Prediction of outcome in newly diagnosed myeloma: a meta-analysis of the molecular profiles of 1905 trial patients

Robust establishment of survival in multiple myeloma (MM) and its relationship to recurrent genetic aberrations is required as outcomes are variable despite apparent similar staging. We assayed copy number alterations (CNA) and translocations in 1036 patients from the NCRI Myeloma XI trial and linked these to overall survival (OS) and progression-free survival. Through a meta-anlysis of these data with data from MRC Myeloma IX trial, totalling 1905 newly diagnosed MM patients (NDMM), we confirm the association of t(4;14), t(14;16), t(14;20), del(17p) and gain(1q21) with poor prognosis with hazard ratios (HRs) for OS of 1.60 (P=4.77 × 10−7), 1.74 (P=0.0005), 1.90 (P=0.0089), 2.10 (P=8.86 × 10−14) and 1.68 (P=2.18 × 10−14), respectively. Patients with ‘double-hit’ defined by co-occurrence of at least two adverse lesions have an especially poor prognosis with HRs for OS of 2.67 (P=8.13 × 10−27) for all patients and 3.19 (P=1.23 × 10−18) for intensively treated patients. Using comprehensive CNA and translocation profiling in Myeloma XI we also demonstrate a strong association between t(4;14) and BIRC2/BIRC3 deletion (P=8.7 × 10−15), including homozygous deletion. Finally, we define distinct sub-groups of hyperdiploid MM, with either gain(1q21) and CCND2 overexpression (P<0.0001) or gain(11q25) and CCND1 overexpression (P<0.0001). Profiling multiple genetic lesions can identify MM patients likely to relapse early allowing stratification of treatment.


INTRODUCTION
While survival for multiple myeloma (MM) has improved over the last decade with the introduction of immunomodulatory drugs and proteasome inhibitors most MM patients will still relapse. 1 Upfront identification of patients who are likely to relapse early offers the prospect of intervening pre-emptively to maintain remission. Furthermore, identifying tumor sub-groups with targetable molecular dependencies has the potential to inform on biologically driven therapy.
Myeloma cells are typified by recurrent chromosomal aberrations, a number of which have been variously associated with poor prognosis, notably t(4;14), t(14;16), t (14;20), deletion 17p and gain of 1q. 2 We and others have recently reported that the cooccurrence of multiple genetic lesions may have greater significance for predicting patient outcome than any single abnormality. 3,4 Since many of the molecular abnormalities in MM are only present at relatively low frequency, robustly establishing the impact of molecular sub-classes on prognosis is contingent on the analysis of large patient series that have been uniformly treated.
Here we report a meta-analysis of the relationship between genetic profile and prognosis in newly diagnosed MM (NDMM) using data from two UK multi-center phase III clinical trials, totalling 1905 patients. This dataset includes previously generated data on the MRC Myeloma IX trial and an expanded analysis of the NCRI Myeloma XI trial. In addition, we analysed molecular copy number profiling in 1036 Myeloma XI patients to identify subgroups with molecular addictions that could be therapeutically targetable. 5,6 MATERIALS AND METHODS Myeloma XI trial patients 1036 patients with NDMM enrolled in the UK NCRI Myeloma XI phase III trial were molecularly profiled. Trial characteristics are described in Supplementary Methods. At the time of analysis, the trial endpoints have not been published. Median follow-up was 36.0 months. The study was undertaken with written informed consent from patients and ethical approval was obtained from the Oxfordshire Research Ethics Committee (MREC 17/09/09, ISRCTN49407852).

Myeloma IX trial patients
Detailed characteristics and main outcomes of MRC Myeloma IX have been reported previously and summarised in Supplementary Methods. 7 The study was undertaken with written informed consent from patients and ethical approval was obtained from the MRC Leukaemia Data Monitoring and Ethics committee (MREC 02/08/95, ISRCTN68454111). For the present analysis we included data from 869 of the 1960 NDMM patients with available clinical and comprehensive cytogenetic data. 3 Median follow-up for this group was 72 months. 3,8 Accompanying gene expression and mapping array data have been previously published (GSE15695). 6,9,10 Samples For both trials myeloma cells from bone marrow aspirate samples were obtained at diagnosis and purified (495%) using immune-magnetic cell sorting (Miltenyi Biotec, Bergisch Gladbach, Germany). RNA and DNA were extracted using RNA/DNA mini kit or Allprep kits (QIAGEN) according to manufacturers' instructions.

Copy number and translocation detection
Technical details about fluorescence in situ profiling of Myeloma IX have been published previously. 11 Myeloma XI cases were centrally analysed using MLPA and qRT-PCR. The SALSA MLPA P425-B1 MM probemix (MRC Holland, Amsterdam, The Netherlands) was used as previously described. 12,13 The newly developed probemix X073-A1 was used to profile 1007 of the 1036 cases in an identical fashion (MLPA Probe Mix: Supplementary Table 1). Copy number at each locus was determined as described previously. 12,13 Multiplexed qRT-PCR was used to determine IGH translocation status using a translocation and cyclin D (TC)-classification based algorithm (Supplementary Methods), as previously described. 10 Statistical methods All statistical analyses were undertaken using R version 3.3 and the 'survival', 'rms', 'metafor', 'survC1', 'JAGS' and 'BayesMed' packages. 14 Progression-free survival (PFS) was defined as the time from the date of randomization to progression, according to IMWG criteria, or death from any cause. Overall survival (OS) was defined as the time from the date of randomization to death from any cause. Kaplan-Meier survival curves were generated and the homogeneity between groups was evaluated with the log-rank test. Cox regression analysis was used to estimate hazard ratios (HRs) and respective 95% confidence intervals (CI) and adjustment for variables was performed by multivariable analysis. Fixed effects metaanalysis was performed using individual patient data. Correlations between structural aberrations were analysed using Bayesian inference. A Bayes

Descriptive patient characteristics and structural aberrations
The clinical characteristics of the 1036 newly profiled Myeloma XI trial patients and the 869 Myeloma Trial IX patients are detailed in Table 1. Overall there were no significant differences between trial patients in terms of gender, age and proportion that had been in receipt of intensive/non-intensive therapy. Although the frequencies of the primary IGH translocations, del(17p), del(1p32), del(13q) and del(16q) in tumours were similar in Myeloma IX and XI trial patients, a higher proportion of Myeloma IX patients had hyperdiploidy (HRD), gain(1q) and del(22q) ( Table 1). Amongst Myeloma XI trial patients, homozygous deletion of CDKN2C (1p32), BIRC2/BIRC3 (11q22) and amplification of CKS1B (1q21) and MYC (8q24) were the commonest focal homozygous copy number changes, which were seen at similar frequencies to those previously reported (Table 1). 15 Relationship between cytogenetic aberrations and survival In both trial series, the archetypical high-risk lesions del(17p), gain (1q) and t(4;14) were each significantly associated with shorter PFS and OS (Table 2). In the combined analysis, respective HR for OS were 2.1 for del(17p) (P = 8.86 × 10 − 14 ), 1.68 for gain(1q) (P = 2.18 × 10 − 14 ) and 1.60 for t(4;14) (P = 4.77 × 10 − 7 ; Table 2; Supplementary Figures 1 and 2). In addition, the t(14;16) and t (14;20) translocations involving MAF and MAFB were also associated with shorter OS with respective HRs of 1.74 (P = 0.0005) and HR 1.90 (P = 0.0089). Respective inference C-statistic estimates for adequacy of risk prediction are shown in Supplementary Tables 5 and 6. Deletion of 1p32 (CDKN2C) was significantly associated with shorter OS (HR 1.46; P = 0.0002; Table 2). This association was confined to patients in receipt of intensive treatment (in the combined analysis: HR 1.89; P = 1.23 × 10 − 5 vs HR 1.05; P = 0.72 for non-intensive treatment). The association of del(1p32) with OS was independent from gain(1q21) by multivariable analysis (P o 0.05) in the intensive treatment groups of both trials.
To examine the relationship between 1q21 status and outcome in more detail we sub-classified Myeloma XI patients (n = 1036) by diploid vs gain vs amplification status. 1q21 gain was confirmed as a high-risk lesion and was associated with significantly shorter PFS (HR 1.56; P = 3.53 × 10 − 7 ) and OS (HR 1.67; P = 3.30 × 10 − 5 ) than  Table 2; Supplementary Figures 1 and 2). Similarly to Myeloma IX, the 'triple-hit' of an adverse translocation, Gain(1q) and del(17p) was associated with a very short median OS of 19 months with a HR of 6.23 (P = 1.31 × 10 − 7 ) vs no adverse lesion (Supplementary Figure 5). In both Myeloma IX and XI trials the impact of a 'double-hit' on patient outcome was independent of International Staging System (ISS; Supplementary Table 3). Moreover, integration of ISS and genetic risk defined 'double-hit'-ISS ultra high risk (ISS II or III and 'double-hit'; 12.0%), intermediate risk (ISS I and 'double-hit'; ISS II and 1 adverse lesion; ISS III and no or 1 adverse lesion; 44.1%) and favourable risk groups (ISS I and no or 1 adverse lesion; ISS II and no adverse lesion; 43.9%). 'double hit'-ISS ultra high risk was associated with HR 2.85 (P = 8.32 × 10 − 31 ) for PFS and HR 4.12 (P = 2.85 × 10 − 36 ) for OS in the meta-analysis ( Table 2).
Genetic markers and survival in intensively treated patients Since young and fit patients are most likely to be considered for intensified combination therapy, we subsequently focused on the relationship between molecular profile and survival of this subgroup of Myeloma XI (n = 598) and Myeloma IX (n = 511) patients.
On the basis of clinical and genetic information (Supplementary Table 4) the 'double-hit'-ISS ultra high-risk group comprising 12.5% of patients were associated with a HR of 3.11 (P = 1.59 × 10 − 20 ) for PFS and HR 4.79 (P = 5.10 × 10 − 23 ) for OS.

Associations of copy number changes with translocations and targetable lesions
We next focused on genetic sub-groups of MM that could be specifically targetable using copy number and translocation data on the 1036 Myeloma XI patients. Figure 2 provides an overview of correlations between CNA and translocations (Supplementary Table 2). Of particular note was a relationship between NFκBpathway CNA and translocation groups.

Molecular sub-classification of hyperdiploid myeloma
We noted heterogeneity within the HRD subgroup in terms of cooccurrence of lesions. Although HRD as a whole group was strongly correlated with gain(11q25) (BF = 1.2 × 10 − 66 ), a subgroup lacking gain(11q25) was characterised by gain(1q) (Figure 3; Supplementary Figure 6).

DISCUSSION
Our analysis confirms the association with outcome in MM for the archetypical high-risk lesions del(17p), gain(1q) and adverse translocations and emphasises the importance of 'double-hit' as a risk biomarker. Importantly, we demonstrate that this information can be combined with the ISS to further refine risk prediction. 3,[18][19][20][21] To our knowledge, this study represents the largest analysis investigating the additive effect of multiple genetic lesions on outcome in NDMM. Importantly, our analysis has been based on trials that recruited between 2003 and 2016, a timeframe during which treatment for MM has undergone significant change. 22 The consistent adverse impact of high-risk genetics on survival in Myeloma IX and XI is striking and highlights the need for intensified efforts to target the biology of high-risk disease. Although survival time increased for all risk groups in Myeloma XI vs IX, absolute improvement was smallest for the 'double-hit' high-risk group. Median PFS for 'double-hit' in Myeloma XI patients receiving intensive treatment was 19.7 months, meaning that about half of patients relapsed 12 months following autologous transplant.
Comprehensive assessment of the inter-relationship of CNAs and translocations in the Myeloma XI trial led to characterisation of genetic sub-groups with putative therapeutic relevance. We found that half of Myeloma XI tumors carried a deletion of NFκBpathway genes, and 10% of tumors had two co-occuring deletions. 23 [26][27][28] Virtually all BIRC2/3 deletions were found in FGFR3-positive tumors. They were mutually exclusive of TRAF3 deletions, which were present in FGFR3negative tumors, a pattern which may indicate convergent evolution. Deletions of FGFR3, which often occur as loss of der14 that includes TRAF3, may constitute 'collateral damage' of NIK addiction in t(4;14). 15,24,25,29,30 Although t(4;14) and del(17p) were not correlated with each other, both groups were strongly associated with hypodiploidyassociated lesions del(12p), del(13q) and del(22q). 16,17,31,32 This suggests the consequences of t(4;14) and del(17p) may share molecular mechanisms. Gain of chromosome 1q21 was strongly associated with t(4;14), but not with del(17p). Gain(1q21) was confirmed as a high-risk lesion that is independent of del (1p32). [33][34][35] HRD MM constitutes the largest genetic sub-group of patients, with substantial heterogeneity. 17 We describe two sub-groups of HRD with either gain(11q25) and CCND1 biology or gain(1q21) and CCND2 overexpression. These groups are similar to the D1 and D2 sub-groups of the TC classification, which pioneered biologic classification of HRD MM. Application of the TC classification in routine diagnostics has unfortunately been restricted due to access limitations to array-based gene expression profiling. 17 Pragmatic classification of HRD based on gain(11q25) and gain(1q) may facilitate sub-grouping in clinical practice and open opportunities for improving therapy for these patients. Recently, activity of bcl-2 inhibitors has been reported in CCND1-driven t(11;14) MM, and CCND1-driven gain(11q25)-HRD may constitute another target group. 36 We also found a high frequency of del (13q) in gain(1q)-HRD, in contrast to gain(11q25)-HRD. Interestingly, del(13q) and gain(1q) also frequently co-occur in t(4;14), suggesting similarities in the genetic sequelae of these pathogenetic groups. 37 An inter-relationship between del(13q) and gain (1q)-HRD was suggested based on GEP in the TC classification by Bergsagel et al., but has been demonstrated here for the first time on a DNA level. 17,38 Moreover, HRD MM without any of risk lesions  . HRD genetic sub-groups in Myeloma XI. (a) Each row represents one of in total 1007 cases. Expression intensity is coded in green for CCND1 and red for CCND2 expression. Gain of 11q25 is shown in dark green, gain of 1q in dark red and deletion 13q in dark blue; white = no abnormality detected. B+C. CCND1 (b) and CCND2 (c) qRT-PCR expression levels (relative quantitative RQ values, GAPDH normalised) for HRD cases with gain(1q), gain(1q)+gain(11q25), gain(11q25) or neither. Gene expression levels were significantly different for all possible groupwise comparisons (two-sided Mann-Whitney U test; ****P o0.0001; ***P o0.001). gain(1q)-HRD, del(17p) and del(1p32) had longer remissions and survival than any other sub-group and may be sufficiently treated with single-novel agent/immunomodulatory drug-based approaches, potentially reducing additional side effects and costs of novel agent combinations. [39][40][41][42] In summary, we demonstrate the utility of profiling multiple molecular genetic lesions to identify patients most likely to benefit from molecularly targeted therapies. The molecular tools used for profiling Myeloma XI are readily applicable within diagnostic settings and should therefore help implementing stratified treatment approaches as part of routine patient care.