Background
Human myeloma cell lines (HMCLs) are widely used for their representation of primary myeloma cells because they cover patient diversity, although not fully [
1]. HMCLs are mainly derived from refractory patients, mostly presenting with extramedullary disease and having thus received numerous classes of drugs inducing DNA damage, proteasome inhibition, immunomodulation, and anti-inflammation (e.g., melphalan, bendamustine, Velcade, Revlimid, and dexamethasone). However, HMCLs harbor the 14q32 abnormality, which occurs early at the MGUS stage, and display frequent mutations in
NRAS and
KRAS, as observed in patients at diagnosis (approximately 50% of patients) [
2,
3]. By contrast, HMCLs display very frequent deletion and mutation in the
TP53 gene that are associated with resistance to treatments [
4]. Indeed, it is well known that hits in the
TP53 gene (deletion and/or mutation) at diagnosis are associated with resistance and shortened survival and that their frequency increases with relapse [
4,
5]. Thus, HMCLs are a mixture of abnormalities occurring both early and late in the time course of disease. Besides hits in the
TP53 and
RAS genes, HMCLs have not been widely characterized for their global mutation profile and gene deletion. In the present work, using whole-exon sequencing (WES) in 33 HMCLs, we report common gene mutations and deletions. We analyzed the frequency of mutations/deletions in comparison with patients at diagnosis and relapse. We further identified hits preferentially associated with 14q32 translocations and analyzed responses to conventional and nonconventional drugs in relation to a mutation and/or deletion profile.
Methods
HMCLs and primary MM cells
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].
Whole-exon sequencing
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.
Functional assays
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.
Statistical analyses
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.
Discussion
WES was performed in 33 HMCLs, including 19 that had not been reported yet. HMCLs selected in this study were established between 1965 and 2015 in Europe, USA, or Japan and in the presence or absence of added recombinant IL6. While HMCLs’ ages spanned from ~ 55 to 3 years old, mutation load was similar among those, even if more recent HMCLs display lower mutation load. Our analysis showed that mutated genes were shared between HMCLs and primary myeloma cells, whatever the organ origin of samples that gave rise to the cell lines. Although HMCLs always emerged from patients with extra-medullary disease, no strong comparison could be made with primary or secondary PCL because of the very low number of sequenced PCLs yet, except for del17p (46% in sPCL) [
29]. We thus compared mutation frequency with primary cells at diagnosis and at relapse (without any indication of medullary or extramedullary disease). Although the number of HMCLs was low as compared with patients, we nevertheless identified genes with a very high tumor load suggesting that some of them are drivers. The HMCLs mutational landscape may thus provide a panorama of mutations in refractory patients. The frequency of mutated “myeloma” genes in HMCLs was identical, lower, or higher when compared to primary cells at diagnosis or relapse [
20‐
22]. While mutation rate in
KRAS was similar between HMCLs and primary myeloma cells, the frequencies of
TP53 (67%),
CDKN2C (33%),
PRKD2 (21%),
FAM46C (15%), and
BRAF (15%) dramatically increased compared to primary myeloma cells, either in DMM or RMM. The high
TP53 abnormality frequency (67%) in HMCLs identified by WES in our study (and confirmed by direct RT-PCR sequencing [
1]) was not in good agreement with a previous WES study reporting a rate of 21% in HMCLs [
40], which was highly underestimated: indeed, well-known
TP53 mutations in L-363, LP-1, and SKMM-2 (COSMIC database and
p53.iarc.fr, Release = 18) were not reported in this study and at least three “HMCLs” were not of myeloma origin (ARH77, MC-CAR, CTV-1) [
41,
42].
Our results clearly confirmed a major alteration in both proliferation control, with either loss of suppressor (
TP53,
CDKN2C,
RB1) or acquisition of activator (
BRAF,
RAS) and in tumor suppression/drug response (
TP53,
FAM46C), as in most if not all cancers [
43]. Because the loss of function of
TP53,
FAM46C, or
CDKN2C are not directly targetable, drugs bypassing these proteins or exploiting their loss consequences are required. Indeed, as shown in Fig.
5, cells lacking
CDKN2C expression were sensitive to the CDK4/6 inhibitor palbociclib, especially in the CCND1 group. This CCND1 impact was surprising because palbociclib is efficient against all CDK-CCND complexes, i.e., CDK4/CCND1, CDK4/CCND3, and CDK6/CCND2 [
44]. Of note,
CCND2 myeloma cells overexpress
CDK6 and
CDK4 while
CCND1 myeloma cells overexpress
CDK4 but not
CDK6, suggesting that CDK4 is “empty” of cyclin D in
CCND2 myeloma cells (Additional file
1: Figure S10). This free CDK4 pool might explain the low efficiency of palbociclib in
CCND2 HMCLs. Palbociclib has shown no global efficiency in MM patients without indication of their subgroup origin and their CCND1 expression [
45]. However, since it is efficient (in combination) in patients with tumors overexpressing CCND1 such as mantle cell lymphoma or HR
+ breast cancer, it might be of interest for patients with t(11;14) without Rb deficiency [
46,
47]. Concerning
TP53, we previously described p53 independent drugs, which were efficient whatever
TP53 status, such as PRIMA-1
Met that targets glutathione or BH3 mimetics that target anti-apoptotic proteins [
6,
11]. We also reported that loss of p53 function favors measles virus replication and cell death in myeloma cells [
48].
FAM46C was recently been shown to encode for a non-canonical poly(A) polymerase and its over expression in MM cells induced cell death [
49].
FAM46C is a type I IFN-stimulated gene, and it might modulate virus replication such as the yellow fever virus (YFV) and the Venezuelan equine encephalitis virus (VEEV) [
50]. Of note, anti-viral type I IFN pathway appeared highly impaired, suggesting defects in infection defense that might be exploited using oncolytic viruses such as measles virus [
48,
51].
Concerning mutations with gain of function such as RAS mutations, we showed that sensitivity of 27 HMCLs to MEK1/2 inhibitor trametinib was associated to RAS mutations (70% of “RAS only” mutated HMCLs were sensitive), but not to FGFR3 (none sensitive HMCL out of three with “FGFR3 only” mutation). Concerning BRAF, four HMCLs out of five with BRAF mutation (and with NRAS mutations for two of them) were sensitive but the low number of “BRAF only” mutated HMCLs prevented definitive conclusions. Although all HMCLs without hit in RAS/BRAF/FGFR3 genes were resistant to trametinib, all HMCLs with NRAS mutation were not sensitive since four NRAS mutated HMCLs were resistant. These data collectively suggest that mutation in RAS/BRAF genes is required but not sufficient for eliciting response to trametinib. The BRAF/RAS impact will be assessed in an ongoing clinical trial (NCT03091257) evaluating dabrafenib and/or trametinib in patients with relapsed and/or refractory multiple myeloma patients according to their BRAF/RAS mutation.
The high percentage of altered genes in DNA/chromatin repair/regulation, Fanconi pathway, and chromatin/DNA modification might be related to the frequency in relapsing patients [
32]. Because of the lack of specific drugs, we could not directly assess the functionality/vulnerability of these pathways, which require a deep investigation. Of note, mutations in Fanconi genes were recently reported in patients at relapse, suggesting that drug escape might involve this pathway. HCLMs exhibiting such “BRCAness” will be a good model for assessing efficiency of drugs like USP1 and/or PARP inhibitors [
23,
31,
52].
On the other hand, no major alteration was found in apoptosis pathway, either extrinsic/intrinsic or executive, showing that resistance to cell death was rather upstream of the mitochondria. In good agreement with the low number of alterations in apoptosis pathway, HMCLs were highly primed for death as shown by their BH3 profiling and their high response rate to BH3 mimetics [
11,
53] (Additional file
1: Table S2). Considering the huge difference between cell responses to DNA damaging drugs and BH3 mimetics, loss of response was not on the mitochondrial side, and BH3 mimetics appear thus of major interest to target MM cells whatever their genomic alterations or responses to classical myeloma drugs.
Conclusions
In summary, WES suggests that HMCLs harbor enriched mutations and defects in cell cycle, p53, recombination/DNA repair, NFκB, and epigenetic genes. Importantly, some very early pathogenic events such as IgH translocations and MAPK pathway mutants are stable over time and are not enriched by in vitro long-term culture, thus making HMCLs a reliable drug screening model for refractory patients at diagnosis or relapse. What is more, detection at diagnosis of mutations/deletions in genes associated with progression and HMCLs (i.e., CDKN2C, FAM46C, TRAF3, PRKD2) might identify particularly aggressive sub-clones warranting adapted treatment strategies and surveillance. WES results suggest that in addition to target apoptosis using BH3 mimetics and the antiviral deficiency using oncolytic viruses, targeting DNA damage, recombination/DNA repair, and epigenetic modifiers should be further investigated and might offer significant options for high-risk and refractory patients, including extramedullary diseases.