Prognostic gene expression analysis in a retrospective, multinational cohort of 155 multiple myeloma patients treated outside clinical trials

Abstract Objectives Typically, prognostic capability of gene expression profiling (GEP) is studied in the context of clinical trials, for which 50%‐80% of patients are not eligible, possibly limiting the generalizability of findings to routine practice. Here, we evaluate GEP analysis outside clinical trials, aiming to improve clinical risk assessment of multiple myeloma (MM) patients. Methods A total of 155 bone marrow samples from MM patients were collected from which RNA was analyzed by microarray. Sixteen previously developed GEP‐based markers were evaluated, combined with survival data, and studied using Cox proportional hazard regression. Results Gene expression profiling‐based markers SKY92 and the PR‐cluster were shown to be independent prognostic factors for survival, with hazard ratios and 95% confidence interval of 3.6 [2.0‐6.8] (P < .001) and 5.8 [2.7‐12.7] (P < .01) for overall survival (OS). A multivariate model proved only SKY92 and the PR‐cluster to be independent prognostic factors compared to cytogenetic high‐risk patients, the International Staging System (ISS), and revised ISS. A substantial number of high‐risk individuals could be further identified when SKY92 was added to the cytogenetic, ISS, or R‐ISS. In the cytogenetic standard‐risk group, ISS I/II, and R‐ISS I/II, 13%, 23%, and 23% of patients with adverse survivals were identified. Conclusions For the first time, this study confirmed the prognostic value of GEP markers outside clinical trials. Conventional prognostic models to define high‐risk MM are improved by the incorporation of GEP markers.


| INTRODUC TI ON
Multiple myeloma (MM) is a heterogeneous hematologic malignancy which is considered incurable. [1][2][3] Despite the significantly improved survival over the last decades, treatment responses remain diverse and not all patients benefit equally. This is especially the case for the 15%-25% of high-risk patients that suffer from rapid progression and poor survival. Various definitions of high-risk disease are described in literature using different molecular and/or clinical variables, thereby typically identifying different patient subsets. Although most highrisk definitions have been confirmed to correlate with poor outcome, their generalizability toward a "real-world" scenario is often ambiguous, since 50%-80% of the MM patients do not meet the strict trial eligibility criteria. [4][5][6] For this reason, the question arises if the risk stratification markers developed in clinical trials settings can also be validated in MM patients treated outside trials.
Prognostication in MM is subject to change. In current practice, cytogenetic aberrations measured by fluorescence in situ hybridization (FISH) and the international staging system (ISS) are most common for MM prognostication. 7,8 Patients with at least one of the aberrations del(17p), t(4;14), or t(14;16) are commonly classified as cytogenetic (CA) high-risk. 8,9 However, other descriptions are used as well. 10 More recently, the revised ISS (R-ISS) has been introduced to estimate risk of MM patients by extending the ISS with CA highrisk and serum lactate dehydrogenase (LDH). 11 There are, however, still a fair number of poor-performing patients that are classified into low-or standard-risk groups and vice versa.
Markers based on gene expression profiling (GEP) have shown promising results to predict clinical outcome. For example, SKY92-a biomarker that classifies MM patients into high-or standard-risk based on the expression of 92 genes-has shown to correlate significantly with survival in multiple trial data sets. [12][13][14][15][16][17] This association turns out to be additive to conventional markers, because in combination with ISS or R-ISS, SKY92 risk classification is further improved. 13,[18][19][20] However, it has remained an open question whether these associations generalize to patients treated outside clinical trials.
In this retrospective, multinational study outside of clinical MM trials, we show that GEP-based markers provide an accurate prognostic distinction between risk groups that better reflects survival, as compared to ISS, R-ISS, and cytogenetics. Moreover, by combining the SKY92 classifier with conventional markers, clinical risk assessment of the MM patient can be improved.

| Cohort composition
Patients in this study were part of the Horizon 2020 funded

| Plasma cell purification and RNA isolation
Plasma cells were purified from bone marrow aspirates using CD138+- Hilden) according to the manufacturer's instructions. RNA concentration was measured using the NanoDrop spectrophotometer (Thermo Fisher Scientific), and quality and purity were assessed by the RNA 6000 assay (Agilent 2100 Bioanalyzer; Agilent Technologies). A minimum of 100 ng total RNA was used as test input.

| Gene expression profiling
Gene expression profiling data were generated at a central laboratory (SkylineDx). RNA processing, target labeling, and hybridization to gene expression arrays were performed on the Human Genome

| Statistical analysis
Univariate and multivariate survival analyses were performed by the Cox proportional hazard model provided that the proportional hazard assumption was met based on weighted residuals ("survival" package v3.1-8 in R-3.6.0). Hazard ratio estimates were expressed relative to the lowest-risk group and assessed by a two-sided Wald test. Analysis of deviance was performed with the "stats" package (v3.6.3). P-values below .05 were considered to be significant through this study.

| GEP-based biomarkers are prognostic for MM patients treated outside of clinical trials
The 155 MM patients, treated outside of clinical trial setting, had a median age of 66 years ( Table 1). Most of them received an immunomodulatory drugs (IMiD; 25%), a proteasome inhibitor (PI; 48%), or a combination of both (15%). No correlation was found between treatment and survival (Table S3, overview of treatment analysis is summarized in Supporting Information).
Survival outcomes (median follow up of 31 months, using censored data only) did not correlate with either one of the three sites of inclusion ( Figure S1). Compared to FISH, GEP-based t(4;14), t (11;14), and t(14;16)/t(14;20) had both a positive and negative percent agreement (PPA, NPA) above 80% (Table S1) corroborating a previous study 27 and therefore were considered equivalent in cases for which FISH was not performed or not available (Table 1).
SKY92 identified 23% of patients as high-risk with a hazard ratio for OS (HR os ) and PFS (HR pfs ), and corresponding 95% confidence  Figure 1D,H).
In the subsequent multivariate analysis-using all significant univariate markers as input (SKY92, PR-cluster, and CA high-risk; SKY92 standard-risk ( Figure S2). In the following analyses, we proceed with only SKY92 because it identified a larger proportion of individuals and it overlapped with the majority of PR-cluster patients.

| GEP-based classifiers identify patients overlooked by other markers
In an analysis of variance, it was shown that adding SKY92 to the Since it was shown that SKY92 is useful when added to other markers, we extended the CA high-risk group by the addition of SKY92 and followed the previously proposed stratifications of combining SKY92 with ISS or R-ISS ( Figure S4). 13  Note: Significant codes: **P < .01, *P < .05.
Abbreviation: NA, Not available; OS, Overall survival; PFS, Progression free survival. a CTA-and NP-clusters both only consist of one sample (Table 1)

| D ISCUSS I ON
In the current study, we demonstrated that gene expression-based analysis adds value to the prognostication of MM patients treated outside clinical trials. In the multivariate analysis, the SKY92 risk classifier and PR-cluster remained independently associated with survival, in contrast to frequently used clinical risk indicators (ISS, R-ISS, and CA).
By incorporating GEP-based classifiers, additional high-risk patients were detected that were not identified by other markers. These highrisk patients had associated adverse outcomes and could potentially be referred to clinical trials focusing on this subpopulation 28 or could qualify for better monitoring and a more intense treatment.
Multiple studies have made it apparent that data from randomized controlled studies may not always reflect results for patients undergoing treatment in routine clinical practice. 5,6,29 Due to a selection of younger and fit patients in clinical studies-reflecting the strict eligibility criteria-the MM population treated outside of clinical trials is enriched with comorbidities, inadequate organ function, and a lower performance status, which is frequently translated into poorer survival outcomes. 30 In our cohort, the median PFS was 31 months for all combined treatments. This is longer compared to real-world registries from Denmark and The Netherlands, who reported 18 months 31,32 and contrasted with recent phase 3 clinical data in which a median PFS over 40 months was described in the newly diagnosed setting. [33][34][35] From the list of evaluated biomarkers, the association of SKY92 high-risk and PR-positive patients with poor OS and PFS had the strongest prognostic effect, demonstrating their accuracy and reliability in this cohort reflecting a "real-world" scenario. (or co-regulated) genes might be more informative. In addition, a single gene classifier is likely to target multiple prognostic mechanisms.
It has been shown that the 92 genes in the SKY92 are enriched for the long arm of chromosome 1. 12 A total of nine genes are located on 1q ( Figure S6), of which S100A6 has been described in relation to 1q21 amplification in MM and other cancer types. 36 (Table S4) nor registry errors, we assume this is a correct result, which might be biased due to the lack of information on a subset of subjects and/or the lower observed numbers in ISS I.
Because it is unavoidable that some patients with adverse survival will be classified into the SKY92 standard-risk group and/or patients with favorable survival in the SKY92 high-risk group, it proved useful to combine multiple prognostic markers to strengthen their prognostic power. Analysis of variance has shown that adding SKY92 to any of CA, ISS, or R-ISS significantly improved the stratification for OS, while only addition of ISS to a model with SKY92 was significant, confirming one of our previous findings. 18 By doing so, we have observed that cytogenetics with SKY92 could identify more high-risk patients, and with a larger hazard ratio for OS compared to the single markers alone. Especially in situations where ISS was weakly associated with prognosis-like in this cohort for OS stratification-SKY92 helped to significantly distinct the stratification of MM patients in a three-level classification. Moreover, the majority overlapping R-ISS stages II and III were also discriminated when adding SKY92, resulting in more distinguished intermediateand high-risk groups.
To conclude, this multinational study in MM patients treated outside clinical trials recapitulates GEP-based information as strongly prognostic for both OS and PFS. In addition, we found that the prog-

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.