1 Identification of therapy-induced clonal evolution and resistance pathways in minimal

2 residual clones in multiple myeloma through single-cell sequencing

3

4 Jian Cui1,2#, Xiaoyun Li1,2#, Shuhui Deng1,2,3#, Chenxing Du1,2, Huishou Fan4, Wenqiang
5 Yan1,2, Jingyu Xu1,2, Xiaoqing Li5, Tengteng Yu1,2, Shuaishuai Zhang1,2, Rui Lv1,2, Weiwei
6 Sui1,2, Mu Hao1,2, Xin Du5, Yan Xu1,2, Shuhua Yi1,2, Dehui Zou1,2, Tao Cheng1,2, Lugui

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

7 Qiu1,2,6*, Xin Gao1,2*, Gang An1,2,6*
8
1
9 State Key Laboratory of Experimental Hematology, National Clinical Research Center for
10 Blood Diseases, Haihe Laboratory of Cell Ecosystem, Institute of Hematology & Blood
11 Diseases Hospital, Chinese Academy of Medical Science & Peking Union Medical College,
12 Tianjin 300020, China;
2
13 Tianjin Institutes of Health Science, Tianjin 301600, China;
3
14 LeBow Institute for Myeloma Therapeutics and Jerome Lipper Center for Multiple Myeloma
15 Center, Dana-Farber Cancer Institute, Harvard Medical School, Boston, MA;
4
16 Department of Hematology, Affiliated Hospital of Qingdao University, Qingdao, China;
5
17 Shenzhen Second People’s Hospital, The First Affiliated Hospital of Shenzhen University,
18 Shenzhen, China;
6
19 Institute of Multiple Myeloma, Beijing GoBroad Boren Hospital, Beijing, China.
20

21 #These authors contributed equally to this work: Jian Cui, Xiaoyun Li and Shuhui Deng.
22

23 Running title:
24 scRNA-seq of therapy-induced clonal evolution in MM
25

26 *To whom correspondence should be addressed:

27 Gang An, M.D., Ph.D.
28 Department of Lymphoma and Myeloma,
29 State Key Laboratory of Experimental Hematology,
30 Institute of Hematology & Blood Diseases Hospital,
1
31 Chinese Academy of Medical Science & Peking Union Medical College,
32 288 Nanjing Road, Heping District,
33 Tianjin, PRC, 300020.
34 Email: [email protected]
35 ORCID: https://orcid.org/0000-0003-4922-4614
36

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

37 Xin Gao,
38 Professor and Director,
39 Department of Bioinformatics,
40 State Key Laboratory of Experimental Hematology,
41 Institute of Hematology & Blood Diseases Hospital,
42 Chinese Academy of Medical Science & Peking Union Medical College,
43 288 Nanjing Road, Heping District,
44 Tianjin, PRC, 300020.
45 Email: [email protected]
46 ORCID: https://orcid.org/0000-0002-9459-0472
47

48 Lugui Qiu, M.D.

49 Professor and Director,
50 Department of Lymphoma and Myeloma,
51 State Key Laboratory of Experimental Hematology,
52 Institute of Hematology & Blood Diseases Hospital,
53 Chinese Academy of Medical Science & Peking Union Medical College,
54 288 Nanjing Road, Heping District,
55 Tianjin, PRC, 300020.
56 Email: [email protected]
57 ORCID: https://orcid.org/0000-0002-8752-0644
58

59 Conflict of interest statement:

60 The authors declare no potential conflicts of interest with respect to the research, authorship,
2
61 or publication of this article.
62

63 Word count:
64 Text word count: 7220
65 Abstract word count: 247
66 Figure count: 6

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

67 Reference count: 66
68 Translational relevance: 145

3
69 Translational relevance
70 Therapy-induced clonal evolution is one of the most important hindrances toward a cure for
71 multiple myeloma. Through single-cell sequencing of paired diagnostic and post-treatment
72 bone marrow samples, we examined how therapy-induced clonal evolution and resistance
73 pathways allow minimal residual clones to survive induction treatment. We demonstrated
74 three primary trajectories of transcriptional evolution after treatment: clonal elimination in

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

75 patients with undetectable minimal residual disease, as well as clonal stabilization and clonal
76 selection in detectable minimal residual disease patients. We noted a metabolic shift towards
77 fatty acid oxidation in cycling resistant plasma cells, while selective plasma cells favored the
78 NF-κB pathway. Finally, we identified variations in cellular interactions between malignant
79 plasma cells and the tumor microenvironment. Our study characterizes a close link between
80 genetic and non-genetic mechanisms behind the clonal evolution of multiple myeloma.
81 Targeting therapy-induced resistance mechanisms may help to avert refractory disease in
82 multiple myeloma.

4
 83 Abstract
84 Purpose: In multiple myeloma (MM), therapy-induced clonal evolution is associated with
85 treatment resistance and is one of the most important hindrances toward a cure for MM. To
86 further understand the molecular mechanisms controlling the clonal evolution of MM, we
87 applied single-cell RNA-sequencing (scRNA-seq) to paired diagnostic and post-treatment
88 bone marrow (BM) samples.

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

89 Experimental Design: scRNA-seq was performed on 38 BM samples from patients with
90 monoclonal gammopathy of undetermined significance (MGUS) (n = 1), MM patients at
91 diagnosis (n = 19), MM post-treatment (n = 17), and one healthy donor. The single-cell
92 transcriptome data of malignant plasma cells and the surrounding immune microenvironment
93 were analyzed.
94 Results: Profiling by scRNA-seq data revealed three primary trajectories of transcriptional
95 evolution after treatment: clonal elimination in patients with undetectable minimal residual
96 disease (MRD-), as well as clonal stabilization and clonal selection in detectable MRD
97 (MRD+) patients. We noted a metabolic shift towards fatty acid oxidation in cycling-resistant
98 plasma cells (PCs), while selective PCs favored the NF-κB pathway. Intriguingly, when
99 comparing the genetic and transcriptional dynamics, we found a significant correlation
100 between genetic and non-genetic factors in driving the clonal evolution. Furthermore, we
101 identified variations in cellular interactions between malignant plasma cells and the tumor
102 microenvironment (TME). Selective PCs showed the most robust cellular interactions with
103 the TME.
104 Conclusions: These data suggest that MM cells could rapidly adapt to induction treatment
105 through transcriptional adaptation, metabolic adaptation, and specialized immune evasion.
106 Targeting therapy-induced resistance mechanisms may help to avert refractory disease in
107 multiple myeloma.
108

109 Keywords:
110 Single-cell RNA sequencing, multiple myeloma, clonal evolution, transcriptional evolution,
111 tumor microenvironment
112
5
113 Introduction
114 The advent of novel therapeutics, such as proteasome inhibitors (PIs) and immunomodulatory
115 drugs (IMiDs), has significantly enhanced survival outcomes for patients with multiple
116 myeloma (MM) over the past two decades (1). Presently, the treatment objective for MM is
117 transitioning from achieving complete response (CR), traditionally assessed using
118 conventional biochemical tests, to realizing undetectable minimal residual disease (MRD-)

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

119 status, monitored via highly sensitive next-generation flow (NGF) or next-generation
120 sequencing (NGS) technologies (1, 2). A wealth of evidence substantiates the correlation
121 between reduced disease progression or mortality in MM patients and the attainment of
122 MRD- (3). Particularly for high-risk MM patients, achieving and maintaining MRD- status is
123 crucial to counterbalance the negative prognostic implications of adverse cytogenetics (4).
124

125 However, recent studies exploring MRD dynamics have highlighted that only a subset of
126 patients (those who have undergone autologous stem-cell transplant (5) or received
127 continuous maintenance treatment (6)) experience a transition from detectable MRD (MRD+)
128 to MRD-. This underscores the urgency of characterizing the biological properties of
129 persistent MRD tumor cells. Such efforts can unveil new insights into the chemoresistance
130 mechanisms of these cells and potentially disclose hitherto unknown vulnerabilities that can
131 be exploited to eradicate persistent MRD tumor cells (4, 7).
132

133 A high degree of intraclonal heterogeneity is a key feature of MM. Patients with newly
134 diagnosed multiple myeloma (NDMM) invariably harbor multiple clones with distinct genetic
135 backgrounds (8). Moreover, treatments exert varying impacts on intraclonal diversity, leading
136 to the elimination of sensitive clones and the selection of resistant ones (7, 8). While our
137 current understanding of the genetic and non-genetic biology of MRD tumor cells has been
138 predominantly gleaned from bulk-level analysis (9-11). The advent of single-cell technologies,
139 especially multi-omics single-cell technologies, has offered us a novel perspective on
140 understanding the clonal evolution of MM (12). A recent single-cell study has revealed that
141 high-risk major clones observed at relapse are already present at diagnosis as minor,
142 undetectable subclones (13). Moreover, a subgroup of plasma cells (PCs) exhibiting 1q
6
143 alterations, TP53 mutations, and elevated expression of the PHF19 has been identified to be
144 associated with the progression of MM (14). By integrating single-cell RNA sequencing
145 (scRNA-seq) with various sequencing platforms, co-evolution of tumor cells and the tumor
146 microenvironment (TME) has also been explored during progression of MM (15).
147

148 However, two main limitations exist in previous MM clonal evolution studies. Firstly, the

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

149 focus has largely been on comparing clonal architecture at diagnosis and relapse using paired
150 samples (16), with only a few studies investigating clonal evolution occurring early in MRD
151 tumor cells. Secondly, clonal evolution has been primarily explored from a genetic
152 perspective, including the acquisition of new genetic abnormalities or clonal competition and
153 expansion through a Darwinian process (16, 17). Conversely, the role of transcriptional,
154 metabolic, or epigenetic adaptation in clonal evolution has received comparatively limited
155 attention. Despite the widespread acceptance that a tumor cell, or a group of tumor cells with
156 a specific genetic alteration, will gain a competitive advantage to evade therapeutic pressure
157 (18), a growing number of scRNA-seq studies are now focusing on the non-genetic
158 determinants of malignant clonal fitness (19). As such, clonal evolution is progressively
159 transitioning into a multidimensional concept, encompassing both genetic and non-genetic
160 determinants that drive the therapeutic adaptation of clones. Despite this, the relationship
161 between genetic and non-genetic resistance mechanisms in MM remains largely uncharted.
162

163 To fill these knowledge gaps, our study examined the transcriptional dynamics of PCs from
164 diagnosis through to post-treatment. We collected and sequenced a total of 38 paired bone
165 marrow (BM) samples. Our findings revealed three primary trajectories of transcriptional
166 evolution post-treatment: clonal elimination in MRD- patients, as well as clonal stabilization
167 and clonal selection in MRD+ patients. Additionally, we observed disparate resistance
168 pathways in residual malignant PCs (mPCs). A comparison of the genetic and transcriptional
169 dynamics unveiled a strong correlation between genetic and non-genetic factors driving the
170 clonal evolution of MM. Lastly, we identified variations in cellular interactions between
171 mPCs and the TME, with selective PCs exhibiting the strongest interactions with the TME.
172
7
173 Taken together, our study offered a deeper understanding of the clonal evolution and
174 resistance mechanisms of MM. By examining these processes at a single-cell resolution on a
175 large scale, we confirmed the link between genetic and non-genetic mechanisms driving
176 clonal evolution, providing insights into potential therapeutic strategies aimed at disrupting
177 the connections between MRD tumor cells and the TME.
178

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

179 Materials and Methods
180 Human samples and cell preparation.
181 Bone marrow (BM) samples were collected from patients with monoclonal gammopathy of
182 undetermined significance (MGUS) (n = 1), multiple myeloma (MM) patients at diagnosis (n
183 = 19), and a single healthy donor (HD). Paired BM aspirates were acquired after 2-5 cycles of
184 induction therapy from 17 MM patients. The detailed clinical and demographic characteristics
185 of the MGUS and MM patients are presented in Supplementary Table S1. All samples were
186 obtained at the Institute of Hematology & Blood Diseases Hospital, Chinese Academy of
187 Medical Sciences & Peking Union Medical College in Tianjin, China.
188

189 Bone marrow mononuclear cells (BMMCs) were isolated from the BM aspirates using the
190 density gradient centrifugation method with Ficoll-Paque PLUS lymphocyte isolation sterile
191 solution (Cytiva, Sweden). For patients MM1-MM8, CD138+ plasma cells (PCs) were
192 isolated using magnetic-activated cell sorting with CD138+ magnetic beads (Miltenyi Biotec,
193 Paris, France). As indicated by several previous studies and ours, a ≥90% plasma cell purity
194 could be achieved after cell sorting (20-22). All samples were viably cryopreserved in
195 dimethyl sulfoxide at a final concentration of 10%. This study was approved by the local
196 institutional ethics committees under the oversight of the Institute of Hematology & Blood
197 Diseases Hospital, Chinese Academy of Medical Science & Peking Union Medical College.
198 In accordance with the guidelines of the Declaration of Helsinki, all patients and the HD
199 provided informed consent prior to enrollment.
200

201 Assessment of minimal residual disease.

202 Minimal residual disease (MRD) was assessed at the time of paired post-treatment BM
8
203 sample acquisition. Out of the 17 MM patients with paired post-treatment BM samples, 14
204 underwent MRD assessment at this defined time point. MRD was evaluated in accordance
205 with the next-generation flow (NGF) method developed by EuroFlow (23, 24). The limit of
206 detection (LOD) achieved by NGF was calculated for each sample using the formula:
207 (20/number of viable nucleated cells) × 100. A minimum of 2,000,000 nucleated cells were
208 recorded, resulting in a sensitivity threshold ranging between 10-5 and 10-6. MRD detection

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

209 was defined as when the percentage of phenotypically aberrant PCs was equal to or greater
210 than the LOD achieved in the respective sample.
211

212 Single-cell RNA library construction and sequencing using the 10x Genomics platform.
213 Frozen BM cells were rapidly thawed, washed, counted, and resuspended in PBS and 0.04%
214 bovine serum albumin to a final concentration of 1,000 cells per μl. In this study, samples
215 were sequenced in two batches. Samples MM1 to MM8 constituted the first batch, while
216 samples MM9 to MM19 comprised the second batch. The Single Cell 3' Library Kit V3 (10x
217 Genomics, CA, USA) was used for the cDNA synthesis of individual cells. These cells were
218 then barcoded and ready for library construction following the manufacturer's instructions.
219 Upon preparation, the indexed cDNA libraries underwent sequencing on an Illumina NovaSeq
220 6000 platform using a paired-end mode. Novogene Co., Ltd. (Beijing, China) performed the
221 sequencing according to the manufacturer's guidelines, targeting an approximate sequencing
222 depth of 20,000 reads per cell.
223

224 Data processing and quality control of scRNA-seq data.

225 The processing of the raw sequencing data marked the initial stage, which involved alignment
226 to the GRCh38 (UCSC) human reference genome utilizing the CellRanger pipeline (version
227 3.0.2, 10x Genomics, CA, USA). Default parameters were employed for the alignment
228 process. Post-alignment, quality control procedures were conducted on the datasets using the
229 Seurat package (version 3.1.1) (25). The first step was the removal of doublet cells using
230 DoubletFinder (version 2.0.3) (26). Subsequently, low-quality cells were filtered out
231 according to the following criteria: (i) numbers of detected UMIs and genes (log10 scaled)
232 that fell outside the median ± 3 median absolute deviation (MAD) range; (ii) mitochondrial
9
233 gene proportions below the median + 3 MAD. Quality control was applied separately to
234 CD138+ PCs and BMMCs, considering factors such as UMI count, gene count, and the
235 mitochondrial ratio. After the exclusion of low-quality cells, the two filtered datasets were
236 merged. Genes that were expressed in at least five cells were selected for downstream
237 analysis.
238

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

239 Dimensionality reduction, cell clustering, and visualization.
240 After the elimination of low-quality single cells, the raw expression matrix was normalized
241 using the NormalizeData function. The ScaleData function was then applied to mitigate
242 variations in gene expression from cell to cell. Principal component analysis (PCA) was used
243 for linear dimensionality reduction, targeting the top 2,000 variable genes. To identify the
244 appropriate number of principal components for subsequent clustering, we used the
245 ElbowPlot function. Next, we applied the FindNeighbors function to calculate shared nearest
246 neighbor graphs, and the Louvain algorithm was used, alongside uniform manifold
247 approximation and projection (UMAP) embedding in two-dimensional space, to cluster the
248 single cells. For the clustered analysis of all PCs depicted in Fig. 1E, we did not perform any
249 integration preprocessing. For all other cells, we applied the Harmony package (version 1.0)
250 (27) for integration prior to analysis, using the following parameters: lambda = 1, theta = 1.
251

252 Cell types were determined by evaluating the average expression levels of established
253 markers. These included SDC1, IGHG1, and MZB1 for PCs; CD79A, CD79B, and MS4A1
254 for B cells; S100A8, S100A9, and LYZ for myeloid cells; GNLY, NKG7, CD3D, and CD3E
255 for NK cells and/or T cells; CD34 for hematopoietic stem and progenitor cells (HSPCs); and
256 HBA1, HBA2, HBB, and PF4 for erythrocytes and/or megakaryocytes, as previously
257 described (28-32).
258

259 Following this, PCs were isolated based on the expression of known marker genes such as
260 SDC1, MZB1, and XBP1. A reclustering process was then implemented. The methods
261 described above were used to cluster PCs into 22 subsets from both HD and patients.
262 Malignant PCs from each patient, both pre- and post-treatment, were integrated for
10
263 subsequent analysis.
264

265 Identification of DEGs and pathway enrichment analysis.

266 In order to identify differential expressed genes (DEGs), we used the FindMarkers or
267 FindAllMarkers functions, using normalized data and setting parameters such as test.use =
268 "wilcox" and logfc.threshold = Log1.5. P-values were adjusted using the Bonferroni

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

269 correction method, which takes into account the total number of genes in the dataset. DEGs
270 with adjusted p-values below 0.05 were considered significant and chosen for subsequent
271 analysis.
272

273 For the differential analysis depicted in Fig. 3, we used a pct threshold of 0.6 to identify
274 DEGs across patients. For other differential analyses, a min.pct threshold of 0.3 was used.
275 These DEGs were then uploaded to the Metascape (33) website (https://metascape.org), and
276 default parameters were used to perform pathway enrichment analyses on the 50 hallmark
277 pathways (34).
278

279 Estimation of CNA profiles in PCs.

280 Copy number alteration (CNA) profiles in individual plasma/myeloma cells were identified
281 using the InferCNV package (version 1.7.1), which calculates the average expression of
282 adjacent genes over large genomic regions. PCs derived from HDs were used as the reference,
283 and we profiled the CNA profile in PCs of every patient individually.
284

285 To identify somatic large-scale CNVs, the gene count matrix was used as the input with the
286 following parameters: cluster_by_groups = False, cutoff = 0.1, analysis_mode = “subclusters”,
287 BayesMaxPNormal = 0.25. This allowed us to identify regions of the genome where there
288 were significant increases or decreases in gene copy numbers, which can often be indicative
289 of cancer-associated genetic alterations.
290

291 Analysis of cell differentiation states.

292 Developmental potential and cellular differentiation states in individual plasma/myeloma cells
11
293 per patient were analyzed using the CytoTRACE package (version 0.3.3) (35), using default
294 parameters.
295

296 Cell-cycle score.

297 To determine the cell-cycle phase of each cell, we employed the CellCycleScoring function in
298 the R package Seurat (version 3.1.1) (25). This allowed us to assign a cell cycle phase to each

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

299 individual cell based on its gene expression profile.
300

301 Calculating the proportion of Lambda+ and Kappa+ PCs.

302 The analysis of the proportion of Lambda+ and Kappa+ PCs was conducted as outlined in
303 Andor et al. (36), albeit with some tailored modifications to the algorithm. To ascertain the
304 expression of immunoglobulin Kappa (IGK) and Lambda (IGL) in PCs, IGK and IGL
305 transcripts were quantified. For each individual cell, 'm' was designated as the quantity of
306 IGKL transcripts and 'n' as the highest quantity of transcripts from the three functional IGLC
307 isotypes (IGLC2, IGLC3, and IGLC7). We calculated the absolute value of Log2(m/n) and
308 retained cells that exceeded 1 for further examination. Cells were categorized as Kappa+ if m >
309 n, and Lambda+ if m < n. Finally, the proportion of Lambda+ cells was calculated as L/(L +
310 K), with 'L' representing the count of Lambda+ cells and 'K' the count of Kappa+ cells.
311

312 Motif enrichment and regulatory network.

313 To explore the specific gene regulatory networks affiliated with PCs, we employed the
314 pySCENIC package (version 0.11.2) (37) in conjunction with the cisTarget database. The
315 gene regulatory network of PCs was assembled using default parameters. To identify
316 differentially activated regulons between cell groups, we utilized Wilcoxon rank-sum tests
317 with 'wilcox.test' and 'p.adjust' for multiple hypothesis correction. Regulators meeting the
318 following criteria were considered significant and selected: fold change > 1 between resistant
319 (selective, cycling resistant group, or other resistant cells) and sensitive cells; adjusted p-value
320 < 0.05 with Bonferroni correction; and regulon frequency > 0.5 in the resistant cells.
321

322 Cross-referencing with myeloma datasets.

12
323 The gene expression from the MMRF CoMMpass dataset was obtained from the publicly
324 available database TCGA. And the other three gene expression datasets along with
325 corresponding clinical information of MM patients were obtained from the publicly available
326 database Gene Expression Omnibus (GEO), under the accessing number of GSE2658,
327 GSE9782 and GSE57317. We then applied gene set variation analysis (GSVA) package
328 (version 1.48.2) (38) to calculate the enrichment score of the cycling-resistant or selective

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

329 gene signature for individual patients in each dataset. A univariate Cox proportional hazard
330 regression model was the implemented to determine the association of the cycling-resistant or
331 selective gene score with overall survival in each cohort. Hazard ratio (HR) from the
332 univariate Cox regression analysis was used to determine the protective (HR < 1) and risky
333 genes (HR > 1). In the Cox univariate analysis, the MMRF CoMMpass dataset was set as the
334 test set, while the other three GEO datasets served as validation sets. The results of these Cox
335 univariate regression analyses in each dataset were independent of each other.
336

337 Receptor-ligand interactions.

338 We analyzed intercellular interactions between cell types using the CellPhoneDB package
339 (version 2.1.4) (39), leveraging ligand-receptor co-expression. The interaction score refers to
340 the total average of the mean expression value of a single ligand-receptor partner in the
341 corresponding interacting cell type. We merged the single-cell dataset of MM14 pre- and
342 post-treatment and normalized the raw expression matrix using the NormalizeData function in
343 Seurat. We then used the normalized counts as input for subsequent analysis, employing
344 default parameters. In this study, we focused on the enriched cytokine/receptor interactions in
345 selective plasma cells or resistant plasma cells and selected the cytokine/receptor interactions
346 associated with highly significant (P value < 0.05) and highly expressed (exp_value > 0.1)
347 cell–cell interaction pairs in selective plasma cells compared to sensitive plasma cells or
348 resistant plasma cells compared to sensitive plasma cells.
349

350 Statistical analyses.

351 All statistical analyses and graph generation were conducted using R (version 3.5.1, R
352 Foundation, Vienna, Austria). All statistical tests employed in this study were two-sided. The
13
353 identification of DEGs was performed using a two-sided Wilcoxon rank-sum test. The
354 Bonferroni correction method was applied to control the false discovery rate (FDR) and to
355 correct p-values for multiple testing. An adjusted p-value less than 0.05 was considered
356 statistically significant. The Kaplan-Meier method was utilized for survival data analysis, and
357 differences in survival were evaluated using the log-rank test. HR and 95% CI were calculated
358 using univariable Cox regression models.

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

359

360 Data sharing statement:

361 The scRNA-seq data presented in this study have been deposited to the Genome Sequence
362 Archive in the National Genomics Data Center (GSA; https://ngdc.cncb.ac.cn/gsa/), under
363 BioProject ID: PRJCA018851 (https://ngdc.cncb.ac.cn/bioproject/browse/PRJCA018851),
364 and Accession ID: HRA005218 (https://ngdc.cncb.ac.cn/gsa-human/browse/HRA005218).
365 Please note that any data usage must fully comply with the Regulations on Management of
366 Human Genetic Resources in China. The scRNA-seq matrix data from the BM of eight HDs
367 were obtained from the Human Cell Atlas database (census of immune cells,
368 [https://data.humancellatlas.org/explore/projects/cc95ff89-2e68-4a08-a234-480eca21ce79]).
369 The scRNA-seq matrix data of PCs utilized for external validation of normal-like PCs can be
370 accessed from the Gene Expression Omnibus (GEO) database with the accession numbers
371 GSE161195 and GSE117156. The bulk RNA-seq data of BM samples, along with clinical
372 data from 858 NDMM samples, were procured from the TCGA (https://tcga-data.nci.nih.gov/)
373 Multiple MMRF CoMMpass dataset. The bulk RNA-seq data used for survival analysis can
374 be found in the GEO database with the accession numbers GSE2658, GSE9782, and
375 GSE57317. All other relevant data supporting the key findings of this study are available
376 within the article and its Supplementary Information files or from the corresponding author at
377 [email protected] upon reasonable request. All non-commercial reagents used in this
378 paper are available from the corresponding author at [email protected] upon reasonable
379 request.
380

381 Ethics statement:

382 All patients included in this study were enrolled in the National Longitudinal Cohort of
14
383 Hematological Diseases in China (ClinicalTrials.gov identifier: NCT04645199). In addition,
384 the study complied with the Declaration of Helsinki and was approved by the Ethics
385 Committee of the Institute of Hematology and Blood Diseases Hospital, Chinese Academy of
386 Medical Science & Peking Union Medical College (Certificate: IIT2020023-EC-1).
387

388 Results

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

389 Patients, treatments, and single-cell transcriptional landscape in HDs and MM patients.
390 To comprehensively investigate the transcriptional dynamics of PCs throughout treatment, we
391 leveraged scRNA-seq to dissect the transcriptional heterogeneity, subclonal structure, clonal
392 evolution of transcriptional clusters and genetic subclones, and cellular interactions in MM at
393 diagnosis and post-treatment. Our study sample set included 38 BM samples from patients
394 with MGUS (n = 1), MM patients at diagnosis (n = 19), MM post-treatment (n = 17), and one
395 HD (Fig. 1A; Table S1).
396

397 We analyzed 161,403 CD138+ PCs and bone marrow mononuclear cells (BMMCs) derived
398 from HD, MM pre-, and post-treatment samples, all of which passed stringent quality controls
399 (Fig. S1A, S1B; Table S2). Additionally, we integrated our dataset with the scRNA-seq
400 dataset from eight HDs from the Human Cell Atlas (HCA) (40). After normalizing the gene
401 expression matrix, we identified six major cell subsets in the BM using graph-based clustering
402 (Fig. S1C, S1D). We observed a distinct separation of plasma/myeloma cells from immune
403 cells, with notable enrichment of CD138+ cells in the MM TME (Fig. 1B-1D). In total, we
404 profiled 76,036 PCs (median = 902) and analyzed a number of immune cells per sample
405 ranging from 839 to 10,137, with a median of 3,059 cells (Table S3).
406

407 Single-cell transcriptional profiles reveal patient-specific patterns and common driver
408 events.
409 To delve deeper into the inter- and intra-patient tumor heterogeneity in MM, we carried out
410 unsupervised clustering for samples with a sufficient number of PCs (≥ 30 PCs). This process
411 yielded 22 clusters of PCs. Interestingly, most PCs derived from HDs were clustered into
412 Cluster 1 (C1), a group that also contained PCs from various patients. Consequently, PCs in
15
413 C1 were identified as HD/normal-like PCs (nlPCs) (Fig. 1E). We observed significant
414 transcriptional differences among patients, with the majority of PC clusters being composed
415 exclusively of cells from a single patient. However, there were some exceptions: Cluster 15
416 (C15), for instance, contained PCs from two patients with distinct translocations (MM12 with
417 t(6;14), and MM13 with t(4;14)) (Fig. 1F; Table S1). Furthermore, longitudinal samples from
418 the same patients were grouped into separate clusters (MM5: C2 and C11; MM6: C5 and C19)

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

419 (Fig. 1F).
420

421 Despite the considerable inter-patient tumor transcriptional heterogeneity, we noticed

422 commonly overexpressed driver genes in mPCs across previously defined MM subgroups
423 (41). The expression of translocation target genes, including CCND1, FGFR3 and NSD2,
424 CCND2, MAFB, MAF, and CCND3, correlated with the translocations detected by interphase
425 fluorescence in situ hybridization (iFISH) (Fig. 1G; Table S1). In addition, LAMP5,
426 previously reported to be upregulated in MM patients with osteolytic lesions (42), was found
427 to be overexpressed in nine out of 18 patients (Fig. 1G).
428

429 nlPCs is a transitional state of PCs from normal PCs to mPCs.

430 In our present study, we identified a PC cluster (C1) that encompassed PCs from HDs and
431 multiple MM patients. Specifically, 9.8% of PCs from MM2 pre-treatment (MM2_pre)
432 (339/3,449), 100% of PCs from MM9_pre, and 98.5% of PCs from MGUS were categorized
433 into this subset (Table S4). To further explore this, we initially characterized the clonotypic
434 identity of PCs within this cluster. We analyzed the relative expression of the immunoglobulin
435 κ and λ light chain variable regions (IGKV and IGLV, respectively), coupled with the
436 immunoglobulin heavy chain variable region (IGHV) for every individual cell. Our findings
437 revealed diverse immunoglobulin sequences from HD PCs. However, we noticed a specific
438 immunoglobulin clonotype for some MM patients, including MM2_pre, MM3 post-treatment
439 (MM3_post), and MM9_pre (Fig. S2A-S2C). To further validate the monoclonal nlPCs in C1,
440 we next determined the light chain restriction of PCs in each sample (see Supplementary
441 Methods) (36). For HD PCs, the frequency of immunoglobulin lambda+ PCs ranged from
442 35.7% to 46.2%. For mPCs from each MM patient, the frequency of immunoglobulin
16
443 lambda+ PCs fell either below 2.3% or exceeded 85.5%. Conversely, nlPCs from MM2_pre,
444 MM3_post, and MM9_pre exhibited an exclusive light chain assignment of 25.8%, 28.0%,
445 and 1.4%, respectively (Fig. S2D). The difference in light chain restriction between MM2_pre
446 and MM9_pre suggested the potential admixture of normal cells within the nlPCs of
447 MM2_pre. Collectively, this light chain restriction for MM9_pre supported our hypothesis
448 that, despite clustering with HD PCs, at least a portion of the MM PCs within the seemingly

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

449 healthy cluster were clonal PCs.
450

451 To further understand the transcriptional differences between nlPCs and mPCs, we
452 subsequently performed unsupervised clustering on the nlPCs and mPCs from MM2_pre. Our
453 results unveiled more transcriptional clusters than copy number variation (CNV) clones.
454 Moreover, nlPCs and mPCs were distinctly separated, while the transcriptome-based
455 clustering displayed a high degree of consistency with CNV clones and cell groups (Fig. 2A).
456 Further analysis revealed unique CNA profiles among three CNV subclones, with S1 and S3
457 demonstrating similar CNA profiles, and S2 exhibiting a seemingly normal CNA profile (Fig.
458 2B). In addition, cellular differentiation states inferred using CytoTRACE indicated a less
459 differentiated state in mPCs and a more differentiated state in nlPCs (Fig. 2C). Further
460 analysis, which included HD PCs, indicated an intermediate differentiated state of nlPCs (Fig.
461 S2E). DEG analysis in nlPCs and mPCs from MM2_pre compared to HD PCs revealed 55
462 shared upregulated DEGs. These genes included some previously described MM-specific
463 targets, such as SLAMF7 (43) and MCL1 (44) (Fig. 2D; Table S5). Pathway enrichment
464 analysis for the shared upregulated DEGs demonstrated significant enrichment for the NF-κB
465 pathway (Fig. 2E). On the other hand, compared to nlPCs, mPCs showed upregulated
466 expression of endoplasmic reticulum (ER) and unfolded protein response (UPR) pathway
467 genes PPIB, mitochondrial stress genes COX7C, COX6B1, and a previously reported
468 MM-related resistance gene MIF (45) (Fig. 2E; Table S5). Furthermore, activation of the
469 MYC signaling pathway was observed in mPCs (Fig. 2F). Selected MM-related driver genes,
470 including ITGB7, LAMP5, KRAS, and BRAF, were expressed at the highest levels in mPCs,
471 while nlPCs only exhibited low-level expression of these genes (Fig. 2G). Analysis of all
472 nlPCs grouped in the C1 cluster demonstrated similar seemingly normal CNA profiles (Fig.
17
473 S2F), while the expression of MM-related genes was exclusively detected in patient-derived
474 nlPCs (Fig. 2H). Further analysis of nlPCs from MM9_pre, MM3_post and MGUS also
475 demonstrated similar seemingly normal CNA profiles, whereas mPCs from MM3_post
476 displayed various loci involved by CNV (Fig. S3). Therefore, nlPCs might represent a
477 transitional state of PCs, evolving from normal PCs to mPCs.
478

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

479 Finally, we validated our identification of nlPCs using two external scRNA-seq datasets (32,
480 46). Our analysis confirmed an exclusive heavy chain assignment in MM12 in the external
481 datasets and validated the abnormal expression of MM-related driver genes in nlPCs from
482 several patients (Fig. S4). In conclusion, we identified a cluster of nlPCs present in several
483 patients. This group of cells is most frequently seen when patients carry only minor clones of
484 cytogenetic abnormalities (MM2_pre, MM3_post, and MM9_pre) or when patients do not
485 exhibit cytogenetic abnormalities (MM12 in the external validation cohort).
486

487 Several distinguishing features were observed in this cluster: it lacked secondary genetic
488 events as shown in the CNA profile, exhibited monoclonal immunoglobulin expression or
489 light chain restriction, and despite clustering with HD PCs at the single-cell level, the nlPCs
490 demonstrated significant transcriptional alterations. Our results suggested that nlPCs
491 represented a transitional state of PCs, transitioning from normal PCs to mPCs. Given the
492 multi-step disease progression characteristic of MM, such nlPCs might represent a potential
493 target for early intervention in MM and warrant further investigation.
494

495 Longitudinal single-cell analysis of MM patients pre- and post-treatment reveals

496 transcriptional evolution and distinct resistance pathways.
497 To gain further insights into the transcriptional evolution and resistance pathways of residual
498 mPCs, we performed unsupervised clustering on the mPCs from each patient separately,
499 specifically focusing on those with sufficient post-treatment PCs (≥ 30 PCs) (Fig. S5). In
500 accordance with the NGF MRD data (Table S1), for patients who achieved MRD- status
501 (sensitivity threshold at 10-5) post-treatment (MM4, MM7, and MM11), we observed only a
502 very low percentage of residual mPCs among all PCs. Conversely, for MRD+ post-treatment
18
503 patients, our results pointed to two distinct patterns: one where 50% of PCs in the BM
504 post-treatment were mPCs, and the other where almost all the PCs were residual mPCs (Fig.
505 S6A).
506

507 However, the percentage of mPCs within the total PC population does not fully reflect the
508 transcriptional changes that occur in mPCs. Thus, we compared the dynamic shifts in mPC

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

509 subclusters before and after treatment. In doing so, we identified two patterns of
510 transcriptional evolution in nine MRD+ patients. The transcriptional selection was observed
511 in four patients (MM2, MM5, MM6, and MM14), with the major transcriptional cluster
512 pre-treatment becoming a minor cluster, while a minor cluster evolved into a predominant
513 cluster post-treatment. In contrast, transcriptional stabilization was observed in the remaining
514 five MRD+ patients (MM3, MM10, MM16, MM17, and MM19) (Fig. 3A; Fig. S5).
515

516 Thus, we categorized mPCs into three groups based on their response to treatment: sensitive
517 PCs, such as those in clusters C0, C1, C3, and C5 in MM5, which were nearly eliminated
518 following treatments; resistant PCs, which did not respond or only partially responded to
519 treatment, such as those in clusters C0, C1, C2, and C4 in MM19; and selective PCs, which
520 showed a growth advantage and became the predominant cluster post-treatment, including C2
521 in MM5, C1 in MM6, and C1 in MM14 (Fig. S6B, S6C).
522

523 Although most of the resistant clusters remained in a state of cell-cycle arrest post-treatment,
524 we identified a cell-cycle-related resistant cluster in five out of seven patients who received
525 bortezomib and lenalidomide in combination with dexamethasone treatment (Fig. 3B). To
526 delve deeper into these cycling-resistant clusters, we carried out a DEG analysis between HD,
527 sensitive, and cycling-resistant PCs. Interestingly, our results demonstrated that
528 cycling-resistant PCs highly expressed several genes previously linked to myeloma drug
529 resistance, such as CKS1B, STMN1, TUBA1B, and TYMS (47-50). We also noted high
530 expression of PHF19, a gene recently implicated in the malignant progression of MM and PC
531 leukemia (14).
532
19
533 The pathways enriched in cycling-resistant cells were associated with UPR, hypoxia, and
534 cell-cycle/proliferation pathways. Notably, in line with recent reports that indicate an
535 upregulation of antioxidant gene programs in cycling persister cells (51), we found that
536 cycling-resistant cells also highly expressed MIF, a key antioxidant gene known to mediate
537 MM cell resistance to PIs (45) (Fig. 3C, 3D; Table S6). However, our analysis of
538 noncycling-resistant PCs highlighted only a few upregulated genes, such as S100A4,

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

539 TXNDC5, and RPS10 (Fig. S6D, S6E; Table S7).
540

541 Compared to both HD PCs and sensitive PCs, the genes displaying upregulation in selective
542 PCs were found to be significantly enriched in pathways such as NF-κB, as demonstrated by
543 genes like FOSB, JUN, and EGR1. Furthermore, they were also enriched in the anti-apoptotic
544 pathway, exemplified by genes such as MCL1. The gene PPP1R15A, crucial for controlling
545 cell viability amid protein folding stress (52), was also observed to be upregulated in selective
546 PCs. Notably, this gene has been previously associated with the progression from MGUS to
547 MM (53). Furthermore, CD74, which is the primary surface receptor for MIF, was
548 significantly upregulated in selective PCs (Fig. 3E, 3F; Table S8). As previous studies have
549 suggested that stimulation of surface CD74 contributes to NF-κB pathway activation (54), we
550 thus hypothesized that CD74 played an important role in the interaction between selective
551 cells and the TME.
552

553 Next, we investigated transcription factors (TFs) that could regulate different drug resistance
554 pathways in cycling-resistant and selective cells. Our analysis indicated that EZH2, E2F1, and
555 E2F2, all of which are cell-cycle/proliferation-regulating TFs, were upregulated in
556 cycling-resistant PCs (Fig. 3G). Conversely, the genes highly expressed in selective PCs were
557 regulated by TFs such as FOS and JUN. It has been reported that the TF AP-1, formed by
558 FOS and JUN, plays a crucial role in enhancing survival but decreasing the proliferation of
559 MM cells (15) (Fig. 3G).
560

561 Lastly, we compared the DEGs between selective PCs and cycling-resistant PCs (Table S9).
562 We found that selective PCs were enriched for pathways including NF-κB, apoptosis and P53
20
563 pathways, while cycling-resistant PCs were enriched for pathways including MYC, E2F target,
564 and UPR pathways (Fig. 3H).
565

566 Predictive value of the gene signature of cycling-resistant PCs.

567 We proceeded to analyze whether the molecular signatures that delineate cycling-resistant
568 PCs (the cycling-resistant signature) and selective PCs (the selective signature) correlate with

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

569 patient outcomes. We employed Gene Set Variation Analysis (GSVA) to calculate the
570 expression of the cycling-resistant and selective signatures in 858 NDMM patients using the
571 independent dataset from the Myeloma Research Foundation (MMRF) CoMMpass dataset.
572 Survival analysis indicated that patients with a high cycling-resistant score had significantly
573 shorter survival times compared to those with a lower score. The median overall survival
574 (mOS) was 39.0 months for the high-score group, compared to an unreached median for the
575 low-score group, with a hazard ratio (HR) of 2.94 and 95% confidence intervals (CI) of
576 2.23-3.88 (P < 0.0001) (Fig. 4A). However, it was observed that patients with a high selective
577 score had a longer OS compared to those without a high cycling-resistant score, although both
578 medians were not reached, and the hazard ratio was 0.56, with a 95% CI of 0.39-0.81 (P =
579 0.0015) (Fig. 4B).
580

581 The adverse prognostic value of the cycling-resistant signature was further validated by
582 utilizing three independent datasets (GSE2658, GSE9782, and GSE57317) consisting of 868
583 MM cases (Fig. 4C). And our results indicated a slightly longer OS for patients with a high
584 selective score compared to those without a high selective score, though the difference did not
585 reach statistical significance in the Log-rank analysis (Fig. 4D).
586

587 Longitudinal single-cell analysis of MM patients pre- and post-treatment reveals genetic
588 evolution and different patterns of metabolic adaptation.
589 To gain deeper insights into the interplay between genetic and non-genetic evolution, we
590 applied inferCNV (55) to delineate the clonal architecture of genetic events in patients. A
591 representative patient, MM5, exhibited branching evolution (16), with analysis of pre- and
592 post-treatment mPCs identifying six CNV subclones among the eight transcriptional clusters
21
593 (Fig. 5A). Consistent with iFISH results (Table S1), we observed a common 13q deletion
594 (13q-) in all six CNV subclones (Fig. 5B).
595

596 To investigate the clonal evolution pattern in MM5, we considered the clonal structure of each
597 CNV subclone. Early genetic events were found in all or most of the CNV subclones, while
598 later events occurred in only some of the subgroups. Additionally, ancestral malignant

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

599 subclones displayed simpler clonal architectures, while descendant subclones exhibited more
600 complexity. Specifically, in MM5, 13q- was identified as the earlier cytogenetic event,
601 whereas 1p- was considered a later event. Moreover, we found that S2, S1, S3, S5, and S6
602 shared similar CNA profiles, with S1, S3, S5, and S6 acquiring some new cytogenetic
603 abnormalities compared to S2. In contrast, the clonal architecture of S4 differed slightly from
604 the S2 branch. Thus, the five main CNV subclones of MM5 at diagnosis could be divided into
605 two branches (S4, S1/S2/S3/S5/S6) (Fig. 5C).
606

607 Further analysis revealed a high degree of concordance between transcriptional clusters and
608 genetic subclones, such as C2 (transcriptional cluster) and S4 (genetic subclone), or between
609 C4 (transcriptional cluster) and S6 (genetic subclone) (Fig. 5D). This observation suggested a
610 strong link between genetic and non-genetic mechanisms driving the clonal evolution of MM.
611 Similarly, the transcriptional selection observed in MM5 post-treatment could also be
612 interpreted from a genetic perspective. Notably, two groups of tiny subclones at diagnosis,
613 namely S4 and S6, became the dominant subclones post-treatment, whereas S1, S2, and S3
614 were virtually eliminated, and S5 was only partly preserved (Fig. 5E).
615

616 Viewed through the genetic lens of clonal evolution, S4 (corresponding to C2) emerged as the
617 predominant major CNV subclone post-treatment, followed by S6 (corresponding to C4).
618 Interestingly, MM5 exhibited a significantly shorter survival compared to other patients with
619 MRD+ status, with an OS of only 10 months. We hypothesized that both the non-genetic
620 resistance mechanisms of C2 and C4 contributed to the inferior outcome in MM5. To
621 investigate this further, we inferred the cellular differentiation states using CytoTRACE,
622 revealing a more differentiated state in C2 and a less differentiated state in C4. However, both
22
623 C2 and C4 displayed higher differentiation compared to other transcriptional clusters (Fig.
624 5F).
625

626 Further DEG analysis between C2 and C4 clusters highlighted significant differences in gene
627 expression patterns. C2 showed high expression of genes related to the "Hypoxia" pathway,
628 such as PPP1R15A and ATF3, while C4 exhibited elevated expression of genes involved in

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

629 "Oxidative phosphorylation", including ATP6V0B, UQCRB, COX6C, and UQCRH.
630 Additionally, the immunotherapeutic target TNFRSF17 (also known as BCMA) was highly
631 expressed in the C4 cluster (Fig. 5G, 5H; Table S10). These distinct patterns of metabolic
632 adaptation in C2 (S4) and C4 (S6) clusters suggested their involvement in promoting
633 prosurvival pathways post-treatment and potentially offering novel therapeutic opportunities
634 to target residual malignant cells.
635

636 Similar genetic selection was observed in MM6 and MM16 (Fig. 5I; Fig. S7), while MM10,
637 MM16, MM17, and MM19 exhibited genetic stabilization (Fig. 5I; Fig. S8). In summary, our
638 findings demonstrated a close link between the genetic landscape and non-genetic
639 mechanisms driving myeloma cell treatment resistance. Moreover, the distinct patterns of
640 metabolic adaptation played crucial roles in driving the clonal selection of residual malignant
641 cells.
642

643 TME remodeling after treatment.

644 To delve deeper into the TME remodeling after treatment, we analyzed 73,424 BMMCs
645 derived from our HD, MM pre-, and post-treatment samples (MM09-MM17, MM19), all of
646 which passed stringent quality controls, identifying 17 major cell types (Fig. 6A, 6B). Notably,
647 we observed significant differences in cellular compositions between the two disease statuses.
648 Consistent with previous studies, MM patients displayed trends towards increased proportions
649 of natural killer (NK) cells, CD16+ monocytes, and CD8+ memory/effector cells
650 pre-treatment compared to HDs, while the proportions of B cells were decreased in MM
651 pre-treatment (28). Intriguingly, a substantial increase in the proportion of CD14+ monocytes
652 was observed in MM post-treatment compared MM at diagnosis (Fig. 6C).
23
653

654 To understand the functional implications of these changes, we conducted a DEG analysis on
655 CD14+ monocytes post-treatment. The results revealed significantly elevated expression of
656 chemokines, such as CCL3, CCL4, and CXCL2, by CD14+ monocytes (Fig. 6D; Table S11).
657 This heightened expression of chemokines suggested an enhanced recruitment effect of
658 CD14+ monocytes on T cells. Conversely, genes highly expressed by CD16+ monocytes

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

659 pre-treatment were primarily related to the "Interferon alpha response" pathway, including
660 IFI6, IFL44L, and IFITM1 (Fig. 6E; Table S12), indicating that anti-myeloma treatment
661 partially mitigated the inflammation in the TME (56).
662

663 Taken together, our results indicated that the treatment not only remodeled the composition of
664 the MM TME but also altered the transcriptional status of immune cells, reflecting a complex
665 interplay between the immune system and myeloma progression.
666

667 Differential cellular interactions between mPCs and TME cells.

668 Considering the potential immunosuppressive effects of the TME on providing shelter for
669 MRD cells from anti-myeloma treatment (56), we conducted a detailed investigation of the
670 cellular interactions between TME immune cells and residual mPCs using the accumulated
671 ligand-receptor interaction database CellPhoneDB (39). Consistent with previous studies (29),
672 our findings highlighted the strongest interactions between mPCs and myeloid cells,
673 particularly monocytes and dendritic cells (DCs) (Fig. 6F, 6G). Further examination revealed
674 even stronger interactions between selective mPCs and TME immune cells, compared to
675 cycling-resistant and/or sensitive mPCs, mediated by APP/CD74, COPA/CD74,
676 FAM3C/HLA-C, MIF/CD74, and SELL/SELPLG interactions (Fig. 6H, S9A). Notably, the
677 most differential interactions we identified involved the upregulated expression of CD74 on
678 the surface of selective mPCs and its corresponding receptors in the TME, suggesting that
679 CD74 overexpression in selective cells might contribute to their immune escape.
680

681 Moreover, we found enhanced interactions between selective mPCs and NK cells, mediated
682 by HLA-E/CD94:NKG2A interactions, which represent inhibitory signals to NK cells (Fig.
24
683 S9B). Additionally, HLA-F/LILRB1 interactions, representing an inhibitory signaling axis in
684 the TME (57), was also enriched between selective mPCs and myeloid cells (Fig. 6I). These
685 findings raised intriguing questions about the potential involvement of these differential
686 cellular interactions in the observed patterns of transcriptional evolution and distinct
687 resistance pathways among the residual mPCs. Further studies are warranted to elucidate the
688 functional implications of these interactions in the immune escape and survival of selective

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

689 mPCs.
690

691 Discussion
692 This scRNA-seq study of paired diagnostic and post-treatment samples represented the most
693 extensive single-cell dataset to date. It elucidated transcriptional dynamics and uncovered
694 resistance pathways within MRD tumor cells in patients with MM. There were five main
695 findings in this study. Firstly, a population of nlPCs was identified in MM. Secondly, three
696 main trajectories of transcriptional evolution post-treatment were observed. These were clonal
697 elimination in MRD- patients, as well as clonal stabilization and clonal selection in MRD+
698 patients. Thirdly, a close link between genetic and non-genetic mechanisms behind the clonal
699 evolution of MM was confirmed, which further supported the multidimensional background
700 of clonal evolution. Fourthly, transcriptional and metabolic adaptations were observed as
701 important activated prosurvival pathways in MRD tumor cells. Finally, differential cellular
702 interactions between mPCs and TME cells were observed, showing that the TME might
703 provide shelter for MRD tumor cells.
704

705 Genetic studies employing iFISH and whole-exome sequencing have demonstrated that MM
706 is not a single disease entity but also a spectrum of molecular conditions, each with distinct
707 clinical outcomes (58). Moreover, gene expression profiling has revealed various subgroups
708 within these conditions, defined by the cytogenetic features of PCs and the aberrant
709 expression of a D-group cyclin (41). Previous phenotypic studies have identified several
710 subclones within patients with MM, each associated with unique chemoresistant profiles,
711 cytogenetic characteristics, and clonogenic potentials (7). Recently, an MGUS-like phenotype
712 has been described in MM patients based on the prevalence of BM PC and clonal PC within
25
713 the BM PC compartment (59). These findings underscore the remarkable inter-tumor
714 heterogeneity of MM and suggest the existence of a cell population evolving from normal to
715 malignant PCs identifiable at the early stages of MM pathogenesis. In alignment with
716 previous scRNA-seq studies (29, 32, 46), our study identified a PC cluster composed of PCs
717 from HDs and multiple MM patients. Interestingly, even clonal PCs from MM patients
718 co-clustered with HD PCs. We designated such clonal PCs as nlPCs. Further analyses

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

719 revealed several distinguishing features of nlPCs, such as the absence of secondary genetic
720 events as shown in their CNA profiles, the expression of monoclonal immunoglobulins or
721 light chain restriction, and extensive transcriptional alterations.
722

723 Therapy-induced clonal evolution plays a crucial role in MM drug resistance (58). As
724 treatments exert selective pressure, subclones either carry or acquire specific genetic
725 alterations, conferring a competitive fitness that enables these PCs to evade therapeutic
726 pressure. Utilizing quantitative multigene fluorescence in situ hybridization, we have charted
727 the phylogeny of tumor subclones and identified four patterns of evolutionary architecture in
728 129 longitudinal samples from 57 MM patients. We found that patients exhibiting clonal
729 stabilization had superior OS, while those with linear evolution had the worst outcomes (16).
730 Increasingly, studies are focusing on clonal evolution in MRD tumor cells (6, 7). Previous
731 research indicates that standard-risk patients are more susceptible to clonal selection events,
732 whereas high-risk patients more often acquire mutations (4). A recent single-cell study has
733 revealed that genetic abnormalities observed at relapse are already present at diagnosis within
734 subclonal populations (13). Based on these findings, we investigated the transcriptional
735 dynamics between disease diagnosis and post-treatment. Our results revealed three main
736 trajectories of transcriptional evolution post-treatment: clonal elimination in MRD- patients,
737 as well as clonal stabilization and clonal selection in MRD+ patients. Furthermore, by
738 delineating the clonal architecture of genetic events per patient using inferCNV (55), we
739 found a high correlation between genetic and non-genetic mechanisms underpinning the
740 clonal evolution of MM. We proposed further investigations to elucidate the clinical
741 significance of nlPCs.
742
26
743 There is growing evidence suggesting that drug resistance in cancer cannot be solely
744 attributed to simple genetic causes (19). The concept that non-genetic mechanisms contribute
745 to therapeutic resistance is increasingly gaining recognition (19). In the context of MM,
746 subclones with identical genetic backgrounds can result in multiple transcriptional clusters
747 (29), which points to the non-genetic mechanisms that influence transcriptional status. A
748 recent scRNA-seq study has reported upregulation of the UPR, ER, and mitochondrial stress

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

749 pathways in primary refractory MM patients (32). Aligning with these findings, we observed
750 a metabolic shift to fatty acid oxidation in resistant PCs in a state of cell cycle arrest, while
751 selective PCs demonstrated a preference for the NF-κB pathway in our study. Through this
752 mechanism of transcriptional selection, MM cells could rapidly adapt to therapeutic pressure
753 and attain a selective advantage within the therapeutic landscape.
754

755 MM progression involves aberrant interactions between PCs and the TME (60).
756 Comprehensive evidence suggests that both hematopoietic and non-hematopoietic cells,
757 including regulatory T cells (61), regulatory B cells (62), myeloid-derived suppressor cells (63,
758 64), and stromal cells (56), contribute to PC proliferation. Our study identified a potential
759 supportive role of the TME for MRD tumor cells. The most pronounced interaction between
760 MRD tumor cells and the TME was observed in selective PCs. Furthermore, the upregulated
761 expression of CD74 by selective PCs and the potential contribution of CD74-related
762 PC-myeloid cell interactions to their immune evasion were noted. Given that myeloid cells or
763 monocyte-derived cells, including osteoclasts (65) and DCs (66), comprise the primary
764 cellular component of the microenvironment that promotes immunosuppression, disrupting
765 the interactions between selective PCs and myeloid cells by blocking CD74 could be a
766 potential therapeutic approach to eradicate MRD tumor cells.
767

768 This study also highlights several remaining challenges. First, since only one MGUS sample
769 was sequenced in our study, the interpretation of the results of our study needs to take into
770 account the limited number of MGUS patients in our cohort. Nevertheless, the identification
771 of nlPCs added substantial weight to the concept of inter-tumor heterogeneity in MM.
772 Moreover, the inclusion of steroids in the induction regimen may also lead to transcriptional
27
773 and metabolic adaptation in residual mPCs. Further experimental studies are necessary to
774 confirm whether the transcriptional and metabolic adaptation in mPCs is in response to a
775 specific drug or represent a general phenomenon.
776

777 In conclusion, single-cell profiling of MRD tumor cells could provide novel insights into the
778 transcriptional evolution and molecular mechanisms underlying therapeutic resistance. This

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

779 knowledge might help identify new vulnerabilities that can be targeted, thereby contributing
780 to improved long-term disease control in MM.
781

782 Acknowledgments:
783 We thank Prof. Nikhil C. Munshi for his insightful suggestions to this study. We thank all the
784 patients and families for participation in this research study. Fig. 1A, and the Graphic abstract
785 were created with the assistance of BioRender. This study was supported by the National
786 Natural Science Foundation (grants 82270218, U22A20291 to G. An), Youth Fund of the
787 National Natural Science Foundation of China (grants 81900214, to S. Deng), the National
788 Natural Science Foundation (grants 32370722 to X. Gao), the CAMS Innovation Fund for
789 Medical Sciences (CIFMS) (grants 2022-I2M-1-022, to L. Qiu; grants 2021-I2M-C&T-B-079,
790 to G. An, and grants 2021-I2M-1-041), the International Cooperation Projects of National
791 Natural Science Foundation (grants 81920108006, to L. Qiu). J. Cui is supported by grant
792 from International Myeloma Society.
793

28
794 References
795
796 1. Rodriguez-Otero P, Paiva B, San-Miguel JF. Roadmap to cure multiple myeloma. Cancer Treat Rev. 2021. 100:
797 102284.
798 2. Anderson KC, Auclair D, Adam SJ, Agarwal A, Anderson M, Avet-Loiseau H, et al. Minimal Residual Disease in
799 Myeloma: Application for Clinical Care and New Drug Registration. Clin Cancer Res. 2021. 27(19): 5195-5212.
800 3. Quach H. MRD end point in myeloma: ready for prime time. Blood. 2022. 139(6): 799-802.
801 4. Goicoechea I, Puig N, Cedena M, Burgos L, Cordón L, Vidriales M, et al. Deep MRD profiling defines outcome and

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

802 unveils different modes of treatment resistance in standard- and high-risk myeloma. Blood. 2021. 137(1): 49-60.
803 5. de Tute RM, Pawlyn C, Cairns DA, Davies FE, Menzies T, Rawstron A, et al. Minimal Residual Disease After
804 Autologous Stem-Cell Transplant for Patients With Myeloma: Prognostic Significance and the Impact of
805 Lenalidomide Maintenance and Molecular Risk. J Clin Oncol. 2022. 40(25): 2889-2900.
806 6. Paiva B, Manrique I, Dimopoulos MA, Gay F, Min CK, Zweegman S, et al. MRD dynamics during maintenance for
807 improved prognostication of 1280 patients with myeloma in the TOURMALINE-MM3 and -MM4 trials. Blood. 2023.
808 141(6): 579-591.
809 7. Paiva B, Corchete LA, Vidriales MB, Puig N, Maiso P, Rodriguez I, et al. Phenotypic and genomic analysis of
810 multiple myeloma minimal residual disease tumor cells: a new model to understand chemoresistance. Blood. 2016.
811 127(15): 1896-1906.
812 8. Morgan GJ, Walker BA, Davies FE. The genetic architecture of multiple myeloma. Nat Rev Cancer. 2012. 12(5):
813 335-348.
814 9. Cui J, Lv R, Yu T, Yan W, Xu J, Fan H, et al. Minor clone of del(17p) provides a reservoir for relapse in multiple
815 myeloma. Haematologica. 2024. 109(2): 591-603.
816 10. Cui J, Yu T, Lv R, Liu J, Fan H, Yan W, et al. Longitudinal genetically detectable minimal residual disease by
817 fluorescence in situ hybridization confers a poor prognosis in myeloma. Ther Adv Med Oncol. 2024. 16:
818 17588359231221340.
819 11. Misund K, Hofste op Bruinink D, Coward E, Hoogenboezem RM, Rustad EH, Sanders MA, et al. Clonal evolution
820 after treatment pressure in multiple myeloma: heterogenous genomic aberrations and transcriptomic convergence.
821 Leukemia. 2022 : -.
822 12. Samur MK, Szalat R, Munshi NC. Single-Cell Profiling in Multiple Myeloma: Insights, Problems, and Promises.
823 Blood. 2023. 142(4): 313-324.
824 13. Lannes R, Samur M, Perrot A, Mazzotti C, Divoux M, Cazaubiel T, et al. In Multiple Myeloma, High-Risk Secondary
825 Genetic Events Observed at Relapse Are Present From Diagnosis in Tiny, Undetectable Subclonal Populations. J Clin
826 Oncol. 2023. 41(9): 1695-1702.
827 14. Johnson TS, Sudha P, Liu E, Becker N, Robertson S, Blaney P, et al. 1q amplification and PHF19 expressing high-risk
828 cells are associated with relapsed/refractory multiple myeloma. Nat Commun. 2024. 15(1): 4144.
829 15. Liu R, Gao Q, Foltz SM, Fowles JS, Yao L, Wang JT, et al. Co-evolution of tumor and immune cells during
830 progression of multiple myeloma. Nat Commun. 2021. 12(1): 2559.
831 16. Yan Y, Qin X, Liu J, Fan H, Yan W, Liu L, et al. Clonal phylogeny and evolution of critical cytogenetic aberrations in
832 multiple myeloma at single-cell level by QM-FISH. Blood Adv. 2022. 6(2): 441-451.
833 17. Croft J, Ellis S, Sherborne AL, Sharp K, Price A, Jenner MW, et al. Copy number evolution and its relationship with
834 patient outcome-an analysis of 178 matched presentation-relapse tumor pairs from the Myeloma XI trial. Leukemia.
835 2021. 35(7): 2043-2053.
836 18. Samur MK, Fulciniti M, Aktas Samur A, Bazarbachi AH, Tai Y, Prabhala R, et al. Biallelic loss of BCMA as a

29
837 resistance mechanism to CAR T cell therapy in a patient with multiple myeloma. Nat Commun. 2021. 12(1): 868.
838 19. Marine J, Dawson S, Dawson MA. Non-genetic mechanisms of therapeutic resistance in cancer. Nat Rev Cancer.
839 2020. 20(12): 743-756.
840 20. Corre J, Perrot A, Caillot D, Belhadj K, Hulin C, Leleu X, et al. del(17p) without TP53 mutation confers a poor
841 prognosis in intensively treated newly diagnosed patients with multiple myeloma. Blood. 2021. 137(9): 1192-1195.
842 21. Shah V, Sherborne AL, Walker BA, Johnson DC, Boyle EM, Ellis S, et al. Prediction of outcome in newly diagnosed
843 myeloma: a meta-analysis of the molecular profiles of 1905 trial patients. Leukemia. 2018. 32(1): 102-110.
844 22. An G, Li Z, Tai YT, Acharya C, Li Q, Qin X, et al. The impact of clone size on the prognostic value of chromosome
845 aberrations by fluorescence in situ hybridization in multiple myeloma. Clin Cancer Res. 2015. 21(9): 2148-2156.

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

846 23. Paiva B, Puig N, Cedena MT, Rosiñol L, Cordón L, Vidriales MB, et al. Measurable Residual Disease by
847 Next-Generation Flow Cytometry in Multiple Myeloma. J Clin Oncol. 2020. 38(8): 784-792.
848 24. Flores-Montero J, Sanoja-Flores L, Paiva B, et al. Next Generation Flow for highly sensitive and standardized
849 detection of minimal residual disease in multiple myeloma. Leukemia. 2017. 31(10): 2094-2103.
850 25. Satija R, Farrell JA, Gennert D, et al. Spatial reconstruction of single-cell gene expression data. Nat Biotechnol. 2015.
851 33(5): 495-502.
852 26. McGinnis CS, Murrow LM, Gartner ZJ. DoubletFinder: Doublet Detection in Single-Cell RNA Sequencing Data
853 Using Artificial Nearest Neighbors. Cell Syst. 2019. 8(4): 329-337.e4.
854 27. Korsunsky I, Millard N, Fan J, et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat
855 Methods. 2019. 16(12): 1289-1296.
856 28. Zavidij O, Haradhvala NJ, Mouhieddine TH, Sklavenitis-Pistofidis R, Cai S, Reidy M, et al. Single-cell RNA
857 sequencing reveals compromised immune microenvironment in precursor stages of multiple myeloma. Nat Cancer.
858 2020. 1(5): 493-506.
859 29. Tirier SM, Mallm JP, Steiger S, Poos AM, Awwad M, Giesen N, et al. Subclone-specific microenvironmental impact
860 and drug response in refractory multiple myeloma revealed by single-cell transcriptomics. Nat Commun. 2021. 12(1):
861 6960.
862 30. Ren X, Wen W, Fan X, et al. COVID-19 immune features revealed by a large-scale single-cell transcriptome atlas.
863 Cell. 2021. 184(7): 1895-1913.e19.
864 31. Han X, Zhou Z, Fei L, et al. Construction of a human cell landscape at single-cell level. Nature. 2020. 581(7808):
865 303-309.
866 32. Cohen YC, Zada M, Wang SY, Bornstein C, David E, Moshe A, et al. Identification of resistance pathways and
867 therapeutic targets in relapsed multiple myeloma patients through single-cell sequencing. Nat Med. 2021. 27(3):
868 491-503.
869 33. Zhou Y, Zhou B, Pache L, et al. Metascape provides a biologist-oriented resource for the analysis of systems-level
870 datasets. Nat Commun. 2019. 10(1): 1523.
871 34. Liberzon A, Birger C, Thorvaldsdóttir H, et al. The Molecular Signatures Database (MSigDB) hallmark gene set
872 collection. Cell Syst. 2015. 1(6): 417-425.
873 35. Gulati GS, Sikandar SS, Wesche DJ, Manjunath A, Bharadwaj A, Berger MJ, et al. Single-cell transcriptional
874 diversity is a hallmark of developmental potential. Science. 2020. 367(6476): 405-411.
875 36. Andor N, Simonds EF, Czerwinski DK, Chen J, Grimes SM, Wood-Bouwens C, et al. Single-cell RNA-Seq of
876 follicular lymphoma reveals malignant B-cell types and coexpression of T-cell immune checkpoints. Blood. 2019.
877 133(10): 1119-1129.
878 37. Van de Sande B, Flerin C, Davie K, et al. A scalable SCENIC workflow for single-cell gene regulatory network
879 analysis. Nat Protoc. 2020. 15(7): 2247-2276.
880 38. Hänzelmann S, Castelo R, Guinney J. GSVA: gene set variation analysis for microarray and RNA-seq data. BMC

30
881 Bioinformatics. 2013. 14: 7.
882 39. Efremova M, Vento-Tormo M, Teichmann SA, Vento-Tormo R. CellPhoneDB: inferring cell–cell communication
883 from combined expression of multi-subunit ligand–receptor complexes. Nat Protoc. 2020. 15(4): 1484-1506.
884 40. Rozenblatt-Rosen O, Stubbington M, Regev A, Teichmann SA. The Human Cell Atlas: from vision to reality. Nature.
885 2017. 550(7677): 451-453.
886 41. Zhan F, Huang Y, Colla S, Stewart JP, Hanamura I, Gupta S, et al. The molecular classification of multiple myeloma.
887 Blood. 2006. 108(6): 2020-2028.
888 42. Merz M, Merz A, Wang J, Wei L, Hu Q, Hutson N, et al. Deciphering spatial genomic heterogeneity at a single cell
889 resolution in multiple myeloma. Nat Commun. 2022. 13(1): 807.

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

890 43. Kikuchi J, Hori M, Iha H, Toyama-Sorimachi N, Hagiwara S, Kuroda Y, et al. Soluble SLAMF7 promotes the growth
891 of myeloma cells via homophilic interaction with surface SLAMF7. Leukemia. 2020. 34(1): 180-195.
892 44. Siu KT, Huang C, Panaroni C, Mukaihara K, Fulzele K, Soucy R, et al. BCL2 blockade overcomes MCL1 resistance
893 in multiple myeloma. LEUKEMIA. 2019. 33(8): 2098-2102.
894 45. Wang Q, Zhao D, Xian M, Wang Z, Bi E, Su P, et al. MIF as a biomarker and therapeutic target for overcoming
895 resistance to proteasome inhibitors in human myeloma. Blood. 2020. 136(22): 2557-2573.
896 46. Ledergor G, Weiner A, Zada M, Wang S, Cohen YC, Gatt ME, et al. Single cell dissection of plasma cell
897 heterogeneity in symptomatic and asymptomatic myeloma. Nat Med. 2018. 24(12): 1867-1876.
898 47. Zamani-Ahmadmahmudi M, Nassiri SM, Soltaninezhad F. Development of an RNA sequencing-based prognostic
899 gene signature in multiple myeloma. Br J Haematol. 2021. 192(2): 310-321.
900 48. Decaux O, Lodé L, Magrangeas F, Charbonnel C, Gouraud W, Jézéquel P, et al. Prediction of survival in multiple
901 myeloma based on gene expression profiles reveals cell cycle and chromosomal instability signatures in high-risk
902 patients and hyperdiploid signatures in low-risk patients: a study of the Intergroupe Francophone du Myélome. J Clin
903 Oncol. 2008. 26(29): 4798-4805.
904 49. Zhou W, Yang Y, Xia J, Wang H, Salama ME, Xiong W, et al. NEK2 induces drug resistance mainly through
905 activation of efflux drug pumps and is associated with poor prognosis in myeloma and other cancers. Cancer Cell.
906 2013. 23(1): 48-62.
907 50. Chang H, Qi X, Trieu Y, Xu W, Reader JC, Ning Y, et al. Multiple myeloma patients with CKS1B gene amplification
908 have a shorter progression-free survival post-autologous stem cell transplantation. Br J Haematol. 2006. 135(4):
909 486-491.
910 51. Oren Y, Tsabar M, Cuoco MS, Amir-Zilberstein L, Cabanos HF, Hütter J, et al. Cycling cancer persister cells arise
911 from lineages with distinct programs. Nature. 2021. 596(7873): 576-582.
912 52. Parzych K, Chinn TM, Chen Z, Loaiza S, Porsch F, Valbuena GN, et al. Inadequate fine-tuning of protein synthesis
913 and failure of amino acid homeostasis following inhibition of the ATPase VCP/p97. Cell Death Dis. 2015. 6(12):
914 e2031.
915 53. Khalili P, Maddah R, Maleknia M, Shateri Amiri B, Forouzani F, Hasanvand A, et al. Evaluation of Genes and
916 Molecular Pathways Involved in the Progression of Monoclonal Gammopathy of Undetermined Significance (MGUS)
917 to Multiple Myeloma: A Systems Biology Approach. Mol Biotechnol. 2023. 65(8): 1275-1286.
918 54. Gotoh N. Novel therapeutic strategies to eradicate tumors by targeting cancer stem-like cells. Ann Oncol. 2016. 27:
919 vii40.
920 55. Puram SV, Tirosh I, Parikh AS, Patel AP, Yizhak K, Gillespie S, et al. Single-Cell Transcriptomic Analysis of Primary
921 and Metastatic Tumor Ecosystems in Head and Neck Cancer. Cell. 2017. 171(7): 1611-1624.e24.
922 56. de Jong M, Kellermayer Z, Papazian N, Tahri S, Hofste Op Bruinink D, Hoogenboezem R, et al. The multiple
923 myeloma microenvironment is defined by an inflammatory stromal cell landscape. Nat Immunol. 2021. 22(6):
924 769-780.

31
925 57. Barkal AA, Weiskopf K, Kao KS, Gordon SR, Rosental B, Yiu YY, et al. Engagement of MHC class I by the
926 inhibitory receptor LILRB1 suppresses macrophages and is a target of cancer immunotherapy. Nat Immunol. 2018.
927 19(1): 76-84.
928 58. Pawlyn C, Morgan GJ. Evolutionary biology of high-risk multiple myeloma. Nat Rev Cancer. 2017. 17(9): 543-556.
929 59. Burgos L, Tamariz-Amador L, Puig N, Cedena M, Guerrero C, Jelínek T, et al. Definition and Clinical Significance of
930 the Monoclonal Gammopathy of Undetermined Significance–Like Phenotype in Patients With Monoclonal
931 Gammopathies. J Clin Oncol. 2023: JCO.22.01916.
932 60. Palumbo A, Anderson K. Multiple Myeloma. N Eng J Med. 2011. 364(11): 1046-1060.
933 61. Feng X, Zhang L, Acharya C, An G, Wen K, Qiu L, et al. Targeting CD38 Suppresses Induction and Function of T

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

934 Regulatory Cells to Mitigate Immunosuppression in Multiple Myeloma. Clin Cancer Res. 2017. 23(15): 4290-4300.
935 62. Zhang L, Tai YT, Ho M, Xing L, Chauhan D, Gang A, et al. Regulatory B cell-myeloma cell interaction confers
936 immunosuppression and promotes their survival in the bone marrow milieu. Blood Cancer J. 2017. 7(3): e547.
937 63. Lv M, Wang K, Huang XJ. Myeloid-derived suppressor cells in hematological malignancies: friends or foes. J
938 Hematol Oncol. 2019. 12(1): 105.
939 64. Ramachandran IR, Martner A, Pisklakova A, Condamine T, Chase T, Vogl T, et al. Myeloid-derived suppressor cells
940 regulate growth of multiple myeloma by inhibiting T cells in bone marrow. J Immunol. 2013. 190(7): 3815-3823.
941 65. An G, Acharya C, Feng X, Wen K, Zhong M, Zhang L, et al. Osteoclasts promote immune suppressive
942 microenvironment in multiple myeloma: therapeutic implication. Blood. 2016. 128(12): 1590-1603.
943 66. Leone P, Berardi S, Frassanito MA, Ria R, De Re V, Cicco S, et al. Dendritic cells accumulate in the bone marrow of
944 myeloma patients where they protect tumor plasma cells from CD8+ T-cell killing. Blood. 2015. 126(12): 1443-1451.

945

32
946 Author contributions:
947 J. Cui: Conceptualization, data curation, formal analysis, investigation, validation,
948 visualization, writing-original draft, writing-review & editing. X. Li: Conceptualization, data
949 curation, formal analysis, investigation, software, visualization, writing-original draft,
950 writing-review & editing. Shuhui Deng: Conceptualization, funding acquisition,
951 investigation, project administration, writing-original draft, writing-review & editing. C. Du:

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

952 Data curation, resources. H. Fan: Data curation, resources. W. Yan: Data curation, resources.
953 J. Xu: Data curation, resources. X. Li: Writing-review & editing. T. Yu: Data curation,
954 resources. S. Zhang: Data curation, resources. R. Lv: Data curation, resources. W. Sui: Data
955 curation, resources. M. Hao: Data curation, resources. X. Du: Writing-review & editing. Y.
956 Xu: Data curation, supervision. S. Yi: Data curation, supervision. D. Zou: Data curation,
957 supervision. T. Cheng: Data curation, supervision. L. Qiu: Conceptualization, data curation,
958 funding acquisition, resources, supervision, writing-original draft, writing-review & editing.
959 X. Gao: Conceptualization, data curation, funding acquisition, resources, supervision,
960 writing-original draft, writing-review & editing. G. An: Conceptualization, data curation,
961 funding acquisition, project administration, resources, supervision, writing-original draft,
962 writing-review & editing.
963

964 Supplementary materials:

965 Supplementary Methods. Supplementary Figure S1-S9. Supplementary Table S1-S15.
966

967 Informed consent statement:

968 Written informed consent was obtained from all patients to publish this paper.
969

33
970 Figure legends
971 Figure 1. Single-cell transcriptional landscape of HD and MM patients.
972 (A) The workflow illustrates the strategy for cell collection from matching diagnosis-treated
973 paired samples for scRNA-seq. (B) Six major single-cell clusters from BM samples of HD,
974 MGUS, NDMM (pre-treatment), and MM patients during induction therapy (post-treatment).
975 Annotation of cell clusters is marked in the bottom part with color codes. Cell embeddings are

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

976 shown by using UMAP1 and UMAP2. (C) Dot plot for expression of marker genes of the
977 identified six major single-cell clusters. Color represents a maximum-normalized mean
978 expression of cells expressing marker genes, and size represents the percentage of cells
979 expressing these genes. (D) Bar plot comparing the proportions of distinctive cell types in BM
980 between HD, MM pre, and MM post samples. (E) UMAP representations of the single-cell
981 transcriptomes of 76,148 BM PCs derived from 37 patients (one MGUS, 19 MM pre-
982 treatment, and 12 MM post-treatment) with PC neoplasms, and nine HDs (eight from Human
983 Cell Atlas). Cells are color-coded by cell clusters (left), by sample group (right top), and by
984 the mPCs or nlPCs (right bottom). (F) Bar plot showing the distribution of cells from subjects
985 with PC neoplasm and HDs by disease stage and sample ID (as in Fig. 1E). Subjects are
986 color-coded according to sample group (green for HD and MGUS samples, blue for MM-pre
987 samples and red for MM-post samples); names on the bottom bars correspond to the
988 individual with the majority of cells in each cluster. (G) Violin plots showing the expression
989 of common MM driver genes across 18 MM patients with mPCs, one MGUS, and four HDs.
990

991 Figure 2. Identification of nlPCs.

992 (A) UMAP visualization of MM2 pre-treatment PCs, color coded by transcriptional cluster,
993 CNV clone, and cell group (from left to right). (B) Heatmap showing the genome-wide
994 inferred CNA profiles for MM2 pre-treatment PCs. Horizontal lines divide CNV subclones.
995 Groups are colored by CNV subclones and cell groups as shown on the left-hand side. (C)
996 UMAP visualization MM2 pre-treatment PCs with CytoTRACE score mapped on. (D) The
997 number of shared upregulated DEGs in HD PCs, mPCs, and nlPCs compared with each other.
998 (E) Enriched hallmark pathways for shared upregulated genes in mPCs versus HD PCs and
999 nlPCs versus HD PCs. (F) Enriched hallmark pathways for upregulated DEGs in mPCs versus
34
1000 nlPCs. (G) Violin plots showing the expression of common MM driver genes in mPCs and
1001 nlPCs from MM2 pre-treatment sample and HD PCs from HDs. (H) Violin plots showing the
1002 expression of common MM driver genes across nine HDs, one MGUS, and 12 MM samples
1003 with PCs in cluster 1 (as in Fig. 1E).
1004

1005 Figure 3. Longitudinal single-cell analysis of MM patients pre- and post-treatment

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

1006 reveals transcriptional evolution and distinct resistance pathways.
1007 (A) Bar plots comparing longitudinal malignant PC transcriptional clonality in pre- and
1008 post-treatment of individual MM patients. Only patients treated with PI + IMiD induction
1009 regimens are presented. (B) Differences in the cell-cycle phases in each transcriptional cluster
1010 are shown in Fig. 3A. (C) Heatmaps showing the 68 shared upregulated genes in
1011 cycling-resistant PCs versus HD PCs and cycling-resistant PCs versus sensitive PCs.
1012 Annotated hallmark and GO pathways related to the shared upregulated DEGs are labeled on
1013 the right-hand side. (D) Violin plots showing the expression of six selected shared upregulated
1014 genes shown in Fig. 3C, in three different PC groups: HD PCs, sensitive PCs, and
1015 cycling-resistant PCs. (E) Heatmaps showing the 16 shared upregulated genes in selective
1016 PCs versus HD PCs and selective PCs versus sensitive PCs. Annotated hallmark pathways
1017 related to the shared upregulated DEGs are labeled on the right-hand side. (F) Violin plots
1018 showing the expression of five selected shared upregulated genes shown in Fig. 3E, in three
1019 different PC groups: HD PCs, sensitive PCs, and selective PCs. (G) Dot plot of the scaled
1020 mean AUC scores of expression regulation by TFs, as estimated using pySCENIC, for
1021 sensitive, cycling-resistant, and selective PCs. Shown are the 10 differentially expressed TFs
1022 in cycling resistant PCs versus sensitive PCs, and the 19 differentially expressed TFs in
1023 selective PCs versus sensitive PCs. Thresholds for differential expression were
1024 Bonferroni-adjusted p-values < 0.05. (H) Hallmark pathway enrichment analysis of the
1025 differentially expressed genes in selective PCs compared to cycling-resistant PCs.
1026

1027 Figure 4. Predictive value of the gene signature of cycling-resistant PCs.

1028 (A) Kaplan-Meier curves and analysis for OS comparing NDMM patients in the CoMMpass
1029 data with high cycling-resistant scores (red) and patients with low cycling-resistant scores
35
1030 (blue). (log-rank test, two-sided p < 0.0001, HR = 2.94, 95% CI = 2.23–3.88). (B)
1031 Kaplan-Meier curves and analysis for OS comparing NDMM patients in the CoMMpass data
1032 with high selective scores (red) and patients with low selective scores (blue). (Log-rank test,
1033 two-sided p= 0.0015, HR = 0.56, 95% CI = 0.39–0.81). (C) Forest plots of HR for median OS
1034 in patients with high cycling-resistant score vs. low cycling-resistant score in different
1035 datasets. (D) Forest plots of HR for median OS in patients with high selective score vs. low

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

1036 selective score in different datasets. NR, not reached.
1037

1038 Figure 5. Longitudinal single-cell analysis of MM patients pre- and post-treatment

1039 reveals genetic evolution and different patterns of metabolic adaptation.
1040 (A) Heatmap showing the genome-wide inferred CNA profiles for MM5 pre- and
1041 post-treatment mPCs. Horizontal lines divide CNV subclones. Groups are colored by CNV
1042 subclones and cell groups, as shown on the left-hand side. (B) Consensus CNA profiles of the
1043 six CNV clones for MM5 mPCs. (C) Plausible phylogeny tree and branching evolutionary
1044 pattern of MM5 CNV clones. (D) UMAP visualization MM5 mPCs, color coded by
1045 transcriptional cluster (left), and CNV clone (right). (E) Circle plot in MM5 comparing the
1046 distribution of cells from CNV clones (middle) across transcriptional clusters (outer). The
1047 inner red layer indicates the post-treatment fraction within each transcriptional cluster. (F)
1048 UMAP visualization MM5 mPCs with CytoTRACE score mapped on. (G) Volcano plot of
1049 DEGs using two-sided Wilcoxon rank-sum test in MM5 C2 and C4 clusters. Thresholds for
1050 differential expression were p-value < 0.05 (Bonferroni-adjusted) and logFC > Log1.5. (H)
1051 Hallmark pathway enrichment analysis of the DEGs in MM5 C2 cluster compared to MM5
1052 C4 cluster. (I) Line plot of subclone fraction derived from the CNA analysis per patient
1053 pre/post-treatment. Purple lines mark selective subclones, yellow lines mark resistant
1054 subclones, and green line mark sensitive subclones. Bottom, post-treatment response.
1055

1056 Figure 6. TME remodeling during induction therapy.

1057 (A) UMAPs of CD138− hematopoietic cells, color-coded by cell types. (B) Gene expression
1058 dot plot of major marker genes for individual cell types. (C) Bar plot of cell type fractions for
1059 HD (left), MM pre- (middle), and post-treatment (right). (D) The number of shared
36
1060 upregulated DEGs between CD14 monocytes (left). Hallmark pathway enrichment analysis of
1061 the shared upregulated genes (right). (E) The number of shared downregulated DEGs between
1062 CD16 monocytes (left). Hallmark pathway enrichment analysis of the shared downregulated
1063 genes (right). (F, G) Dot plots showing the number of ligand-receptor (F) and receptor-ligand
1064 (G) interactions inferred by CellPhoneDB between hematopoietic cells and sensitive,
1065 cycling-resistant, or selective PCs. (H, I) Dot plots showing the mean expression (color key)

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024

1066 and significance (size key) of ligand-receptor (H) and receptor-ligand (I) pairs between
1067 hematopoietic cells and sensitive, cycling-resistant, and selective PCs. Only interactions that
1068 showed differences across the three types of PCs are shown.

37
Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024
Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024
Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024
Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024
Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024
Downloaded from http://aacrjournals.org/clincancerres/article-pdf/doi/10.1158/1078-0432.CCR-24-0545/3464142/ccr-24-0545.pdf by guest on 26 June 2024