Single-cell profiling of multiple myeloma reveals molecular response to FGFR3 inhibitor despite clinical progression

Abstract

Genomic characterization of cancer has enabled identification of numerous molecular targets, which has led to significant advances in personalized medicine. However, with few exceptions, precision medicine approaches in the plasma cell malignancy multiple myeloma (MM) have had limited success, likely owing to the subclonal nature of molecular targets in this disease. Targeted therapies against FGFR3 have been under development for the past decade in the hopes of targeting aberrant FGFR3 activity in MM. FGFR3 activation results from the recurrent transforming event of t(4;14) found in ∼15% of MM patients, as well as secondary FGFR3 mutations in this subgroup. To evaluate the effectiveness of targeting FGFR3 in MM, we undertook a phase 2 clinical trial evaluating the small-molecule FGFR1–4 inhibitor, erdafitinib, in relapsed/refractory myeloma patients with or without FGFR3 mutations (NCT02952573). Herein, we report on a single t(4;14) patient enrolled on this study who was identified to have a subclonal FGFR3 stop-loss deletion. Although this individual eventually progressed on study and succumbed to their disease, the intended molecular response was revealed through an extensive molecular characterization of the patient's tumor at baseline and on treatment using single-cell genomics. We identified elimination of the FGFR3-mutant subclone after treatment and expansion of a preexisting clone with loss of Chromosome 17p. Altogether, our study highlights the utility of single-cell genomics in targeted trials as they can reveal molecular mechanisms that underlie sensitivity and resistance. This in turn can guide more personalized and targeted therapeutic approaches, including those that involve FGFR3-targeting therapies.

INTRODUCTION

The treatment landscape for multiple myeloma (MM) has changed dramatically over the past two decades. Immune-based therapies have become the standard of care for this hematological malignancy, especially at relapse. Advances in molecular profiling and sequencing have enabled a better understanding of the molecular drivers underlying the initiation and progression that underlie myeloma pathogenesis. The main initiating events are hyperdiploidy and immunoglobulin heavy chain (IgH) translocations with one of five recurrent partner loci: CCND1 in t(11;14), MMSET/FGFR3 in t(4;14), MAF in t(14;16), MAFB in t(14;20), and CCND3 in t(6;14). Subsequent progression of the disease is marked by the acquisition of additional somatic mutations, indels, and copy-number alterations, which underlies evolution and therapeutic resistance (Pawlyn and Davies 2018).

The t(4;14) translocation is found in ∼15% of MM patients and is believed to be a high-risk molecular marker with adverse survival rates (Abdallah et al. 2020). The translocation results in ectopic and overexpression of FGFR3 in nearly 80% of newly diagnosed t(4;14) patients, with the remainder lacking expression because of the loss of the der(14) chromosome (Benard et al. 2017). The lack of FGFR3 expression in a subset of t(4;14) MM has raised questions regarding its relevance in oncogenic transformation. Meanwhile, whole-exome sequencing (WES) of 80 t(4;14) cases from the Multiple Myeloma Research Foundation (MMRF) CoMMpass study (NCT01454297) identified nonsynonymous mutations of FGFR3 in 20% of patients and posited that these mutations are likely to occur after the primary translocation event (Benard et al. 2017). Common FGFR3 mutations in t(4;14) patients occur in the extracellular domain (p.R248), transmembrane domain (p.Y373), kinase domain (p.K650), and stop codon (p.J807G, p.J807R, and p.J807C) resulting in augmented receptor dimerization and ligand-independent signaling and have been shown to be strongly transforming in several experimental models (Plowright et al. 2000; Chesi et al. 2001; Wesche et al. 2011). Supporting the clinical relevance of FGFR3 mutations in myeloma patients with t(4;14), FGFR3^mut-expressing patients had a median survival of 2.8 years compared to not reached for t(4;14) FGFR3^wt-expressing patients (Benard et al. 2017). Given the various ways that FGFR3 can be dysregulated in MM and the associated adverse prognosis conferred to this patient population, new treatments tailored to this molecular subgroup are needed, as is a more in-depth understanding of the molecular dynamics of treatment response.

Based on preclinical studies demonstrating antimyeloma activity of FGFR3 inhibition in t(4;14) myeloma (Trudel et al. 2004), we initiated a phase 2 clinical trial to evaluate the efficacy of erdafitinib, a potent tyrosine kinase inhibitor of FGFR1–4, plus dexamethasone in t(4;14) relapsed or refractory myeloma (NCT02952573). Participants in the study were assigned to one of two groups depending on whether they expressed wild-type FGFR3 or harbored a somatic mutation of FGFR3 (Supplemental Fig. 1). We report the changes in genomic landscape pretreatment and posttreatment in a t(4;14) FGFR3^mut myeloma patient who was enrolled in this clinical trial.

RESULTS

Clinical Presentation and Molecular Profiling

A 52-yr-old female was diagnosed with IgA lambda MM, t(4;14) positive, ISS-III in February 2014 (Supplemental Fig. 1A). Her initial treatment included triplet induction (cyclophosphamide-bortezomib-dexamethasone) followed by autologous stem cell transplantation (ASCT) and four cycles of lenalidomide/dexamethasone consolidation followed by lenalidomide maintenance. She experienced disease progression within 1 year of ASCT and subsequently received multiple lines of therapy including carfilzomib-based (carfilzomib-lenalidomide-dexamethasone), selinexor-based (selinexor-pomalidomide-dexamethasone), and daratumamab-based (daratumamab-pomalidomide-dexamethasone) regimens, as well as intensive therapy with DPACE (dexamethasone-cisplatin-adriamycin-cyclophosphamide-etoposide). Following DPACE chemotherapy, she experienced prolonged bone marrow suppression, eventually demonstrating biochemical progression upon count recovery. Clinical molecular profiling performed prior to subsequent treatment (Supplemental Fig. 1A) confirmed high expression of FGFR3, the presence of IGHA1-FGFR3 fusion transcripts indicative of t(4;14), and an activating stop-loss, in-frame 30-bp deletion in FGFR3 (c.2404_*12delGGGGGCTCGCGGACGTGAAGGGCCACTGGT, p.G802_X807del). The patient was therefore enrolled onto a phase 2 clinical trial evaluating the oral, small-molecule FGFR3 inhibitor, erdafitinib, in patients expressing FGFR3 with or without mutation (NCT02952573, Supplemental Fig. 1B). Two months into treatment with erdafitinib, the patient demonstrated evidence of biochemical progression (Supplemental Table 1), eventually succumbing to the disease 1 month later.

Clinical tumor sequencing for this patient was performed as part of the MMRF's Michigan Oncology Sequencing Center (MI-ONCOSEQ) clinical sequencing study (NCT0288410) using the OncoSeq 1500 assay (CLIA-certified laboratory-developed test), which targets 1500 cancer-related genes (4.6 Mb). Briefly, a bone marrow (BM) aspirate and peripheral blood sample were collected from the patient and shipped overnight to the Michigan Center for Translational Pathology (Supplemental Fig. 1A). Tumor (CD138⁺-enriched cells) and normal (CD138-depleted peripheral blood mononuclear cells) target capture libraries and tumor whole transcriptome capture libraries were prepared and sequenced according to the study protocol on a HiSeq2500 (500× coverage for each tumor sample, 200× coverage for matched normal, and 40 million paired end reads for tumor transcriptome sequencing). Sequencing quality control (QC) reported an estimated 46% tumor content after sequencing and passing scores for nucleic acid quality, sequencing quality, library quality, and single-nucleotide polymorphism (SNP) fingerprinting. In the cancer cells, the OncoSeq assay detected a gain of Chromosome 1, loss of Chromosomes 13 and X, a somatic subclonal in-frame stop-loss mutation in FGFR3 (p.G802_X807del, AF = 20%), and four nonsynonymous, subclonal somatic mutations of unknown significance: EED (p.H258N, AF = 10%), LRP1B (p.T1099A, AF = 15%), MYC (p.A416S, AF = 20%), and BCOR (p.A184V, AF = 19%) (Table 1; Supplemental Table 2). Although the assay does not directly detect IgH translocations, the sample had evidence of the t(4;14) rearrangement at FGFR3/WHSC1 as supported by gene fusion transcripts with IGHA1 and outlier expression of FGFR3 from RNA-seq analysis.

Table 1.

Summary of variants detected by the OncoSeq 1500 assay as part of the MMRF MI-ONCOSEQ clinical sequencing study (NCT0288410)

Gene	Chromosome	HGVS DNA reference	HGVS protein reference	Variant type	Predicted effect	dbSNP/dbVar ID	Genotype
FGFR3	Chr 4:1808969	c.2404_*12delGGGGGCTCGCGGACGTGAAGGGCCACTGGT	p.Gly802_Ter807del	In-frame deletion	Stop-loss	N/A	Somatic (AF = 19%)
EED	Chr 11:85977170	c.772C > A	p.His258Asn	Missense	Substitution	N/A	Somatic (AF = 10%)
LRP1B	Chr 2:141680558	c.3295A > G	p.Thr1099Ala	Missense	Substitution	N/A	Somatic (AF = 15%)
MYC	Chr 8:128753085	c.1246G > T	p.Ala416Ser	Missense	Substitution	N/A	Somatic (AF = 20%)
BCOR	Chr X:39934048	c.551C > T	p.Ala184Val	Missense	Substitution	N/A	Somatic (AF = 19%)

(HGVS) Human Genome Variation Society, (N/A) not applicable, (AF) allele frequency.

Single-Cell Expression Analysis

To understand the molecular underpinnings of the response to erdafitinib treatment, BM samples were collected prior to erdafitinib treatment initiation (pretreatment) and after 28 days of erdafitinib treatment at cycle 2 day 1 (C2D1). Mononuclear cells isolated from the BM were then profiled using the 10× Genomics single-cell RNA-sequencing platform. This resulted in high-quality single-cell expression profiles for 1130 and 910 cells, respectively, for pretreatment and C2D1 samples (Supplemental Table 3). Myeloma cell clusters (C1, C2, C5, C7, n = 905 cells) were subset bioinformatically from the full BM mononuclear cell data set (Supplemental Fig. 2A) based on cluster enrichment for expression of myeloma cell markers SDC1 (CD138), TNFRSF17 (BCMA), and FGFR3 (Supplemental Fig. 2B). This subset of myeloma cells were then reclustered to reveal transcriptional heterogeneity within the malignant cell compartment, with four unique clusters (Fig. 1A). From the pretreatment sample, myeloma cells predominantly clustered into two populations: C1 and C2 (Fig. 1A–C). As shown in Figure 1D, C1 was characterized by the expression of genes associated with plasma cell differentiation, the endoplasmic reticulum, and protein processing (JCHAIN, SSR4, XBP1, CD48), whereas C2 up-regulated expression of the FGFR3 target gene CCL3 (Masih-Khan et al. 2006). Cells from the C2D1 sample also colocalized in C1, as well as in two other predominant clusters, C0 and C3 (Fig. 1A–C), which expressed genes related to programmed cell death prevention (BCL2, PIM2, TRADD, BEX2) and cell cycle/proliferation (MKI67, TOP2A, BIRC5), respectively (Fig. 1D). This supports the presence of transcriptional variability in the malignant cell compartment, and a shift in expression profiles upon treatment that may reflect cellular programs of resistance to erdafitinib.

Figure 1.

Delineation of transcriptional heterogeneity in myeloma cells before and after treatment with erdafitinib. (A) t-distributed stochastic neighbor embedding (t-SNE) visualization of 965 myeloma cells from erdafitinib pretreatment and C2D1 samples, colored by cluster identity or sample identity (B). (C) Distribution of myeloma cells profiled by single-cell RNA sequencing (scRNA-seq) across transcriptional clusters for pretreatment and C2D1 samples. (D) Heatmap depicting top 20 marker genes up-regulated in each myeloma transcriptional cluster as determined by differential expression analysis (P < 0.05). A subset of 75 randomly selected cells per transcriptional cluster is shown and data represents scaled expression values (any values outside a range of −2 to 2 were clipped).

We next evaluated how the observed transcriptional heterogeneity relates to the copy-number variations (CNVs) reported by the clinical sequencing conducted for this patient. To do this, we inferred CNVs in each myeloma cell using the sciCNV method with 60 normal plasma cells from two donors from published data sets as a reference (Zheng et al. 2017). As seen in Figure 2A and consistent with clinical tumor sequencing results (Supplemental Table 2), single-cell inferred chromosomal copy-number variation (sciCNV) accurately detected the gain of Chromosome 1q and loss of Chromosome 13 in both the pretreatment and C2D1 samples. However, we did not detect deletion of Chromosome X, which we suspect may be attributed to sex differences between the patient and that of the subjects used for normal plasma cell references, because one of the donors is male based on Chr Y gene expression (Supplemental Fig. 3A).

Figure 2.

Subclonal copy-number alterations inferred from individual myeloma cells. (A) Heatmap of genome-wide copy-number variation inferred from single-cell RNA sequencing (scRNA-seq) data of malignant plasma cells as determined using single-cell inferred chromosomal copy-number variation (sciCNV) (Mahdipour-Shirayeh et al. 2021) using normal plasma cells (NPCs) from BMMC as reference. Columns represent genome position across chromosomes. (B) Gene set scoring for Chromosome 17p module score (MSigDB Positional Geneset) calculated using Seurat's AddModuleScore in malignant cells across treatment groups. Boxplots represent the distribution of each measurement within treatment groups, where the central rectangle spans the interquartile range and then the central line represents the median, and “whiskers” above and below the box show the value 1.5× the interquartile range (IQR). Cells from the pretreatment sample scoring less than 1.5× below the IQR are highlighted red (outlier) and inferred to be 17p deleted subclones. (C) Fluorescence microscopy images of representation cells from top three cytogenetic subclones. Probes are color-coded above each image with corresponding karyotypes below. Counted cells are listed to the right of each subclone expressed relative to the total t(4;14) malignant cells counted. (D) Hypothetical evolutionary path of CNV subclones based on karyotypes from fluorescence in situ hybridization (FISH) analysis. Subclonal proportions are listed to the right of each subclone expressed relative to total cells counted (normal and malignant).

In addition to the clinical results from bulk DNA sequencing, sciCNV also detected a focal gain of Chromosome 22 in both samples, a gain of Chromosome 4 in the pretreatment sample and the loss of Chromosome 17p in the C2D1 sample. We in turn used shallow whole-genome sequencing (WGS) of CD138-selected cells from the pretreatment sample, which validated the presence of a gain of Chromosome 4 (Supplemental Fig. 3B). However, this validation method did not detect the gain in Chromosome 22, which we hypothesize may be an artifact from scRNA-seq expression-based CNV inference related to high expression of immunoglobulin light chain genes, which are localized in this genomic region (Supplemental Fig. 3C).

Given that deletion of Chromosome 17p is a known poor prognostic event in myeloma, we hypothesized that erdafitinib treatment led to expansion of an aggressive del17p subclone. As predicted, we identified three cells from the pretreatment sample that scored lower for a Chromosome 17p gene set (Fig. 2B, <25th percentile of Chromosome 17p score in C2D1 cells) suggesting that a rare subpopulation of cells preexisting before treatment contained deletion of Chromosome 17p, which then expanded upon treatment with erdafitinib. To validate this hypothesis, we profiled pretreatment and C2D1 cells using fluorescence in situ hybridization (FISH) with probes for Chromosomes 1, 4, 13, 14, and 17. Consistent with our scRNA-seq data, del17p was detected in t(4;14)-positive cell populations at a frequency of 0.63% pretreatment and 84.6% at C2D1 (Fig. 2C,D; Supplemental Table 4). Thus, our data confirms that a Chromosome 17p deletion subclone preexisted and expanded rapidly during erdafitinib therapy (Fig. 2C,D; Supplemental Table 4).

In addition to copy-number variants, the clinical panel sequencing of this tumor revealed the presence of a subclonal FGFR3 in-frame stop-loss mutation (p.G802_X807del), which is predicted to result in protein elongation as the last five amino acids are replaced by aberrant amino acids. Because this deletion occurs in the 3' end of the FGFR3 gene, which is enriched in scRNA-seq, we evaluated whether the p.G802_X807del stop-loss aberration could be detected in myeloma cells from our scRNA-seq data. Indeed, we identified five reads with coverage of the mutation site, which corresponded to four cells (three cells with one mutant read and one cell with two mutant reads; Fig. 3A). We then mapped the cell barcodes back to our clustering results and found that all four cells with the FGFR3 stop-loss mutation were located in the pretreatment sample, cluster C2 (Fig. 3B). Notably, our single-cell analysis of transcriptional heterogeneity revealed FGFR3 target genes such as CCL3 were particularly up-regulated in C2 relative to all other clusters suggesting that FGFR3 p.G802_X807del may confer enhanced or ligand-independent activity to the FGFR3 receptor tyrosine kinase. Because 90.8% of the C2 cluster is comprised of cells from the pretreatment sample (109/120 cells) compared to 9.2% cells (11/120 cells), from the sample taken after 28 days of erdafitinib treatment, we suspect that cells containing this aberration may be particularly sensitive to erdafitinib. Thus, although this patient did not display a reduction of M protein and free light chains as conventional markers of disease response, molecular profiling revealed a marked reduction in the small cell cluster containing the FGFR3 mutation and CCL3 up-regulation indicative of a molecular response to erdafitinib (Fig. 3C).

Figure 3.

Identification of FGFR3 stop-loss deletion in single-cell RNA sequencing (scRNA-seq) reads from preerdafitinib-treated myeloma cells. (A) Integrative Genomics Viewer (IGV) screenshot of FGFR3 gene loci on Chromosome 4 depicting reads from scRNA-seq that contain the p.G802_X807del stop-loss loss deletion (grouped by cell barcode). (B) t-distributed stochastic neighbor embedding (t-SNE) visualization of 965 myeloma cells from erdafitinib pretreatment and C2D1 samples with cells containing the p.G802_X807del stop-loss deletion colored according to corresponding cell barcode colors from Figure 3A or gray if no reads coverage at the p.G802_X807del stop-loss deletion loci. (C) Proposed model for molecular response to erdafitinib as determined by scRNA-seq profiling.

DISCUSSION

Characterization of human cancers using genomic technologies has allowed the identification of numerous molecular targets, which has led to significant advances in drug development and personalized medicine. However, with few exceptions (e.g., venetoclax in t(11;14) MM patients; Kaufman et al. 2021), precision medicine approaches for MM have had limited success, likely owing to the subclonal nature of molecular targets. Our study demonstrates that the use of scRNA-seq to profile the molecular landscape of tumors extracted pretreatment and posttreatment with targeted agents can reveal significant insights into the underlying genomic and transcriptomic features of response and resistance. Our analysis of the patient's tumor at multiple timepoints revealed multiple subclonal malignant populations and rapid selection of a rare, resistant subclone indicating the necessity for combinatorial treatment approaches in patients with subclonal FGFR3 mutations. In our study, this resistant subclone demonstrated acquired deletion of Chromosome 17p, which is associated with reduced overall survival in patients with MM (Lakshman et al. 2019). Notably, bulk clinical sequencing prior to trial enrollment did not identify the subclonal deletion of Chromosome 17p in this patient, likely because of limits of detection given the low allele frequency of this variant (as confirmed by FISH). Thus, our data provide rationale for the use of single-cell profiling in patients receiving targeted therapy, as it may permit the identification of subclones likely to escape the activity of certain targeted therapies. Indeed, our study and the use of scRNA-seq provides insight into mechanisms of drug resistance suggesting that TP53 deletion via loss of 17p may confer FGFR3 inhibitor resistance, consistent with similar reports for EGFR inhibitors in lung cancer (Huang et al. 2011). The integration of single-cell genomics into precision medicine trials may allow for implementation of more personalized and rational combinatorial approaches that target the various subclones of a given myeloma tumor. Indeed, investigating combination therapies using a genomics-driven approach for molecularly targeted therapy matching is currently being taken by the MMRF MyDrug study (NCT03732703) (Kumar et al. 2022).

Our use of scRNA-seq indirectly permitted multimodal profiling of genetic alterations in individual myeloma cells and linkage of an FGFR3 sequence mutation to disruption of gene expression programs in a subclonal population. More specifically, we identified the FGFR3 p.G802_X807del aberration in the scRNA-seq reads of our data and showed that these cells clustered in the same transcriptional subgroup (C2). Although this specific aberration has not yet been described in the literature, several missense stop-loss mutations in the stop codon loci have been reported (p.X807G, p.X807R, and p.X807C) and result in the same read-through of the stop codon (Rousseau et al. 1995). Several activating FGFR3 mutations have been described in myeloma (Chesi et al. 2001; Ronchetti et al. 2001; Onwuazor et al. 2003) and render FGFR3 ligand-independent through changes in the intracellular domain activation loop (p.K650E), or in the extracellular domain (p.Y373C). Functional analysis of various FGFR3 mutants in the context of bone growth disorders has been reported including those in the intracellular domain (p.K650E, p.X807R), the transmembrane domain, and the extracellular domain (p.Y373C), which revealed that receptor phosphorylation is greater for FGFR3 mutations in the intracellular domain, which includes p.X807R, and that p.X807R results in a constitutively active FGFR3 (Gibbs and Legeai-Mallet 2007). Although a functional analysis of the downstream effects of p.G802_X807del is beyond the scope of this report, our scRNA-seq data support indirectly that this aberration would have similar downstream effects to p.X807R, because cells expressing the FGFR3 stop-loss allele in our study demonstrated up-regulation of FGFR3-target genes CCL3 and CCL4 (Masih-Khan et al. 2006) beyond that seen in cells with the t(4;14) translocation alone. Further, it is intriguing that the cells predicted to lack the mutation harbored an expression profile of increased differentiation and endoplasmic reticulum/protein processing because our previous work has demonstrated functional maturation of myeloma cells when mutated FGFR3 is inhibited (Trudel et al. 2004). This would further support that the constitutive activation of FGFR3 in cluster C2 prevents the completion of normal differentiation programs and onset of high-level antibody secretion, as we have previously shown.

Our data revealed that the C2 malignant subpopulation, which contained p.G802_X807del-expressing cells, was almost completely eliminated by treatment with erdafitinib. However, given that t(4;14) is an initiating event in myeloma and thus presumably all cells expressed this translocation, including the erdafitinib-resistant subclone, our data supports that targeting wild-type FGFR3 on a t(4;14) background may not be effective. Rather, our data suggest that the mechanism of action for erdafitinib is targeted toward aberrant FGFR3 activity via FGFR3 mutation. Thus, although this patient succumbed to clinical disease progression, a dramatic molecular response was achieved with erdafitinib as evidenced by elimination of the intended FGFR3-mutant subclone. This is particularly important in light of several FGFR3-targeted clinical trials in MM showing best responses of stable disease, leading some to question whether this class of agents has a role in myeloma therapy (Arnulf et al. 2007; Trudel et al. 2012; Scheid et al. 2015). However, our study shows that measuring molecular responses at the single-cell level may more accurately reflect the effectiveness of targeted agents such as erdafitinib. Further, the incorporation of scRNA-seq companion studies is critically important for informing whether these agents are indeed effective against their intended target and presents an opportunity to dissect responses to combination therapies directed against different subclonal populations. Taken together, our findings support the continued exploration FGFR3-targeted agents in the t(4;14) subtype of myeloma that express FGFR3 mutations and highlight some key molecular insights provided by clinical single-cell profiling approaches in cancer therapeutics.

METHODS

Bone Marrow Collection and Processing

Bone marrow aspirates were obtained with consent under the study protocol approved by the Research Ethics Board at University Health Network (CAPCR #16-5997). Samples were collected using fine-needle aspiration of the iliac crest into EDTA tubes and immediately transferred to the research laboratory for processing. Samples were diluted with PBS, and the mononuclear cell fraction was enriched for using density-based cell separation (Ficoll-Paque PLUS, GE Healthcare). Residual contaminating red blood cells were subsequently removed using ACK lysis buffer. After washing in PBS, cells were examined for quantity and viability using trypan blue and a Countess II automated counter (Thermo Fisher Scientific). Cells were then loaded into the 10× Genomics Chromium device according to the manufacturer's instructions.

Single-Cell Sequencing Using 10× Genomics Chromium Device

Single-cell libraries were constructed using the 3' V2 chemistry kit from 10× Genomics according to manufacturer's instructions. Libraries were sequenced on an Illumina HiSeq 2500 targeting 60,000 reads/cell. The 10× Genomics CellRanger software suite (v2.0.1) was used for processing raw sequencing reads, alignment (GRCh38) and to generate a digital gene expression (DGE) matrix of genes-by-cell unique molecular identifier (UMI) counts. Sequencing metrics can be found in Supplemental Table 3. The resulting raw gene matrices were used as input for downstream analyses using R v3.6.1.

Bioinformatic Processing of scRNA-seq Data

Cell-associated barcodes were determined based on the inflection point of read counts as described previously (Croucher et al. 2021). Low-quality cells were defined as having fewer than 200 detected genes and/or >20% mitochondrial transcripts and removed from downstream analysis. All subsequent steps in the clustering analysis were performed using Seurat v3.2.1. Briefly, NormalizeData() was used to calculate log-normalized expression values which were inputted to FindVariableGenes() to identify highly variably expressed genes for data scaling using ScaleData(). RunPCA() was then used to compute the top principal components and the top 10 principal components were used as input for nonlinear dimensionality reduction (t-SNE) and graph-based clustering as implemented by Seurat (res = 0.6). The plasma cell clusters (C1, C2, C5, C7, n = 905 cells) were then subset from the full data set of BM mononuclear cells based on expression of the myeloma marker, SDC1 (CD138), TNFRSF17 (BCMA), and FGFR3, and reclustered according to the method described above.

Inferred Copy-Number Analysis Using sciCNV

The CNV profiles were inferred in pretreatment (n = 340) and posttreatment (n = 555) tumor single cells against pooled control normal plasma cells (NPCs) (n = 60) using the sciCNV method (Mahdipour-Shirayeh et al. 2021) with a sliding-window of 143-gene size and without any baseline correction (see Mahdipour-Shirayeh et al. 2021 for more details). The input data to sciCNV pipeline was RTAM-normalized (Mahdipour-Shirayeh et al. 2021) in which cells with fewer than 250 expressed genes and genes expressed in <2% of cells were filtered out. The sciCNV profiles of test and control single cells were, then, scaled at each genomic locus using the mean sciCNV result of test cells to set CNV values to integers. The scaled CNV signals at each locus were denoised against a static noise threshold of 0.2. Then the sciCNV profile of each cell was standardized against the median of its nearest neighbors to reduce the stochastic noise in the sciCNV output. The final sciCNV results were represented as a heatmap using heatmap.3 R package and by separating pretreatment and posttreatment tumor cohorts.

FISH

Ten thousand CD138⁺ cells (enriched using magnetic beads, KIT) were centrifuged onto slides with a Shandon Cytospin and fixed in ice-cold 3:1 methanol/acetic acid. Fixed slides were incubated in 2× SSC (pH 7.0) for 10 min at 37°C, digested with 0.005% pepsin for 10 min at 37°C, dehydrated in a series of ethanol, and air-dried. Slides were then denatured in 70% formamide/2× SSC (pH 5.3) for 5 min at 72°C and hybridized overnight in a humidified chamber with denatured FISH probes. The next day, hybridized slides were washed and mounted in Vectashield/DAPI. Images were acquired on a BioView Allegro fully automated microscope and analyzed with BioView Solo analysis software (BioView USA Inc.). Hybridized slides were then washed, denatured for 3 min, and rehybridized with new FISH probes twice to obtain a total of three matching sets of images with seven different FISH probes. Probes used were Vysis LSI IGH/FGFR3 Dual Color Dual FISH probes, Cytocell Aquarius CKS1B/CDKN2C (P18) Amplification/Deletion Probe, Kreatech DLEU1 (13q14)/TP53 (17p13) FISH probe, and Agilent SureFISH 13q34 LAMP1 598kb Aqua probe. At least 300 cells from each slide were scored.

ADDITIONAL INFORMATION

Data Deposition and Access

The novel reported FGFR3 variant has been registered to the ClinGen Allele Registry (Pawliczek et al. 2018) under the Canonical Allele Identifier CA2573332805, and has been deposited to the Clinical Interpretation of Variants in Cancer (CIViC) database (Griffith et al. 2017) under the evidence item EID11033. Single-cell data generated from this study is available through the interactive single-cell portal CReSCENT (CRES-P34; https://crescent.cloud/) (Mohanraj et al. 2020).

Ethics Statement

Bone marrow aspirates were obtained with consent under the study protocol approved by the Research Ethics Board at University Health Network, Toronto, Canada (CAPCR #16-5997). The authors declare no competing interests.

Acknowledgments

We gratefully acknowledge the myeloma clinical trials group at Princess Margaret Cancer Centre for their contributions to the FGFR3 clinical trial and coordination of research samples. We thank the staff of the Princess Margaret Genomics Centre (www.pmgenomics.ca; Troy Ketala, Neil Winegarden, Julissa Tsao, and Nick Khuu) and Bioinformatics and High-Performance Computing Core (Carl Virtanen, Zhibin Lu, and Natalie Stickle) for sample coordination and their expertise in generating the single-cell sequencing data used in this study.

Author Contributions

D.C.C., T.J.P., and S.T. conceived the idea and designed the study. D.C.C. prepared bone marrow samples for sequencing and performed all related data analysis for scRNA-seq. A.M.-S. performed inferred CNV analysis of the scRNA-seq data using sciCNV. A.J.D. prepared and provided all clinical data for the manuscript. D.D.A. provided support for shallow whole-genome sequencing analysis. Z.L. provided support for sample preparation and performed CD138 selection. N.E. and R.T. performed cytogenetic FISH analysis. D.D.A., A.M.-S., R.T., T.J.P., and S.T. provided guidance in data analysis and interpretation of the results. D.C.C. and A.J.D. drafted the manuscript, with input from T.J.P. and S.T. All authors reviewed the manuscript prior to submission.

Funding

The phase 2 clinical trial was supported by Janssen Research and Development and the Multiple Myeloma Research Foundation. Additional research funding was provided by the Princess Margaret Cancer Foundation. Infrastructure support was received from the Canada Foundation for Innovation—John R. Evans Leaders Fund (CFI #38401) and the Ministry of Colleges and Universities Ontario Research Fund—Research Infrastructure Program. D.C.C. was supported by an Ontario Graduate Scholarship, the David Rae Graduate Student Scholarship from the University of Toronto, and a Graduate Fellowship in Cancer Research from the Princess Margaret Hospital Foundation. T.J.P. holds the Canada Research Chair in Translational Genomics and is supported by a Senior Investigator Award from the Ontario Institute for Cancer Research and the Gattuso-Slaight Personalized Cancer Medicine Fund.

Competing Interest Statement

The authors have declared no competing interest.

Referees

Yiming Zhong

Anonymous