Single‐cell RNA sequencing reveals chemokine self‐feeding of myeloma cells promotes extramedullary metastasis

In this study, we aimed to determine the mechanisms underlying the initial extramedullary translocation of myeloma cells from bone marrow into peripheral blood. We found that clonal circulating plasma cells (cPCs) are more frequently detected by flow cytometry in extramedullary plasmacytoma (EMP) patients and worsen their prognosis. It is technically much easier to collect single cPCs using FACS than it is to perform EMP biopsy. Therefore, combining EMP imaging with cPC detection may be a promising strategy for prognostic stratification. Here, using single‐cell transcriptome analysis, we found that the chemokine CXCL12, a key molecule involved in CXCR4‐dependent cell retention in the bone marrow, is abnormally upregulated in cPCs and might initially enable cPCs to evade bone marrow retention and translocate into the bloodstream.

In this study, we aimed to determine the mechanisms underlying the initial extramedullary translocation of myeloma cells from bone marrow into peripheral blood. We found that clonal circulating plasma cells (cPCs) are more frequently detected by flow cytometry in extramedullary plasmacytoma (EMP) patients and worsen their prognosis. It is technically much easier to collect single cPCs using FACS than it is to perform EMP biopsy. Therefore, combining EMP imaging with cPC detection may be a promising strategy for prognostic stratification. Here, using single-cell transcriptome analysis, we found that the chemokine CXCL12, a key molecule involved in CXCR4-dependent cell retention in the bone marrow, is abnormally upregulated in cPCs and might initially enable cPCs to evade bone marrow retention and translocate into the bloodstream.
Keywords: chemokine self-feeding; circulating plasma cells; extramedullary plasmacytoma; multiple myeloma Multiple myeloma (MM) is a malignant proliferation of monoclonal plasma cells that produce monoclonal immunoglobulins and is defined by the presence of 10% of clonal plasma cells (PCs) in the bone marrow or a biopsy-proven extramedullary plasmacytoma (EMP). MM remains an incurable disease, representing the second most common hematologic malignancy worldwide. Its incidence has been increasing in recent years [1,2]. MM often presents with hypercalcemia, renal function impairment, anemia, and osteolytic lesions [3,4]. Over the past decade, the median survival of myeloma patients has increased to 8 years. This remarkable improvement is mostly because of the use of high-dose therapy followed by autologous stem cell transplantation, in addition to the widespread incorporation of novel agents, including immunomodulatory drugs (thalidomide and lenalidomide) and proteasome inhibitors (bortezomib). MM is a heterogeneous entity; the overall survival (OS) time of MM ranges from several months to several years.
For most myeloma patients, the proliferation of PCs is restricted to the bone marrow. However, a subset of MM patients develops EMPs, defined by the presence of clonal PCs outside the bone marrow, which generally involve soft tissue, cortical bone, and centrum, and may be some of the most prominent clinical features of MM [5]. EMPs are present at the time of initial diagnosis in 9.6-19.6% of MM patients and develop during treatment in 6-20% patients [6,7]. MM patients with EMP invasion have a poor prognosis and are difficult to treat [5]. In such patients, late detection and delayed treatment can seriously diminish quality of life and can be life-threatening. Therefore, it is clinically important to optimize the detection of this entity.
To date, the mechanisms of extramedullary spread in MM are poorly understood. In one hypothetical model of EMP pathogenesis, metastatic myeloma cells initially exit the bone marrow, translocate into blood as clonal circulating plasma cells (cPCs), and finally settle in peripheral tissues and form an EMP [8][9][10]. However, the mechanistic evidence for this hypothesis is limited.
Here, we focused on the initial extramedullary translocation of myeloma cells from bone marrow into peripheral blood. The transcriptome can dynamically reflect intrinsic and environmental fluctuations [11,12]. We took transcriptome snapshots to provide insights into the translocation mechanism and applied singlecell RNA-seq to explore transcriptomic differences between myeloma cells in bone marrow (BMMCs) and peripheral cPCs in each EMP-positive patient. We found a consensus of upregulated auto-secretion of the chemokine CXCL12 in cPCs. This CXCL12 upregulation may contribute to the initial extramedullary translocation by evasion of CXCR4-dependent cell retention in bone marrow. Upregulation of another bone marrow retention chemokine, CXCL7, was previously published [13], further supporting the notion of chemokine auto-secretion relating to extramedullary translocation in myeloma.

Patients
The participants were 11 patients with primarily untreated MM (median age: 58 years, range: 49-76 years) and 10 patients with EMP (median age: 64.5 years, range: 44-75 years). All patients were diagnosed according to the 2009 International Myeloma Working Group (IMWG) consensus criteria [14]. The patients enrolled in this study were hospitalized during the period of October 2014 to June 2015 at the Peking University Third Hospital and Beijing Chao-Yang Hospital. The end point of the analyses was OS, which was defined as the time from diagnosis to death (regardless of the cause) or date of the last follow-up evaluation.
All patients received combination chemotherapy regimens, including BAD (bortezomib, doxorubicin, dexamethasone), VCD (bortezomib, cyclophosphamide, dexamethasone), TD (thalidomide, dexamethasone), RD (lenalidomide, dexamethasone), RCD (lenalidomide, cyclophosphamide, dexamethasone), and DCEP (dexamethasone, cyclophosphamide, etoposide, cisplatin). Some patients received additional radiation therapy, surgical therapy, and autologous hematopoietic stem cell transplantation. Efficacy criteria for MM were defined according to International Myeloma Working Group (IMWG) consensus criteria for response and minimal residual disease assessment in MM [15]. This research was approved by the ethics committee of Peking University Third Hospital (IRB00006761-2016120). All patients provided written informed consent in accordance with the Declaration of Helsinki.

Sample collection and preparation
Typically, 4 mL bone marrow aspirate or peripheral blood was drawn from patients in a sterile manner. Peripheral blood mononuclear cells (PBMCs) and bone marrow cells were isolated by Ficoll-Paque Plus (GE Healthcare Bio-Sciences, Pittsburgh, PA, USA). A 2-min lysis of RBC (red blood cell) was performed if necessary. Samples were passed through 30-lm nylon mesh to remove cell clumps. Filters were wetted with buffer before use. Purification of CD138 + plasma cells from the single-cell suspension was performed using human CD138 MACS MicroBeads (Miltenyi Biotec, Bergisch Gladbach, Germany) according to the manufacturer's instructions. Cell number was determined, and the cell suspension was centrifuged at 300 g for 10 min. The supernatant was aspirated, and the cell pellet was resuspended in 80 lL of buffer per 2 9 10 7 total cells. Then, 20 lL of CD138 MicroBeads per 2 9 10 7 total cells was added, and mixed well, and cells were incubated for 15 min in the a refrigerator (2-8°C). Cells were washed by adding 1-2 mL of buffer per 2 9 10 7 total cells and centrifuging at 300 g for 10 min. The supernatant was then aspirated, and up to 10 8 cells were resuspended in 500 lL of buffer. The column was placed in the magnetic field of a MACS Separator, and the MS column was prepared by rinsing with 500 lL of buffer. Cell suspension was applied onto the column, and unlabeled cells that passed through were collected. The column was washed three times with 500 lL of buffer, and total eluent was collected; this eluent is the unlabeled cell fraction. The column was removed from the separator and placed on a collection tube. Onto the column, 1 mL of buffer was pipetted and the magnetically labeled CD138 + cells were immediately flushed out by firmly pushing the plunger into the column. The single-cell suspension was then stained with antibodies and analyzed or sorted by flow cytometry.

Single-cell RNA sequencing and analysis
Single myeloma cells or cPCs were sorted into lysis buffer in 96-well plates with index sorting mode on FACS. The SmartSeq2 protocol was immediately performed to rescue mRNAs from cell lysate [16]. Libraries were constructed with a Tn5-based TruePrep DNA Library Prep Kit (Vazyme Biotech, Nanjing, China) and were sequenced on the HiSeq 2500 platform (Illumina, Inc., San Diego, CA, USA). Sequencing reads were aligned to the human reference genome GENCODE GRCh37 v25 transcriptome using SALMON v0.8.2 [17]. Cells with sufficient quality were selected (mitochondria reads < 20%, TPM > 1 genes over 1k). Thirty-two single BMMCs were prepared for every patient (Fig. 5); generally, > 50% cells passed QC.
Principal component analysis (PCA) was carried out using FactoMineR in R. Four hundred highly variable genes were selected from first four principal components (PCs). MM-related gene groups were chosen, and their loading value or score in PC1 and PC2 was plotted ( Fig. 4C-E) and summarized (Fig. 4F,G). To explore the homogeneity and heterogeneity of myeloma among patients, 126 BM cells were analyzed by SEURAT (version 2.3.4: New York Genome Center, New York, NY, USA, default parameters) for highly variable genes and unsupervised hierarchical clustering (Fig. 5A). Selected gene groups were plotted in patient-oriented ordering ( Fig. 5B-D, Fig. S3). In addition to unsupervised PCA, differential expression analysis between cPCs and BMMCs in the same patient was performed with DESeq2 and plotted in GRAPHPAD PRISM (Fig. 6C-E). Statistics of genes related to BM ingress, BM adhesion, and BM egress are summarized in Fig. 6F. The same analysis was performed for Fig. S5.

Statistical analysis
All calculations were performed using the Windows version of IBM SPSS software, version 18.0.1 (PASW Statistics for Windows; IBM, Armonk, NY, USA). OS and PFS were determined by the Kaplan-Meier analysis. Differences between survival curves were tested for statistical significance using 2-tailed log-rank tests. Numerical data were analyzed by GRAPHPAD PRISM (GraphPad Software, Inc., La Jolla, CA, USA). Adjusted P-values were analyzed in R (R Core Team, Vienna, Austria). Nonparametric Mann-Whitney U-test was used for two-group comparisons in Fig. 1D. P < 0.05 (log-rank test) was considered statistically significant.

Circulating plasma cells worsen prognosis for EMP-positive patients
Twenty-one MM patients were enrolled. In Table 1, clinical characteristic analysis is provided for the 10 EMPpositive and 11 EMP-negative patients. Compared with EMP À patients, more EMP + patients were female (80% versus 27.3%). There was an even distribution of immunoglobulin subtype, similar ISS stages, and higher ESR-LDH-ALB detection. EMP was located primarily in soft tissue, centrum, and the chest wall.
To explore the hypothesis that EMP occurrence might have been associated with BMMCs or cPCs, 19 bone marrow aspirate samples and 16 peripheral blood samples were successfully collected. Both BMMCs and cPCs expressed high levels of malignant activation marker CD38 and plasma cell marker CD138, as expected (Fig. S1). Consistent with MM diagnosis, BMMCs were positive in all 19 bone marrow aspirates, but there was no difference between EMP-positive and EMP-negative aspirate samples. cPCs were identified after MACS enrichment. Five out of 9 (55.5%) EMP-positive patients had detectable cPCs in blood, whereas 2 out of 7 (28.6%) EMP-negative patients had detectable cPCs in blood (Fig. 1, Table 2). These results indicate that cPCs may be correlated with EMP progression from MM.
Circulating plasma cells have predictive power for survival of newly diagnosed MM [18]. However, it is still unknown whether cPCs contribute to EMP prognosis [19]. To address this issue, we first summarized OS of 21 patients. EMP-positive patients had significantly shorter survival than did EMP-negative patients (median survival: 12 months versus 74 months, P = 0.0002). The same survival pattern was observed from retrospective data of 180 MM patients in our hospital during the period from 2008 to 2016 (median survival: 21 months versus 74 months, P = 0.0001), suggesting minimal biased sampling of these 21 patients. In contrast, simply checking the presence of cPCs failed to predict survival in a significant way (P = 0.20), whereas combining EMP and cPC information showed significant improvement in prognosis (P = 0.017) (Fig. 2). EMPnegative patients showed maximal survival, regardless of cPCs presence. EMP-positive patients without cPCs in blood showed similar survival to 180 retrospective data with median survival of 21 months. However, EMP-positive patients with cPCs had extremely shortened median survival of 8 months, and all passed away within 2 years. Therefore, cPCs may provide a stratification insight for EMP as well as MM patients. Together, these results indicate that cPCs were more frequently detected in EMP-positive patients, and worsened their prognosis.
Circulating plasma cells showed a cluster differentiation marker phenotype similar to that of BM myeloma cells Compared with normal plasma cells, myeloma cells dynamically regulate expression of surface cluster differentiation (CD) markers, such as downregulation of B-cell markers CD45 and CD19, upregulation of adhesion molecules CD56 (N-CAM) and CD81, and variable regulation of tumorigenic marker CD117 (c-kit) [20]. As expected, BMMCs were CD45 À/dim CD19 À/dim CD117 À CD56 + CD81 + phenotype (Fig. 3).
To explore the possibility that CD markers may contribute to cPCs' exit from bone marrow, we compared cPCs with BMMCs in the same patient (P017 and P020 are EMP-positive; P013 is EMP-negative) ( Table 2, Fig. 3). In general, CD markers were similarly expressed in both BMMCs and cPCs, and CD56 (N-CAM) was slightly downregulated in cPCs in P017 (BMMC 93800 to cPC 34100 in MFI, 2.8-fold) and P020 (BMMC 9093 to cPC 6524 in MFI, 1.4-fold), confirming existing reports of lower CD56 expression in cPCs [20]. The finding that cPCs are similar to BMMCs in terms of surface CD markers does not help in understanding the extramedullary translocation. We then applied single-cell RNA-seq (scRNAseq) to these cells, making effort to discover the major difference between cPCs and BMMCs through wholetranscriptome snapshots, overcoming limited marker space of immunophenotyping.

Single-cell RNA-seq identifies myeloma cells from normal cells
Each cell contains only a few pgRNAs, which are easy to degrade and sensitive to contaminations [21]. We used the SmartSeq2 protocol to capture and amplify single-cell transcriptomes for sequencing. SmartSeq2 has been shown to have high sensitivity, high reproducibility, and full-length profiling ability [16,22]. These features were confirmed by using MM cell lines H929 and U266 and observing high gene recovery, highly reproducible gene expression results, and high coverage and evenness across the gene body (Fig. S2). Twenty-one out of 24 sequenced single CD38 + CD138 + BMMCs, with 5 out of 6 sequenced CD38 À CD138 À nonmyeloma control cells, were plotted. In PCA, PC1 (principal component 1) represented 22.6% of the total variance and provided separation between the two types of cells (Fig. 4A). The hierarchical clustering of the highly variable genes among cells showed that the two types of cells have distinct gene expression patterns and tend to cluster separately, as expected (Fig. 4B, Table S1).

Patient-oriented clustering of single BMMCs
Next, we explored whether EMP patients have common traits in the single-cell transcriptome, which may in turn benefit EMP diagnosis or intervention. Single-cell analysis simultaneously monitors disease heterogeneity and homogeneity in multidimensional comparisons, such as between disease situations (EMP-positive or EMP-negative), between patients, and between single cells within the same patient. Therefore, to achieve a global view of myeloma, single-cell RNA-seq was performed on 126 single BMMCs from six patients (four EMP-positive, two EMP-negative), and hierarchical clustering of their highly variable genes was analyzed (Fig. S3, Table S2). The heat map depicts the structure of subgroups with both cells and their marker genes. In general, most of the cells clustered according to the patients. Each patient had a unique set of marker genes. Thus, no obvious similarity was observed among EMP-positive patients (P014, P017, P020, and P023) or between EMP-negative patients (P019 and P021). BMMCs clustered together within individual patients.
Growth factors were reported to feed myeloma progression through surface receptors [40]. Universal transcription of FGFRs (fibroblast growth factor receptor 3), HDGF (hepatoma-derived growth factor), IGF1 (insulin-like growth factors 1), and TNFRSF17 (B-cell maturation antigen, BCMA) emphasized their values in myeloma diagnosis and intervention. Among these genes, we found no obvious differentiations between EMP-positive and EMP-negative patients (Fig. 5C).
Oncogenes and tumor suppressors are also relevant to myeloma progression, and we also observed patientoriented clustering of single BMMCs (Fig. S4). We further compared our data with previously published gene expression profiles GEP92 [41] and GEP70 [42] using microarrays and confirmed that the gene expression of single BMMCs is highly patient-dependent (Fig. S5).
Abnormal chemokine auto-secretion may contribute to cPC extramedullary translocation from the bone marrow As stated above, extramedullary translocation is the initial step of cPC egress from the bone marrow. To explore possible factors related to this translocation, we compared BMMCs and cPCs within the same patient (P017 and P020, both EMP-positive). In general, cPCs and BMMCs tended to cluster within the same cell type (Fig. 6A,B). Next, differentially expressed genes (DEG) were identified by DEG2 analysis, comparing cPCs to BMMCs. Then, shared DEGs of both EMP-positive patients were analyzed (Fig. 6C, Table S3). Interestingly, among shared genes, CXCL12 was identified, indicating its significance (Fig. 6D,E).
Chemokine CXCL12 and its sole receptor CXCR4 play pivotal roles in bone marrow cell retention [43,44]. CXCL12, functionally designated as stromal-derived factor-1 (SDF-1), is typically expressed in bone marrow stromal cells or vascular endothelial cells to attract CXCR4 + cells in bone marrow and brain [45][46][47], but is not expressed in immune cells, such as normal plasma cells. Therefore, this novel finding of abnormal CXCL12 transcription in cPCs may be utilized to circumvent CXCR4-dependent bone marrow retention.
As abnormal upregulation of chemokine CXCL12 expression in cPCs was found in both patients, this finding suggests a self-feed loop generated by cPCs, consequently releasing themselves from the CXCR4dependent bone marrow retention and translocating freely into the circulation.
To further support this chemokine auto-secretion and self-feed notion, we incorporated analysis of a published dataset. Pioneer genetic interrogation of cPCs at singlecell resolution has been reported [13]. Although no information about EMP diagnosis was provided, we can still compare the transcriptomes between cPCs and BMMCs across patients for common genes, as in Fig. 6. Both patients showed the distinction between cPCs and BMMCs by unsupervised clustering. Common genes in cPCs versus BMMCs were analyzed with the same significance criteria (FC > 10 and P < 10 À5 ). We found only one gene, CXCL7 (pro-platelet basic protein, PPBP), which was another bone marrow retention-related chemokine that is normally expressed by endothelial cells and platelets, but not by immune cells, including normal plasma cells (Fig. S6).
Although CXCL7's direct link to myeloma is unknown, it has been reported to be expressed in cancer cells [48]. Furthermore, its chemokine receptors CXCR1 and CXCR2 are related to bone marrow retention and extramedullary translocation of immune cells, such as neutrophils. Similar to CXCL12, the chemokine secretive pathway was upregulated for CXCL7 (Fig. S6). Again, auto-secretion of CXCL7 generated a possible self-feed loop that provided additional freedom for cPC extramedullary translocation.

Discussion
To explore the initiation of EMP pathogenesis through extramedullary translocation, we monitored the correlation of cPCs with EMP-positive patients in terms of detection frequency and survival prognosis. By comparing the single-cell transcriptome between cPCs and BMMCs, CXCL12 emerged in EMP-positive patients and was strongly correlated with cell retention in bone marrow. Previously published data identified that CXCL7, another bone marrow attracting chemokine, was auto-secreted in cPCs. Although the transient extramedullary translocation event was not captured in this study, since the very first alteration of myeloma cells into circulating metastatic cells occurs long before EMP diagnosis, the same mechanism may exist.
Establishment of a causal relationship between chemokine auto-secretion and EMP incidence is still challenging. A possible validation method is to block target chemokines, such as CXCL12 or CXCL7, in peripheral blood (e.g., with monoclonal antibody), and then test cPC presence and EMP incidence. The development of animal models for EMP would also benefit from validation of this mechanism.
As rare as 1/10 000 to 1/1000 in PBMCs, cPC detection is an invaluable method for MM diagnosis, followup, and prognosis [49][50][51][52][53][54]. Compared with bone marrow biopsy, cPC detection is cost-efficient and feasible for multitesting, which is crucial for following treatment response. Compared with imaging-based EMP diagnosis, the single-cell resolution of cPC detection increases diagnostic sensitivity and may contribute to identification of early metastasis. In addition, it is technically much easier to collect single cPCs using FACS than it is to perform an EMP biopsy. We have shown that cPCs are more frequently detected in EMP-positive patients, confirming their clinical significance. Furthermore, by combining imaging EMP diagnosis and cPC detection, we found a possible strategy for prognostic stratification. Larger scale studies on this leading correlation between cPCs and EMP diagnosis are yet to be done.
Understanding of cancer biology has recently been brought into single-cell resolution. scRNA-seq has become a powerful tool with which to dissect the cellular state at the finest scale. This approach retains the benefits of conventional bulk RNA-seq in providing a global yet detailed view of the transcriptome and surpasses bulk RNA-seq by generating much more informative data reflecting intercellular heterogeneity, which is highly biomedically relevant.
Interventions targeting chemokines instead of receptors should be explored. In MM, antagonists and antibodies targeting CXCR4 are under clinical trials, with the rationale that after chemotherapy and irradiation, hematopoietic stem cells are retained in bone marrow by the CXCL12/CXCR4 axis, delaying the exit of mature cells, such as neutrophils and monocytes, to the periphery, and disrupting this axis will benefit patients with faster blood recovery. However, here we provided evidence that chemokines, not chemokine receptors, are related to EMP metastasis, emphasizing the long-neglected significance of chemokine upstream regulators, such as hypoxia, as targets.

Conclusions
In summary, an abnormal chemokine auto-secretion was found to be upregulated specifically in cPCs. Both chemokines (CXCL12 and CXCL7) are strongly related to bone marrow retention and extramedullary translocation. Therefore, we deduce that a chemotactic self-feed mechanism may contribute to initial extramedullary translocation of cPCs from bone marrow, and the eventual formation of soft-tissue EMP. Our findings increase understanding of metastatic MM and identify a new chemokine target for intervention.
Ms Fei Wang at the National Center for Protein Sciences Beijing (Peking University) for the assistance in flow cytometry experiments, and Dr Yun Zhang and her team at the Peking University High Throughput Sequencing Center for next-generation sequencing. The work was supported by the Natural Science Author contributions SG, JW, YH, and HJ designed the study and prepared the manuscript. JW, JZ, YZ, and HJ contributed to patient enrollment and follow-up. SG, XL, and WW carried out single-cell RNA-seq analysis. XZ and FW carried out scRNA-seq data analysis. YP, JW, YH, and HJ critically reviewed the manuscript.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Fig. S1. Gating strategy of flow cytometry. Fig. S2. Protocol validation. Fig. S3. Patient-oriented clustering of BMMCs. Fig. S4. Oncogenes, tumor suppressor genes supported patient-oriented clustering of BMMCs. Fig. S5. Gene expression profile (GEP) genes supported patient-oriented clustering of BMMCs. Fig. S6. CXCL7 were upregulated in cPCs compared with BMMCs in published data. Table S1. PCA result for Fig. 4B. Table S2. Seurat result for Fig. S3. Table S3. DEG for Fig. 6C.