Identification of High-Risk Multiple Myeloma With a Plasma Cell Leukemia-Like Transcriptomic Profile

PURPOSE Primary plasma cell leukemia (pPCL) is an aggressive subtype of multiple myeloma, which is distinguished from newly diagnosed multiple myeloma (NDMM) on the basis of the presence of ≥ 20% circulating tumor cells (CTCs). A molecular marker for pPCL is currently lacking, which could help identify NDMM patients with high-risk PCL-like disease, despite not having been recognized as such clinically. METHODS A transcriptomic classifier for PCL-like disease was bioinformatically constructed and validated by leveraging information on baseline CTC levels, tumor burden, and tumor transcriptomics from 154 patients with NDMM included in the Cassiopeia or HO143 trials and 29 patients with pPCL from the EMN12/HO129 trial. Its prognostic value was assessed in an independent cohort of 2,139 patients with NDMM from the HOVON-65/GMMG-HD4, HOVON-87/NMSG-18, EMN02/HO95, MRC-IX, Total Therapy 2, Total Therapy 3, and MMRF CoMMpass studies. RESULTS High CTC levels were associated with the expression of 1,700 genes, independent of tumor burden (false discovery rate < 0.05). Of these, 54 genes were selected by leave-one-out cross-validation to construct a transcriptomic classifier representing PCL-like disease. This not only demonstrated a sensitivity of 93% to identify pPCL in the validation cohort but also classified 10% of NDMM tumors as PCL-like. PCL-like MM transcriptionally and cytogenetically resembled pPCL, but presented with significantly lower CTC levels and tumor burden. Multivariate analyses in NDMM confirmed the significant prognostic value of PCL-like status in the context of Revised International Staging System stage, age, and treatment, regarding both progression-free (hazard ratio, 1.64; 95% CI, 1.30 to 2.07) and overall survival (hazard ratio, 1.89; 95% CI, 1.42 to 2.50). CONCLUSION pPCL was identified on the basis of a specific tumor transcriptome, which was also present in patients with high-risk NDMM, despite not being clinically leukemic. Incorporating PCL-like status into current risk models in NDMM may improve prognostic accuracy.


INTRODUCTION
For over a century, the level of circulating tumor cells (CTCs) has been assessed in multiple myeloma (MM) to identify aggressive disease. 1 Although MM is characterized by an intramedullary outgrowth of malignant plasma cells, the degree of hematogenous tumor cell dissemination is highly variable between patients. At diagnosis, CTCs are routinely quantified in peripheral blood by morphology and can be detected in the majority of patients with MM if flow cytometry is used. 2 However, in only 2% of patients, these levels are $ 20% or $ 2 3 10 9 /L, which is pathognomonic for primary plasma cell leukemia (pPCL). 3,4 Symptomatic MM patients with lower CTC levels at diagnosis are classified as newly diagnosed MM (NDMM), but these patients may still develop secondary PCL (sPCL) after treatment. 5 Clinically, pPCL is considered a high-risk disease entity within MM. 6 Patients commonly present with a large tumor burden and extensive morbidity, show poor response to standard treatment, and have a dismal overall survival (OS). 5,[7][8][9] Yet, several reports have suggested that certain patients with NDMM experience an equally aggressive disease course to patients with pPCL, without having CTC levels $ 20%. [10][11][12] The International Myeloma Working Group has therefore challenged the current diagnostic criteria for pPCL, which has prompted ongoing research efforts to identify these PCL-like patients in alternative ways. 8 Disease aggressiveness in pPCL is considered to be reflected by the presence of significantly higher CTC levels than that in NDMM. Although this was previously hypothesized to be the result of a spillover from a large intramedullary tumor, evidence is accumulating that altered molecular features involved in cell adhesion, evasion of apoptosis, migration, bone marrow (BM) independence, and RNA metabolism are associated with this phenotype. [13][14][15][16][17][18][19][20][21][22] Still, molecular determinants of PCL-like disease remain poorly understood, with conventional highrisk markers in NDMM (ie, t(4;14), t(14;16) and deletion of chromosome 17p13 (del17p13)) only being detectable in a subset of pPCL tumors. [23][24][25][26][27][28][29] We therefore hypothesized that by molecularly classifying PCL-like disease, a novel high-risk biology could be unveiled that may already be detectable in patients with NDMM, despite not being clinically leukemic.

Study Design
This study was conducted in two phases: (1) transcriptomic classifier construction for PCL-like disease, and (2) assessment of its prognostic significance in NDMM. All human investigations in this study were performed after approval by medical ethical committees. Patient samples were obtained with written informed consent, in accordance with the Declaration of Helsinki.
PCL-like status prevalence per disease stage was assessed in 2,492 plasma cell samples (prevalence cohort). The prognostic value of PCL-like status was determined in a subset of 2,139 patients with NDMM from the prevalence cohort (survival cohort) with tumor transcriptomic, follow-up data and known age from the HOVON- 65

CTC Level Quantification
Baseline CTC levels were determined by Next-Generation Flow (NGF; EuroFlow) in 297 patients with NDMM and by morphological assessment in 51 patients with pPCL. 40,41 Tumor Cell Transcriptomics Transcriptomic profiles were generated from CD138enriched BM tumor cells, using microarray or RNA Seq protocols. For technical validation, PCL-like scores were generated with both protocols in a subset of 123 samples.

Classifier Construction
First, a linear regression model was used to rank genes on the basis of their association with high CTC levels, independent of tumor burden, defined as BM plasmacytosis. Second, the optimal number of genes to distinguish NDMM from pPCL was determined with a leave-one-out crossvalidation analysis. Third, a cutoff was chosen for PCL-like disease by selecting the minimal PCL-like score to detect all pPCL tumors in the discovery cohort.
Classifier performance was tested in the validation cohort. Predicted CTC levels were calculated by fitting a linear model with both tumor burden data and PCL-like scores. The contribution of each term to the variance in observed CTC levels was determined by analysis of variance.   and SKY92 high-risk, UAMS70 high-risk, and double-hit status were calculated as previously described. [43][44][45][46][47] Group Comparisons Groups were compared using the two-sided Wilcoxon signedrank or Wilcoxon rank-sum test for paired or unpaired continuous scenarios, respectively. The Fisher's exact test was used for categorical variables. Associations between continuous variables were tested by linear regression. P values were corrected for multiple testing according to the Benjamini-Hochberg procedure. P values and false discovery rates (FDRs) , .05 were considered statistically significant.

Survival Analysis
For progression-free survival (PFS), an event was defined as either progressive disease (PD) or death from any cause. For OS, an event was defined as death from any cause.   (100) 118 (97) 87 (100) 217 (98) Positive

Baseline Characteristics of pPCL Versus NDMM
To investigate clinical and molecular determinants of PCLlike disease, baseline patient and tumor characteristics were collected for 297 NDMM and 51 pPCL patients with available CTC level data (CTC cohort; Fig 1 and Table 1). NGF was performed to quantify CTCs in patients with NDMM, which could be detected in 257 of 297 (87%) patients (range, 0.00028%-36%), with 40 of 40 (100%) CTC-negative assays reaching a limit of detection , 10 -5 (Data Supplement, Supplementary Fig 1).
Both baseline CTC levels (median, 31% v 0.016%, P , .0001) and tumor burden as reflected by BM plasmacytosis (median, 64% v 32%, P , .0001) were higher in patients with pPCL than in patients with NDMM (Figs 2A and 2B). Tumor burden and CTC levels showed a positive, yet weak association (Adj. R 2 , 0.16, P , .0001), with all pPCL samples having higher CTC levels than expected on the basis of their tumor burden ( Fig 2C).

A Transcriptomic Profile Representing PCL-like Disease
To enable a more comprehensive screening of tumor cell aberrations that are associated with PCL-like disease, transcriptomic profiling was performed for BM tumor cells in a subgroup of 154 patients with NDMM and 29 patients with pPCL from the CTC cohort (Fig 1 and Supplementary Fig 4]).

Identification of PCL-like MM Tumors
Since the PCL-like score is a reflection of PCL-like disease, we hypothesized that this information could be leveraged to identify NDMM tumors with a similar transcriptome to pPCL tumors. To this end, a threshold for the PCL-like classifier   was set ( Fig 3B). With this threshold, 13 of 14 (93%) pPCL tumors in the validation cohort were correctly classified as PCL-like ( Fig 3C). Of note, a subgroup of NDMM tumors was also classified as PCL-like on the basis of this threshold, despite presenting with CTC levels as low as 0.083%: PCLlike MM (Figs 3B and 3C). PCL-like MM had both lower CTC levels (median, 3.0% v 35%; P , .0001) and a lower tumor burden (median, 36% v 71%; P 5 .045) than pPCL (Fig 3D). Patients with NDMM who had a PCL-like score below 3.55 were referred to as intramedullary MM (i-MM).

Genes Tumor Samples
In all 10 NDMM cohorts of the prevalence cohort, a PCLlike transcriptome was consistently identified, with a prevalence ranging from  6]).

Molecular and Clinical Determinants of PCL-Like MM
A comparison of ssGSEA scores between subgroups showed that pPCL and i-MM were highly distinct at the transcriptomic level, whereas PCL-like MM and pPCL were very similar (Fig 4B). A total of 1,160 pathways were differentially expressed between PCL-like MM and i-MM, which were among others involved in p53 signaling, Rho GTPase activity, mitosis, and binding and uptake of ligands (Figs 4C and 4D).
Comparing all three transcriptomic classifiers head-to-head showed that the PCL-like classifier had the highest sensitivity to detect pPCL (93%  Supplementary Fig 9]).

PCL-like Status as an Independent Prognostic Marker in NDMM
The

DISCUSSION
In this study, a molecular classifier representing PCL-like disease was constructed and validated. On the basis of tumor transcriptomic data alone, this classifier not only identifies pPCL with a high sensitivity but also detects a PCLlike transcriptome in 10% of patients with NDMM, despite not meeting current diagnostic criteria for pPCL. PCL-like status had significant prognostic value in the context of conventional risk factors in NDMM and thereby represents a novel diagnostic tool to identify high-risk patients.
pPCL is an aggressive subtype of MM that in contrast to NDMM is characterized by high CTC levels. To the best of our knowledge, this is the first comprehensive study to demonstrate that pPCL can not only be classified at the clinical but also at the molecular level. This finding allowed for the identification of NDMM tumors with a PCL-like transcriptome, which demonstrated an enrichment of many characteristics that have been associated with pPCL previously, eg, increased proliferation, increased hypoxia, decreased expression of adhesion markers, assignment to the MF cluster, and the presence of t (11;14). 13,23,27,28,52,[67][68][69] A similar observation has recently been described with respect to the detection of NDMM-like mutational and copy number profiles in a subgroup of SMM, raising the question whether treatment decisions in MM should be driven by disease manifestations or rather by their molecular classification. [70][71][72] Although PCL-like disease is associated with high CTC levels, it should be emphasized that a different group of patients with NDMM is identified with the PCL-like classifier than with clinically relevant CTC level thresholds (Data Supplement [ Supplementary Fig 14]). 4,10-12,73-78 Although 53% of PCL-like MM patients in our cohort had CTC levels $ 2%, which may represent underdiagnosed pPCL because of circadian fluctuations in CTC levels or atypical CTC morphology, a subgroup of NDMM patients with CTC levels as low as 0.083% also had BM tumor cells that transcriptionally resembled pPCL. 79,80 We demonstrated that the lower CTC levels observed in PCL-like MM versus pPCL could be explained by a smaller tumor burden. Yet, about one third of CTC variance remains unexplained, which warrants further exploration. This information could be of particular relevance when studying the evolutionary biology of MM, as high CTC levels represent advanced disease. 81,82 The observation that paired NDMM and PD samples in our study did not differ in terms of PCL-like score could suggest that PCL likeness is an inherent, rather than an acquired feature of MM tumors. In this regard, PCL-like status could be one of several conditions that need to be met for the development of sPCL, in combination with additional factors such as immune evasion or increased BM angiogenesis. 83,84 Although PCL-like status is associated with both an inferior PFS and OS in univariate analyses, it is prognostically most valuable in combination with other risk models. Combining PCL-like status with R-ISS stage improved prognostic accuracy and enabled the identification of a subgroup of NDMM patients with exceptionally     high-risk disease, as evidenced by a median OS of 13.2 months (95% CI, 6.8 to 41.1) for PCL-like MM patients with R-ISS stage III. Considering that most patients with pPCL in our study also presented with R-ISS stage III in combination with a PCL-like tumor transcriptome, it is remarkable to see that this OS in patients with NDMM agrees well with recently reported survival rates in pPCL. [85][86][87][88][89][90][91][92][93] This suggests that a combination of R-ISS stage with PCL-like status may be used to identify borderline pPCL patients although it should be noted that many diagnostic laboratories still have limited experience with transcriptomic classifiers.  In the conducted meta-analyses, PCL-like status conferred high-risk disease irrespective of received treatment. Its association with an inferior PFS in 7 of 8 trial cohorts and an inferior OS in 6 of 8 trial cohorts underlines the unmet need to develop effective treatment strategies for PCL-like disease. 7, [94][95][96] Although preliminary reports suggest that novel agents including daratumumab and venetoclax are effective in pPCL, this study has identified additional potential therapeutic vulnerabilities in PCL-like tumors, which warrants further translational efforts. [97][98][99] Examples include high expression of the nuclear export gene XPO1 in PCL-like tumors, which is associated with sensitivity to selinexor, increased PHF19 expression, and PRC2 pathway activity, which can be targeted by PRC2 inhibitors, and high levels of phosphorylating enzyme gene DCK, which has been found to confer a good response to nucleoside analogs. [100][101][102] In conclusion, in this study a transcriptomic classifier for PCLlike disease was constructed and validated, which enables the identification of NDMM tumors with a similar molecular composition to pPCL and improves prognostic accuracy in NDMM in combination with conventional risk markers.