Abstract

   Multiple myeloma (MM) remains an incurable disease primarily due to the
   emergence of drug resistance, and the underlying mechanisms remain
   unclear. Extrachromosomal circular DNAs (eccDNAs) are prevalent in
   cancer genomes of both coding and non‐coding regions. However, the role
   of non‐coding eccDNA regions that serve as enhancers has been largely
   overlooked. Here, genome‐wide profiling of serum eccDNAs from donors
   and MM patients who responded well or poorly to
   bortezomib‐lenalidomide‐dexamethasone (VRd) therapy is characterized. A
   high copy number of eccDNA ANKRD28 (eccANKRD28) predicts poor therapy
   response and prognosis but enhanced transcriptional activity.
   Established VRd‐resistant MM cell lines exhibit a higher abundance of
   eccANKRD28, and CRISPR/Cas9‐mediated elevation of eccANKRD28
   desensitizes bortezomib and lenalidomide treatment both in vitro and in
   vivo. Integrated multi‐omics analysis (H3K27ac ChIP‐seq, scRNA‐seq,
   scATAC‐seq, CUT&Tag, et al.) identifies eccANKRD28 as an active
   enhancer involved in drug resistance driven by the key transcription
   factor, POU class 2 homeobox 2 (POU2F2). POU2F2 interacts with
   sequence‐specific eccANKRD28 as well as RUNX1 and RUNX2 motifs to form
   the protein complex, which activates the promoter of oncogenes,
   including IRF4, JUNB, IKZF3, RUNX3, and BCL2. This study elucidates the
   potential transcriptional network of enhancer eccANKRD28 in MM drug
   resistance from a previously unrecognized epigenetic perspective.

   Keywords: multiple myeloma, extrachromosomal circular DNA, drug
   resistance, enhancer, transcriptional regulation
     __________________________________________________________________

   Enhancer eccANKRD28‐manipulated MM cells have been demonstrated to
   facilitate drug resistance and promote MM progression by activating the
   key transcription factor, POU2F2. POU2F2 interacts with
   sequence‐specific eccANKRD28 as well as RUNX1 and RUNX2 motifs to form
   the protein complex, which activates the promoter of oncogenes (IRF4,
   JUNB, IKZF3, et al.) both in vitro and in vivo.

   graphic file with name ADVS-12-2415695-g003.jpg

1. Introduction

   The current therapeutic approach for newly diagnosed multiple myeloma
   (NDMM) utilizing proteasome inhibitors (PI) and immunomodulatory drugs
   (IMiDs), especially the most commonly used
   bortezomib‐lenalidomide‐dexamethasone (VRd) targeted therapy, has shown
   efficacy in achieving deep remission,^[ [48]^1 ^] yet most patients
   eventually develop resistance.^[ [49]^2 ^] This drug resistance is
   closely associated with the intricate epigenomic landscape of MM.^[
   [50]^3 ^] Therefore, the mechanism that induces VRd resistance in MM
   and identifying novel targets are urgently needed.

   Extrachromosomal circular DNAs (eccDNAs) are circular DNA molecules
   derived from chromosomes but exist as separate entities from
   chromosomal DNA upon formation.^[ [51]^4 ^] A comprehensive pan‐cancer
   analysis of ATAC‐sequencing libraries from 23 tumor types identified
   over 18000 eccDNAs, many of which harbor well‐known cancer‐promoting
   genes.^[ [52]^5 ^] EccDNA has already been found in both circulating
   and hematopoietic cells and has been suggested to reflect disease
   status and promote malignant phenotypes.^[ [53]^6 , [54]^7 ^] EccDNA
   can contain entire gene sequences, specific oncogenes with amplified
   copy numbers, and heightened transcription levels. The upregulation of
   oncogenes by eccDNA is implicated in tumor growth.^[ [55]^8 ^] EccDNAs
   containing oncogenes can drive intratumoral heterogeneity and
   contribute to extreme copy number amplification through uneven
   segregation. A recent study demonstrated that approximately half of the
   amplifications in MM can be attributed to eccDNA formation,
   highlighting the intersection of MM genomics and epigenetics.^[ [56]^9
   ^] In particular, microDNA, eccDNAs <1000 bp in size derived from
   chromosomal genomic sequences, is released from normal and tumor cells
   into the circulation and is often more stable than linear cell‐free DNA
   (cfDNA).^[ [57]^10 ^] The precise contribution of microDNA,
   particularly non‐coding DNA elements that do not contain open reading
   frames (ORFs), to resistance mechanisms remains inadequately
   elucidated, with little experimental evidence currently available to
   establish their oncogenic and epigenetic roles definitively.

   Ankyrin repeat domain 28 (ANKRD28) has been confirmed as an oncogene in
   acute myeloid leukemia (AML).^[ [58]^11 , [59]^12 ^] Analogously,
   super‐enhancers‐regulated ANKRD28 correlates with poor prognosis in
   patients with germinal center B‐cell diffuse large B‐cell lymphoma
   (GCB‐DLBCL).^[ [60]^13 ^] Despite the evidence in other hematologic
   neoplasms, the role of ANKRD28, especially its non‐coding DNA regions
   that served as enhancers, remains unclear in MM. This study elucidated
   the crucial role of serum eccANKRD28 extrachromosomal amplification in
   facilitating drug resistance, indicating a novel extracellular nucleic
   acid biomarker for assessing therapy response.

2. Results

2.1. Genome‐Wide Characterization of Serum EccDNAs in Multiple Myeloma

   To investigate serum eccDNA from NDMM patients on a genomic scale, we
   employed Circle‐seq and RNA‐seq. Serum samples were collected before
   administering three cycles of the VRd regimen, followed by evaluating
   therapeutic responses (Figure [61]1A). EccDNAs were initially isolated
   from the serum of three distinct groups: healthy donors (HD, n = 5),
   individuals with high sensitivity (HS, n = 5), and those with low
   sensitivity (LS, n = 5) to VRd. To validate the existence of eccDNAs in
   the patient's serum, atomic force microscopy (AFM) was used to
   visualize the circular structure of eccDNAs (Figure [62]1B).
   Subsequently, the chromosomal origins of eccDNAs were analyzed through
   gene annotation. In total, 19218 genes were identified, 6343 eccDNAs
   were mapped to genic regions, and 52.25% of eccDNAs overlapped with
   intergenic regions (Figure [63]S1A, Supporting Information).

Figure 1.

   Figure 1
   [64]Open in a new tab

   Genome‐wide characterization of serum eccDNAs in multiple myeloma. A)
   Flowchart illustrating sample collection process within the study
   design. B) AFM images of extracted eccDNAs in MM patient's serum. C)
   Stack bar chart of the most frequent genomic events in HS and LS
   samples, broken down by cytogenetic group. D) EccDNA frequency relative
   to chromosomes. EccDNA counts per chromosome in healthy donors (HD,
   green circle), high‐sensitivity individuals (HS, achieving sCR/CR after
   three cycles of VRd, yellow square), and low‐sensitivity individuals
   (LS, achieving PR/MR/SD/PD after three cycles of VRd, blue triangle). n
   = 5 in each group. E) Circos plot showing the chromosomal distribution
   of eccDNAs. The four circles in the plot, from outermost to innermost,
   represent the human chromosomes and the distributions of eccDNAs in the
   HD (green), HS (yellow), and LS (blue) samples. F) Length distribution
   of eccDNAs in the three groups, with significant peaks observed at 100
   and 1000 bases. G) Circle scores of eccDNAs in the HD (green), HS
   (yellow), and LS (blue) groups. H) Genomic distributions of eccDNAs in
   the HD (green), HS (yellow), and LS (blue) groups. Abbreviations: sCR,
   stringent complete response; CR, complete response; PR, partial
   response; MR, minimal response; SD, stable disease; PD, progressive
   disease.

   The formation of eccDNA reflects genomic instability, with focal
   amplification often resulting from the fusion of rearranged DNA
   segments from different chromosomal locations. To further test this
   assertion, we examined whether eccDNA preferentially originates from
   chromosomal abnormalities, including IgH translocations such as t
   (4;14), t (11;14), and t (14;16), as well as other mutations like
   del(17p) and gain(1q). These alterations contribute to MM progression
   and are, in some cases, associated with treatment resistance.^[ [65]^14
   , [66]^15 ^] Among these 10 MM patients, 1q gain was the most common
   adverse chromosomal aberration (Figure [67]1C). We analyzed the
   chromosomal distribution of average eccDNA frequency to gain further
   insight. Interestingly, chr1 exhibited the highest average count of
   eccDNAs compared to other chromosomes, surpassing even the gene‐rich
   chr19, which has been identified as significant in different cancer
   types (Figure [68]1D).^[ [69]^16 ^] Paired‐end reads were aligned to
   chromosomes to characterize eccDNA and detect structural variations
   resulting from DNA circularization (Figure [70]1E). Consistent with
   previous observations, eccDNAs were distributed strikingly consistently
   between donors and MM samples.

   Alterations in fragmentation patterns reflect changes in the chromatin
   structure before the release of cfDNA, as well as other genomic and
   epigenomic alterations occurring during tumorigenesis. We subsequently
   analyzed the unique structural characteristics of these 15 sequencing
   samples, focusing on their eccDNA length, circle score, and genomic
   distribution. We observed that the average length of eccDNA was ranked
   in descending order: HD > HS > LS, with a statistically significant
   difference observed between the HD and LS groups (P<0.05, Figure
   [71]S1B, Supporting Information). It can be observed that the
   distribution characteristics of eccDNA length are very similar to those
   of cfDNA length (shorter in cancer patients and longer in healthy
   individuals).^[ [72]^17 , [73]^18 ^] Here, the eccDNA identified using
   the Circle‐seq method ranged in length from 10 bp to 10000 kb, with the
   majority falling between 100 to 1000 bp, suggesting that shorter
   eccDNAs are more prevalent in serum (Figure [74]1F). The circle score,
   a scoring scheme that considers alignment quality, split segment
   length, and the number of split reads supporting circular DNA, was
   utilized to evaluate the quality of eccDNA among the three groups.^[
   [75]^19 ^] We observed that the proportion of eccDNA in the HD group
   was more significant than that in the HS group and that the LS group
   was the lowest at the same circle score level. As the circle score
   increased, the proportion of eccDNA initially increased but
   subsequently decreased. Specifically, within the lower range of the
   Circle score (≈0–1000), an increase in detected eccDNAs correlates with
   an increase in the Circle score. Conversely, when the circle score
   exceeded 1000, the proportion of eccDNA in all the samples continuously
   decreased to markedly low levels (Figure [76]1G). To explore the
   possible origin of eccDNAs, we mapped the overall population of serum
   eccDNAs to different classes of genomic elements via tagmentation.
   Non‐coding sequences, including intronic and intergenic regions, often
   constitute enhancers, and compared with nontumor samples, eccDNAs in
   tumors present a more significant proportion of cis‐acting elements,
   particularly enhancers.^[ [77]^20 , [78]^21 ^] Notably, eccDNAs were
   significantly enriched in intergenic and genetic regions and short
   interspersed elements (SINEs) (Figure [79]1H), suggesting that these
   regions, rather than gene‐rich areas, were more likely to generate
   eccDNAs. Furthermore, in the intergenic region, the distribution of
   eccDNA in the LS group was significantly greater than that in both the
   HD and HS groups, indicating that the development of therapeutic
   response in the LS group may be correlated with enhancer activity,
   which is often affected by genetic perturbations and epigenetic
   changes.

2.2. Differential EccDNAs and EccGenes between HS and LS Patients

   Furthermore, we sought to investigate whether there were differences in
   the content of eccDNAs between HS and LS patients. 2484 eccDNAs were
   preliminarily screened with a threshold of Log2FC >2 and adjusted p
   < 0.05. Compared with those in the LS patients, 1803 eccDNAs were
   upregulated, and 681 eccDNAs were downregulated in the HS patients
   (Figure [80]2A). Furthermore, KEGG enrichment analyses were performed
   based on the differential eccDNA‐overlapping genes (eccGenes).
   Differential eccGenes between the HS and LS groups were enriched
   predominantly in processes such as human disease terms (cancer and
   antineoplastic drug resistance) and genetic information processing
   terms (replication, repair, and transcription) (Figure [81]2B). Among
   these genes, 44 eccDNAs were co‐upregulated (hyper‐upregulated), and 21
   were co‐downregulated (hypo‐downregulated) with eccGenes
   (Figure [82]2C). The locations of eccDNA shedding indicate specific
   chromosomal regions excised from linear chromosomes and subsequently
   circularized to form eccDNAs.^[ [83]^22 ^] We focused on 18 hyper‐up
   and 9 hypo‐down eccDNAs with eccGenes enriched in cancer‐ or
   myeloma‐related processes. These 27 eccDNAs revealed more shedding
   locations in LS patients, particularly those enriched on chromosome 3.
   Conversely, HS samples exhibited more eccDNA shedding locations on
   chromosomes 1, 5, 11, and 19 (Figure [84]2D). Then, we selected 10
   eccDNA candidates by higher‐ranking Log2FC and restricted the
   Circle‐seq signal by considering only those that were detectable in
   more than half of the samples. To further validate our sequencing
   results, qPCR and outward PCR were performed in another 10 NDMM cohorts
   (Figure [85]S2A,B, Supporting Information). Among them, the abundance
   of eccANKRD28 (chr3:15768491‐15768703, 213 bp) exhibited statistically
   significant differences according to both Circle‐seq and qPCR analysis
   (Figure [86]2E). We also intersected eccDNAs and eccGenes that were
   co‐regulated between the HS and LS groups and between the HS+LS and HD
   groups, resulting in two candidate eccDNAs (eccANKRD28 and eccMORC1,
   Figure [87]S2C, Supporting Information). Compared with those in the HS
   group, the proportions of eccANKRD28 and eccMORC1 were both more
   significant in the LS group, and relative to those in the HD group,
   they were also both greater in the MM group. In contrast, eccMORC1 did
   not show a statistically significant difference between groups in the
   validation cohort in Figure [88]2E, indicating that eccANKRD28 is
   involved in both the pathogenesis of MM and VRd resistance. Based on
   these integrative analyses, eccANKRD28 was chosen for further
   investigation.

Figure 2.

   Figure 2
   [89]Open in a new tab

   Differential eccDNAs and eccGenes between HS and LS patients. A)
   Volcano plot of differential eccDNAs detected between HS and LS
   patients. The upregulated and downregulated eccDNAs in HS and LS
   patients are marked in green and red, respectively (Log2FC > 2, P value
   < 0.05). B) KEGG pathway analysis of total differential eccGenes in
   NDMM patients. C) Venn diagrams showing the overlap of all differential
   eccGenes and eccDNAs. Left, upregulated targets in both sequences
   (blue); Right, downregulated targets in both sequences (red). D)
   Chromosomal distribution of 44 hyper‐up (green square) and 21 hypo‐down
   (orange circle) eccDNAs. The color intensity indicates differences in
   eccDNA number. E) Bar plot and split violin plot for qPCR validation of
   the differences in normalized counts of 10 candidate eccDNA Circle‐seq
   cohorts and another 5 HS (yellow) and 5 LS (blue) samples. ^* P < 0.05;
   ^** P < 0.01; ^*** P < 0.001. F) Outward and inward PCR detection of
   eccANKRD28 in RPMI‐8226, U266, and OPM‐2 cells. The DNA bands marked
   with blue boxes were gel‐purified and sequenced. G) Detection of
   eccANKRD28 in the RPMI‐8226 cell line (gray, left) by FISH: metaphase
   spread of chromosome (blue) was stained with a probe (green) against
   the eccANKRD28 locus chr3:15768491‐15768703 (middle). C4‐2B (right) is
   the negative control cell line, for its spread does not have an
   extrachromosomal DNA signal. Yellow arrows mark the eccDNA signals.

   To verify the specific circular structure of eccANKRD28, inward PCR and
   outward PCR were conducted in the RPMI‐8226, U266, and OPM‐2 cell lines
   (Figure [90]2F), with specific primers designed to target the junction
   sites of each candidate. The elimination of linear DNA was validated
   through PCR analysis of the COX5B gene, which was found to be absent in
   eccDNA, as previously reported (Figure [91]S2D, Supporting
   Information).^[ [92]^23 ^] The results revealed that eccANKRD28 was not
   present in the genomic DNA. The outward PCR products were utilized for
   Sanger sequencing to identify the junction sites of the eccDNA
   (Figure [93]S2E, Supporting Information). To confirm its existence, we
   also performed metaphase FISH using junction‐specific probes in
   RPMI‐8226 cells compared with C4‐2B (prostate cancer cell line) as a
   negative control validated in CytoCellDB (Figure [94]2G).^[ [95]^24 ^]
   For C4‐2B, the spreads do not show an eccANKRD28 signal. The positive
   signals in RPMI‐8226 representing eccANKRD28 were mainly observed off
   the chromosomes, confirming the extrachromosomal feature of eccANKRD28.

2.3. Potential Role of eccANKRD28 in the Prediction and Prognostic
Stratification of NDMM Patients

   In response to selective stress, such as carcinogenesis and drug
   resistance during cancer treatment, the copy number of eccDNA can
   change dynamically.^[ [96]^4 ^] To determine the origin of serum
   eccANKRD28, the fluorescence intensity in five CD138+ cells and paired
   CD138‐ cells was determined via FISH. We found that CD138+ MM cells had
   a much greater intensity of eccANKRD28 than other cells did, indicating
   that serum eccANKRD28 was derived mainly from CD138+ plasma cells in
   the bone marrow (Figure [97]S3A,B, Supporting Information). We
   conducted pathway enrichment analysis on the genes overlapping with
   highly abundant eccDNA in the LS group, revealing their enrichment in
   pathways related to canonical homologous recombination (cHR) and
   non‐homologous end joining (NHEJ) (Figure [98]3A), suggesting that the
   functions of cHR and NHEJ are correlated with increased eccDNA levels
   and amplified oncogene copy numbers within tumor cells, resulting in
   resistance to chemotherapy.

Figure 3.

   Figure 3
   [99]Open in a new tab

   Potential role of eccANKRD28 in the prediction and prognostic
   stratification of NDMM patients. A) Pathway enrichment of all
   differential eccGenes. B) eccANKRD28 copy number across junction sites
   was measured by qPCR. Spearman correlation analysis of the serum free
   light chain ratios of involved/uninvolved [FLCR(i/u)], β2‐microglobulin
   (β2‐MG), the percentage of bone marrow plasma cells (BMPC%), albumin
   (ALB), lactic dehydrogenase (LDH) and C‐reactive protein (CRP) with the
   serum eccANKRD28 copy number in 98 NDMM patients in the Renji‐VRd
   cohort. Each dot represents an independent individual. The gray shaded
   area represents the 95% CI. The Spearman correlation coefficient (r), P
   value (p), and sample number (n) are indicated. C) Stack bar chart of
   chromosomal abnormality events between the eccANKRD28‐high and
   eccANKRD28‐low groups, broken down by cytogenetic group. D) Bar plot
   reporting the number of high‐ and standard‐risk cytogenetics subgroups
   between the eccANKRD28‐high (red) and eccANKRD28‐low (blue) (p = 0.001,
   Fisher's exact test). The high‐risk was defined as the presence of at
   least one of the following: t(4;14), t(14;16), del(17p), and amp(1q).
   E) Stack bar chart of VRd responses between the eccANKRD28‐high (red)
   and eccANKRD28‐low (blue) groups. F) ISS stage and R‐ISS stage
   correlation with eccANKRD28 copy number. ^* P < 0.05; ^** P < 0.01;
   ^**** P < 0.0001. G) Kaplan–Meier curves depicting PFS (p = 0.0007) and
   OS (p = 0.17) in the 98 NDMM patients in the Renji‐VRd cohort
   stratified by the copy number of eccANKRD28. H) Kaplan–Meier curves of
   PFS (p = 0.029) and OS (p = 0.062) in the 98 NDMM patients in the
   Renji‐VRd cohort stratified by eccGene ANKRD28. P values were derived
   from a two‐sided log‐rank test without correction for multiple
   hypotheses.

   To elucidate the correlation between eccANKRD28 and myeloma status, we
   investigated the correlation between eccANKRD28 CNV and the clinical
   characteristics of the serum of NDMM patients. The detection of
   circular DNA in liquid biopsies is potentially helpful in identifying
   tumorigenesis and progression; however, several difficulties still
   exist. Most importantly, reliable methods for quantifying the abundance
   of circular DNA are lacking.^[ [100]^25 ^] Thus, we validated the
   abundance of eccANKRD28 through copy number variation (CNV)
   quantification via qPCR. Among the parameters, eccANKRD28 CNV was
   positively correlated with the involved/uninvolved free light chain
   ratio [FLCR(i/u)] (p = 1.0e‐9, r = 0.62, n = 80), β2‐MG (p = 0.04, r =
   0.24, n = 78), and bone marrow plasma cell percentage (BMPC%) (p =
   2.6e‐14, r = 0.79, n = 61), where FLCR(i/u) and BMPC% are both key
   parameters used to assess the tumor burden in myeloma. Notably,
   eccANKRD28 CNV was negatively correlated with ALB (p = 0.04, r = ‐0.24,
   n = 74) (Figure [101]3B). The distribution of chromosomal abnormalities
   varied little between the eccANKRD28‐high and eccANKRD28‐low groups
   (Figure [102]3C). However, there were significant differences between
   high‐ and standard‐risk cytogenetics subgroups (p = 0.001)
   (Figure [103]3D). Furthermore, the distribution among serial VRd
   responses also differed significantly between the low‐ and
   high‐eccANKRD28 groups. Most notably, patients who achieved MR/SD/PD
   were exclusively in the eccANKRD28‐high group (Figure [104]3E). Next,
   we explored the potential application of eccANKRD28 CNVs for the
   prognostic stratification of NDMM. Higher ISS and R‐ISS stages
   predicted relapse and mortality. The eccANKRD28 CNV relatively
   increased with increasing ISS and R‐ISS stages (Figure [105]3F),
   indicating the broad involvement of eccANKRD28 amplification in
   predicting treatment response.

   The amplification of eccDNA‐based oncogenes is associated with
   aggressive cancer behavior, including tumor growth, metastasis, and
   drug resistance, resulting in poor outcomes in cancer. Thus, we
   reinforced the above observations in an NDMM cohort, including 98
   patients assigned to the VRd regimen at Renji Hospital from Jan 5,
   2018, to Dec 28, 2023. At a median follow‐up of 30 months (IQR 8–60 m),
   patients with high eccANKRD28 copy numbers at diagnosis had
   significantly poor PFS (median PFS 29 months versus not reached), with
   a striking hazard ratio (HR) of 3.14 (95% CI 1.64–5.99, p = 0.0007).
   The median overall survival (OS) was not reached in either group (HR =
   4.06, 95% CI 0.70–23.41; p = 0.17; Figure [106]3G). Similarly, we also
   found that serum ANKRD28, the eccGene amplified from eccANKRD28, was
   inversely correlated with poor PFS (median PFS 30 months versus 45
   months, HR = 2.06, 95% CI 0.93–4.54; p = 0.029) in the Renji‐VRd
   cohort. The eccGene corresponding median OS was not reached in either
   group (HR = 4.68, 95% CI 0.61–36.12; p = 0.062; Figure [107]3H). Then
   we collected 5 CD138+ BMPC samples from the Renji‐VRd cohort and found
   that eccANKRD28 was not present in the genomic DNA of MM patients
   (Figure [108]S3C, Supporting Information). Therefore, we used the
   CoMMpass‐VRd data to further support the conclusions from our center
   work, described above (Figure [109]S3D, Supporting Information). Our
   findings suggest that the serum eccANKRD28 level could serve as a risk
   factor and a promising marker for monitoring MM progression.

2.4. EccANKRD28 Dynamics Correlate with the Development of VRd Resistance and
Amplify Transcription Activity

   To further identify the key role of eccANKRD28 dynamics in VRd‐related
   therapy resistance, we aimed to quantify the evolution of eccANKRD28
   from its inception experimentally. We chose CRISPR technology to
   generate endogenous eccANKRD28 in RPMI‐8226 cells (Figure [110]4A),
   which we term RPMI‐8226‐EC.^[ [111]^26 ^] We utilized lentiviral
   vectors to introduce Cas9 and single‐guide RNAs (sgRNAs) into cells and
   subsequently measured cell proliferation and eccDNA copy number. The
   transfection efficiency evaluated with mCherry+ cells was ≈80%, and the
   eccDNA copy number was calculated via a qPCR standard curve (Figure
   [112]S4A, Supporting Information). In addition, using parental cells
   (RPMI‐8226‐WT), we constructed bortezomib‐resistant (RPMI‐8226‐BR) and
   lenalidomide‐resistant (RPMI‐8226‐LR) cells, for which the IC[50] of
   the drug was increased by at least 4 fold (Figure [113]S4B, Supporting
   Information). The colony formation capacity (Figure [114]4B) was
   markedly enhanced. The γH2AX is a specific biomarker for DNA
   double‐strand breaks (DSBs). DSBs are the most deleterious DNA damage
   that contributes to genomic instability and increases tumorigenesis
   risk. Consequently, dysregulation of γH2AX is associated with human
   cancer. To elucidate the role of eccANKRD28 in MM and its relationship
   with γH2AX, we investigated the impact of elevated eccANKRD28 levels on
   the drug resistance of MM cells. As shown in Figure [115]4C, bortezomib
   resistance led to an increase in the number of residual γH2AX foci. On
   the other hand, γH2AX was also a key regulator of the DNA repair system
   after DSBs, thereby contributing to drug resistance of MM cells. This
   also indicated that the changes in eccANKRD28 abundance in MM‐resistant
   cell lines were due to endogenous DNA damage. As expected, the
   eccANKRD28 copy number was successfully increased via CRISPR/Cas9
   editing. Compared to RPMI‐8226‐WT cells, the IC[50] values were
   significantly increased in both RPMI‐8226‐BR and RPMI‐8226‐LR cells.
   Furthermore, CRISPR/Cas9 editing of eccANKRD28 desensitized parental
   RPMI‐8226 cells to bortezomib and lenalidomide treatment
   (Figure [116]4D). Correlative analyses of eccANKRD28 copy numbers and
   IC[50] values demonstrated high concordance with our models
   (Figure [117]4E), and we observed a substantial, dose‐dependent
   increase in eccANKRD28 copy number in response to bortezomib treatment
   (Figure [118]4F), which suggested strong positive selection for
   oncogene‐bearing eccANKRD28 in myeloma.

Figure 4.

   Figure 4
   [119]Open in a new tab

   EccANKRD28 dynamics are correlated with the development of VRd
   resistance and amplify transcription activity. A) CRISPR/Cas9‐mediated
   generation of custom eccDNA in RPMI‐8226 cells. Abbreviations:
   CRISPR/Cas9, clustered regularly interspaced short palindromic repeats;
   NHEJ, non‐homologous end joining. The graphical scheme was created with
   BioRender. B) Left: The representative colony formation images of
   RPMI‐8226, U266, and OPM‐2 cells between WT and BR groups. Right:
   Colony formation rates were expressed as the percentage of colonies in
   BR cultures compared with that in WT cultures. Each assay was performed
   in triplicate. C) Representative dual immunofluorescence‐FISH images
   for eccANKRD28 (green) and γH2AX (red) protein expression in U266
   wild‐type (WT) and bortezomib‐resistant (BR) cells. Scale bar, 10 µm.
   D) Dose‒response curves for the RPMI‐8226 wild‐type (WT), RPMI‐8226
   bortezomib‐ or lenalidomide‐resistant (BR or LR) and RPMI‐8226
   eccDNA‐CRISPR/Cas9 (EC) cell lines used to determine sensitivity to
   bortezomib and lenalidomide. The cells were incubated with bortezomib
   and lenalidomide at serial dilutions, starting at doses of 5 nm
   bortezomib and 5 µm lenalidomide, for 48 h before cell viability was
   determined via a CCK‐8 assay. Results are shown as the mean ± SD of
   three independent experiments. Statistical significance was assessed
   using two‐way ANOVA (^* P < 0.05; ^** P < 0.01; ^*** P < 0.001). E)
   Multiple correlation analysis between the IC[50] and eccANKRD28 copy
   number in RPMI‐8226, U266, and OPM‐2 cells. F) Mean eccANKRD28 copy
   number after eccANKRD28 induction on day 0 ± BTZ treatment beginning on
   day 3. G) Interphase FISH microscopy for eccANKRD28. Case #1 and Case
   #2 were representative images of CD138+ sorted plasma cells from NDMM,
   Case #1 was evaluated as MRD‐neg (HS group, n = 5), and Case #2 was
   evaluated as SD (LS group, n = 5) after three cycles of the VRd
   regimen. Representative images from a series of 10 images of each
   subject. Scale bar, 10 µm. H‐I) Copy number of eccANKRD28 and ANKRD28
   mRNA expression determined by qPCR in the HD/HS/LS group. ^** P < 0.01;
   ^*** P < 0.001.

   Interestingly, DNA FISH for eccANKRD28 in the nuclei of CD138+ BMPCs
   from patients with NDMM confirmed this finding (Figure [120]4G). In
   interphase nuclei, signal intensity analysis indicated that eccANKRD28
   signal intensity was preferentially located toward the nuclear
   periphery. Interestingly, when 5 HDs were compared with 10 HS and LS
   patients, the eccANKRD28 copy number was significantly correlated with
   the VRd‐low sensitivity. Still, it was not significantly associated
   with the VRd‐high sensitivity. Additionally, the eccANKRD28 DNA signal
   pattern was confirmed in both BMPCs. Moreover, the eccANKRD28 copy
   number was positively correlated with therapy response, accompanied by
   increased transcriptional activity (Figure [121]4H,I). Taken together,
   both the cell line and patient data agreed with models of eccDNA, along
   with their oncogenic content, which were subjected to significant
   selective pressure, affecting the mean eccDNA‐containing oncogene
   amplification.

2.5. H3K27ac Enrichment Visualization and Enhancer EccANKRD28 Validation

   Recent findings have indicated that enhancer‐containing eccDNAs may
   function as mobile regulatory elements for chromosomal transcription.^[
   [122]^27 , [123]^28 ^] The enhancer activity promotes oncogene
   expression and facilitates chromosomal structural rearrangements. Since
   the chromosome region is entirely non‐coding, we mapped the active
   regulatory elements via H3K27ac ChIP‐seq. We identified strong H3K27ac
   signals directly within the ANKRD28 region of skew across 10 MMCLs
   (Figure [124]5A,B). Multiple lines of evidence support their functional
   activity: the enhancer eccANKRD28 interval chr3:15768491–15768703 is
   predicted to have significant binding overlap with transcription
   factors, including H3K27ac on the Cistrome DB
   ([125]http://cistrome.org/db/#/) and ENCODE screen (Figure [126]5C;
   Figure [127]S5A, Supporting Information).

Figure 5.

   Figure 5
   [128]Open in a new tab

   H3K27ac enrichment visualization and enhancer eccANKRD28 validation. A)
   H3K27ac ChIP‐seq track for signals in the ANKRD28 gene region of skew
   across 10 MMCLs (EJM, KMS‐12, LP‐1, MM.1S, MOLP‐2, MOLP‐8, NCI‐H929,
   OPM‐2, RPMI‐8226, and U266). B) Zoomed‐in image of the H3K27ac signal
   for eccANKRD28. The region (pink) in (A) and (B) is highlighted for the
   eccANKRD28 interval. C) Predicted factors bound in the eccANKRD28
   interval via the Cistrome Data Browser. D) ChIP‒qPCR validation of
   eccANKRD28 binding with the H3K27ac protein in RPMI‐8226, U266, and
   OPM‐2 cells. E) Confocal dual immunofluorescence‐FISH microscopy of
   eccANKRD28 (green) and H3K27ac (red) colocalization in U266 metaphase
   cells. Box (yellow) is magnified (right). Scale bar, 5 µm.

   Furthermore, ChIP‒qPCR revealed that eccANKRD28 could directly bind to
   H3K27ac in three different MMCLs (Figure [129]5D; Figure [130]S5B,C,
   Supporting Information). Next, we assessed whether eccDNA foci
   colocalize within high local concentrations of the transcriptional
   machinery to create transcription hubs. Using dual
   immunofluorescence‐FISH, a distinct H3K27ac highly enriched signal was
   observed at the eccANKRD28 locus (Figure [131]5E). This observation
   provided the foundation to investigate the function of eccANKRD28 in
   gene regulation from an epigenetic perspective.

2.6. Enhancer EccANKRD28 Further Boosts VRd Resistance via Interaction with
the POU and RUNX Family in RRMM

   In eukaryotic genomes, the fundamental function of gene regulatory
   elements is the modulation of chromatin accessibility to facilitate the
   binding of transcription factors (TFs) and RNA polymerase. To identify
   the key TF mediating the activity of enhancer eccDNA, we integrated
   scATAC‐seq and scRNA‐seq datasets from 8 RRMM patients (Figure
   [132]6A–C). In agreement with previous reports,^[ [133]^29 ^] we found
   that the enhancer eccANKRD28 exhibited increased chromatin
   accessibility (open chromatin) and ultralong‐range chromatin contact,
   as measured by scATAC‐seq, demonstrating the correlation between the
   presence of numerous chromatin open sites and the effectiveness of
   targeted drugs (Figure [134]6D,E). A greater extent of coverage and
   larger chromatin open areas are associated with increased
   transcriptional activity in cells. Moreover, analysis of the
   genome‐wide distribution of cis‐regulatory elements showed that the
   three VRd response groups were all located mainly within the intronic
   regions, unlike those previously reported,^[ [135]^10 ^] which contain
   sequences of exons that form si‐like RNAs to repress the endogenous
   gene (Figure [136]6F). To understand whether different groups of peaks
   were enriched for binding sites of specific TFs, we performed motif
   enrichment mapping to the ENCODE consortium. Notably, the RUNX family
   (RUNX1/RUNX2/RUNX3) and CBFβ, the RUNX1/3‐binding partner,^[ [137]^30
   ^] were enriched mainly in the VRd‐S group, whereas the POU family
   (POU2F2/POU2F3/POU5F1B/POU3F4/POU2F1) was enriched in the VRd‐R group
   (Figure [138]6G). In brief, footprinting was used to call the
   differential motif binding of POU2F2 and RUNX1 within promoters, and we
   found that the eccDNA binding activity of POU2F2 was significantly
   decreased in the VRd‐S group and that the motif activity in the V/Rd‐R
   and VRd‐R groups was also more significant than that in the VRd‐S group
   (Figure [139]6H). Survival analysis was performed among the POU and
   RUNX gene families, and POU2F2, RUNX1, and RUNX2 were negatively
   associated with PFS and OS in the CoMMpass‐VRd cohort
   (Figure [140]6I–K). Super‐enhancers (SEs) and typical‐enhancers (TEs)
   are two genomic loci classes regulating cell identity genes. We thus
   used ROSE to identify SEs and TEs based on H3K27ac signals across 10
   MMCLs (Figure [141]6L; Figure [142]S7A, Supporting Information).
   Interestingly, the key VRd resistance‐associated TF genes POU2F2,
   RUNX1, and RUNX2 presented significantly higher expression levels in
   SEs than in TEs. Hence, we concluded that eccANKRD28 interacted with
   POU2F2‐RUNX1/2 to initiate enhancer effects and drove MM's aberrant
   gene regulatory program.

Figure 6.

   Figure 6
   [143]Open in a new tab

   Enhancer eccANKRD28 promotes VRd resistance via interaction with the
   POU and RUNX family. A) Characteristics for relapsed/refractory
   multiple myeloma (RRMM) subjects available in both scRNA‐seq and
   scATAC‐seq. Patient IDs in this manuscript are renumbered with IDs used
   in Tirier et al.^[ [144]^3 ^] Before treatment, drugs against which a
   patient has been refractory are marked with an R while marked with an S
   if sensitive. B,C) Uniform manifold approximation and projection (UMAP)
   projection of cell clusters detected in 8 RRMMs of paired scATAC‐seq
   and scRNA‐seq data. Cell clusters are labeled based on patient ID in
   (B) and VRd response in (C). D) QC filtering plot for cells of RRMM3
   and RRMM8 showing the TSS enrichment score versus unique nuclear
   fragments per cell. The dot color represents the density in AU of
   points in the plot. Dashed lines represent the filters for high‐quality
   single‐cell data (3000 unique nuclear fragments and TSS score greater
   than or equal to 7). E) Co‐accessibility shows eccANKRD28 track
   coverage identified by the scATAC‐seq signal by different responses. F)
   Bar plot showing the number of final peaks identified across patients
   with different VRd responses from the scATAC‐seq dataset. Bars are
   colored based on the location of the peak in the promoter, distal,
   exonic, or intronic regions. G) ComplexHeatmap enrichment for motifs
   enriched in scATAC‐seq peaks up‐ or downregulated in different VRd
   responses. H) TF footprint plot of POU2F2 and RUNX1 by different VRd
   responses. The red line represents the V/Rd‐R, the blue line represents
   the VRd‐R, and the green line represents the VRd‐S group. I–K)
   Kaplan–Meier plot for PFS and OS comparing 312 patients in the
   CoMMpass‐VRd cohort showing increased POU2F2 expression (I) (P = 0.027
   for PFS and P = 0.047 for OS), increased RUNX1 expression (J) (P =
   0.018 for PFS and P = 0.062 for OS). Increased RUNX2 expression (K) (P
   < 0.0001 for PFS and P < 0.0001 for OS) was associated with poor
   survival in MM patients. The log‐rank test evaluated significance. L)
   ROSE calling via H3K27ac ChIP‐seq revealed super‐enhancers (SEs, red)
   and typical enhancers (TEs, blue) in U266 and OPM‐2 cells (another 8
   MMCLs in Figure [145]S7A, Supporting Information). SEs were determined
   by the inflection point of the plot, VRd resistance‐associated SE‐TFs
   were annotated along the vertical axis, and VRd resistance‐eccDNAs were
   annotated along the horizontal axis. Box and whisker plots displaying
   the FPKMs for VRd resistance‐associated TF genes between SEs and TEs.
   Abbreviations: VRd‐R, bortezomib and lenalidomide double resistant;
   V/Rd‐R, either bortezomib or lenalidomide resistant; VRd‐S, bortezomib
   and lenalidomide double sensitive.

2.7. POU2F2‐RUNX1/2‐Based Network Regulates VRd‐Associated Cancer Epigenetics

   POU2F2 (POU class 2 homeobox 2), also known as OCT2, is widely
   expressed in B cells and B‐cell lineage tumor cells but has not been
   previously studied in MM.^[ [146]^31 , [147]^32 ^] To further elucidate
   the potential functional significance of POU2F2, we next performed a
   CUT&Tag assay. As expected, POU2F2 upregulated its downstream RNA level
   directly through RNA‐seq and CUT&Tag chromatin profiling, and POU2F2
   peak levels were increased near transcription start sites (TSSs) in
   U266‐BR cells compared to U266‐WT cells (Figure [148]7A; Figure
   [149]S7B, Supporting Information). In line with our CUT&Tag results,
   the POU2F2 protein level was also significantly increased in the BR
   cell lines RPMI‐8226 and U266 (Figure [150]7B; Figure [151]S7C,
   Supporting Information). To elucidate the mechanism of POU2F2 in VRd
   resistance, the target genes that POU2F2 may directly regulate were
   analyzed via macs2. Motif analysis for POU2F2‐bound peaks shows that
   under the influence of the insulator CTCF, a strong enrichment was
   recognized by RUNX1 and RUNX2 transcription factor motifs
   (Figure [152]7C). To validate their interaction, we conducted the
   coimmunoprecipitation (Co‐IP) assay. POU2F2 revealed a pull‐down of
   RUNX1 and RUNX2 proteins in U266 and OPM‐2 cells (Figure [153]7D). In
   other words, POU2F2 interacted with RUNX1/2 to form the protein
   complex. Subsequently, we sought to decipher the structural basis of
   the POU2F2–eccANKRD28 interaction. As shown in Figure [154]7E, the
   sequence‐specific POU2F2‐eccDNA interaction is mediated by numerous
   base pair‐specific hydrogen bonds. We conducted ChIP‐qPCR
   (Figure [155]7F) and dual immunofluorescence‐FISH (Figure [156]7G)
   experiments to verify the molecular docking result and found POU2F2
   could bind to enhancer eccANKRD28 with a strong relationship of
   colocalization. Further, analysis of sequencing data from MMRF‐CoMMpass
   found that POU2F2, RUNX1, and RUNX2 gene mutations were relatively rare
   (1.36%, 2.09%, and 0.83%, respectively) among MM patients compared to
   other tumor types profiled in TCGA, suggesting that their functions are
   preserved in MM (Figure [157]7H). To probe the molecular mechanism of
   VRd resistance, we next set out the target up‐regulation genes of the
   network of TFs downstream of POU2F2‐RUNX1/2. By integrating CUT&Tag and
   RNA‐seq datasets, we identified the activation and enhanced
   transcription of multiple oncogenes, including IRF4, RUNX3, IKZF3,
   BCL2, and JUNB (Figure [158]7I; Figure [159]S7D, Supporting
   Information), which were known to be related to drug resistance and
   validated in the CoMMpass‐VRd dataset (Figure [160]7J) as well as
   Renji‐VRd cohort (Figure [161]7K). Our findings showed that the
   enhancer eccANKRD28 could bind to POU2F2, which directly acts on RUNX1
   and RUNX2 and their downstream drug resistance‐related oncogenes,
   forming a cascade transcriptional regulatory network.

Figure 7.

   Figure 7
   [162]Open in a new tab

   The POU2F2‐RUNX1/2 transcriptional network regulates VRd‐associated
   cancer epigenetics. A) Left: Heatmap of normalized CUT&Tag tag
   densities for POU2F2 binding sites called from U266‐WT and U266‐BR
   cells. Right: The Histogram showing the average tag density of CUT&Tag
   peaks is displayed below the heatmaps. B) Immunoblotting analysis of
   POU2F2 in RPMI‐8226‐WT, BR, and EC cells. C) Motif analysis revealed
   that the CTCF, RUNX1, and RUNX2 binding sites are within the POU2F2
   locus. D) The immunoprecipitation was performed using the anti‐POU2F2
   antibody in U266 and OPM‐2 cells and then blotted for the indicated
   proteins. E) Predicted interaction of protein structures of POU2F2 with
   RUNX1/2 and eccANKRD28. The docking module highlights the hydrogen
   bonds (yellow) connecting POU2F2 and eccANKRD28. F) ChIP‒qPCR
   validation of eccANKRD28 binding with the POU2F2 protein in RPMI‐8226,
   U266, and OPM‐2 cells. G) Representative dual immunofluorescence‐FISH
   microscopy of eccANKRD28 (green) and POU2F2 (red) colocalization in
   U266 cells. Box (yellow) is magnified (right). Scale bar, 10 µm. H)
   POU2F2, RUNX1, and RUNX2 mutation frequency in TCGA and MMRF‐CoMMpass.
   I) Venn diagrams showing the overlap of differentially up‐regulation
   genes in U266‐BR cells compared to U266‐WT cells through CUT&Tag and
   RNA‐seq datasets. J‐K) Oncogene expressions in the CoMMpass‐VRd cohort
   (J, n = 312) and Renji‐VRd cohort (K, n = 98), separately, grouped by
   individuals with VRd high sensitivity (HS), median sensitivity (MS,
   achieving VGPR) and low sensitivity (LS). ^* P < 0.05; ^** P < 0.01;
   ^*** P < 0.001.

2.8. Enhancer EccANKRD28 Accelerates Tumor Growth and Mediates Drug
Resistance in MM Xenograft Models

   It was also interesting to validate the ability of eccANKRD28
   manipulated MM cells to enhance drug resistance to bortezomib,
   lenalidomide, and their combination in vivo. Therefore, we used
   immunodeficient mice and RPMI‐8226‐WT/EC cells to develop a xenograft
   model. Two weeks after inoculation, these mice were randomized to
   treatment with intraperitoneal injections of vehicle or bortezomib,
   oral gavage of lenalidomide, or a combination of both (Figure [163]8A).
   As shown in Figure [164]8B, eccANKRD28 manipulated MM cells could
   effectively promote the growth of xenograft tumors compared with the
   parental cells. We subsequently tested if a higher abundance of
   eccANKRD28 could induce drug resistance to bortezomib or lenalidomide.
   Notably, tumor acceleration was observed in all animals injected with
   RPMI‐8226‐EC cells. Furthermore, the following treatments on
   tumor‐bearing mice demonstrated that eccANKRD28 could desensitize
   RPMI‐8226 cells to bortezomib. Although lenalidomide alone did not
   induce tumor growth delay, bortezomib alone did show some activity in
   this setting (Figure [165]8C,D), and the combination of both
   significantly enhanced the therapeutic effects of suppressing tumor
   growth compared to either agent alone (Figure [166]8D,E). The
   histopathological evaluation confirmed the upregulation of POU2F2 in
   RPMI‐8226‐EC manipulated tumors and indicated regimens mediated target
   protein degradation to some degree (Figure [167]8F). Western blot
   analysis further supported the ability of eccANKRD28 to induce
   resistance, with a marked increasing level of POU2F2 compared with
   parental cells (Figure [168]8G). Increased copy numbers of eccANKRD28
   were also observed in mice serum from MM‐bearing mice injected with
   RPMI‐8226‐EC cells with indicated regimens, including vehicle,
   bortezomib, lenalidomide, or a combination of both (Figure [169]8H).
   Similarly, RPMI‐8226‐EC manipulated xenografts displayed higher mRNA
   expression of irf4, junb, ikzf3, runx3, and bcl2 than RPMI‐8226‐WT
   MM‐bearing mice (Figure [170]8I). Taken together, the above results
   strongly suggest that eccANKRD28 could desensitize human MM cells to
   bortezomib/lenalidomide and promote tumor growth.

Figure 8.

   Figure 8
   [171]Open in a new tab

   Enhancer eccANKRD28 promotes tumor progression and desensitizes MM
   cells to VRd in vivo. A) Overview of the eccANKRD28 function in
   mediating drug resistance using MM subcutaneous xenograft model. The
   graphical scheme was created with BioRender. B) Images of subcutaneous
   tumors formed by RPMI‐8226‐WT or RPMI‐8226‐EC cells (n = 16 per group).
   C) Tumor volume assessments among the eight groups of mice at the
   endpoint. The groups of 1 to 8 are the group of RPMI‐8226‐WT+Vehicle
   (WT+Veh), RPMI‐8226‐WT+bortezomib (WT+BTZ), RPMI‐8226‐WT+ lenalidomide
   (WT+LEN), RPMI‐8226‐WT+bortezomib+lenalidomide (WT+B+L),
   RPMI‐8226‐EC+Vehicle (EC+Veh), RPMI‐8226‐EC+bortezomib (EC+BTZ),
   RPMI‐8226‐EC+lenalidomide (EC+LEN), and
   RPMI‐8226‐EC+bortezomib+lenalidomide (EC+B+L), respectively (n = 4 per
   group). D) The tumor volume of each group was measured twice a week. E)
   Tumor weights were analyzed at the endpoint. F) Representative IHC
   analyses for POU2F2 in subcutaneous tumors (scale, 10 µm). G)
   Immunoblot illustrating levels of POU2F2 in eccANKRD28 xenografts with
   vehicle or indicated treatments. H) The eccANKRD28 abundance in mice
   serum was calculated with qPCR. I) Heatmap of oncogene mRNA expressions
   in eccANKRD28 xenografts calculated with RT‐qPCR. The oncogene mRNA
   levels were first normalized to the GAPDH control and then compared to
   wild‐type (WT) levels, which were arbitrarily set at 1.0. The data from
   RT‐qPCR analysis were presented as mean ± SD of three individual
   experiments, and the data from animal experiments were presented as
   mean ± SEM. Statistical analysis was performed using a two‐way
   ANOVA.Veh, vehicle; BTZ, bortezomib; LEN, lenalidomide; B+L,
   bortezomib+lenalidomide. ^* P < 0.05, ^** P < 0.01, ^**** P < 0.0001.

3. Discussion

   Drug resistance is the most important problem in cancer treatment.
   Genetic abnormalities are important initiation factors for the
   development of MM, and the role of epigenetic changes in drug
   resistance has received increasing attention. Here, we report for the
   first time that enhancer eccANKRD28 manipulated MM cells have been
   demonstrated to facilitate drug resistance and promote MM progression
   both in vitro and in vivo. The POU2F2‐RUNX1/2 transcription factor
   cascade elucidates the epigenetic regulatory basis through which the
   enhancer eccANKRD28 induces VRd resistance (summarized in Figure [172]9
   ). These findings hold significant promise for developing critical
   prognostic and therapeutic clinical tools.

Figure 9.

   Figure 9
   [173]Open in a new tab

   Schematic representation for the role of eccANKRD28 in VRd‐resistant MM
   cells (created with FigDraw).

   The POU Class 2 homeobox 2 (POU2F2)‐encoded protein binds the octamer
   sequence 5′‐ATTTGCAT‐3′, a common transcription factor‐binding site in
   immunoglobulin gene promoters. Since NF‐κB activity can also contribute
   to developing chemotherapy resistance, inhibiting NF‐κB is a potential
   strategy for overcoming chemotherapy resistance in MM.^[ [174]^33 ^]
   Prior research has demonstrated that NF‐κB directly activates POU2F2
   transcription in other cancer cells, thus leading to POU2F2
   overexpression.^[ [175]^34 , [176]^35 ^] Its expression and binding
   activity to cis‐regulatory elements are associated with decreased
   patient survival.^[ [177]^36 ^] Bortezomib mainly inhibits the activity
   of NF‐kB through the ubiquitin‐proteasome system (UPS), thus exerting
   an anti‐MM effect. Our observations indicated a correlation between
   bortezomib sensitivity and POU2F2 activity, suggesting that POU2F2 is a
   second target of this medication. Furthermore, the activity of the
   POU2F2 pathway may serve as a biomarker for predicting the therapeutic
   efficacy of bortezomib and other drugs that inhibit NF‐κB. Ikaros
   family zinc fingers (IKZFs) can form a trimeric complex with the
   RUNX1‐binding partner CBFβ, suggesting that RUNX1 and RUNX3 desensitize
   MM to lenalidomide via protection of IKZFs from degradation.^[ [178]^30
   ^] IRF4 orchestrates oncogenic transcription in MM. The IMiDs, such as
   lenalidomide, can indirectly reduce IRF4 expression through
   CRBN‐dependent degradation of IKZF1 and IKZF3.^[ [179]^37 , [180]^38 ^]
   Myeloma cell‐derived RUNX2 promotes myeloma progression,^[ [181]^39 ^]
   and the hypermethylation or alteration of RUNX2 contributes to B‐cell
   lineage resistance.^[ [182]^40 , [183]^41 ^] JUNB can protect MM cells
   against Vd‐induced cell death.^[ [184]^42 ^] BCL2 is a common, highly
   expressed prosurvival (antiapoptotic) protein in neoplastic plasma
   cells.^[ [185]^43 ^] In summary, activating the transcription of these
   oncogenes directly regulated by POU2F2 results in drug resistance. The
   CCCTC‐binding factor (CTCF) functions as a versatile transcriptional
   regulator capable of binding to boundary elements and facilitating the
   establishment of chromatin loops in conjunction with cohesin proteins.
   CTCF is also necessary for the maintenance of enhancer‐centric and
   NF‐κB‐dependent chromatin interactions.^[ [186]^44 ^] The predominant
   genomic sites where CTCF binds in MM cells are in intergenic regions,
   with some sites overlapping with transcriptional enhancers.

   To improve the therapeutic efficacy of PI or IMiDs, endogenous DNA
   generation inhibition should be first prioritized. Environmental stress
   causes genomic instability and drives local chromosomal breakage,
   leading to eccDNA formation by self‐joining DNA segments.^[ [187]^45 ,
   [188]^46 ^] In the absence of centromeres, eccDNA elements are
   inherited unequally, contributing to intratumoral heterogeneity and
   conferring a proliferative or survival advantage to cancer cells under
   drug pressure. Following this idea, we used CRISPR/Cas9 to generate a
   genic and intergenic eccDNA, eccANKRD28, to improve our understanding
   of the role of eccANKRD28 in drug resistance, which may facilitate the
   modeling of human genetic disorders caused by eccDNA. Both cHR and NHEJ
   participate in eccDNA genesis and the suppression of eccDNA exclusion
   by fostering cell cycle arrest at the G2/M checkpoint.^[ [189]^47 ^]
   Inhibition of cHR or NHEJ activity may reduce the copy number of
   eccDNA. In addition, eccDNA‐based strategies can disturb the binding of
   eccDNA, disrupt critical TF‐eccDNA interactions, and restrict binding
   at the epigenetic level by modulating chromatin accessibility.
   Approaches to directly target TFs, such as POU2F2, RUNX1, and RUNX2,
   include inhibiting their expression (RNA interference or miRNAs). They
   are among the most promising anti‐MM strategies with a potentially high
   therapeutic index. However, large surface areas for TF‒DNA
   interactions, which are difficult to target, and predominant nuclear
   localization, which makes TFs less accessible to therapeutic agents,
   remain challenging.

   Although our NSG xenograft MM mouse model demonstrated that
   eccANKRD28‐manipulated MM cells mediate drug resistance in vivo, this
   approach has limitations compared to the Cre‐loxP system. A recent
   study successfully utilized the Cre‐loxP system to simulate ecDNA
   generation and its role in mouse tumorigenesis; however, this approach
   is currently limited to large ecDNA fragments (>1 Mbp).^[ [190]^48 ^]
   No established system can safely and stably introduce and maintain
   specific small eccDNA fragments (such as the 213 bp eccANKRD28 in our
   study) into animal models via gene editing. Additionally, the NSG model
   lacks a functional immune system, failing to recapitulate the complex
   tumor microenvironment (TME) critical for drug resistance. The TME,
   including immune cells, stromal cells, and extracellular matrix
   components, plays a critical role in tumor progression and drug
   resistance. In contrast, the Cre‐loxP model allows for conditional
   manipulation of eccANKRD28 in specific cell types within an intact
   immune system, providing a more physiologically relevant context.
   Differences in drug metabolism between NSG mice and humans may also
   affect the interpretation of drug resistance mechanisms.

   The overexpression of POU2F2 is correlated with aggressive cancer
   phenotypes and unfavorable prognoses across various malignancies.
   Consequently, our observations could apply to other
   hematologic malignancies, including DLBCL and AML, characterized by
   chromosome instability.^[ [191]^49 ^] We highlighted that serum eccDNA
   could predict the treatment response, which makes them ideal candidates
   for non‐invasive biopsy biomarkers. Nonetheless, it is essential to
   recognize the constraints due to this research's relatively modest
   sample size. Our findings thus require further investigation and
   validation in larger cohorts with uniform treatment, as eccANKRD28 has
   the potential to introduce a minimally invasive drug sensitivity
   methodology to identify individuals with VRd‐resistance and
   POU2F2‐RUNX1/2 transcriptional cascade‐targeted therapies in patients
   with high eccANKRD28 copy number. Our study highlights how integrative
   multi‐omic analysis can be applied to explore and identify novel
   eccDNA‐oriented targets against multi‐drug resistance from an
   epigenetic perspective.

4. Experimental Section

Human Samples

   A total of 108 serum samples from 10 healthy donors and 98 NDMM
   patients were analyzed in this study, and the clinical information was
   described in Table [192]S1 (Supporting Information). Briefly, serum
   samples were obtained before administering three cycles of the VRd
   regimen. Afterward, their therapeutic responses were evaluated.
   Peripheral blood samples were collected and centrifuged at 1600 × g for
   10 min at 4 °C. The serum portion was centrifuged at 16000 × g for
   10 min at 4 °C to remove residual cells and debris. Three distinct
   groups were selected from the cohort: healthy donors (HD, n = 10) and
   NDMM patients who either responded well (achieving sCR/CR, high
   sensitivity, HS, n = 28) or poorly (achieving PR/MR/SD/PD, low
   sensitivity, LS, n = 30) to VRd treatment. Five individuals per group
   were subsequently used for the discovery cohort (NDMM1‐10) via
   Circle‐seq and RNA‐seq. Another five individuals per group were used
   for the validation cohort (NDMM11‐20) via qPCR and Sanger‐seq, and the
   clinical information of these two cohorts was described in Table
   [193]S2 (Supporting Information). The remaining samples were used in
   subsequent experiments, especially for copy number quantification. All
   experiments of candidate eccDNA function in this study comparing the HS
   and LS groups were carried out in separate cohorts. Among the 98 NDMM
   patients, 84 were enrolled in the clinical trial, which was registered
   at [194]www.chictr.org.cn as #ChiCTR2000036105, and the other 14 were
   retrospectively registered. The cohort was selected to assess the
   response to the VRd regimen in NDMM patients and to screen for
   associated biomarkers for more accurate prediction. Written informed
   consent was obtained from all participants. Patient clinical response
   was established according to the IMWG guidelines,^[ [195]^50 , [196]^51
   ^] and cytogenetic risk classifications were based on the Mayo
   Stratification of Myeloma and Risk‐Adapted Therapy (mSMART) criteria.^[
   [197]^52 ^]

Cell Culture and Reagents

   The human MM cell lines RPMI‐8226, U266, and OPM‐2 were purchased from
   the American Type Culture Collection (ATCC, Manassas, VA). All MM cell
   lines were maintained in RPMI‐1640 medium (#C11875500BT, Gibco, Suzhou,
   China) supplemented with 10% FBS (#F0193, Sigma‐Aldrich, Missouri, USA)
   and 1x penicillin‐streptomycin (P/S) (#15140‐122, Gibco). C4‐2B and
   HEK293T cell lines were provided by Dr. Chen Yang (Fudan University,
   Shanghai, China) and maintained in DMEM medium ((#C11995500BT, Gibco)
   supplemented with 10% FBS and 1x P/S. KaryoMAX Colcemid (#15212012) was
   purchased from Gibco (Grand Island, NY, USA). Bortezomib and
   lenalidomide were provided by Chia Tai Tianqing Pharmaceutica Group Co.
   (Nanjing, China).

   Mouse monoclonal anti‐POU2F2 (F‐5, #sc‐377475X), anti‐RUNX1 (A‐2,
   #sc‐365644), and anti‐RUNX2 (F‐2, #sc‐390351) were purchased from Santa
   Cruz Biotechnology (CA, USA). Rabbit monoclonal anti‐POU2F2 (#ab179808)
   antibody was purchased from Abcam (Cambridge, UK). Anti‐H3K27ac
   (#A2771) antibody was obtained from Abclonal Technology (Wuhan, China).
   Anti‐gamma H2AX (#9718T) was purchased from Cell Signaling Technology
   (Beverly, USA). Anti‐histone H3 (#06‐599) was obtained from
   Sigma–Aldrich (St. Louis, MO, USA). Mouse (#LF101) or rabbit (#LF102)
   secondary HRP‐conjugated antibodies, RIPA lysis buffer (#PC101), and
   IP/Co‐IP lysis buffer (#PC105) were purchased from Epizyme Biotech
   (Shanghai, China). The mouse IgG antibody (#B900620) was from
   Proteintech (Wuhan, China).

Serum EccDNA Purification, Circle‐Seq, and AFM Characterization

   The purification and enrichment of eccDNA from human serum were carried
   out as described previously with minor modifications.^[ [198]^53 ^]
   Briefly, serum DNA extraction was performed via a QIAamp Circulating
   Nucleic Acid Kit (#55114, Qiagen). DNA was digested with an exonuclease
   (Plasmid‐Safe ATP‐dependent DNase, #E3101K, Epicentre) at 37 °C for
   5 min to remove linear DNA, followed by column purification using
   QIAquick PCR Purification Kit (#28104, Qiagen). The complete removal of
   linear DNA was confirmed by PCR of the COX5B gene (Figure [199]S2D,
   Supporting Information). The enriched circular DNA was subjected to
   library preparation with a GenSeq Rapid DNA Lib Prep Kit (#GS‐LC‐004,
   GenSeq Inc, China). Sequencing was performed on an Illumina NovaSeq
   6000 with 150 bp paired‐end mode. Atomic force microscopy (AFM)
   measurements were performed with Dimension icon ScanAsyst (Bruker,
   USA).

Bulk RNA‐Seq

   Total RNA was used to remove the rRNAs with a GenSeq rRNA Removal Kit
   (#GS‐LC‐010, GenSeq Inc). According to the manufacturer's instructions,
   the rRNA‐depleted samples were subsequently subjected to library
   construction with the GenSeq Low Input RNA Library Prep Kit
   (#GS‐LC‐032, GenSeq Inc). Libraries were controlled for quality and
   quantified via the BioAnalyzer 2100 system (Agilent Technologies, Santa
   Clara, CA, USA). Library sequencing was performed on an Illumina
   NovaSeq instrument with 150 bp paired‐end reads.

MM Cell EccDNA Purification

   The cells were suspended in L1 solution (Plasmid Mini AX, #01050, A&A
   Biotechnology) supplemented with proteinase K (#25530049, ThermoFisher
   Scientific, USA) before incubation overnight at 50 °C with agitation.
   After lysis, the samples were treated with an alkaline solution,
   followed by the precipitation of proteins and the separation of
   chromosomal DNA from circular DNA through an ion exchange membrane
   column. Column‐purified DNA was treated with FastDigest MssI (#FD1344,
   ThermoFisher Scientific) to remove mitochondrial circular DNA and
   incubated at 37 °C for 16 h. The remaining linear DNA was removed by
   exonuclease (#E3101K, Epicentre) at 37 °C in a heating block, and the
   enzyme reaction was carried out continuously for 1 week, with
   additional ATP and DNase added every 24 h (30 units per day) according
   to the manufacturer's protocol. EccDNA‐enriched samples were used as
   templates for phi29 polymerase amplification reactions (REPLI‐g Midi
   Kit, #150043, Qiagen), which amplified eccDNA at 30 °C for 2 days,
   followed by purification using QIAquick PCR Purification Kit (#28104,
   Qiagen).

Establishing VRd‐Resistant Cell Lines

   Cells were grown in their respective culture media and passaged for at
   least two generations after thawing to ensure proper viability. The
   parental drug‐naive cells were imitated by 5 nm bortezomib and/or 5 µm
   lenalidomide and enhanced by doubled dosage every 1 month up to 6
   months total. The acquisition of a resistant phenotype was monitored
   and confirmed by calculating the IC[50] using the CCK‐8 assay. Cells
   with IC[50] over five times were kept for further experiments. The
   resistant RPMI‐8226 cells reached a maximum concentration of 30.28 nm
   bortezomib and 55.20 µm lenalidomide tolerance, the U266 cells reached
   a maximum periodic dose of 16.79 nm bortezomib, and the OPM‐2 cells’
   most significant sustainable periodic dose of bortezomib was 57.19 nm
   (Figure [200]S4B, Supporting Information). All cells were tested
   negatively for mycoplasma contamination, and STR was authenticated if
   cultured over 6 months.

Outward PCR and Inward PCR

   Outward directing PCR oligos were designed in Primer3web (v4.0) and
   devised to yield products across the junctions of 10 detected circular
   DNA structures (Tables [201]S3,S4, Supporting Information). Each 10 µL
   PCR mixture typically included 120 ng of phi29‐amplified template (2
   µL), 10 µm primer, and 2 × Phanta Max Master Mix (5 µL), and the
   reaction conditions were 10 min at 95 °C followed by 40 cycles of
   15 sec at 95 °C and 60 sec at 60 °C. Eight serial dilutions of purified
   human plasmid DNA were used to produce standard curves for each run,
   and eccDNA copy numbers were calculated. Concentrations of standards
   were measured using the Qubit dsDNA High Sensitivity Assay (#32854,
   ThermoFisher Scientific). Inward‐designed oligos (Table [202]S5,
   Supporting Information) represent positive controls for PCRs with
   circular and linear DNA templates. Size separation of the PCR products
   via 2% agarose gel electrophoresis and Sanger sequencing confirmed the
   circular structure of the selected eccDNAs. The median copy number was
   used as a cutoff to categorize high and low eccANKRD28.

Quantitative RT–PCR (qRT‒PCR)

   Total RNA was extracted using TRIzol (Cat#15596018, Invitrogen, USA)
   and reverse‐transcribed into complementary DNA (cDNA) with PrimeScript
   RT Master Mix (#RR036A, Takara, Tokyo, Japan). qRT‒PCR was carried out
   using a CFX96 real‐time PCR instrument (Bio‐Rad, California, USA) and
   TB Green Premix Ex Taq II (#RR820A, Takara) according to manufacturer
   instructions. A melting curve was created for each amplicon to verify
   its accuracy. The levels of target mRNAs were normalized to those of
   GAPDH.

Metaphase Chromosome Spreads

   Chromosome spreads were obtained by dropping mitotic cells on glass
   slides. Briefly, cells were concentrated in metaphase by treatment with
   100 ng mL^−1 KaryoMAX Colcemid (#15212012, Gibco, USA) for 3 h and then
   incubated overnight. The cells were washed once with PBS, and a
   single‐cell suspension was incubated in 75 mm KCl for 15 min at 37 °C.
   Cells were then fixed with Carnoy's fixative (3:1 methanol: acetic
   acid) and spun down. Cells were washed with fixative three additional
   times. Cells were then dropped onto humidified glass slides.

DNA Fluorescence In Situ Hybridization (FISH)

   CD138+ plasma cell or CD138‐ BM cell fractions were isolated using
   magnetic‐activated cell sorting technology (#130051301, Miltenyi
   Biotec, Auburn, CA). For interphase FISH, cells were directly
   harvested, washed with PBS, and fixed using Carnoy's fixative. The
   green SweAMI‐FISH probe (Servicebio, Wuhan, China) was specially
   designed for eccANKRD28
   (5′‐ATTTATTTACTTATTTAGATAATTATATGTTACACACATT‐3′). Briefly, slides were
   denatured at 85 °C for 10 min, then the FISH probe was diluted in
   hybridization buffer, added to the sample, and hybridized at 40 °C
   overnight in a humid and dark chamber. Slides were then washed with
   2 × SSC, 1 × SSC, and 0.5 × SSC at 37 °C for 5 min each. Drop the
   pre‐warmed secondary probe mixture and hybridize at 40 °C for 45 min.
   Slides were then washed successively with pre‐warmed 2 × SSC, 1 × SSC,
   0.5 × SSC, and 0.1 × SSC at 40 °C for 5 min each. Add pre‐warmed signal
   probe mixture and hybridize at 40 °C for 45 min and then wash again
   with pre‐warmed 2 × SSC, 1 × SSC, 0.5 × SSC, 0.1 × SSC for 5 min each.
   For dual immunofluorescence‐FISH, slides were followed by blocking with
   3% BSA, and primary antibodies against H3K27ac (#A2771, Abclonal) or
   γH2AX (#9718T, Cell Signaling Technology) were added at a final
   dilution of 1:100 followed by overnight incubation at 4 °C. The
   secondary antibody goat anti‐rabbit Cy3 (#A0516, Beyotime) was added at
   a final dilution of 1:200 for 1 h at room temperature. The slides were
   washed and mounted with DAPI. Images were captured under a confocal
   microscope (Olympus, FV3000, Japan).

CUT&Tag Assay

   U266‐WT and U266‐BR cells were harvested, counted, and centrifuged at
   600 × g for 5 min at room temperature. ≈100 000 cells per sample were
   subjected to CUT&Tag as previously described.^[ [203]^54 ^] Briefly,
   the cells were bound to Concanavalin A‐coated magnetic beads, and the
   cell membrane was permeabilized with digitonin. The enzyme pA‐Tn5
   transposase precisely binds the DNA sequence near the target protein
   under POU2F2 antibody (#ab179808, Abcam) guidance and results in
   factor‐targeted tagmentation. The DNA sequence was tagmented, with
   adapters added simultaneously at both ends, which could be enriched by
   PCR to form sequencing‐ready libraries. After PCR, the libraries were
   purified with AMPure beads, and library quality was assessed on an
   Agilent Bioanalyzer 2100 system. The library preparations were
   sequenced on an Illumina NovaSeq platform at Tianjin Novogene
   Bioinformatic Technology Co., Ltd. (Beijing, China), and 150 bp
   paired‐end reads were generated.

ChIP‒qPCR Assay

   A SimpleChIP Plus Enzymatic ChIP Kit (#9003, Cell Signaling Technology)
   was used to detect the ability of H3K27ac or POU2F2 to recruit the
   eccANKRD28 fragment. The cells were fixed with formaldehyde and then
   lysed, and the chromatin was partially digested with microcecal
   nuclease to form fragments. The chromatin fractions were incubated with
   one of the following antibodies: H3K27ac (#A2771, Abclonal) or normal
   rabbit IgG (#4620, Cell Signaling Technology), POU2F2 (#sc‐377475X,
   Santa Cruz Biotechnology) or normal mouse IgG (#B900620, Proteintech)
   with ChIP‐grade protein G magnetic beads. After protein‐DNA was
   decrosslinked, purification was performed using DNA purification
   centrifuge columns, and enrichment of eccANKRD28 or oncogenes was
   detected by qPCR. The qPCR system consisted of nuclease‐free H[2]O,
   5 µm primer, and SimpleChIP universal qPCR premix. A total of 40 cycles
   of the standard PCR program were performed according to the
   manufacturer's protocol. The signal obtained from each
   immunoprecipitation was expressed as a percentage of the total input
   chromatin: % of input = 1%× 2 (Ct 1% input sample−Ct IP sample), Ct =
   Threshold cycle of the PCR. The qPCR primers used for the ChIP assay
   are listed in Table [204]S6 (Supporting Information).

Cell Viability

   MM cell viability was measured using a CCK‐8 assay kit (Dojindo, Japan)
   following the manufacturer's instructions. The transfected cells were
   seeded into culture plates, and the incubated cells were treated with
   10 µL of CCK‐8 reagent. The absorbance was measured at 450 nm. IC[50]
   values were derived by a dose‐response inhibition curve and were fitted
   using nonlinear regression.

Immunoblotting and Coimmunoprecipitation (Co‐IP)

   All procedures followed the standard protocol previously reported.^[
   [205]^15 ^] The U266 and OPM‐2 cell lines were used for Co‐IP analysis.
   Cells were lysed in lysis buffer for 30 min on ice. The lysates were
   incubated overnight at 4 °C on a rotator with 5 µg of anti‐POU2F2
   (#sc‐377475X, Santa Cruz Biotechnology) and mouse IgG antibodies
   (#B900620, Proteintech). 30 µL of protein A/G magnetic beads
   (#HY‐K0202, MCE, USA) were transferred to the protein‐antibody
   complexes. Immunoprecipitates were collected after 2 h incubation. PBST
   was used to wash the pellet fraction four times before resuspending it
   in 50 µL of 1 × SDS‐PAGE loading buffer and western blotting with
   monoclonal antibodies against POU2F2 (#179808, Abcam), RUNX1
   (#sc‐365644, Santa Cruz Biotechnology), and RUNX2 (#sc‐390351, Santa
   Cruz Biotechnology), histone H3(#06‐599, Sigma–Aldrich).

Soft Agar Clonogenicity Assay

   RPMI‐8226, U266, and OPM‐2 cells were harvested after 180 days of
   treatment with dimethylsulfoxide (DMSO), bortezomib, or lenalidomide
   and seeded into a 6‐well plate at 1000 cells per well. After 24 h, the
   cells from each condition were treated with either DMSO, 50 nm
   bortezomib, or 5 µm lenalidomide over 20 days in triplicate. At 20 d,
   the cell culture medium was aspirated; the cells were washed gently
   with PBS, fixed in 4% PFA in PBS for 20 min, stained with 1% crystal
   violet, washed once with PBS, and dried for 30 min. The area intensity
   was calculated using ImageJ (NIH, USA). Three independent biological
   replicates were run for each cell line.

CRISPR/Cas9 Editing

   The construction of the RPMI‐8226‐CRISPR/Cas9‐eccDNA (RPMI‐8226‐EC)
   cell line was conducted through CRISPR–cas9‐mediated approaches with
   minor modifications.^[ [206]^26 ^] The insertion region was selected
   for the eccANKRD28 interval chr3:15768491‐15768703 (213 bp). The eccDNA
   sequences were retrieved from the UCSC Genome Brower, and repetitive
   and low‐complexity DNA sequences were annotated and masked by
   RepeatMasker. The guide sequences of sgRNAs were designed with the
   online software tool CRISPRCasFinder (Table [207]S7, Supporting
   Information). The guide sequence selected was constructed into
   U6‐CMV‐mCherry‐P2A‐puro‐WPRE. EccANKRD28 guide and control RNA
   lentiviral particles were produced by transfecting HEK293T cells
   together with packaging plasmids using the jetPRIME transfection
   reagent (#101000046, Polyplus, Illkirch, France) according to the
   manufacturer's instructions and following the standard lentivirus
   production protocol.^[ [208]^55 ^] The medium was replaced with RPMI
   medium without additives 4 h post‐transfection, and medium containing
   virus particles were collected 48 and 72 h post‐transfection. RPMI‐8226
   cells with eccANKRD28 region were infected with lentivirus; 3 days
   after infection, puromycin (0.75ug mL^−1) was added to the culture
   medium for 48 h. One week after transfection, the puromycin‐resistant
   cells were harvested, and mCherry+ single cell was sorted using a BD
   FACS Arial III and cultured in 96‐well until 70–80% confluence. Single
   clones typically evolve after 21 days in culture. The stable clones
   were further expanded, and their genomic DNA (gDNA) was extracted and
   subjected to genotyping with a pair of primers flanking the eccDNA
   region. The PCR product of the genotyping results was subjected to
   Sanger sequencing to confirm the insertion at the predicted cutting
   site. Establishing the stable RPMI‐8226‐CRISPR/Cas9‐eccDNA cell line
   could be validated by imaging eccANKRD28 with metaphase FISH, and
   RT‐qPCR assessed the copy number of the generated eccANKRD28.

Xenograft Experiments

   All animal experimental protocols were approved by the Medical
   Experimental Animal Care Commission of Renji Hospital (RJ2025‐007B).
   The investigation was conducted according to the ethical standards, the
   national and international guidelines, and the Committee for Ethical
   Review of Research Involving Animal Subjects at Renji Hospital
   (approval number: RJ2025‐007B). 2 × 10^6 RPMI‐8226‐WT or RPMI‐8226‐EC
   cells were injected subcutaneously into 6‐week‐old female NSG mice
   (strain: NOD.Cg‐Prkdc^scidIl2rg^em1Smoc , n = 16 mice per group). At
   day 14 after tumor establishment, the RPMI‐8226‐WT/EC tumor‐bearing
   mice were randomized into four groups [vehicle, bortezomib (0.5 mg
   kg^−1), lenalidomide (30 mg kg^−1), or combination of both. n = 4 mice
   per group], and treatments were started. Bortezomib (0.5 mg kg^−1) was
   administrated intraperitoneally three times per week, while
   lenalidomide (30 mg kg^−1 in 0.9% CMC‐Na) was administered daily via
   oral gavage with 0.9% CMC‐Na as vehicle control. Tumor growth was
   monitored twice a week by measuring tumor size in two orthogonal
   dimensions, and tumor volume was calculated by the following equation:
   tumor volume = (length × width^2)/2. After 10 days of treatment, the
   mice were euthanized, and the tumors were isolated and weighed. The
   tumors were then divided into two parts, one fixed in formalin for
   immunohistochemistry staining and the other stored in liquid nitrogen
   for immunoblotting and RNA exaction.

Immunohistochemistry (IHC)

   Tumor xenografts derived from in vivo mouse models were fixed in
   formalin for 24 h. Microarray slides were cut at 4 microns on plus
   slides. Slides were allowed to air dry. Slides were placed in a
   55–60 °C oven for 30 min. The IHC stains were performed on the Ventana
   automated immunostainer at 37 °C. Tissue sections were incubated with
   anti‐POU2F2 (#sc‐377475X, Santa Cruz Biotechnology) at a final dilution
   of 1:200 overnight at 4 °C. The evaluation of stained proteins was
   performed by assessing both the intensity (scored as 0, 1, 2, or 3) and
   the extent of staining (scored as 0, 0%; 1, <10%; 2, 10–50%; 3, >50%)
   as previously described.^[ [209]^56 ^]

Protein–DNA Interaction Docking Study

   Protein structures were obtained from PDB ([210]http://www.rcsb.org/;
   POU2F2: 1HDP, RUNX1: 3WTS, RUNX2: 6VG8). Docking of protein‐protein
   interactions was performed using ClusPro ([211]https://cluspro.org/),
   and protein‐DNA docking was conducted using HDOCK
   ([212]http://hdock.phys.hust.edu.cn/). PyMOL was used to generate 3D
   illustrations.

CUT&Tag/ChIP‐Seq Data Analysis

   CUT&Tag/ChIP‐seq data analysis was performed as previously described
   with slight modifications.^[ [213]^57 , [214]^58 ^] In summary, raw
   data in fastq format were first processed using fastp. Clean reads were
   mapped to the human genome version hg38 via Bowtie2. Only uniquely
   mapped (MAPQ ≥ 13) and deduplicated reads were used for further
   analysis. All peak calling was performed with macs2 using “macs2 ‐q
   0.05 ‐f AUTO –call‐summits –nomodel –shift ‐100 –extsize 200 –keep‐dup
   all”. Peaks with a q‐value threshold of 0.05 were used for all the
   datasets by default. Peaks were adjusted to the exact size‐centered
   peak sums, and motif discoveries of these loci sequences were performed
   using findMotifsGenome.pl program in HOMER v4.11 software with “‐len
   8,10,12,14 ‐gc ‐size given ‐homer2 ‐dumpFasta”. ChIPseeker was used to
   retrieve the nearest genes around the peak and annotate the genomic
   region of the peak.^[ [215]^59 ^] Peaks of different groups were merged
   via “bedtools merge”. The mean RPM of each group in the merged peak was
   calculated. Only peaks with a fold change in the RPM greater than 2
   were considered differential. Responses associated with different peaks
   were identified using ChIPseeker. Heatmaps were plotted using the
   computeMatrix and plotHeatmap commands in deepTools.^[ [216]^60 ^] The
   normalized CUT&Tag‐seq data were visualized using the Integrative
   Genomics Viewer (IGV).

scATAC‐Seq and scRNA‐Seq Analysis

   scATAC‐seq and scRNA‐seq downstream were reanalyzed with Seurat and
   ArchR, respectively.^[ [217]^3 , [218]^61 ^] All cells with a total
   number of <3000 fragments, a TSS enrichment score <7, a
   doublet‐enrichment score >6, and a predicted doublet score >100 were
   excluded (Figure [219]S6, Supporting Information). Myeloma cells from
   all patients were normalized, and the dimensionality was reduced via
   iterative latent semantic indexing and clustering. Peak calling was
   performed with macs2 motif deviations calculated via the JASPAR
   database. For each TF motif, a mean value was calculated.

Statistical Analysis

   Statistical analysis used GraphPad Prism 10, RStudio (version 4.2.2),
   and Python 3. The RESOURCES TABLES in [220]Supporting Information could
   provide details of the software used. Survival probabilities were
   estimated through the Kaplan‒Meier method. For mutation analyses, for a
   given gene, the somatic mutation frequency in each TCGA and the
   MMRF‐CoMMpass cohort (the unique mutated cases/total unique cases in
   the cohort) were calculated using the GDC Mutation Frequency Tool
   provided on the GDC portal ([221]https://portal.gdc.cancer.gov/).
   Two‐way ANOVA was used to calculate the difference in IC50 with various
   treatments. The significance of the variations was assessed using
   either the Kruskal‒Wallis test or the Student's t‐test. All p values
   reported were two‐tailed, with a threshold of P < 0.05 deemed
   statistically significant across all analyses. Data were presented as
   mean ± SD in cell experiments and mean ± SEM in animal studies, ^* P
   < 0.05, ^** P < 0.01, ^*** P < 0.001, ^**** P < 0.0001, and all above
   experiments have been repeated three times. All analyses were performed
   using the human genome reference sequence GRh38. Detailed information
   regarding the statistical tests employed for each experiment can be
   found in the corresponding figure legends.^[ [222]^62 , [223]^63 ^]

Ethical Statement

   The study was approved by the Ethics Committee of Renji Hospital,
   Shanghai Jiao Tong University School of Medicine (Approval No:
   SK2020‐086) and conducted following the Declaration of Helsinki.

Conflict of Interest

   The authors declare no conflict of interest.

Author Contributions

   B.C. and J.H. conceived and designed the project. J.H. supervised the
   research. B.C. wrote the manuscript, performed all the experiments
   and/or bioinformatics analysis. Y.Z. performed the Circle‐seq analysis.
   B.C. and C.S. performed the CRISPR/Cas9 cell line editing. L.N., T.L.
   and J.H. gave advice and revised the manuscript. F.X., L.Z., M.Z and
   H.H. recruited and followed up with patients. J.L., D.Z. and G.Z.
   collected samples. All authors approved the final manuscript.

Supporting information

   Supporting Information
   [224]ADVS-12-2415695-s001.docx^ (6.3MB, docx)

Acknowledgements