Targeted sequencing of refractory myeloma reveals a high incidence of mutations in CRBN and Ras pathway genes

Significant efforts have recently been made to better understand the genetic landscape of multiple myeloma (MM), with more than 1000 MM exomes sequenced to date.^1-5  However, most of the data have been obtained from untreated or heterogeneously treated cohorts. Recurrent mutations have been identified only in a limited number of genes, a subset of which seems to affect survival.^1,6,7  Patterns of clonal evolution were identified; however, these findings are still based on a relatively small number of intensively longitudinally studied patients.^2,8-10 

Although the novel agents, namely immunomodulatory drugs (IMiDs) and proteasome inhibitors (PIs) paved the road toward a prolonged survival and durable responses, most MM patients eventually relapse and become drug refractory. The mechanisms underlying resistance to IMiDs, PIs, and corticosteroids, however, remain poorly understood. Cereblon (CRBN) was recently identified as being essential for the anti-MM activity of IMiDs.^11-18  To date, the reported incidence of mutations in proteasome subunits, the CRBN pathway, or the steroid receptor is low, and mutations in these genes, except in single case studies^19,20  or in vitro cell line analyses,^20,21  have not yet been identified as a major source of primary or acquired drug resistance in larger MM data sets.

The acquisition or the selection of mutants under the selective pressure of targeted therapy has been shown be a key driver of the development of drug resistance in other hematologic malignancies (eg, BCR-ABL mutations in imatinib-treated chronic myeloid leucemia²²  or BTK mutations in chronic lymphocytic leukemia patients treated with the selective inhibitor ibrutinib).²³  Of note, genomic data from large series of homogeneously pretreated patients with refractory MM (rMM) are still missing.

Therefore, in this study, 88 frequently mutated or drug-resistance pathway MM genes with a focus on genes relevant to IMiD and PI interactions were investigated in a cohort of 50 heavily pretreated rMM patients by using our updated disease-specific Multiple Myeloma Mutation Panel (M³P v3.0). Our approach yielded rapid, robust, and sensitive variant screening and allowed for the detection of small subclonal mutations. Strikingly, we were able to identify mutations in potentially targetable genes and in genes related to drug resistance in the majority of patients.

Between November 2013 and April 2015, samples from 50 rMM patients were collected at the University Hospital Heidelberg (Heidelberg, Germany). Samples were obtained from bone marrow aspirates in 44 of the patients. In 6 patients with extramedullary disease, tumor DNA was obtained from tissue block (n = 2; patients #1 and #7) and from peripheral blood (n = 4; secondary plasma cell leukemia; #2, #3, #16, and #17). In all but one patient (#30), the corresponding germ line was available for analysis, and in 3 patients with acquired IMiD resistance at tumor sampling (#12, #24, and #32), prior samples from earlier IMiD-sensitive disease were available for comparative analysis. Patients in our study had received a minimum of 2 previous lines of therapy and had been exposed to at least one PI and one IMiD. All patients gave written informed consent to the sampling, analysis of tumor and germ line probes, and collection of clinical data in accordance with the Declaration of Helsinki. Data collection and analysis were approved by the institutional ethics committee (Heidelberg #229/2003, S-152/2010).

Clinical data were collected retrospectively from internal digital and external paper patient records. Refractory disease was defined according to International Myeloma Working Group criteria²⁴  as either progressive disease (PD) under therapy or within 60 days after discontinuation of therapy.

Mononuclear cells were isolated by density centrifugation (Biocoll Separating Solution; Biochrom GmbH, Berlin, Germany). Magnetic cell separation of CD138⁺ and CD138^– subsets from the mononuclear cells was performed by using Microbeads CD138-Human and Manual MACS Cell Separation Units (Miltenyi Biotec, Bergisch Gladbach, Germany). Sample purity was assessed by flow cytometry to be >85%. The CD138^– subsets were further cultured to obtain the stroma cells. Sample purity of stroma cells was assessed by flow cytometry to contain <1% CD138⁺ cells. Sample purity of tissue blocks was estimated by the infiltration rate of clonal plasma cells assessed by immunohistochemistry for CD138. DNA from isolated cells was extracted by using the QIAamp DNA Mini Kit (QIAGEN GmbH, Hilden, Germany). DNA extraction from tissue blocks was performed at the Institute of Pathology (Heidelberg, Germany) as previously described.²⁵  Germ line controls for sequencing studies were obtained from stroma cells, as described above, or from buccal swabs, which were processed by using the blackPREP Swab DNA Kit (Analytikjena; Biometra, Berlin, Germany) and stored at −20°C.

Targeted sequencing was performed on the personal genome machine (Thermo-Fisher, Waltham, MA) by using 20 ng DNA from flow-sorted CD138⁺ plasma cells as well as from corresponding germ line samples. IonOneTouch automated systems were used to generate and enrich 400 bp libraries. Our 185-kb customized M³P v3.0 includes 1327 amplicons covering the coding regions of 88 disease-specific genes (supplemental Table 1, available on the Blood Web site). We selected key genes in pathways previously identified as being important in MM (MAPK, nuclear factor κB, interleukin-6, cell cycle, MYC) as well as genes known to be expressed in MM and being recurrently mutated in >2% of patients in previously published genome-wide MM studies. Furthermore, we selected genes that provide a potential drug target and genes that have been associated with drug resistance to the most commonly used drug families (IMiDs, PIs, and corticosteroids). Three paired tumor and germ line samples were barcoded by using the Ion Xpress Barcode Adapters, pooled, and sequenced on a 318TM chip by using the Ion PGM Hi-Q Seq Kit (Thermo-Fisher). An average of ∼800× depth sequencing coverage was generated per nucleotide. Mutation analysis was performed by using Ion Reporter Software v4.4 (Thermo-Fisher), and further data curation was performed by using the Biological Reference Repository (BioR, Mayo Clinic, Rochester, MN).²⁶  BioR includes gene annotations that use various database resources, including National Center for Biotechnology Information/Ensembl; University of California, Santa Cruz (UCSC); Single Nucleotide Polymorphism Database (dbSNP); HapMap; 1000 Genomes Project; the Catalogue of Somatic Mutations in Cancer (COSMIC; Wellcome Trust Sanger Institute); SIFT (J. Craig Venter Institute)²⁷ ; and PolyPhen-2 (Harvard Medical School).²⁸  Germ line variants were filtered out by comparing each tumor sample with its germ line counterpart. We then filtered synonymous variants and variants identified outside exonic and splicing site regions and visualized the remaining calls by using the Integrative Genomics Viewer (Broad Institute, Cambridge, MA). Finally, we eliminated from further analyses the variants that were either supported with unidirectional reads, covered with <20× depth, or found in <3% of reads in the tumor sample. For the sample for which we had no germ line sample available, we applied the same filters; however, we included only those findings that were listed in the COSMIC database and excluded those that were present in dbSNP or in 1000 Genomes Project with a prevalence of more than 1%.

For the comparative analysis of longitudinal samples, we used a specific 400 bp gene panel containing a reduced number of 24 amplicons including CRBN and batched 3 CD183 tumor samples on a 318 chip. This allowed us to ensure the maximum sensitivity of the personal genome machine technology (∼10⁻³). All sequencing data are deposited in the Sequence Read Archive database (http://www.ncbi.nlm.nih.gov/sra); study accession number is SRP066263.

To establish mutant CRBN-expressing OCI-MY5 cell lines, we created lentivector pCDHPuroCRBN by subcloning wild-type (wt) CRBN from pCDHGFPCRBN and used this pCDHPuroCRBN as a template to introduce CRBN mutations by polymerase chain reaction (see supplemental Data for details). The ligated plasmid was amplified and isolated by using a Miniprep Plasmid Isolating Kit (QIAGEN) and then cloned into lentivector pCDH-CMV-MCS-EF1-Puro (SBI, Mountain View, CA) and packaged into a lentivirus. The lentivirus constructs were then used to infect OCI-MY5 myeloma cells to overexpress mutant CRBN protein. Cell viability was assessed in triplicate by using 3-(4,5-dimethylthiazol-2-yl)-2,5-dimethyltetrazolium bromide (MTT) after a 5-day incubation time with increasing doses of lenalidomide.

Baseline characteristics of the patient cohort are described in Table 1. Cytogenetic results were available in 39 patients (78%) and included gain 1q21 (>2 copies), deletion 17p, and translocations t(4;14) and t(14;16) in 62%, 33%, 13%, and 8% of screened patients. Overall, 82% of patients carried at least 1 adverse cytogenetic aberration (Figure 1). Patients received a median of 5 prior therapies before tumor sampling (range, 2 to 15 prior therapies). All patients were exposed to at least 1 IMiD and 1 PI, and 88%, 78%, and 68% of patients were refractory to at least 1 IMiD, PI, or PI/IMiD, respectively. At time of sampling, 82% of patients had PD and 78% were refractory to the last line of treatment. The median time between PD and sampling was 12 days (range, 0 to 164 days). Previous therapies and refractory status immediately before sampling are described in Table 1 and displayed in Figure 1.

In our cohort of 50 patients, we detected 136 mutations, resulting in an average mutation incidence of 2.7 per patient (range, 0 to 13 mutations), of which there were 114 missense (84%), 9 nonsense (7%), 1 stop lost (1%), 11 frameshift deletions (8%), and 1 frameshift insertion (1%), and 6 were located in splice donor or acceptor sites (4%). We found mutations in at least 1 of the selected M³P genes in 98% of patients, and at least 1 mutation was identified in 47 (53%) of the 88 investigated genes (Figure 1). Multiple mutations within the same gene were observed in 5 patients, 1 of whom harbored 4 mutations within the steroid receptor NR3C1 (variant reads [VRs], 7% to 17%). Other genes harboring multiple mutations in a single patient included BRAF (VRs, 54%/41%), CUL4B (VRs, 19%/6% and 5%/5%), NFKB2 (VRs, 26%/6%), and TP53 (VRs, 18%/8% and 14%/8%) (supplemental Table 2). Poly-Phen2,²⁹  SIFT,²⁷  and PROVEAN³⁰  were applied to estimate the functional impact of each mutation. In total, predictions for 115 mutations were generated and in 93% of cases a damaging, potentially damaging, or probably damaging impact was assessed by at least 1 of the algorithms. The mutational spectrum observed was broad, with a significant number of small subclonal mutations; VRs did not reach 10% in more than a quarter (26%) of all mutations, and almost half (45%) of the mutations identified were below a VR of 25% (supplemental Table 2).

The incidence of mutations in the RAS pathway in our refractory cohort was quite high, with 36 patients (72%) having at least 1 mutation in NRAS, KRAS, and/or BRAF, significantly exceeding previously published frequencies in newly diagnosed patients (Figure 1).^1-4  We observed 9 BRAF-mutated patients (18%), with 4 (8%) having the mutation at known activating and druggable site p.Val600Glu.^31,32  In 4 patients, kinase-inhibiting BRAF mutations were identified, known to paradoxically activate the MAPK pathway via c-RAF (Gly466Val,³³  Gly469Arg,³⁴  and Asp594Asn³⁵  [present in 2 patients]). Two patients showed additional activating mutations in NRAS and KRAS, respectively, and 1 patient harbored 2 BRAF mutations (Lys499Asn and Glu695Lys) (supplemental Table 2). Finally, 2 patients (4%) with RASA2 mutations and 1 patient with an activating mutation in PTPN11 were identified; both of the genes had potentially activating effects on the MAPK pathway.^36,37 

We saw an increased incidence of TP53 mutations in our patient cohort (26%), more commonly found in patients with del17p (8 [62%] of 13) compared with patients without the deletion (4 [15%] of 26) (Figure 1). Two patients harbored 2 mutations within the gene. The majority of mutations were found within subclonal ranges (9 of 15 mutations: 16% VRs or less; supplemental Table 2).

Of note, the cereblon pathway, important for the anti-MM action of IMiDs,^12-14,17  was found mutated in almost one-quarter of patients (22%). This includes CRBN (12%), CUL4B (6%), IRF4 (4%), and IKZF1 (2%) (Figure 1; supplemental Figure 1). All CRBN mutations found in our cohort were detected in patients with IMiD-refractory disease (patients #5, #12, #24, #32, #34, and #38; Figure 1) as was the case for mutant CUL4B (#6, #15, and #38) and IKZF1 (#2). The mutation found in IKZF1 (p.Ala152Thr) is located inside a sequence of 30 amino acids known from a previous study to be essential for lenalidomide sensitivity.¹⁷  All CRBN mutations occurred at critical sites, potentially having an impact on CRBN-IMiD interactions by either truncating the protein (p.Ile393Metfs*10 [#5], p.Pro241Argfs*10 [#38], p.Gln327* [#12], and splicing acceptor c.551-2T>C [#34]; supplemental Table 2) or by being located within the thalidomide-binding domain (p.Phe381Cys [#24] and p.Pro411His [#32]) as illustrated in Figure 2.^38,39 

Furthermore, the cumulative time on IMiD treatment among the 6 CRBN-mutant patients (median, 1146 days; range, 154 to 1912 days) was longer compared with CRBN wt patients (median, 479 days; range, 40 to 2182 days; P = .02) .

In 3 CRBN-mutated patients with acquired resistance to IMiDs (#12, #24, and #32), earlier tumor samples collected before refractory lenalidomide or pomalidomide treatment were available (Figure 3; supplemental Figure 2). In all 3 patients, the mutation was undetectable at the earlier time point, even with an increased sequencing depth.

To assess the impact of these 3 CRBN mutations on protein function and sensitivity to IMiD treatment, the mutations were introduced into the myeloma cell line OCI-MY5. OCI-MY5 is characterized by a very low expression of wt CRBN, rendering the cell line IMiD resistant. When wt CRBN is introduced and overexpressed, OCI-MY5 regains IMiD sensitivity.¹²  We introduced and overexpressed all 3 CRBN mutations into OCI-MY5 in independent experiments, as well as the wt CRBN (positive control) and the viral vector alone (negative control). Cell viability was assessed by using an MTT assay (in triplicate) after a 6-day treatment with increasing doses of lenalidomide (Figure 4). The OCI-MY5 wt cells regained lenalidomide sensitivity, whereas the OCI-MY5 cells containing the CRBN mutants or the viral vector alone remained lenalidomide resistant. The expression of exogenous wt and mutated CRBN was detected by an immunoblotting assay using an anti-CRBN antibody.

Only limited data are currently available regarding mutations in proteasomal subunits associated with resistance to PI-based therapies in MM. Thus, only a limited number of genes coding for proteasomal subunits were selected for our current M³P design (supplemental Table 1). Mutations of XBP1 or PI subunits (PSMB8 and PSMD1) were found in 2 patients (4%) in our cohort, with 1 plasma cell leukemia patient carrying a mutation in both PSMB8 and PSMD1. The response of the latter patient to PIs was unavailable. However, another patient with a p.L314Ffs*38 XBP1 had known PI resistance at the time of tumor sampling.

Besides BRAF, other mutations in potentially actionable genes were identified in a subset of patients, including IDH1/2, and FGFR3 (Figure 1). IDH mutations are frequently found in glioma⁴⁰  or acute myeloid leukemia.⁴¹  An IDH1-activating mutation at amino acid position 132 (p.Arg132His) found in 1 patient provides a druggable target, provided that it is of functional relevance in MM, and specific inhibitors for activating IDH1 mutations are in advanced clinical testing.^42,43  FGFR3 has been identified as a druggable target, and we identified an activating p.Arg248Cys mutation^44,45  and an activating read-through stop lost mutation *809Cys⁴⁶  in our cohort. Various specific FGFR3-inhibiting agents are in clinical and preclinical investigation in MM. The most promising results have been obtained in solid tumors.^47,48  Furthermore, 1 FGFR3 mutation is present in a t(4;14) patient (#49), whereas the translocation status is unknown in the second patient (#40). However, whether the sequelae conferred by the translocation in general or whether specific FGFR3 mutations are essential for successful anti-FGFR3 treatment remains to be confirmed.

Despite the fact that more than 1000 MM genomes have already been sequenced, the mutational landscape of drug-refractory MM remains largely unknown, and our understanding of drug resistance mechanisms to the most commonly used drug families is limited. We have therefore designed and established a disease-specific MM gene panel (M³P) for rapid and sensitive variant screening in key MM genes. Our panel includes a subset of disease-specific key MM genes that we used to screen a selected cohort of 50 drug-refractory MM patients. We provide evidence that M³P sequencing can generate valuable information within potentially meaningful timeframes for clinical decision-making in the vast majority of rMM patients.

Mutations in potentially targetable genes (MAPK, FGFR3, and IDH1) were detectable in 76% of patients, whereas mutations in genes affecting the prognosis as described in previous studies of newly diagnosed MM (TP53, EGR1, and IRF4)¹  were identified in 32% of patients. Mutations in genes potentially associated with drug response (CRBN pathway, XBP1, proteasome subunits) were identified in 24% of patients.

MAPK mutations provide a broad potential target for advanced MM patients. BRAF or MEK inhibitors are already in clinical testing and seem promising.^31,32,49,50 NRAS and KRAS mutations occurred mostly within known activating hotspots of the gene, and 4 of 9 BRAF mutations were druggable p.Val600Glu. However, BRAF mutations with paradoxical activation of the pathway via c-raf were also found in 4 patients, and in 1 of them (patient #41), 2 simultaneous BRAF mutations were identified: 1 at amino acid position 499 that has been suggested to be ERK activating and has been described in cardiofaciocutaneous syndrome,⁵¹  and 1 in p.Glu695Lys, located in the protein kinase domain within the activator loop, previously reported in skin cutaneous melanoma.⁵²  In addition, a Gly503Arg mutation in PTPN11, another gene of the RAS pathway that has been reported as being frequently mutated in RAS-opathies (juvenile myelomonocytic leukemia, adult acute myelogenous leukemia, gastric cancer, glioblastoma, and anaplastic large cell lymphoma) was identified.^36,37,53  This gene is thought to play a role in acquired resistance to targeted therapy⁵⁴  and therefore represents a potential drug target. PTPN11/SHP2 inhibitors have already been identified, and compounds are in preclinical testing.⁵⁵  Furthermore, RASA2 was found to be mutated in 2 patients. Mutations in this gene have been described in various malignancies, including colorectal, skin, lung, and endometrial cancers; however, the biological function of the gene has not yet been fully deciphered, and it is hypothesized to be a RAS inhibitor.⁵⁶  Of note, 1 of the patients (#1) with a Val600Glu BRAF mutation harbored an additional nonsense mutation in RASA2, which may contribute to further MAPK activation in this patient.

The most remarkable finding of our study is the identification of direct mutations in CRBN in 12% of patients and mutations in the CRBN pathway in almost one-quarter (22%) of the cohort. Although a role of mutant IRF4 in IMiD resistance remains to be established, CRBN is the central mediator of IMiD action, and its expression is essential for the anti-MM effects of IMiDs.^12-16  Notably, all CRBN-mutated patients (n = 6) and 10 (91%) of 11 of the CRBN pathway–mutated patients were unresponsive to IMiD-based treatment at the time of tumor sampling (the IMiD-refractory status of 1 patient was unknown) suggesting an association of these mutations with IMiD response. Thus far, CRBN mutations are largely absent in the publically available MM sequencing data that, of note, include mostly untreated patients. For example, within 511 of the most recently analyzed patients in the Multiple Myeloma Research Foundation (MMRF) CoMMpass trial, only 1 CRBN mutation (p.Pro85Ser) was identified (https://research.themmrf.org/). Similarly, in 349 untreated M³P-sequenced patients at the Mayo Clinic, we identified only 1 CRBN mutation (p.Asn316Lys) located in close proximity to the IMiD-binding site of the gene. Of interest, this patient did not respond to either lenalidomide induction or subsequent pomalidomide treatment, suggesting a potential impact of that mutation on the patient’s IMiD response. Although primary and secondary resistance to IMiDs is observed in MM patients, mutations in this gene or pathway as the underlying cause for IMiD resistance have not yet been described in the disease apart from a single case report of a highly drug-resistant patient.¹⁹  The increased incidence of CRBN and CRBN pathway–associated mutations in our cohort might be explained by cohort selection. In addition, the targeted sequencing approach allowed us to screen with increased sensitivity for variants and, in fact, 3 of the 6 CRBN mutations in our cohort were identified with VRs below 10% of the reads. In these 3 patients with subclonal CRBN mutations, we had earlier tumor samples available. All 3 patients had responded to IMiD treatment after earlier sampling (Figure 3) and even with increased sequencing sensitivity, we were unable to detect any CRBN mutations in the earlier, IMiD-sensitive phase of their disease. In addition, our data on the exposure to IMiDs before the acquisition of CRBN mutations indicated that the development of CRBN mutations might be correlated with long-term exposure to IMiDs, but this has to be taken with great caution because of the limited number of patients in this analysis. Regrettably, no prior samples were available from the other 3 patients with mutant CRBN, and larger studies on sequential assessments in patients under IMiD treatment are needed.

To further investigate the functional effect of these mutations on the response to lenalidomide, we introduced them into OCI-MY5 cells and confirmed that all 3 mutations caused in vitro IMiD resistance. These findings support a causal relationship between the appearance of these mutations and the lack of response to IMiDs in the patients. Tumor heterogeneity and the impact of subclonal somatic mutations on the response to targeted cancer therapies have been frequently discussed in relation to various cancers.⁵⁷  Of interest, in a recent publication, there was evidence that low-frequency mutations in KRAS may cause resistance to anti-epidermal growth factor receptor therapy in colorectal cancer, even if mutant allele frequency was as low as 0.4% to 17%,⁵⁸  and similar reports were published for gefitinib resistance in lung cancer.^59,60  The authors discuss potential explanations, including amplification of the gene locus and the possibility of parallel evolution of distinct resistance mechanisms within different subclones of a tumor. Whether such explanations might hold true in rMM remains to be addressed.

Including the mutation described above, we have detected 15 CRBN mutations in 9 MM patients thus far by M³P sequencing at the Mayo Clinic, all of the patients being clinically resistant to IMiD treatment.^7,19  This includes an additional patient who was initially responsive to thalidomide and lenalidomide treatment, who acquired resistance to lenalidomide over the disease course, and who was finally unresponsive to pomalidomide treatment at the time of tumor sampling. In this patient, we detected a total of 6 mutations within CRBN: (p.Lys401Glu, 40% VR; p.His397Tyr, 6% VR; p.Pro352Ser, 2% VR; p.Ile314Ser, 11% VR; p.Leu308Phe, 5% VR; and p.Leu190Phe, 7% VR). Of note, all mutations but 1 were located within the thalidomide-binding domain, critical for the response to IMiDs. We had 2 prior samples available from earlier time points and, as with the 3 patients described above, we were unable to detect the mutations in CRBN when the patient was still sensitive to IMiD treatment.

In summary, our study supports the hypothesis that CRBN mutations predominantly, but not exclusively, occur in advanced MM. These CRBN mutations are probably evoked by clonal selection under prolonged IMiD-based therapy, and their appearance probably contributes to IMiD-resistant MM disease. Clearly, the subclonal appearance of CRBN mutations in some patients requires further investigation. In addition, our study provides evidence of an altered mutational landscape in drug-resistant MM that presents most significantly with an increased number of mutations in the MAPK pathway. Whether these MAPK mutations are associated with resistance to IMiD treatment or are, more likely, positively selected by repeated drug exposures in advanced disease, remains to be determined.

This study was supported by research funding from Oncospire (A.K.S.), by Grant No. KO 4604/1-2 [705 136/807 015] from the Deutsche Forschungsgemeinschaft (K.M.K.), by a grant within the Max-Eder Program of German Cancer Aid (Deutsche Krebshilfe) (M.S.R.), and by a research grant from Novartis GmbH (M.S.R.).

Contribution: A.K.S., E.B., E.K.M., K.M.K., M.S.R., and P.L.B. designed the research; A.J., C.-X.S., J.X., K.M.K., L.B., M.A., N.H.H., P.J., and Y.-X.Z. performed experiments; A.K.S., E.B., E.K.M., K.M.K., M.S.R., N.H.H., R.A.S., and S.B. analyzed data; A.K.S., E.B., E.K.M., K.M.K., and M.S.R. wrote the manuscript; and E.K.M., M.M., J.H., and H.G. recruited patients and provided clinical information.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: A. Keith Stewart, Mayo Clinic in Arizona, 13400 E. Shea Blvd, CRB Room 1-001, Scottsdale, AZ 85259; e-mail: stewart.keith@mayo.edu.