

Additional file 1: Figure S7 summarizes hits in the most dysregulated pathways. While pathway dysregulations were globally well balanced among the recurrent translocation subgroups, genes encoding helicases were more frequently encountered in t(11;14) cell lines (p = 0.006). On the other hand, intrinsic apoptosis mutants were more frequent in t(4;14), 66% vs 12% in non t(4;14) cell lines (p = 0.004).

Thirteen HMCLs displayed one or several mutations in the apoptotic pathway (extrinsic, intrinsic, and executive), which were heterozygous (except in LP1 that displayed a bi-allelic BCL2L11/BIM deletion) (Fig. 3). To assess the impact of mutations, we analyzed the cell death response through either the extrinsic, i.e., response to Fas/Trail-R agonist receptors (CH11, mapatumumab, or lexatumumab [9]) or the intrinsic pathway of apoptosis, i.e., response to BH3-mimetics (ABT-199, A-1210477, or A-1155463) [10, 11, 34, 35] (Additional file 1: Table S2). Pathway hit scores were calculated according to the number of hits in each pathway. No correlation could be drawn between the sensitivity to Trail-R agonists or Fas ligands and extrinsic apoptosis hits. Similarly, ABT-199/-737 responses did not correlate with intrinsic apoptosis hits. Moreover, intrinsic pathway hits were not associated with BH3-profiling, i.e., cytochrome C release in response to BIM peptide (Additional file 1: Table S2), confirming that the heterozygous mutations did not affect the upstream apoptotic responses (Additional file 1: Figure S8).

Eighty-two percent of HMCLs displayed at least one variant of the MAPK pathway, with 60% of HMCLs bearing a K-/H-/N-RAS variant (Fig. 3 and Additional file 1: Figure S9). FGFR3 mutations were present in 12% of HMCLs but in 50% of t(4;14) HMCLs with FGFR3 overexpression (KMS11, LP1, OMP2). Five HMCLs expressed a BRAF mutation with BCN displaying six non-silent mutations in the PKc-like domain (without evidence of a frameshift). BRAF mutations mostly occurred in the PKc-like domain (four of five mutants), outside of the V600 codon. Two samples displayed both NRAS and BRAF mutations (NAN10 and NAN3). As shown in Fig. 4a, RAS-mutated HMCLs displayed hypersensitivity to the MEK-1/2 inhibitor trametinib (Mann–Whitney test, p = 0.002), while the three FGFR3-mutated HMCLs did not display significant sensitivity. Two HMCLs with BRAF mutations (without concomitant RAS mutation) out of three displayed hypersensitivity to trametinib. Of note, the two BRAF^mut HMCLs displaying high sensitivity to trametinib had variants around V600. Finally, the MAPK pathway hit score was associated with an increased sensitivity to trametinib (β₁ = − 1.17, 95% CI = [− 1.41; − 0.61]), which was mainly related to RAS mutations (Fig. 4b).

Approximately half (55%) of HMCLs showed impaired cell cycle regulation, mostly bi-allelic deletion of the CDKN2C locus (33% of HMCLs), RB1 alterations (12%) or CDKN2A alterations (9%) (Additional file 1: Figures S3 & S5-“CellCycle”). Because deletions in CDKN2A or CDKN2C, by contrast to RB1 deletion/mutation, were reported to favor cell cycle inhibition by the CDK4/CDK6 inhibitor palbocicib, we assessed whether altered HMCLs were sensitive to the inhibitor [36]. Palbociclib induced an inhibition of cell cycle that was correlated to pRb inhibition (β = − 25, 95% CI = [− 33.8; − 2.9]) (Fig. 5a–c). However, no correlation could be found between palbociclib sensitivity and the CDKN2C (p = 0.17), CDKN2A (p = 0.47) or RB1 (p = 0.33) status (Mann–Whitney test, Fig. 5d). HMCLs overexpressing CCND1 with an unaltered RB1 showed a trend for better sensitivity than that of other HMCLs (p = 0.1) (Fig. 5e). Conversely, no correlation was found between palbociclib sensitivity and either the cell cycle pathway score (β = − 0.123, 95% CI = [− 0.89; 0.53]) or CDK4/CDK6 expression levels (β = 0.49, 95% CI = [− 1.47; 1.97] and β = − 0.02, 95% CI = [− 0.46; 0.5]), respectively).

The NFκB pathway was altered in 45% of HMCLs, mostly by inactivating TRAF3 (frame-shift, non-sense mutation, insertion or deletion) or BIRC2/BIRC3 (homozygous deletion) as previously reported in primary myeloma cells [20, 23]. Alterations in the NFκB pathway were associated with the overexpression of NFκB signature genes [37] (i.e., CD74, TNFAIP3, IL2RG, BIRC3, and PLEK), and we further identified that NFE2L3 (a downstream target of TNF-α signaling displaying a NFκB site in the promoter) was highly expressed in samples harboring NFκB pathway hits (Fig. 6). Furthermore, we found no correlations between NFκB pathway dysregulation and sensitivity to proteasome inhibitors (β = 0.2, 95% CI = [− 0.25; 0.65] for bortezomib and β = − 0.09, 95% CI = [− 0.5; 0.16] for carfilzomib, Additional file 1: Figure S8).

We previously reported that the sensitivity to alkylating drugs was impaired by p53 deficiency [12]. We further assessed whether hits in pathway(s) were associated with the response to drugs reported to be related to p53 deficiency. As shown in Fig. 7, p53 pathway alterations were associated with a lesser response to melphalan (β = 0.58, 95% CI = [0.08; –0.9]) and bendamustine (β = 0.63, 95% CI = [0.1; –1]). Of note, HMCLs harboring a high DNA damage kinase sensor score had high sensitivity to melphalan (β = − 0.83, 95% CI = [− 1.38; − 0.27]) and bendamustine (β = − 0.93, 95% CI = [− 1.52; − 0.28]), which was related to TP53 status.

On the other hand, hits in the p53 pathway were also associated with reduced sensitivity to lexatumumab (β = 0.71, 95% CI = [0.01; 1.95]) and with a trend for increased sensitivity to mapatumumab (β = − 0.79, 95% CI = [− 1.24; − 0.08]). These correlations were related to the direct and indirect p53-mediated regulation of TNFRSF10B and TNFRSF10A expression, as previously reported [9].

Correlations between other clinically commonly used drugs (Imids, dexamethasone, and proteasome inhibitors) and pathways were not significant (Additional file 1: Figure S8).

We assessed whether efficacy of DNA targeting drugs could be related to specific alterations. We analyzed sensitivity to RITA and CX-5461, which induce DNA crosslinking and stabilize DNA G-quadruplex, respectively, and are known to involve DNA repair during DNA replication [13, 38, 39]. Since p53 is involved in DNA repair, we analyzed drug responses according to TP53^status of HMCLs. Sensitivity to RITA in TP53^wt HMCLs was enhanced by mutations in helicases (β = − 0.59, 95% CI = [− 1.04; − 0.01]) (Fig. 7). HMCL sensitivity to CX-5461 was not associated with any genes or pathway alterations (Additional file 1: Figure S8) although XG11, which displayed a homologous BRCA2 mutation, showed the highest sensitivity to CX-5461.