Plasma cell development-based prediction of prognosis and treatment response in individual multiple myeloma

Abstract

We hypothesized that gene networks converged between plasma cell development and multiple myeloma (MM) pathogenesis may better explain the clinical heterogeneity in individual MM patients. Here, we transformed the reciprocal expression pattern between a gene module centered on the expression of MCL1 and another gene module represented by the expression of TLR10 from individual MM transcriptomes into a myeloma classification score (MCS). MCS enabled prognostication in 1738 patients with newly diagnosed MM (NDMM) and 319 patients with relapsed/refractory MM (RRMM) from our own observational cohort, from MMRF, and from five prospective trials. MCS also enabled the discrimination between progressive and stable disease states in the longitudinal samples from 85 patients in MMRF and the PETHEMA/GEM2012MENOS65 trial. MCS further enabled prediction of the responses to bortezomib-based treatment in 425 patients enrolled in the HOVON-65/GMMG-HD4 and APEX (039) trials, and the response to carfilzomib-based treatment in 86 transplantation-eligible patients from MMRF. MCS-defined risk groups harbored distinct targetable pathways, including unfolded protein response, replication stress, genomic instability and interferon responses. However, MCS-based individualized prediction of prognosis and treatment response was independent of the hitherto available prognostic models. Thus, MCS represents an alternative biomarker for risk stratification and prediction of treatment response in individual MM patients.

Keywords

multiple myeloma / individualized molecular subtyping / risk stratification / treatment response / proteasome inhibitor

1 Introduction

Overall survival (OS) of patients with multiple myelomas (MM) ranges from less than two years to more than ten years [1,2]. Much effort has been expended in the development of risk stratification models, which are expected to improve treatment. A revised International Staging System (R-ISS) and the updated R2-ISS models have been developed to stratify newly diagnosed MM (NDMM) patients into three or four risk groups at the population level [1,2]. These risk groups are based on the levels of β2-microglobulin and lactate dehydrogenase (LDH/LDHA) in serum, and on the presence of high-risk chromosomal abnormalities (CA), including 1q+, del(17p), and translocations of t(4;14) or t(14;16). However, survival variabilities within the risk groups are unexplained, and high-risk factors do not necessarily predict responses to therapy [3–6]. Up to 20%–25% of MM patients lack detectable CA [7], and hence their risk groups could not be precisely assessed. Moreover, whether these risk models are applicable throughout the MM disease course is insufficiently tested.

The introduction of immunomodulatory derivatives (IMiD) and proteasome inhibitors, together with autologous stem cell transplantation (ASCT), has generally improved the OS of patients with NDMM [8]. However, treatment responses vary substantially among individual patients, and less than 50% of treated patients benefit from these treatments [9–12]. Without robust tools to predict treatment response in individual patients, patients with naturally good prognosis could be wrongly perceived as if they would have benefited from the treatment. Moreover, bortezomib- or IMiD-based treatment regimens are associated with considerable adverse effects [13,14]. Tools simultaneously stratifying risk and predicting treatment response in individual patients are urgently needed.

Gene expression profiles (GEP) of CD138⁺ MM cells can be used to classify MMs into molecular subtypes. Eight GEP subtypes of NDMMs were initially proposed in the Translocation/Cyclin D (TC) classification scheme [15]. Using unbiased approaches, further GEP studies identified seven to 12 GEP subtypes of NDMMs [16–18]. Prognosis-guided transcriptome analyses have led to the development of the GEP-70 [19] and EMC-92 [6] signatures. Moreover, gene sets potentially differentiating patients with NDMM who may or may not benefit from bortezomib- or IMiD-based treatments have been proposed [10,20–22]. The Individualized Risk Model for Myeloma (IRMMa) is a recent extension of the TC classification scheme that utilizes information on genomic alterations to assign NDMM patients into 12 groups [23]. These studies are instrumental toward establishing a molecular classification scheme that is both prognostic and predictive for MM. However, questions regarding the generality of their application and capacity in individualized prediction of treatment response encourage alternative strategies.

MM classification should be anchored in MM pathogenesis [24]. Without adequate anchorage in MM pathogenesis, classification schemes may still correlate prognosis with patient groups but will fail to assign individual MM cases into subtypes driven by distinct pathogenic processes, and thus being incapable of predicting treatment response. We explored an approach that expression pattern of gene module(s) centered on key signaling pathway(s) conserved between plasma cell development and MM genesis may enable molecular classification of MM. Based on disease network approach [25] and with the key regulators of germinal center (GC) formation [26,27] as the seed genes, we constructed gene co-expression modules in MM transcriptomes [28]. We identified an MCL1 module (MCL1-M) containing 87 genes consistently co-expressed with MCL1 in MM transcriptomes [28]. Besides crucial roles of MCL1 in long-term survival of plasma cells and MM cells [29,30], MCL1-M members are also co-expressed in GC B cells [28]. Additional MCL1-M members are involved in MM aggressiveness (LDH/LDHA), response to the bone marrow microenvironment (IL6R), osteoclast formation (ANXA2), proteasome activity (PSMB4, PSMD4), and regulation of chromatin structure (ANP32E, H3F3A) [28]. The pattern of MCL1-M expression is complemented by the expression of a reciprocally expressed module of 46 genes represented by TLR10, hereafter named the TLR10 module (TLR10-M). High expression of MCL1-M and low expression of TLR10-M are associated with the signature of pre-plasmablasts, while low expression of the MCL1-M and high expression of TLR10-M is associated with the aberrant expression of B cell genes [28].

In this study, we developed a single sample prediction (SSP) algorithm that transforms the pattern of reciprocal expression between MCL1-M and TLR10-M in the transcriptome of CD138⁺ MM cells into a myeloma classification score (MCS). We assessed the capacity of the MCS from individual MM samples in (1) prognostication throughout the disease course, (2) discrimination between stable disease (SD) and progressive disease (PD) states during treatment, (3) prediction of response to bortezomib- or carfilzomib-based treatment, and (4) prediction of targetable pathways. We demonstrate the potential of MCS in individualized prognostication and prediction of response to bortezomib-based treatment.

2 Materials and methods

2.1 Identification of the classifiers and molecular clusters

MM classifiers and molecular clusters used in this study have been reported previously [28]. Using transcriptome data of the NDMM samples from total therapy (TT) 2 and 3 trials (GSE2658) [14,31], we computationally screened gene co-expression modules centered on key regulators of GC formation. This screening identified the MCL1-M (87 genes) as the only gene module that enabled clustering of MM samples into subtypes with both prognostic and predictive features. The MM clusters were defined based on the reciprocal expression pattern between MCL1-M and its anti-correlated gene set, named TLR10-M (46 genes) (Fig. 1A and 1B).

2.2 Development of SSP algorithm

Using the transcriptomes of NDMM samples from the TT2 and TT3 trials [14,31], we refined our initially reported classifier genes [28] to 75 MCL1-M members and 22 TLR10-M members (Table S1).

Data of all samples from TT2/TT3 (n = 559) and Multiple Myeloma Research Foundation (MMRF) (n = 840) were used as the training cohorts for constructing an SSP algorithm for microarray or RNA-seq platform (Table 1). Translating our cohort-level clustering scheme [28] into the SSP algorithm involved three main steps. We first used unsupervised consensus clustering to generate MCL1-M or TLR10-M subtype labels for all samples (Fig. 1C). These subtype labels were then used along with the classifier’s GEP data to train a support vector machine (SVM)-based subtyping model, which generates the MCS for all samples (Fig. 1D). Finally, we constructed a multiple linear regression-based SSP algorithm using the normalized classifier’s GEP as the explanatory variable and the MCS as the response variable (Fig. 1E). The SSP algorithm generated the MCS for individual samples from the validation datasets. These MCS values enabled individualized assignment of samples into subgroups with significantly enriched expression of MCL1-M, and hence a low value for the MCS, or with significantly enriched expression of TLR10-M, and hence a high value for the MCS. Intermediate expression of both modules yields intermediate values for the MCS (Fig. S1). The analytical robustness of the SVM model and SSP algorithm was validated in eight independent datasets containing 1404 patients (Table 1; Figs. S2–S4).

2.3 Examination of the clinical utilities of the MCS

We assessed the MCS as both a categorical and continuous predictor of survival in patients with NDMM (n = 1814) from the observational cohorts of our own (Chaoyang Hospital dataset, Table S2) and MMRF, and from the TT2/TT3 and HOVON-65/GMMG-HD4 trial cohorts (Table S3) [32]. The MCS was also assessed as a categorical predictor of survival for patients with RRMM (n = 319) from the APEX and TT6 trials (Table S3) [11,33]. Using data from all paired longitudinal samples from 45 patients from MMRF and 40 patients from the PETHEMA/GEM2012MENOS65 [34] trial, we assessed the capacity of the MCS to predict disease states during treatment. Using the data from the HOVON-65/GMMG-HD4 and from the APEX (039) trials (n = 592) [11,32], we assessed the capacity of the MCS to predict responses to bortezomib-based treatments. Using the data from MMRF, we assessed the capacity of the MCS to predict responses to carfilzomib-based treatments. Using the data from MMRF and the HOVON-65/GMMG-HD4 trial, we compared MCS’s prognostication and predictive capacities with other risk stratification and GEP models [1,2,6,7,19–21,23].

The complete workflow for developing the SSP algorithm, survival analysis, gene set enrichment analysis and genomic analysis are described in Supplementary Methods.

3 Results

3.1 Individualized prognostication in patients with NDMM or RRMM

In the observational cohorts of our own (Chaoyang Hospital) and MMRF (Tables 1 and S2), NDMM patients with low MCS values showed poorer OS compared with patients with high MCS values (Fig. 2A–2C; Table S4). This result held, regardless of whether the patients had received ASCT treatment. Notably, among the MMRF patients who received ASCT treatment (n = 390), patients with high MCS values showed a 5-year OS of 93.2% (95% confidence interval (CI) 86.4%–100.0%), while those with low MCS values showed a 5-year OS of 62.9% (95% CI 48.3%–81.9%) and those with intermediate values had a 5-year OS of 77.5% (95% CI 65.9%–91.1%). These differences are highly significant (log-rank test, P = 0.0001) (Fig. 2B; Table S4).

We next determined MCS values for data from published trials. These were the TT2 trial that incorporated thalidomide (GSE2658) [14], the TT3 trial (GSE136400) that incorporated bortezomib together with thalidomide or lenalidomide [31,35], and the HOVON-65/GMMG-HD4 trial (GSE19784) that incorporated bortezomib [32] (Tables 1 and S3). Once again, NDMM patients with low MCS values consistently showed a poorer PFS and OS than those with high MCS values (Fig. 2D–2H; Table S4). In the APEX trial (GSE9782) [11], which incorporated bortezomib treatment without ASCT, and in the TT6 trial (GSE57317) [33], which incorporated bortezomib and lenalidomide treatment with ASCT, RRMM patients with low MCS values showed poorer OS compared with those with higher MCS values (log-rank test, P = 0.003 and P = 0.0001, respectively) (Fig. 2I and 2J; Table S4).

With the exception of patients who received ASCT in the MMRF dataset and patients in the HOVON-65/GMMG-HD4 trial, patients with NDMM and intermediate MCS had OS values similar to that of patients with high MCS, in the other datasets examined (Table S4). These results from 2622 patients indicate that the individual MCS values enabled assignment of 22.6% of NDMM patients to the high-risk subgroup, and 24.3% to the low-risk subgroup. The remaining 53.1% NDMM were categorized as medium-risk, and their OS was closer to that of the low-risk subgroup. The same trend applied to the data from the RRMM patients.

Using an additive Cox proportional hazard model, we assessed the relationship between the MCS and the 2-year and 3-year OS. In the combined data of the TT2, TT3a/TT3b and HOVON-65/GMMG-HD4 trials (n = 894), we observed a steep decline of the 2-year and 3-year OS that parallels the decline of MCS in the high-risk subgroup (Fig. 3A). The steep MCS-correlated decline of the 2-year and 3-year OS in the high-risk subgroup was also evident in the MMRF dataset, in both ASCT and non-ASCT subsets (Fig. 3B and 3C). We note that patients without ASCT treatment were significantly older than patients with ASCT treatment (Fig. S5). Among the high-risk subgroup, the 2-year OS values of patients with or without ASCT treatment were 0.80 (95% CI 0.70–0.91) vs. 0.57 (95% CI 0.54–0.70) (hazard ratio (HR) 0.30, 95% CI 0.16–0.56; P = 0.00030); the 3-year OS values were 0.69 (95% CI 0.56–0.84) vs. 0.54 (95% CI 0.43–0.68) (HR 0.38, 95% CI 0.21–0.70; P = 0.0016). These findings suggest that MCS-defined high-risk NDMM in non-ASCT-eligible patients comprise an ultra-high-risk patient subgroup.

MCS-based individualized prognostication is independent of the prognosis associated CAs and other genomic alterations (see below, Fig. 6). However, high-risk NDMMs harbored markers of genomic instability, including high mutational burden and widespread loss of heterozygosity (LOH). In addition, there was an increased expression of the CDC20 co-expression module (CDC20-M), which is a surrogate marker of replication stress and genomic instability [36] (Fig. 3D–3F). High-risk NDMMs were also enriched in a signature of unfolded protein response (UPR) (Fig. 3G; Fig. S6). Furthermore, MCS-based high-risk NDMMs harbored an APOBEC mutational signature (SBS2) (Fig. 3H). Moreover, cell proliferation-related signaling pathways (e.g., E2F targets, G2M checkpoints, and mTORC1 signaling) were enriched among the high-risk NDMMs, while pathways of interferon-\alphaand interferon-\gamma responses were enriched in low-risk NDMMs (Fig. S7). In addition, transcriptomes of the bone marrow samples from the low-risk NDMMs were enriched in a T cell-inflamed tumor microenvironment signature, which has been suggested as a pan-tumor biomarker for PD-1 blockade-based immunotherapy (Fig. S8) [37,38].

Overall, these findings demonstrate that MCS is a robust survival predictor for patients with NDMM or RRMM under different treatment regimens. MCS-defined risk groups are associated with distinct targetable pathways.

3.2 Prediction of disease states during treatment

Integrated analyses of the longitudinal samples from 45 patients in the MMRF dataset indicate that the disease states of MM patients under treatment could be predicted by changes in the MCS (∆MCS = MCS in the follow-up sample − MCS in the paired baseline sample; or ∆MCS between the consecutive follow-up samples) (Fig. 4A). Patients with a ∆MCS ≥ −2 had a longer OS compared with patients with a ∆MCS < −2 (median OS: not reached vs. 2.42 years; HR 0.36, 95% CI 0.15–0.87; P = 0.029) (Fig. 4B). These two subgroups did not correlate with age or with ASCT treatment (Fig. S9), nor with risk-associated chromosomal translocations (Figs. 4C, 4E, S10, and S11; Tables S5 and S6). In general, the karyotype in the baseline samples was largely maintained in those samples with a ∆MCS ≥ −2; 9 of the 17 stable cases harbored 1q gain in baseline sample (Figs. 4C, 4D and S10; Table S5). Follow-up samples with a ∆MCS < −2 acquired additional copy number variations (CNV), most of which were subclonal CNVs; among the 15 cases with karyotype results, 1q gain was not involved in 5 cases, and increase in 1q copy number in the follow-up samples was not detected in 4 cases with 1q gain in the baseline sample (Figs. 4E, 4F, and S11; Table S6). For follow-up samples with a ∆MCS < −2 (Fig. 4G and 4H), there was an elevation in the CDC20-M expression level. Finally, the ∆MCS-based prediction of disease state was replicated using data from 40 patients in the phase 3 PETHEMA/GEM2012MENOS65 trial (Tables 1 and S4) [34]. Here, MCS from the measurable residual disease (MRD) samples after the induction treatment were compared with the MCS from the corresponding diagnostic samples. The ∆MCS was significantly higher in patients with complete remission (CR) compared to patients with partial response (PR) or with PD; however, 1q+ was not associated with disease states (Fig. 4I). MMRF CoMMpass network has recently identified that progression in a small fraction of patients is associated with a transition from the non-proliferation to the proliferation subtype [18]. Among the 20 MCS-defined progressive cases, 5 cases showed a transition to the proliferation subtype during progression and 3 cases remained as the proliferation subtype throughout the disease course. Transition to proliferation subtype was observed in 1 of the 19 MCS-defined stable cases.

These findings together indicate that, independent of 1q+ and risk-associated chromosomal translocations, ∆MCS between the follow-up and baseline samples or between the consecutive follow-up samples potentially enables discrimination between SD and PD states during treatment.

3.3 Prediction of response to bortezomib- or carfilzomib-based treatment in individual patients

The predictive capacity of the MCS with respect to the response to bortezomib was first analyzed in the data of phase 3, HOVON-65/GMMG-HD4 trial, in which NDMM patients were randomly assigned to bortezomib-based PAD (bortezomib, doxorubicin, dexamethasone) or non-bortezomib-based VAD (vincristine, doxorubicin, dexamethasone) induction arms, followed by ASCT and maintenance treatment [32]. Among NDMM patients with low MCS values (the MCL1-M^high subtype), the median PFS was 26.6 months in the PAD arm, and 18.4 months in the VAD arm (HR 0.61, 95% CI 0.41–0.92; P = 0.015); the median OS was 69.3 months in the PAD arm, and 36.8 months in the VAD arm (HR 0.63, 95% CI 0.4–1.00; P = 0.045). Among NDMM patients with high MCS values (the TLR10-M^high subtype), the PAD and VAD treatment arms showed indistinguishable PFS and OS (Figs. 5A and S12). Analyses of the 2-, 3-, and 5-year OS as a continuous function of MCS, showed that patients with the lowest MCS derived the most benefit from the PAD regimen. For example, patients with an MCS of −10 showed a 29.1% improvement in the 5-year OS. This improvement was lost in patients with an MCS > −1.3 (Fig. 5B).

We then examined data from the phase 3 APEX trial (039), in which patients with RRMM were randomly assigned to bortezomib or dexamethasone treatment arms [9,11]. Among RRMM patients with low MCS values, the median PFS was 5.1 months in the bortezomib arm and 2.9 months in the dexamethasone arm (HR 0.44, 95% CI 0.21–0.90; P = 0.0060, log-rank test); among RRMM patients with high MCS values, no survival differences were observed between these two treatment arms (Figs. 5A and S12).

Using MMRF dataset, we also examined the OS of the NDMM patients treated with carfilzomib-containing regimens as the first-line therapy. Among ASCT-treated patients, the OS of high-risk patients treated with the carfilzomib-containing regimens was improved to a level similar to the OS levels seen in the low-risk or medium-risk patients (Fig. 5C). However, these findings were not observed among the patients without ASCT treatment (Fig. 5D), which are consistent with the trial results showing that carfilzomib-containing regimens could not improve the survival of NDMM patients who are ASCT-ineligible or without intent for immediate ASCT [39,40].

These findings together show that the MCS enabled prediction of response and the extent of response to bortezomib- or carfilzomib-based treatment in individual MM patients.

3.4 Comparison to other schemes

Among the NDMM patients with R-ISS or R2-ISS of the HOVON-65/GMMG-HD4 trial, OS of the MCS low subtype was significantly poorer compared with the MCS high subtype (Fig. S13). Adjusted for the variables of R-ISS or the top features of R2-ISS [1,2], multivariate COX regression analyses show that the MCS was an independent prognostic factor only among the patients treated with the VAD regimen (Tables S7 and S8). This is in line with the subtype-specific benefit to bortezomib-containing treatment shown in Fig. 5A and 5B; under the PAD regimen, the OS difference between the patients with high-risk MM and the patients with low-risk MM was markedly smaller compared to the OS difference for patients treated under the VAD regimen. 1q+ and del(17p) were also only prognostic among the patients treated with the VAD regimen (Table S8). Analyses of the MMRF cohort show that at the cohort level, abnormalities of 13q genes, 1q gain/amplification, risk-associated translocations, recurrent RAS mutations, and HRD were enriched in high-risk or low-risk subgroups. However, these genetic abnormalities did not allow assignment of individual MM cases into MCS-based risk subgroups, because medium-risk NDMMs also harbor these abnormalities (Fig. 6A; Fig. S14). For example, 28.7% of the high-risk NDMMs did not harbor gain/amplification of 1q21; however, medium-risk NDMMs also harbored gain/amplification of 1q21, though at lower frequencies (Fig. 6A; Table S9). The same trend was observed in the GSE2658, GSE19784, and Chaoyang Hospital datasets (Fig. S14). NDMM patients harboring a t (4; 14) translocation, 13q loss, 1q amplification or gain, MYC gain/translocation, or the proliferation subtype proposed by the MMRF CoMMpass Network [18] could be further stratified according to their MCS-based risk subgroups (Fig. 6B–6G). A similar trend was observed for NDMM patients harboring mutations in KRAS, NRAS, or abnormalities in TP53 (data not shown). Among the ASCT-treated patients from MMRF cohort, MCS was an independent prognostic predictor, whereas 1q21 gain/amplification was not significantly associated with prognosis (Table S10). Thus, MCS-defined risk subgroups are independent of the available risk stratification models [1,2].

At the individual sample level, recently recognized genetic driver alterations [41] were largely not specifically involved in MCS-defined risk subgroups (Fig. S15 and S16). However, combined analyses of single nucleotide variation and CNV data [41] show that the RAS pathway was affected in 38.3% of MCS-defined high-risk and 89.7% of low-risk NDMMs. These involved mutations in KRAS, NRAS, FGFR3, PTPN11, NF1, RASA2, or PIK3CA with variant allele frequencies (VAF) ranging from 20% to 50% and gain or amplification of RRAS2. Mutually exclusive mutations of KRAS and NRAS were only observed in the low-risk NDMM cases (Fig. S16 and S17). Furthermore, co-occurring inactivation of genes located at 13q (DIS3, RB1, and TGDS) was seen in 70.2%–77.7% of MCS-defined high-risk NDMM cases (Table S9). Thus, the RAS pathway is preferentially involved in the MCS defined low-risk subgroup and 13q loss-associated pathways in the high-risk subgroup.

Gene classifiers of the MCS did not significantly overlap with genes defining EMC-92 [6], GEP-70 [19], GEP-80 [21] or IMiD-14 [20] signatures (Table S11). Not surprisingly, therefore, patients assigned to the MCS-based risk subgroups were different from those assigned to the risk groups defined by the EMC-92 [6], GEP-70 [19], GEP-80 [21], or IMiD-14 [20] signatures (Table S12). Results of C-index analysis with the data of HOVON-65/GMMG-HD4 trial show that MCS is not superior to ISS, R-ISS, R2-ISS and the prognostic signature in predicting prognosis (Table S13); however, no survival differences were observed between the PAD and VAD regimens within the risk groups defined by the prognostic signatures (Fig. S18). Notably, for ASCT treated patients in MMRF (n = 384), C-index of MCS reached 0.72 (95% CI 0.63–0.81), while C-index of ISS was 0.62 (95% CI 0.53–0.70) (Table S13).

Overall, these findings indicate that the MCS provides an alternative means for MM prognostication.

4 Discussion

We here developed a SSP algorithm which transforms the patterns of reciprocal expression between MCL1-M and TLR10-M in transcriptomes of CD138⁺ plasma cells from MM patients into a MCS. This score enables individualized prognostication for patients with NDMM or RRMM treated with various regimens. Comparison of the MCS between the consecutive time points predicts stable or progressive disease state during treatment. The MCS also predicts the responsiveness and the extent of responsiveness, of individual MM patients to bortezomib-based treatment. Furthermore, MCS-defined risk groups are enriched with different targetable pathways.

The MCS is independent of hitherto developed risk stratification and molecular classification schemes, and differs from the hitherto developed prognostic or predictive models in the following respects:

First, unlike previous risk stratification and molecular classification models [1,2,5,7,10,23,42,43], we did not use the data of prognosis and genomic alterations in the training phase of our SSP algorithm. Compared to the currently used cohort-level prognostic factors including R-ISS stages, chromosome translocations, and 1q+ , our findings show that MCS provides an alternative means in assigning individual patients at different disease stages into distinct risk subgroups.

Second, the patterns of reciprocal expression between MCL1-M and TLR10-M in individual transcriptome, not the expression levels of MCL1-M and TLR10-M per se, are transformed into the MCS; the extent of reciprocal expression between these two modules was used to define the cutoffs for MCS-based risk subgroups (Fig. S1). These features enable the MCS to provide a robust prognostic output for individual NDMM or RRMM patients from seven independent cohorts. In addition, our previous studies show that MCS-based MM molecular subtypes resemble distinct stages of plasma cell development [28], it can therefore be speculated that MCS may enable prognostication in individual patients treated with regimens not analyzed in this study.

Third, the determination of the MCS at consecutive time points provides a prediction of disease progression: a stable MCS indicated a SD state, whereas a significant decline in the MCS indicates a progressive worsening of the disease state. Thus, changes in the MCS in the follow-up samples from individual patients may provide an alternative means for early detection of relapse and facilitate the adjustment of treatment regimens. In contrast, 1q+ is insufficient to discriminate between disease states in individual patients.

Fourth, unlike prognostication-oriented comparator classification models [1,2,6,7,10,15–20,22,23,44], the MCS is predictive for the response to bortezomib-based treatment in individual patients with NDMM or RRMM. Benefit from bortezomib-based treatment is restricted to patients with MCS-defined high-risk MM. With declining MCS (and increasing risk) in NDMM, increasing benefit to bortezomib-based treatment appeared to be maintained throughout the treatment period. In patients with medium- or low-risk NDMM (accounting for 60% of NDMM cases), bortezomib-based treatment was not associated with an improvement of survival. Bortezomib is an WHO essential medicine, despite the rapid evolvement of treatment options, recently reported clinical trial data indicate that bortezomib will remain part of the backbone of upfront treatment for patients with NDMM [40]. Extending previous cohort-level studies [9–12], our findings indicate that MCS-based discrimination between individual patients who may or may not benefit from bortezomib treatment may prevent overtreatment in > 50% of the 130 000 MM patients yearly diagnosed worldwide [45]. Though high-risk group specific benefit from carfilzomib-based treatment remains to be further validated in trial cohorts, enrichment of the UPR signature suggests that patients with MCS-defined high-risk MM may also benefit from the treatment with next generation proteasome inhibitors.

A limitation of this study is that our SSP algorithm has not yet been prospectively validated. Nevertheless, the analytical robustness of the SSP algorithm was validated using data from ~1400 patients who had not been used in the model development, and the prognostic capacity of MCS was consistent among 2057 MM patients from three continents. These patients were at initial diagnosis or relapse, were from real-world cohorts or prospective clinical trials originating from different regions and ethnic backgrounds and were treated with different regimens. The predictive capacity of the MCS was replicated in two previously reported phase 3 clinical trials for patients with NDMM or RRMM. MCS-defined high-risk MM cases were also enriched in molecular markers associated with replication stress, genomic instability, and an APOBEC mutational signature. A major fraction of the MCS-defined low-risk MM cases harbored mutually exclusive mutations of RAS genes and interferon signaling activities. These features are reported to be predictive for response to immunotherapies and chemotherapy in different types of cancers [37,38,46–49].

Our findings also address a common challenge in the development of molecular classification schemes for MM and other types of cancers. Using unbiased clustering analysis or prognosis-correlated algorithms, high number of prognostic models have been proposed. However, prognostic models mostly do not predict treatment response, their clinical utility is consequently limited. We believe that the clinical utility of a given classification scheme is to the first hand dependent on the relevance of classifier genes to tumor pathogenesis. We hypothesized that gene networks conserved between plasma cell development and MM pathogenesis are powerful biomarkers for defining biologically motivated molecular subtypes. Our previous work identified gene co-expression modules conserved between brain development and glioma pathogenesis, which enabled the identification of brain development-correlated EM/PM molecular subtypes of adult gliomas and the identification of a developmentally anchored vulnerability that could otherwise not have been revealed under the unbiased or prognosis-correlated classification schemes [36,50,51]. As members of MCL1-M are crucially involved in both plasma cell development and MM pathogenesis [28,52], MCS based individual subtyping may not only provide a tool for risk stratification and prediction of treatment response in MM patients, it may also provide a platform for studying subtype-specific pathogenesis and therapeutic targets.

In summary, our findings suggest that addition of MCS to traditionally used clinicopathological factors may provide a means for biological definition of survival and treatment responses in MM patients. Application of MCS-based SSP model will help improve the treatment of MM patients by increasing therapeutic efficacy and reducing toxicity under the existing treatment regimens, while also facilitating the development of new treatment strategies, particularly for the high-risk MM patients.

4.0.0.0.1 Data availability and compliance statement

The authors declare that the acquisition and subsequent use of all data presented in this manuscript comply fully with all relevant local, national, and international laws, regulations, ethical guidelines, and the terms of use associated with the original data sources.

The authors bear full legal responsibility for ensuring the legality of data acquisition and all subsequent uses.

The codes for single sample prediction algorithm are freely available to the research communities upon request to the corresponding author. Gene expression data of Chaoyang Hospital dataset is deposited at OMIX with accession code OMIX001137.

References

[1] D’Agostino M , Cairns DA , Lahuerta JJ , Wester R , Bertsch U , Waage A , Zamagni E , Mateos MV , Dall’Olio D , van de Donk NWCJ , Jackson G , Rocchi S , Salwender H , Bladé Creixenti J , van der Holt B , Castellani G , Bonello F , Capra A , Mai EK , Dürig J , Gay F , Zweegman S , Cavo M , Kaiser MF , Goldschmidt H , Hernández Rivas JM , Larocca A , Cook G , San-Miguel JF , Boccadoro M , Sonneveld P . Second Revision of the international staging system (R2-ISS) for overall survival in multiple myeloma: a European Myeloma Network (EMN) report within the HARMONY project. J Clin Oncol 2022; 40(29): 3406–3418

[2] Palumbo A , Avet-Loiseau H , Oliva S , Lokhorst HM , Goldschmidt H , Rosinol L , Richardson P , Caltagirone S , Lahuerta JJ , Facon T , Bringhen S , Gay F , Attal M , Passera R , Spencer A , Offidani M , Kumar S , Musto P , Lonial S , Petrucci MT , Orlowski RZ , Zamagni E , Morgan G , Dimopoulos MA , Durie BGM , Anderson KC , Sonneveld P , San Miguel J , Cavo M , Rajkumar SV , Moreau P . Revised international staging system for multiple myeloma: a report from International Myeloma Working Group. J Clin Oncol 2015; 33(26): 2863–2869

[3] Pawlyn C , Davies FE . Toward personalized treatment in multiple myeloma based on molecular characteristics. Blood 2019; 133(7): 660–675

[4] Caro J , Al Hadidi S , Usmani S , Yee AJ , Raje N , Davies FE . How to treat high-risk myeloma at diagnosis and relapse. Am Soc Clin Oncol Educ Book 2021; 41(41): 291–309

[5] Usmani SZ , Rodriguez-Otero P , Bhutani M , Mateos MV , Miguel JS . Defining and treating high-risk multiple myeloma. Leukemia 2015; 29(11): 2119–2125

[6] Kuiper R , Broyl A , de Knegt Y , van Vliet MH , van Beers EH , van der Holt B , el Jarari L , Mulligan G , Gregory W , Morgan G , Goldschmidt H , Lokhorst HM , van Duin M , Sonneveld P . A gene expression signature for high-risk multiple myeloma. Leukemia 2012; 26(11): 2406–2413

[7] Perrot A , Lauwers-Cances V , Tournay E , Hulin C , Chretien ML , Royer B , Dib M , Decaux O , Jaccard A , Belhadj K , Brechignac S , Fontan J , Voillat L , Demarquette H , Collet P , Rodon P , Sohn C , Lifermann F , Orsini-Piocelle F , Richez V , Mohty M , Macro M , Minvielle S , Moreau P , Leleu X , Facon T , Attal M , Avet-Loiseau H , Corre J . Development and validation of a cytogenetic prognostic index predicting survival in multiple myeloma. J Clin Oncol 2019; 37(19): 1657–1665

[8] Kumar SK , Dispenzieri A , Lacy MQ , Gertz MA , Buadi FK , Pandey S , Kapoor P , Dingli D , Hayman SR , Leung N , Lust J , McCurdy A , Russell SJ , Zeldenrust SR , Kyle RA , Rajkumar SV . Continued improvement in survival in multiple myeloma: changes in early mortality and outcomes in older patients. Leukemia 2014; 28(5): 1122–1128

[9] Richardson PG , Sonneveld P , Schuster MW , Irwin D , Stadtmauer EA , Facon T , Harousseau JL , Ben-Yehuda D , Lonial S , Goldschmidt H , Reece D , San-Miguel JF , Bladé J , Boccadoro M , Cavenagh J , Dalton WS , Boral AL , Esseltine DL , Porter JB , Schenkein D , Anderson KC . Bortezomib or high-dose dexamethasone for relapsed multiple myeloma. N Engl J Med 2005; 352(24): 2487–2498

[10] Ubels J , Sonneveld P , van Vliet MH , de Ridder J . Gene networks constructed through simulated treatment learning can predict proteasome inhibitor benefit in multiple myeloma. Clin Cancer Res 2020; 26(22): 5952–5961

[11] Mulligan G , Mitsiades C , Bryant B , Zhan F , Chng WJ , Roels S , Koenig E , Fergus A , Huang Y , Richardson P , Trepicchio WL , Broyl A , Sonneveld P , Shaughnessy JD Jr , Leif Bergsagel P , Schenkein D , Esseltine DL , Boral A , Anderson KC . Gene expression profiling and correlation with outcome in clinical trials of the proteasome inhibitor bortezomib. Blood 2007; 109(8): 3177–3188

[12] Richardson PG , Xie W , Mitsiades C , Chanan-Khan AA , Lonial S , Hassoun H , Avigan DE , Oaklander AL , Kuter DJ , Wen PY , Kesari S , Briemberg HR , Schlossman RL , Munshi NC , Heffner LT , Doss D , Esseltine DL , Weller E , Anderson KC , Amato AA . Single-agent bortezomib in previously untreated multiple myeloma: efficacy, characterization of peripheral neuropathy, and molecular correlations with response and neuropathy. J Clin Oncol 2009; 27(21): 3518–3525

[13] Mehta J . Total therapy 2 in treatment of multiple myeloma: questions about gene expression profiling and treatment-related mortality. J Clin Oncol 2011; 29(5): e124

[14] Barlogie B , Tricot G , Anaissie E , Shaughnessy J , Rasmussen E , van Rhee F , Fassas A , Zangari M , Hollmig K , Pineda-Roman M , Lee C , Talamo G , Thertulien R , Kiwan E , Krishna S , Fox M , Crowley J . Thalidomide and hematopoietic-cell transplantation for multiple myeloma. N Engl J Med 2006; 354(10): 1021–1030

[15] Bergsagel PL , Kuehl WM , Zhan F , Sawyer J , Barlogie B , Shaughnessy J Jr . Cyclin D dysregulation: an early and unifying pathogenic event in multiple myeloma. Blood 2005; 106(1): 296–303

[16] Zhan F , Huang Y , Colla S , Stewart JP , Hanamura I , Gupta S , Epstein J , Yaccoby S , Sawyer J , Burington B , Anaissie E , Hollmig K , Pineda-Roman M , Tricot G , van Rhee F , Walker R , Zangari M , Crowley J , Barlogie B , Shaughnessy JD Jr . The molecular classification of multiple myeloma. Blood 2006; 108(6): 2020–2028

[17] Broyl A , Hose D , Lokhorst H , de Knegt Y , Peeters J , Jauch A , Bertsch U , Buijs A , Stevens-Kroef M , Beverloo HB , Vellenga E , Zweegman S , Kersten MJ , van der Holt B , el Jarari L , Mulligan G , Goldschmidt H , van Duin M , Sonneveld P . Gene expression profiling for molecular classification of multiple myeloma in newly diagnosed patients. Blood 2010; 116(14): 2543–2553

[18] Skerget S , Penaherrera D , Chari A , Jagannath S , Siegel DS , Vij R , Orloff G , Jakubowiak A , Niesvizky R , Liles D , Berdeja J , Levy M , Wolf J , Usmani SZ , Rifkin R , Meehan K , Benson D , Zonder J , Ascensao J , Gasparetto C , Hernández M , Trudel S , Dakhil S , Bahlis N , Paganini JV , Rios P , Sampol A , Mannem S , Silbermann R , Lunning M , Chu M , Milner C , Harroff A , Graham M , Shao S , Dodlapati J , de Larrea CF , Klein L , Kuzma C , Fonseca R , Azaceta G , Granell M , Martinez-Chamorro C , Balaraman R , Silva C , Chinea A , Costello C , Kambhampati S , Lewis DQ , Grossbard M , Yost K , Robles R , Sebag M , Harris W , Ngaiza J , Bar M , Shieh M , Min F , Onitilo A , Volterra F , Wachsman W , Yalamachili M , Abella E , Anderson L , Bargay J , Hassoun H , Hsu G , Kaya H , Menter A , Patel D , Richards D , Solomon W , Anderson R , Chandra S , Conde M , Girnius S , Matkiwsky M , Krsnik I , Kumar S , Oriol A , Rodríguez P , Roy V , Srinivas S , Steis R , Christofferson AW , Nasser S , Aldrich JL , Legendre C , Benard B , Miller C , Turner B , Kurdoglu A , Washington M , Yellapantula V , Adkins JR , Cuyugan L , Boateng M , Helland A , Kyman S , McDonald J , Reiman R , Stephenson K , Tassone E , Blanski A , Livermore B , Kirchhoff M , Rohrer DC , D’Agostino M , Gambella M , Collison K , Stumph J , Kidd P , Donnelly A , Zaugg B , Toone M , McBride K , DeRome M , Rogers J , Craig D , Liang WS , Gutierrez NC , Jewell SD , Carpten J , Anderson KC , Cho HJ , Auclair D , Lonial S , Keats JJ . Comprehensive molecular profiling of multiple myeloma identifies refined copy number and expression subtypes. Nat Genet 2024; 56(9): 1878–1889

[19] Shaughnessy JD Jr , Zhan F , Burington BE , Huang Y , Colla S , Hanamura I , Stewart JP , Kordsmeier B , Randolph C , Williams DR , Xiao Y , Xu H , Epstein J , Anaissie E , Krishna SG , Cottler-Fox M , Hollmig K , Mohiuddin A , Pineda-Roman M , Tricot G , van Rhee F , Sawyer J , Alsayed Y , Walker R , Zangari M , Crowley J , Barlogie B . A validated gene expression model of high-risk multiple myeloma is defined by deregulated expression of genes mapping to chromosome 1. Blood 2007; 109(6): 2276–2284

[20] Bhutani M , Zhang Q , Friend R , Voorhees PM , Druhan LJ , Barlogie B , Sonneveld P , Morgan GJ , Symanowski JT , Avalos BR , Copelan EA , Usmani SZ . Investigation of a gene signature to predict response to immunomodulatory derivatives for patients with multiple myeloma: an exploratory, retrospective study using microarray datasets from prospective clinical trials. Lancet Haematol 2017; 4(9): e443–e451

[21] Shaughnessy JD Jr , Qu P , Usmani S , Heuck CJ , Zhang Q , Zhou Y , Tian E , Hanamura I , van Rhee F , Anaissie E , Epstein J , Nair B , Stephens O , Williams R , Waheed S , Alsayed Y , Crowley J , Barlogie B . Pharmacogenomics of bortezomib test-dosing identifies hyperexpression of proteasome genes, especially PSMD4, as novel high-risk feature in myeloma treated with Total Therapy 3. Blood 2011; 118(13): 3512–3524

[22] Chapman MA , Sive J , Ambrose J , Roddie C , Counsell N , Lach A , Abbasian M , Popat R , Cavenagh JD , Oakervee H , Streetly MJ , Schey S , Koh M , Willis F , Virchis AE , Crowe J , Quinn MF , Cook G , Crawley CR , Pratt G , Cook M , Braganza N , Adedayo T , Smith P , Clifton-Hadley L , Owen RG , Sonneveld P , Keats JJ , Herrero J , Yong K . RNA-seq of newly diagnosed patients in the PADIMAC study leads to a bortezomib/lenalidomide decision signature. Blood 2018; 132(20): 2154–2165

[23] Maura F , Rajanna AR , Ziccheddu B , Poos AM , Derkach A , Maclachlan K , Durante M , Diamond B , Papadimitriou M , Davies F , Boyle EM , Walker B , Hultcrantz M , Silva A , Hampton O , Teer JK , Siegel EM , Bolli N , Jackson GH , Kaiser M , Pawlyn C , Cook G , Kazandjian D , Stein C , Chesi M , Bergsagel L , Mai EK , Goldschmidt H , Weisel KC , Fenk R , Raab MS , Van Rhee F , Usmani S , Shain KH , Weinhold N , Morgan G , Landgren O . Genomic classification and individualized prognosis in multiple myeloma. J Clin Oncol 2024; 42(11): 1229–1240

[24] Maley CC , Aktipis A , Graham TA , Sottoriva A , Boddy AM , Janiszewska M , Silva AS , Gerlinger M , Yuan Y , Pienta KJ , Anderson KS , Gatenby R , Swanton C , Posada D , Wu CI , Schiffman JD , Hwang ES , Polyak K , Anderson ARA , Brown JS , Greaves M , Shibata D . Classifying the evolutionary and ecological features of neoplasms. Nat Rev Cancer 2017; 17(10): 605–619

[25] Goh KI , Cusick ME , Valle D , Childs B , Vidal M , Barabasi AL . The human disease network. Proc Natl Acad Sci USA 2007; 104(21): 8685–8690

[26] Klein U , Dalla-Favera R . Germinal centres: role in B-cell physiology and malignancy. Nat Rev Immunol 2008; 8(1): 22–33

[27] De Silva NS , Klein U . Dynamics of B cells in germinal centres. Nat Rev Immunol 2015; 15(3): 137–148

[28] Samo AA , Li J , Zhou M , Sun Y , Yang Y , Zhang Y , Li J , van Duin M , Lu X , Fan X . MCL1 gene co-expression module stratifies multiple myeloma and predicts response to proteasome inhibitor-based therapy. Genes Chromosomes Cancer 2018; 57(8): 420–429

[29] Peperzak V , Vikstrom I , Walker J , Glaser SP , LePage M , Coquery CM , Erickson LD , Fairfax K , Mackay F , Strasser A , Nutt SL , Tarlinton DM . Mcl-1 is essential for the survival of plasma cells. Nat Immunol 2013; 14(3): 290–297

[30] Gupta VA , Matulis SM , Conage-Pough JE , Nooka AK , Kaufman JL , Lonial S , Boise LH . Bone marrow microenvironment-derived signals induce Mcl-1 dependence in multiple myeloma. Blood 2017; 129(14): 1969–1979

[31] Nair B , van Rhee F , Shaughnessy JD Jr , Anaissie E , Szymonifka J , Hoering A , Alsayed Y , Waheed S , Crowley J , Barlogie B . Superior results of Total Therapy 3 (2003–33) in gene expression profiling-defined low-risk multiple myeloma confirmed in subsequent trial 2006–66 with VRD maintenance. Blood 2010; 115(21): 4168–4173

[32] Sonneveld P , Schmidt-Wolf IG , van der Holt B , El Jarari L , Bertsch U , Salwender H , Zweegman S , Vellenga E , Broyl A , Blau IW , Weisel KC , Wittebol S , Bos GMJ , Stevens-Kroef M , Scheid C , Pfreundschuh M , Hose D , Jauch A , van der Velde H , Raymakers R , Schaafsma MR , Kersten MJ , van Marwijk-Kooy M , Duehrsen U , Lindemann W , Wijermans PW , Lokhorst HM , Goldschmidt HM . Bortezomib induction and maintenance treatment in patients with newly diagnosed multiple myeloma: results of the randomized phase III HOVON-65/GMMG-HD4 trial. J Clin Oncol 2012; 30(24): 2946–2955

[33] Heuck CJ , Qu P , van Rhee F , Waheed S , Usmani SZ , Epstein J , Zhang Q , Edmondson R , Hoering A , Crowley J , Barlogie B . Five gene probes carry most of the discriminatory power of the 70-gene risk model in multiple myeloma. Leukemia 2014; 28(12): 2410–2413

[34] Goicoechea I , Puig N , Cedena MT , Burgos L , Cordon L , Vidriales MB , Flores-Montero J , Gutierrez NC , Calasanz MJ , Ramos MLM , Lara-Astiaso D , Vilas-Zornoza A , Alignani D , Rodriguez I , Sarvide S , Alameda D , Garcés JJ , Rodriguez S , Fresquet V , Celay J , Garcia-Sanz R , Martinez-Lopez J , Oriol A , Rios R , Martin-Sanchez J , Martinez-Martinez R , Sarra J , Hernandez MT , de la Rubia J , Krsnik I , Moraleda JM , Palomera L , Bargay J , Martinez-Climent JA , Orfao A , Rosiñol L , Mateos MV , Lahuerta JJ , Blade J , San Miguel J , Paiva B . Deep MRD profiling defines outcome and unveils different modes of treatment resistance in standard- and high-risk myeloma. Blood 2021; 137(1): 49–60

[35] Danziger SA , McConnell M , Gockley J , Young MH , Rosenthal A , Schmitz F , Reiss DJ , Farmer P , Alapat DV , Singh A , Ashby C , Bauer M , Ren Y , Smith K , Couto SS , van Rhee F , Davies F , Zangari M , Petty N , Orlowski RZ , Dhodapkar MV , Copeland WB , Fox B , Hoering A , Fitch A , Newhall K , Barlogie B , Trotter MWB , Hershberg RM , Walker BA , Dervan AP , Ratushny AV , Morgan GJ . Bone marrow microenvironments that contribute to patient outcomes in newly diagnosed multiple myeloma: A cohort study of patients in the Total Therapy clinical trials. PLoS Med 2020; 17(11): e1003323

[36] Zhang Y , Li J , Yi K , Feng J , Cong Z , Wang Z , Wei Y , Wu F , Cheng W , Samo AA , Salomoni P , Yang Q , Huang Y , Kang C , Jiang T , Fan X . Elevated signature of a gene module coexpressed with CDC20 marks genomic instability in glioma. Proc Natl Acad Sci USA 2019; 116(14): 6975–6984

[37] Cristescu R , Mogg R , Ayers M , Albright A , Murphy E , Yearley J , Sher X , Liu XQ , Lu H , Nebozhyn M , Zhang C , Lunceford JK , Joe A , Cheng J , Webber AL , Ibrahim N , Plimack ER , Ott PA , Seiwert TY , Ribas A , McClanahan TK , Tomassini JE , Loboda A , Kaufman D . Pan-tumor genomic biomarkers for PD-1 checkpoint blockade-based immunotherapy. Science 2018; 362(6411): eaar3593

[38] Ayers M , Lunceford J , Nebozhyn M , Murphy E , Loboda A , Kaufman DR , Albright A , Cheng JD , Kang SP , Shankaran V , Piha-Paul SA , Yearley J , Seiwert TY , Ribas A , McClanahan TK . IFN-gamma-related mRNA profile predicts clinical response to PD-1 blockade. J Clin Invest 2017; 127(8): 2930–2940

[39] Facon T , Lee JH , Moreau P , Niesvizky R , Dimopoulos M , Hajek R , Pour L , Jurczyszyn A , Qiu L , Klippel Z , Zahlten-Kumeli A , Osman M , Paiva B , San-Miguel J . Carfilzomib or bortezomib with melphalan-prednisone for transplant-ineligible patients with newly diagnosed multiple myeloma. Blood 2019; 133(18): 1953–1963

[40] Kumar SK , Jacobus SJ , Cohen AD , Weiss M , Callander N , Singh AK , Parker TL , Menter A , Yang X , Parsons B , Kumar P , Kapoor P , Rosenberg A , Zonder JA , Faber E Jr , Lonial S , Anderson KC , Richardson PG , Orlowski RZ , Wagner LI , Rajkumar SV . Carfilzomib or bortezomib in combination with lenalidomide and dexamethasone for patients with newly diagnosed multiple myeloma without intention for immediate autologous stem-cell transplantation (ENDURANCE): a multicentre, open-label, phase 3, randomised, controlled trial. Lancet Oncol 2020; 21(10): 1317–1330

[41] Walker BA , Mavrommatis K , Wardell CP , Ashby TC , Bauer M , Davies FE , Rosenthal A , Wang H , Qu P , Hoering A , Samur M , Towfic F , Ortiz M , Flynt E , Yu Z , Yang Z , Rozelle D , Obenauer J , Trotter M , Auclair D , Keats J , Bolli N , Fulciniti M , Szalat R , Moreau P , Durie B , Stewart AK , Goldschmidt H , Raab MS , Einsele H , Sonneveld P , San Miguel J , Lonial S , Jackson GH , Anderson KC , Avet-Loiseau H , Munshi N , Thakurta A , Morgan GJ . Identification of novel mutational drivers reveals oncogene dependencies in multiple myeloma. Blood 2018; 132(6): 587–597

[42] Walker BA , Mavrommatis K , Wardell CP , Ashby TC , Bauer M , Davies F , Rosenthal A , Wang H , Qu P , Hoering A , Samur M , Towfic F , Ortiz M , Flynt E , Yu Z , Yang Z , Rozelle D , Obenauer J , Trotter M , Auclair D , Keats J , Bolli N , Fulciniti M , Szalat R , Moreau P , Durie B , Stewart AK , Goldschmidt H , Raab MS , Einsele H , Sonneveld P , San Miguel J , Lonial S , Jackson GH , Anderson KC , Avet-Loiseau H , Munshi N , Thakurta A , Morgan G . A high-risk, Double-Hit, group of newly diagnosed myeloma identified by genomic analysis. Leukemia 2019; 33(1): 159–170

[43] Ubels J , Sonneveld P , van Beers EH , Broijl A , van Vliet MH , de Ridder J . Predicting treatment benefit in multiple myeloma through simulation of alternative treatment effects. Nat Commun 2018; 9(1): 2943

[44] Decaux O , Lode L , Magrangeas F , Charbonnel C , Gouraud W , Jezequel P , Attal M , Harousseau JL , Moreau P , Bataille R , Campion L , Avet-Loiseau H , Minvielle S . Prediction of survival in multiple myeloma based on gene expression profiles reveals cell cycle and chromosomal instability signatures in high-risk patients and hyperdiploid signatures in low-risk patients: a study of the Intergroupe Francophone du Myelome. J Clin Oncol 2008; 26(29): 4798–4805

[45] Cowan AJ , Allen C , Barac A , Basaleem H , Bensenor I , Curado MP , Foreman K , Gupta R , Harvey J , Hosgood HD , Jakovljevic M , Khader Y , Linn S , Lad D , Mantovani L , Nong VM , Mokdad A , Naghavi M , Postma M , Roshandel G , Shackelford K , Sisay M , Nguyen CT , Tran TT , Xuan BT , Ukwaja KN , Vollset SE , Weiderpass E , Libby EN , Fitzmaurice C . Global burden of multiple myeloma: a systematic analysis for the Global Burden of Disease Study 2016. JAMA Oncol 2018; 4(9): 1221–1227

[46] Sistigu A , Yamazaki T , Vacchelli E , Chaba K , Enot DP , Adam J , Vitale I , Goubar A , Baracco EE , Remédios C , Fend L , Hannani D , Aymeric L , Ma Y , Niso-Santano M , Kepp O , Schultze JL , Tüting T , Belardelli F , Bracci L , La Sorsa V , Ziccheddu G , Sestili P , Urbani F , Delorenzi M , Lacroix-Triki M , Quidville V , Conforti R , Spano JP , Pusztai L , Poirier-Colame V , Delaloge S , Penault-Llorca F , Ladoire S , Arnould L , Cyrta J , Dessoliers MC , Eggermont A , Bianchi ME , Pittet M , Engblom C , Pfirschke C , Préville X , Uzè G , Schreiber RD , Chow MT , Smyth MJ , Proietti E , André F , Kroemer G , Zitvogel L . Cancer cell-autonomous contribution of type I interferon signaling to the efficacy of chemotherapy. Nat Med 2014; 20(11): 1301–1309

[47] He Y , Wang L , Wei T , Xiao YT , Sheng H , Su H , Hollern DP , Zhang X , Ma J , Wen S , Xie H , Yan Y , Pan Y , Hou X , Tang X , Suman VJ , Carter JM , Weinshilboum R , Wang L , Kalari KR , Weroha SJ , Bryce AH , Boughey JC , Dong H , Perou CM , Ye D , Goetz MP , Ren S , Huang H . FOXA1 overexpression suppresses interferon signaling and immune response in cancer. J Clin Invest 2021; 131(14): e147025

[48] von Locquenghien M , Rozalen C , Celia-Terrassa T . Interferons in cancer immunoediting: sculpting metastasis and immunotherapy response. J Clin Invest 2021; 131(1): e143296

[49] Borden EC . Interferons alpha and beta in cancer: therapeutic opportunities from new insights. Nat Rev Drug Discov 2019; 18(3): 219–234

[50] Sun Y , Zhang W , Chen D , Lv Y , Zheng J , Lilljebjorn H , Ran L , Bao Z , Soneson C , Sjögren HO , Salford LG , Ji J , French PJ , Fioretos T , Jiang T , Fan X . A glioma classification scheme based on coexpression modules of EGFR and PDGFRA. Proc Natl Acad Sci USA 2014; 111(9): 3538–3543

[51] Wei Y , Li G , Feng J , Wu F , Zhao Z , Bao Z , Zhang W , Su X , Li J , Qi X , Duan Z , Zhang Y , Vega SF , Jakola AS , Sun Y , Carén H , Jiang T , Fan X . Stalled oligodendrocyte differentiation in IDH-mutant gliomas. Genome Med 2023; 15(1): 24

[52] Yang Y , Jiang S , Du H , Tang J , Xiao P , Wu Y , Li J , Feng J , Wei Y , Samo AA , Lu X , Fan X . Distinct pathway activities are associated with prognosis and response to bortezomib-containing treatment in MCL1-M based molecular subtypes of multiple myeloma. Ann Hematol 2024; 103(12): 5495–5510

RIGHTS & PERMISSIONS

Higher Education Press