Acquired resistance to a GPRC5D-directed T-cell engager in multiple myeloma is mediated by genetic or epigenetic target inactivation

Bispecific antibodies targeting GPRC5D demonstrated promising efficacy in multiple myeloma, but acquired resistance usually occurs within a few months. Using a single-nucleus multi-omic strategy in three patients from the MYRACLE cohort (ClinicalTrials.gov registration: NCT03807128), we identified two resistance mechanisms, by bi-allelic genetic inactivation of GPRC5D or by long-range epigenetic silencing of its promoter and enhancer regions. Molecular profiling of target genes may help to guide the choice of immunotherapy and early detection of resistance in multiple myeloma. Derrien et al. analyze three patients relapsing on talquetamab, a bispecific antibody against CD3 and GPRC5D, and show that acquired resistance is mediated by genetic inactivation or by long-range epigenetic silencing of GPRCD5.

Bispecific antibodies are monoclonal antibodies that redirect T cells by targeting both the T-cell co-receptor CD3 and markers expressed on the surface of tumor cells 1 .Bispecific antibodies targeting B-cell maturation antigen (BCMA; teclistamab) or G protein-coupled receptor, family C, group 5, member D (GPRC5D; talquetamab) demonstrated favorable safety profiles and promising efficacy in relapsed or refractory multiple myeloma (RRMM) in recent phase 1-2 studies 2,3 ; however, ~30% of patients did not respond and half of the responders experienced disease progression within 12 months.Thus, it is crucial to understand the mechanisms of innate and acquired resistance to bispecific antibodies.
Here, we explore the mechanisms of acquired talquetamab resistance in three patients with RRMM from the MYRACLE cohort (ClinicalTrials.govregistration: NCT03807128) (ref.4).The three patients had previously received multiple therapy lines including immunomodulatory agents, proteasome inhibitors, dexamethasone and anti-CD38 antibodies (Supplementary Table 1).We conducted a multi-omic characterization of myeloma cells, before talquetamab treatment (t1) and after relapse (t2), by whole-genome sequencing (WGS) and single-nucleus Multiome, allowing simultaneous profiling of gene expression (RNA-seq) and open chromatin (ATAC-seq) from the same cells.MM-01-0288 was a female patient diagnosed with immunoglobulin G (IgG) myeloma at 64 years old.After four previous therapy lines in 5.3 years, she was enrolled in the phase 1 MonumenTAL-2 trial (ClinicalTrials.govregistration: NCT05050097) and received talquetamab (weekly subcutaneous injection of 400 mg kg −1 until progression) in combination with carfilzomib.She achieved a very good partial response but the disease progressed after 6 months.WGS at t1 revealed clonal t(4;14), 1q gain, 13q loss and a focal heterozygous deletion at 12p encompassing GPRC5D locus, so that a single allele remained in tumor cells (Fig. 1a and Extended Data Fig. 1a,b).This tumor was heavily mutated (31,410 somatic mutations), with a high contribution of clock-like and APOBEC mutational processes (Extended Data Fig. 2a-c).Driver mutations included KRAS G12R, RB1 R787* and CDKN1B N61I (Supplementary Table 2).These mutations were present in a subclone, representing 91% of tumor cells at t1 but only 24% at t2 (subclone A).In contrast, subclone B, consisting of RB1 L572fs and HUWE1 P3841L mutations, was minor at t1 (8% of tumor cells) but became dominant at t2 (70%).Finally, the t2 sample had acquired seven subclonal GPRC5D alterations: three frameshift indels (E27fs, S125fs and F158fs), two nonsense mutations (W217* and W237*), an in-frame deletion of four amino acids within transmembrane helix 3 (G97-F100) and a 10-kb deletion encompassing

Brief Communication
https://doi.org/10.1038/s43018-023-00625-9 (Extended Data Fig. 10a).By contrast, bulk RNA-seq revealed a notable silencing of GPRC5D expression in both relapses (Fig. 2b,c), although the gene was highly expressed at t1 in MM-01-0221 (210 transcripts per million versus 1.3 at relapse, fold change = 0.0062, no t1 data for MM-01-0302).Single-nucleus RNA-seq confirmed the lack of GPRC5D transcript in >99.8% of both MM-01-0221 and MM-01-0302 cells at t2 (Extended Data Fig. 10b).In absence of genetic insult, we examined the chromatin landscape at GPRC5D locus.Healthy bone marrow and seven other multiple myeloma (MM) samples used as controls displayed a major ATAC-seq peak (open chromatin) at GPRC5D promoter and five minor peaks along the gene, all of which were closed in MM-01-0221 and MM-01-0302 t2 samples (Fig. 2b and Extended Data Fig. 10c).This significant loss of chromatin accessibility was not restricted to GPRC5D but involved 51 chromatin accessibility peaks in an extended 1.15-Mb region around the gene (Fig. 2d and Supplementary Table 5).Thirty-five of these peaks were annotated as enhancers in ENCODE cis-regulatory element database 5 , 18 were found to interact with GPRC5D promoter in B-cell promoter-capture Hi-C data 6 and 6 were annotated as GPRC5D enhancers in the GeneHancer database 7 (Fig. 2d).Finally, peak-togene expression linkage analysis in the 65,700 cells (two normal plasma cell and nine MM samples) revealed 12 peaks linked with GPRC5D expression, 9 of which were closed in our 2 GPRC5D-silenced relapses, including 7 ENCODE enhancers.Thus, acquired talquetamab resistance in this patient involved long-range epigenetic silencing of GPRC5D locus, with coordinated chromatin silencing of its promoter and enhancer regions.
In conclusion, we report two mechanisms of acquired talquetamab resistance resulting from the genetic or epigenetic inactivation of the target.The simultaneous growth of seven resistant subclones in MM-01-0288 illustrates the ease with which tumor cells with a pre-existing chromosome 12p deletion can acquire resistance with a second hit in GPRC5D.These findings mirror previous results indicating that chromosome 16p deletion may increase the risk of BCMA CAR-T resistance 8,9 .In 752 instances of newly diagnosed MM analyzed by WGS 10 , chromosome arm 12p was lost in 10%, and arm 16p in 18% of cases.Future studies of larger cohorts will be needed to establish whether these deletions predict response to bispecific antibodies and must be screened to select the best treatment option.GPRC5D is surrounded by inactive chromatin domains (H3K27Me3; Fig. 2d).This may favor its silencing by redistribution of inactive chromatin marks, as previously reported in other cancers 11 .Future studies will need to address whether GPRC5D expression before treatment influences talquetamab response.We anticipate that a thorough genetic, epigenetic and transcriptomic profiling of the targets will be important to select the most adapted bispecific antibody for each patient and to detect early the emergence of resistant subclones.
the transcription start site (TSS) (Extended Data Fig. 1b).As a single copy of GPRC5D remains in tumor cells, each alteration was present in a distinct subclone, representing between 6.2% and 31.6% of tumor cells (Fig. 1b).These alterations are predicted to prevent the translation of a functional GPRC5D protein able to be embedded in the cell membrane.Although messenger RNA expression was maintained (Extended Data Fig. 3a-c), flow cytometry confirmed the absence of GPRC5D protein at the surface of t2 cells (Extended Data Fig. 4 and Extended Data Fig. 5a-c).
We next mapped clonal and subclonal genetic mutations (Extended Data Fig. 6a) in single-nucleus data to refine the clonal architecture of MM-01-0288 and its evolution during treatment.Single-nucleus RNA-seq revealed heterogeneous subclones, in particular at t2 (Extended Data Fig. 6b).Trunk somatic mutations were detected in all cells.In contrast, subclonal mutations allowed to follow subclones before and after treatment: subclone A mutations were detected in the dominant t1 subclone and two minor subclones at t2, whereas subclone B mutations were detected in the minor t1 subclone and the three largest t2 subclones (Extended Data Fig. 6c).Virtual copy-number analysis revealed clonal and subclonal chromosome aberrations (−1p, −10p, −14q, −14qtel, +17q and −21), consistent with WGS data (Extended Data Fig. 7a-c), which defined five subclones (cn1 to cn5; Fig. 1c and Extended Data Fig. 8a-d).Copy-number subclones largely overlapped transcriptomic clusters and refined the clonal relationships inferred from somatic mutations (Fig. 1d).Differentially expressed genes and pathways were consistent with subclonal genetic events, for example with an up-regulation of 17q genes and the Ras signaling pathway in the cn1 subclone that displays +17q and KRAS mutation (Supplementary Tables 3 and 4).Finally, we mapped GPRC5D alterations in single-nucleus data (Fig. 1e and Extended Data Fig. 9).Each alteration was detected in a single copy-number subclone: S125fs in cn3, F158fs in cn4 and all other alterations in the subclone cn5.Talquetamab resistance in cn2, the only subclone without a second hit in GPRC5D, could involve a combination of low frequency genomic alterations undetectable at this coverage and/or post-transcriptional mechanisms precluding the protein to be translated or presented on the cell surface.Overall, integrating WGS and single-nucleus data allowed us to reconstruct the clonal evolution of this tumor, with the emergence of seven resistant subclones harboring distinct GPRC5D alterations (Fig. 1f).Thus, in this patient with a 12p deletion at baseline, talquetamab resistance involved the independent acquisition of second hits in GPRC5D by several subclones, leading to the bi-allelic inactivation of the target.
Tumor response was evaluated according to international consensus criteria 12 .The duration of response was defined as the time between initial response and the first documented evidence of progressive disease per International Myeloma Working Group criteria.
All samples were analyzed by WGS and single-nucleus Multiome, except MM-01-0221-t1 (bulk RNA-seq only) and MM-01-0302 t1 (WGS only).Five other patients with myeloma from the MYRACLE cohort were included for comparison (Supplementary Table 6).They were analyzed by single-nucleus Multiome to examine the expression and chromatin  5 for the complete definition and annotation of deregulated peaks.

Brief Communication
https://doi.org/10.1038/s43018-023-00625-9 accessibility of GPRC5D in MM not treated with talquetamab and by WGS to determine their driver alterations.All BM samples were collected by the Hematology Department of CHU Nantes during diagnosis or follow-up visits.All patients provided signed informed consent to join the study (without compensation).Sex was not considered in the study design.The sex of participants was determined based on self-report and validated by the number of sexual chromosomes in WGS data.Normal plasma cells were also retrieved from the femoral canal of individuals with isolated hip osteoarthritis, who were otherwise healthy and signed informed consent, and analyzed by single-nucleus Multiome.

Myeloma and normal plasma cell purification
Myeloma cells were purified from BM or peripheral blood mononuclear cells using StraightFrom Whole Blood and Bone Marrow CD138 Micro-Beads (Miltenyi, 130-105-961) or FACS-sorted using a BD FACSAria III as CD38/CD319 positive and CD235a/CD3/CD33 negative (antibodies from Becton Dickinson, 562444, 564338, 740785, 555332 and 555450).BM mononuclear cells were purified from control BM samples by Ficoll.Normal plasma cells were enriched by immunomagnetic sorting using anti-CD319 Antibody (Miltenyi, 130-099-575) and then FACS-sorted using the same procedure as myeloma cells.

Whole-genome sequencing and analysis
Genomic DNA was extracted from myeloma cells with a NucleoSpin Tissue kit (Macherey Nagel, 740952).Matched constitutional DNA was extracted from peripheral blood of the same patients with a NucleoSpin Blood kit (Macherey Nagel, 740951) or NucleoSpin Tissue kit (Macherey Nagel, 740952).Tumor and constitutional DNA were sequenced at Integragen (Evry, France) on an Illumina NovaSeq 6000 as paired-end 150-bp reads, with an average depth of ~45-fold for tumors and ~30-fold for matched constitutional control samples, except for the two tumors of MM-01-0288 patient that were sequenced at 135× to identify minor subclones.We used Illumina DRAGEN software (v.3.10.8) to align sequences on the hg38 version of the human genome and call somatic mutations.We then selected single-nucleotide variants (SNVs) and indels with a DRAGEN flag equal to 'PASS', or 'weak_evidence' with Tumor.SQ ≥ 10.To better assess the clonal evolution between t1 and t2 in MM-01-0288, we used BAM-readcount (v.1.0.0, https://github.com/genome/bam-readcount) to quantify the number of mutated reads in both samples for all mutations identified in either t1 or t2.This allowed to rescue variants identified in one sample and present in the other at low variant allele fraction (VAF) hence not considered reliable by the variant caller.Only variants matching the following criteria were finally retained: sequencing depth ≥8 in the tumor and reference sample, VAF ≥ 1% in the tumor with ≥3 variant reads, ≤1 variant read in the control sample and a VAF ratio >5 between the tumor and control samples.We used Palimpsest 13 to identify mutational signatures (from COSMIC database) operative in each sample.We used MANTA software 14 as implemented in DRAGEN to identify somatic structural variants (SVs).To keep only the most reliable events, we selected only SVs with the 'PASS' flag, supported by at least ten reads in the tumor with a VAF ≥10% and ≤1 variant read in the control sample.We used Facets R package (v.0.6.2) (ref.15) to reconstruct copy-number profiles from WGS bam files of tumor and paired non-tumor liver samples.Single-nucleotide polymorphism (SNP) count matrices for tumor and non-tumor samples were obtained by processing the BAM files with snp-pileup.We used the preProcSample function to generate log-R-ratio and B allele frequency data from SNP matrices, ProcSample to estimate the wild-type two-copy state and emcncf to estimate tumor ploidy, purity, allele specific copy-numbers and cellular fraction for each segmented chromosome segment.We used Palimpsest (v.2.0.0) (ref.13) to estimate the cancer cell fraction (CCF), which is the proportion of tumor cells carrying each mutation or SV, from its VAF, taking into account tumor purity and absolute copy number, as previously described 16 .We annotated alterations affecting MM driver genes identified in previous studies 17,18 .We generated CIRCOS plots summarizing driver mutations, copy-number alterations and SVs using a modified version of the RCircos package (v.1.2.2) (ref.19).

Single-nucleus Multiome
Cryopreserved cells were thawed and nucleus permeabilization was performed following recommendations from 10x

Single-nucleus RNA-seq analysis
We filtered the feature-barcode gene expression matrix to keep only high-quality cells (>3,000 UMI counts, >1,000 genes detected and <15% of mitochondrial reads) and expressed genes (detected in three or more cells).Detailed quality control metrics are provided in Supplementary Table 7. Secondary analyses were performed using Seurat package (v.4.3.0)(refs.22).We used SCTransform with default parameters to normalize the filtered UMI count matrix.We generated UMAP visualizations of all cells from MM-01-0288 patient (t1 and t2, n = 14,849), and all cells from the 11 samples analyzed (n = 65,700).We first used the 3,000 most-variant genes to perform principal-component analysis with run-PCA, and ran Louvain graph-based clustering on the 50 first principal components before projecting the resulting data on a two-dimensional UMAP with runUMAP.We used FindNeighbors and FindClusters with a resolution of 0.3 to identify transcriptomic clusters.We used SingleR (v.1.8.1) (ref.23) to annotate immune cell types, with the Single-Cell Tumor Immune Atlas (https://doi.org/10.5281/zenodo.4263972).We removed cells with SingleR annotations other than 'Plasma B cells', 'Proliferative B cells' or 'B cells', as well as cell clusters with low expression of plasma cell markers (CD38, TNFRSF17, MZB1, PECAM1, SDC1, SSR4, TXNIP and XBP1).These cells correspond to non-plasma cells that were not filtered out during sample preparation.We obtained on average 95.2% of plasma cells per sample.We used the InferCNV package (v.1.9.1, https://github.com/broadinstitute/infercnv)with default parameters to reconstruct virtual copy-number profiles and identify tumor subclones with specific copy-number changes.Plasmocytes from healthy BM samples were used as reference cells, and genes with an average read count >0.1 in reference cells were used for the analysis.We used the Seurat FindMarkers function (with a Wilcoxon rank-sum test) for differential expression analyses.Genes with an adjusted P < 0.05 and an absolute log (fold change) > 0.6 were considered significantly differentially expressed.We used enrichR R package (v.3.2) to identify pathways (from 'KEGG_2021', 'Reactome_2022', 'WikiPathways_2019' and 'MSigDB_hallmark_2020' databases) significantly enriched (adjusted P < 0.05) among differentially expressed genes.

Brief Communication
https://doi.org/10.1038/s43018-023-00625-9 of fragments per nucleus and TSS enrichment in each sample, and we applied ad hoc filters (number of fragments >10 3.3 -10 3.7 , TSS enrichment score >12-14, adapted to each sample) to retain only the most reliable nuclei for further analysis.Detailed quality control metrics are provided in Supplementary Table 7.We used ArchR addRepro-duciblePeakSet function to identify chromatin accessibility peaks, with MACS2 (ref.25) method and grouping cells by sample.We used ArchR getMarkerFeatures function to identify peaks with differential accessibility (adjusted P < 0.05 and |log 2 (fold change)| > 1) between cells from the two GPRC5D-silenced relapses (MM-01-0221-t2 and MM-01-302-t2) versus cells from other samples.We used ArchR peak2genelinkage function to identify peaks whose accessibility is correlated to the expression of GPRC5D among the 65,700 cells.

Clonal evolution of MM-01-0288 myeloma
To reconstruct the clonal architecture of MM-01-0228, and its evolution between t1 and t2, we first compared the CCF of somatic mutations between the two time points (Extended Data Fig. 6a).K-means clustering revealed three subsets of mutations: trunk mutations (clonal in both t1 and t2), subclone A (dominant in t1 and minor in t2) and subclone B (minor in t1 but dominant in t2).Among driver alterations only the t(4;14) translocation and ADCY2 mutation were trunk.KRAS G12R, RB1 R787* and CDKN1B N61I belonged to subclone A. HUWE1 P3841L and RB1 L572 fs belonged to subclone B. Finally, SF3B1 R1075K and the seven alterations of GPRC5D were specific to t2 subclones, with CCF ranging from 6.2 to 31.6%.We used SCReadCounts (v.1.3.1)(ref.26) to detect somatic mutations in single-cell data, and we projected the number of mutated reads for each subset of mutations on the snRNA-seq UMAP (Extended Data Fig. 6b,c).As expected, trunk mutations were found in all cells.By contrast, subclone A mutations were restricted to the major cell cluster in t1 and two minor cell clusters in t2.Subclone B mutations were restricted to the minor cell cluster in t1 and three cell clusters in t2.The proportion of cells harboring each mutation subset was consistent with the CCF estimated from WGS data.These data reveal which subclone in t1 gave rise to each subclone in t2, as represented by arrows in Extended Data Fig. 6c.
InferCNV's hierarchical clustering revealed nine copy-number clusters (Extended Data Fig. 8a).Careful examination of WGS and InferCNV data revealed six reliable subclonal copy-number alterations (CNAs) in this patient: four identified by Facets in WGS data (−1p, −14q, −14qtel and +17q) and two (−10p and −21) detected in single-cell data only and clearly related to a specific subclone.We used InferCNV's ChromHMM tool to assign the status of each subclonal CNA in each cluster (Extended Data Fig. 8b).Of note, del14q (present in 77% of t2 cells according to Facets) was poorly detected by InferCNV.We thus used Numbat tool 27 , which takes into account SNP data in addition to expression ratios, to identify this alteration (Extended Data Fig. 8c).Finally, InferCNV clusters with identical copy-number profiles were merged to obtain the final classification in five copy-number subclones (Fig. 1c and Extended Data Fig. 8d).Copy-number subclones matched transcriptomic clusters on the snRNA-seq UMAP and allowed us to refine the clonal architecture reconstructed from somatic mutations (Fig. 2d).Note that subclone cn5 appears as two transcriptomic clusters corresponding to cells in G1 versus cycling cells.Finally, we mapped GPRC5D alterations onto the snRNA-seq UMAP.Point mutations were identified with scReadCounts, whereas reads harboring indels and supporting the deletion of exon 1 were manually identified in the BAM files using the Integrative Genomics Viewer 28 .Each alteration was encountered in a single genetic subclone of the t2 sample: F158fs in cn4, S125fs in cn3 and all others in cn5 (Fig. 2e).Finally, we integrated mutations and CNAs to reconstruct the relationships between genetic subclones, and the timing of acquisition of driver alterations.We used the fishplot package (v.0.5.1) (ref.29) to represent clonal evolution over time in patient MM-01-0288 (Fig. 2f).

Bulk RNA-seq
We performed bulk RNA-seq for myeloma cells of MM-01-0221 (t1 and t2) and MM-01-0302 (t2).Total RNA was purified using direct-zol RNA MicroPrep kit (Ozyme, ZR2060) with on-column DNase treatment according to manufacturer's instructions.Libraries were generated from 200 ng of total RNA using NEBNext Poly(A) mRNA Magnetic Isolation Module (New England Biolabs, E7490S) and NEBNext Ultra II Directional RNA Library Prep (New England Biolabs, E7765S).Libraries were sequenced on a NovaSeq 6000 as paired-end 100-bp reads.Full Fastq files were aligned to the reference human genome hg38 using STAR (v.2.7.9a).We used featureCount to obtain the number of raw read counts associated to each gene in the ENSEMBL database (release-87 gtf file), and we calculated TPM scores (transcripts per millions of mapped reads) by normalizing the count matrix for the library size.We used COMMPASS RNA-seq data (https://research.themmrf.org)to compare GPRC5D expression in MM-01-0221 and MM-01-0302 with a large series of MM (bulk RNA-seq from 859 MM samples), as shown in Fig. 2c.We used two-sided Wilcoxon rank-sum tests to compare the expression of GPRC5D between molecular subgroups in the COMMPASS series.

Statistics and reproducibility
No statistical method was used to predetermine sample size.The sample size (three patients) was chosen based on sample availability.These patients were not selected and correspond to the first three patients who relapsed after talquetamab at CHU Nantes with a BM aspirate sufficiently rich in tumor cells to conduct whole-genome and single-cell sequencing.No data were excluded from the analyses except poor quality cells (as explained in the 'Single-nucleus RNA-seq analysis' section).The experiments were not randomized.Data collection and analysis were not performed blind to the conditions of the experiments.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.Each point represents a cell.They are grouped according to the similarity of their transcriptomic profiles, and colored by sample of origin.c, Projection of somatic mutations identified by WGS onto single-cell RNA-seq UMAPs.For each set of mutations, the total number of mutated reads per cell is represented with a color code.Trunk mutations are detected in all cells.Subclone A mutations are detected in the major subclone of t1 and 2 minor subclones of t2.Subclone B mutations are detected in the minor subclone of t1 and 3 subclones of t2.The proportion of cells harboring each set of mutations is consistent with their CCF estimated from WGS data in each tumor.

Brief Communication
Extended Data Fig. 7 | Clonal and subclonal copy-number changes identified by whole genome sequencing in MM-01-0288.Pangenomic copy-number profiles of MM-01-0288 before (a) and after (b) talquetamab treatment.The two samples display 10 common clonal copy-number changes: +1q, +3q, −4p, −4q, +6p, −6q, −7p, −12p, −20p and −X.One aberration (−1p) is absent in t1 and subclonal in t2, and 3 aberrations (−14q, −14qtel and +17q) are subclonal with different cancer cell fractions (CCF) in each sample.Subclonal copy-number number losses and gains are highlighted in blue and red, respectively.For each sample, the upper panel shows the coverage log-ratio between the tumor and normal samples, the middle panel shows the B Allele Frequency of common polymorphisms, and the bottom panel shows the absolute copy-number estimated by FACETs algorithm.c, Detailed characterization of subclonal copynumber changes, including the detailed coordinates and the cancer cell fraction (CCF, for example proportion of tumor cells harboring the alteration) at t1 and t2.We indicated a CCF '<20%' for alterations visible on the profile but below FACETs detection threshold.
Extended Data Fig. 8 | Identification of 5 copy-number subclones in MM-01-0288.a, Unsupervised classification by hierarchical clustering as implemented in InferCNV (Ward.D method) reveals 9 clusters.b, Proportion of cells within each cluster carrying each of the 6 reliable subclonal CNAs identified in the patient.We used InferCNV HMM predictions to determine the presence of aberrations in each cell, except for the -14q deletion that was idenfied using Numbat tool.c, Numbat identifies in single-cell data a subclone carrying the same 14q deletion identified by FACETs in WGS data.The upper track shows the log(fold-change) of expression in cells of the subclone vs. reference cells, and the bottom track shows the allele frequencies of polymorphisms.d, Assignment of final copy-number subclones.InferCNV clusters with identical copy-number profiles were merged: InferCNV 1 and 7 into cn4, InferCNV 3-6 into cn5.MM-01-0221-t2 and MM-01-0302-t2 (in red) and 5 other MM samples (in blue), together with H3K27Ac (enhancer mark) and H3K4Me3 (promoter mark) ChIP-seq signals in 2 MM samples from BLUEPRINT project.All samples display a major peak of accessibility at GPRC5D promoter, and up to 5 minor peaks in the gene body.All loci are closed in MM-01-0221-t2 and MM-01-0302-t2, as demonstrated by the lack of ATAC-seq reads.The ATAC-seq peak overlapping GPRC5D promoter has both H3K27Ac and H3K4Me3 marks in BLUEPRINT samples, consistent with an active promoter role.Other ATAC-seq peaks match H3K27Ac peaks, indicating their putative enhancer role.

Fig. 1 |
Fig. 1 | Polyclonal talquetamab resistance due to bi-allelic genetic inactivation of GPRC5D.a, CIRCOS plots showing (from outer to inner layer) the driver mutations, copy-number changes and structural variants (abnormal junctions) identified by WGS in the pre-(left) and post-talquetamab (right) samples of patient MM-01-0288.b, GPRC5D mutations detected in the relapse are shown along GPRC5D protein sequence, with a color code indicating mutation consequence.The CCF (proportion of tumor cells harboring the mutation) is indicated below each mutation.c, Virtual copy-number profiles reconstructed from single-nucleus RNA-seq data.Each row represents a cell, with regions of increased (resp.decreased) expression in red (resp.blue).The sample of origin of each cell is indicated on the left.Cells are grouped by their copy-number cluster (cn1 to cn5, right legend) identified by InferCNV tool.Subclonal copy-number changes are annotated and framed in the cn clusters in which they were detected.d, Projection of copy-number subclones onto the single-nucleus RNA-seq Uniform Manifold Approximation and Projection (UMAP).Copy-number subclones nicely correlate with transcriptomic clusters.Note that subclone cn5 is split in two groups corresponding to cell cycle phases.e, Projection of GPRC5D alterations onto the single-nucleus RNA-seq UMAP.Each alteration is represented with a specific color code.As expected, GPRC5D alterations are encountered in t2 only.Each alteration is detected in a single copy-number cluster, indicating the subclone in which it appeared.f, Clonal evolution between pre-treatment sample (t1) and post-talquetamab relapse (t2).Driver mutations, t(4;14) translocation and copy-number changes are indicated.Copy-number subclones at t1 and t2 are colored as in c,d.GPRC5D mutations are colored as in e.

Fig. 2 |
Fig. 2 | Talquetamab resistance due to epigenetic inactivation of GPRC5D locus.a, Clinical history of MM-01-0221 and MM-01-0302 patients.yo, years old.b, Gene expression (RNA-seq, top) and chromatin accessibility (ATAC-seq, bottom) at GPRC5D locus in healthy bone marrow (BM), MM-01-0221 and MM-01-0302 t2 tumor samples.MM-02-0221 and MM-02-0302 t1 samples could not be analyzed by ATAC-seq so another t(11;14) myeloma (MM-01-0286) is shown for comparison.Single-nucleus RNA-seq and ATAC-seq profiles of other healthy BM and MM samples are shown in Extended Data Fig. 10.PC, plasma cells.c, Bulk RNAseq expression of GPRC5D in MM-01-0221 and MM-01-0302 samples as compared to the COMMPASS series (859 myelomas).While the expression in MM-01-0221 at t1 is close to the average in COMMPASS, the expression in MM-01-0221 and MM-01-0302 t2 samples is extremely low.d, Large-scale epigenomic landscape around GPRC5D.Shown is a 500-kb region around GPRC5D with coordinated loss of many chromatin accessibility peaks (highlighted by red arrows).The top shows the three-dimensional interactions identified in B cells by promoter-capture Hi-C 6 .ATAC-seq data are shown for the two GPRC5D-silenced relapses (in red), seven other MM samples (in blue) and two normal plasma cell samples (in green).MM ChIP-seq signals for H3K27Ac (enhancers), H3K36Me3 (transcribed genes) and H3K27Me3 (inactive chromatin) were obtained from the BLUEPRINT database and are depicted below, together with the methylation level in healthy BM.See also Supplementary Table5for the complete definition and annotation of deregulated peaks.
https://doi.org/10.1038/s43018-023-00625-9Extended Data Fig. 3 | Single-cell GPRC5D mRNA expression in MM-01-0288.a, Projection of GPRC5D mRNA expression onto the single-nucleus RNA-seq UMAP of patient MM-01-0288.b, Violin plot showing the distribution of single-nucleus GPRC5D expression in cells from t1 and t2 time points.c, Violin plot showing the distribution of single-nucleus GPRC5D expression in the 5 copynumber subclones.https://doi.org/10.1038/s43018-023-00625-9Extended Data Fig. 4 | Flow cytometry gating.The gating strategy is shown on top and the results for each sample below.Extended Data Fig. 5 | Cell surface protein expression of GPRC5D analyzed by flow cytometry.GPRC5D cell-surface protein expression was analyzed in 2 MM cell lines (a), sorted tumor cells from 4 multiple myelomas (b) and sorted tumor cells from multiple myelomas with genetic (MM-01-0288) or epigenetic (MM-01-0221) inactivation of GPRC5D (c).Celles were labeled using an anti-GPRC5D antibody (red line; 571961 clone; R&D Systems) or an isotype control antibody (gray line; IC0041P clone; R&D Systems) for 60 minutes at 4 °C and analyzed by flow cytometry.Extended Data Fig. 6 | Integration of WGS and single-cell data reveals the clonal evolution of MM-01-0288 a, Clonality of somatic mutations identified by whole genome sequencing in MM-01-0288.Each point represents a mutation, with its cancer cell fraction (CCF) in t1 on the x-axis, and in t2 on the y-axis.Trunk mutations (CCF ~ 1 in both t1 and t2) are highlighted, together with two subclones: subclone A (dominant in t1 and minor in t2) and subclone B (minor in t1 and dominant in t2).Private mutations encountered in only one sample are located on the axes.In addition to these, 4,049 mutations were identified with low CCF in both samples.b, Uniform Manifold Approximation and Projection (UMAP) of the 14,849 cells analyzed from MM-01-0288 t1 and t2 samples.
https://doi.org/10.1038/s43018-023-00625-9Extended Data Fig. 10 | See next page for caption.Extended Data Fig. 10 | Genomic landscape and regulation of GPRC5D in the two post-talquetamab relapses with epigenetic silencing.a, CIRCOS plot showing (from outer to inner layer) the driver mutations, copy-number changes and structural variants identified by whole genome sequencing in the posttalquetamab sample of patient MM-01-0221.b, UMAPs showing the classification of 65,700 cells from 11 samples (2 normal bone marrow samples and 9 multiple myelomas) based on their snRNA-seq profiles.Cells are colored according to their sample of origin (left) or their level of GPRC5D expression (right).c, Chromatin landscape at GPRC5D locus.ATAC-seq signals are showing two normal bone marrow samples (in green), the 2 GPRC5D-silenced post-talquetamab relapses We modified the genes.gtffile used for snRNA-seq analysis by removing genes without an HGNC ID that overlap HGNC genes.This prevents CellRanger-arc from discarding reads overlapping well established protein-coding genes and noncoding RNAs (for example antisense), as CellRanger-arc discards reads mapping to several genes in the gtf file.Our modified gtf included 35,010 genes (versus 36,601 for the default gtf file).