Systems medicine dissection of chr1q-amp reveals a novel PBX1-FOXM1 axis for targeted therapy in multiple myeloma

: Understanding the biological and clinical impact of copy number aberrations (CNA) for the development of precision therapies in cancer remains an unmet challenge. Genetic amplification of chromosome 1q (chr1q-amp) is a major CNA conferring adverse prognosis in several types of cancer, including in the blood cancer multiple myeloma (MM). Although several genes across chr1q portend high-risk MM disease, the underpinning molecular aetiology remains elusive. Here, with reference to the 3D chromatin structure, we integrate MM patient multi-omics datasets with genetic variables to obtain an associated clinical risk Abstract Understanding the biological and clinical impact of copy number aberrations (CNA) for the development of precision therapies in cancer remains an unmet challenge. Genetic amplification of chromosome 1q (chr1q-amp) is a major CNA conferring adverse prognosis in several types of cancer, including in the blood cancer multiple myeloma (MM). Although several genes across chr1q portend high-risk MM disease, the underpinning molecular aetiology remains elusive. Here, with reference to the 3D chromatin structure, we integrate MM patient multi-omics datasets with genetic variables to obtain an associated clinical risk map across chr1q and to identify 103 adverse prognosis genes in chr1q-amp MM. Prominent amongst these genes, the transcription factor PBX1 is ectopically expressed by genetic amplification and epigenetic activation of its own preserved 3D regulatory domain. By binding to reprogrammed super-enhancers, PBX1 directly regulates critical oncogenic pathways and a FOXM1-dependent transcriptional programme. Together, PBX1 and FOXM1 activate a proliferative gene signature which predicts adverse prognosis across multiple types of cancer. Notably, pharmacological disruption of the PBX1-FOXM1 axis with existing agents (thiostrepton) and a novel PBX1 small-molecule inhibitor (T417) is selectively toxic against chr1q-amplified myeloma and solid tumour cells. Overall, our systems medicine approach successfully identifies CNA-driven oncogenic circuitries, links them to clinical phenotypes and proposes novel CNA-targeted therapy strategies in multiple myeloma and other types of cancer. The finding that pharmacological abrogation of the PBX1-FOXM1 axis selectively impacts survival of chr1q-amp myeloma cells is one of the most notable findings of this work. As well as providing proof-of-principle for developing CNA-specific therapeutic approaches, our data strongly support the central role of PBX1 and FOXM1 in regulation of the transcriptional programme driving the proliferative phenotype and adverse prognosis in chr1q-amp MM. In addition, these findings support our recent efforts for development of T417, a small molecule inhibitor of PBX1 binding to its cognate DNA motif 45 , and suggest the potential benefit of its use in MM and other cancers with chr1-amp and PBX1 over-expression. Indeed, along with our previously reported pre-clinical activity of T417 against ovarian cancer 45 , the data herein demonstrate selective targeting against MM, breast, lung, liver and brain cancer cells with chr1-amp. These findings not only validate the presence of a common, PBX1-FOXM1 axis underlying chr1q-amp that is active in many types of cancer, but also provide the basis for clinical development of T417 as a chr1q-amp-targeting therapy. In summary, we showed that our systems medicine of CNA in cancer, which includes integration of genetic, epigenetic, transcriptional and 3D-chromatin profiles, is a powerful strategy for discovery of genes and cellular oncogenic pathways of biological significance and clinical impact. show the ectopically expressed PBX1, in co-operation with FOXM1, critical driver of the proliferative phenotype in chr1q-amp MM and several other cancers, and we provide proof-of-principle for selective therapeutic targeting of chr1q-amp, the most prevalent CNA in MM and in cancer in general.

Previous studies, often guided by low resolution methodologies (e.g., FISH against 1q21 locus 12 ), identified several chr1q21 genes associated with adverse prognosis in MM, including the CKS1B, PDKZ1, ILF2, ARNT, ADAR1, IL6R, MCL1, BCL9 and MDM2 genes [13][14][15][16][17][18][19][20] . However, genetic amplification that extends beyond chr1q21 has been previously reported in a small cohort of MM patients 21 , and non-chr1q21 genes (e.g., CD1D, FCER1G) have been linked to the biology and prognosis of MM [22][23][24] . These observations raise the prospect that several areas across chr1q may contribute to the biological profile and clinical impact of chr1q-amp. Further, how genetic amplification affects the 3D chromatin architecture of chr1q and influences biological processes that promote high risk disease is not known. Understanding these processes could inform novel anti-cancer therapeutic approaches targeted to chr1q-amp that are currently lacking.
Here we employed a comprehensive systems medicine approach to resolve the 3D genome landscape of chr1q-amp and to integrate it with multi-omics MM patient datasets. This approach led to the identification of adverse prognosis genes across the whole chr1q arm, and particularly in the 1q22 and 1q23.3 bands. Amongst 1q23.3-associated genes, we identified the transcription factor PBX1, which, in co-operation with FOXM1, regulates myeloma PC proliferation and generates a selective therapeutic vulnerability in chr1q-amp MM that can be targeted by a novel PBX1 inhibitor.

Cell cultures
All cell lines and culturing conditions used in this study are detailed in the Supplemental methods section.

Primary samples
Bone marrow aspirate samples from MM patients and peripheral blood samples from normal donor were obtained upon written informed consent and under research ethics committee approval (Research Ethics Committee Reference: 11/H0308/9). Bone marrow aspirates were subjected to red cell lysis. MM plasma cells were purified after two rounds of CD138 immunomagnetic selection (Miltenyi Biotech) following the manufacturer's instructions, as previously described 25 . Pre-and post-selection purity was assessed by flow-cytometric analysis (BD LSR-Fortessa) using a panel of fluorochrome-labelled anti-CD138, -CD45, -CD19, -CD56 and -CD38 monoclonal antibodies. Purified cells were immediately processed for ATAC-seq and RNA-seq analysis or stored in FBS + 10% DMSO at -150 o C for later use.
Mononuclear cells from normal donor peripheral blood were isolated by Ficoll-Hypaque (Sigma-Aldrich) density centrifugation following the manufacturer's instructions, as described before 26 . The mononuclear cell interphase layer was aspirated, washed with 1ml PBS, centrifuged at 300g for 5min and resuspended in 100μl PBS. Peripheral blood B cells (PBBC) were isolated using the human Total B cell isolation kit II (Miltenyi Biotec) as per manufacturer's instructions.
Molecular cloning of the overexpression plasmid vector is described in the Supplemental methods section.

RNA-seq
Total RNA was extracted from FACS-sorted myeloma plasma cells using the Nucleospin RNA kit (Macherey-Nagel). The Qubit RNA Assay kit (Life Technologies) was used to determine the RNA quantity. Quality of RNA extracts was assessed on the Bioanalyser using the RNA pico kit (Agilent). Samples with RIN value higher than 8 were processed using the NEBNext poly(A) mRNA Magnetic Isolation kit and the NEBNext Ultra II RNA Library Prep kit for Illumina (New England Biolabs), following manufacturer's instructions. The Qubit High Sensitivity DNA kit (Life Technologies) was used for library quantification; library size was evaluated using the Bioanalyser High Sensitivity DNA kit (Agilent). Libraries from the same experiment were diluted to 5nM, pooled together and sequenced at the BRC Genomics Facility (Imperial College London) using the Illumina HiSeq 4000 platform to obtain paired-end 75bp reads.

ATAC-seq
ATAC-seq was performed as previously described 28 . Briefly, 50,000 purified myeloma plasma cells or myeloma cell lines, were washed with cold PBS (Sigma) at 500g at 4 0 C for 5 min. The cells were resuspended in 50 μL of cold Lysis Buffer (10 mM Tris-HCl, pH 7.4, 10mM NaCl, 3 mM MgCl 2 , 0.1% IGEPAL CA-630) and washed at 500g at 4 0 C for 10min. The nuclei were subjected to transposase reaction for 30min at 37 0 C; termination of the reaction and DNA purification was performed using a MiniElute Kit (Qiagen) and eluted twice with 10 μL. The purified DNA was amplified as described before with NEBNext High-Fidelity 2x PCR Master Mix (New England Biolabs). The PCR amplified product was cleaned twice with (0.9X) AMPure beads (Beckman). The quality of the libraries was assessed with the Bioanalyzer High Sensitivity DNA kit (Agilent). The libraries were quantified using the NEBNext Library Quant Kit for Illumina (New England Biolabs) on a StepOne Plus Real-Time PCR (Applied Biosystems). The libraries were sequenced at the Genomics Facility at ICL using the Illumina HiSeq 4000 platform to obtain paired-end 75bp reads.

Bioinformatics and clinical informatics analysis
All methods used for statistical, bioinformatics and clinical informatics analysis are described in the Supplemental methods section.

Data and code availability
High-throughput sequencing data generated during this study have been deposited to the Gene Expression Omnibus repository (GEO): MMCL ChIP-seq and RNA-seq files (GSE165060) and primary MM ATAC-seq and RNA-seq files (GSE153381).
Code used in this study can be accessed from the specified github page: https://github.com/nikostrasan/PBX1-project Additional materials and methods information are described in the Supplemental methods section.

Results
Distinct patterns of amplification within chr1q shape its 3D chromatin architecture We first explored whether and how genomic structural changes might impact the 3D chromatin structure of chr1q-amp myeloma cells. For this purpose, we constructed a correlation matrix of copy number scores across the chr1q arm (2D genome co-amplification map) using whole genome sequencing (WGS) data from MM patients (MMRF CoMMpass database 29 , n=896) and compared it with the 3D genome Hi-C contact maps of two chr1q-amp MM cell lines (MMCL; U266, RPMI8226 30 ; Figure 1A). By applying to the 2D genome map the same computational method used for topologically associated domain (TAD) discovery 30 , we found four main blocks (B1-B4) of genomic co-amplification (termed topologically co-amplified domains; TCDs), which define distinct amplification patterns across MM patients ( Figure 1A and Supplemental Figure S1A). Next, to understand the relationship between chr1q-amp and 3D chromatin structure, we compared the myeloma TCD and TAD maps, along with the Hi-C map of the reference GM12828 B cells 31 (Supplemental Figure S1B). First, we found that TAD organization was highly similar between the U266 and RPMI8226 myeloma cells but highly dissimilar amongst them and the GM12828 B cells (Supplemental Figure S1B), suggesting extensive re-organization of the chr1q-amp 3D genome in myeloma cells. Second, we identified weak overall similarity between the WGS co-amplification map and the 3D genome maps of U266, RPMI8226 and GM12828 cells, when we compared their insulation score profiles (Supplemental Figure S1B). In a complementary analysis that aimed to identify the genetic amplification breakpoints in all patients, we found that approximately 65% of the 135 GM12828 reference TADs are disrupted by chr1q-amp breakpoints (Supplemental Figure  S1C) supporting further the notion that genetic amplification extensively disrupts the 3D chromatin architecture of chr1q. In both analyses, we also identified a pericentromeric cluster of multiple breaks as previously described 32 .
Finally, despite the extensive 3D chromatin re-organization within the TCDs, we observed that the B1-B4 borders coincide with TAD borders in chr1q-amp MM, suggesting that organization of these four hyper-domains evolves in parallel with some level of 3D chromatin structure retention and preservation. (Supplemental Figure S1A, S1D)

Systems medicine analysis identifies adverse prognosis drivers beyond 1q21
Next, to identify all genes across chr1q that could potentially drive high-risk phenotype in MM and with reference to the 3D chromatin structure, we combined genomic (WGS; WES; whole exome sequencing), epigenomic (H3K27ac-seq), and transcriptomic (RNA-seq, DNA microarray) data with genetic variables from three studies: MMRF (n= 896); Arkansas (n=414); and Jin2018 (n=12) 29,30,33,34 ( Figure 1B). Of the 2,215 chr1q genes, we considered as candidate drivers of adverse prognosis only those fulfilling each of the following criteria: (1) their genetic amplification predicts adverse prognosis, independent of the prognostic impact of 73 other molecular markers (MMRF dataset; Supplemental Figure S1E); (2) their genetic amplification is significantly associated with their transcriptional overexpression (MMRF dataset); (3) their overexpression is significantly correlated with adverse prognosis (MMRF and Arkansas datasets); (4) their genetic amplification is accompanied by epigenetic activation (i.e., H3K27ac signal gain compared to non-amplified MM; Jin2018 dataset); ( Figure 1C and Supplemental Table S1). This stepwise analysis identified 103 candidate genes residing exclusively in B1 and B4 hyperdomains, including the previously known MCL1, CKS1B, ILF2 and ARNT genes located in chr1q21.3 [13][14][15] (Figure 1C). Pathway analysis of all 103 genes showed significant enrichment for cell cycle-related processes, suggesting their involvement in the proliferative phenotype that is associated with chr1qamp MM 17 (Supplemental Figure S1F). Interestingly, we identified 1q22 and 1q23.3 as the two cytogenetic bands that, relative to their gene density, contain the highest number of candidate adverse prognosis genes (Supplemental Figure S1G), with 1q23.3 displaying the highest association with adverse prognosis (Figure 1C and 1D). Therefore, there are additional regions, other than 1q21, which contribute to the high-risk, proliferative phenotype linked to chr1-amp in MM.

PBX1 is a novel biomarker of chr1q genetic amplification
Amongst 1q23.3 genes, the transcription factor PBX1 previously reported to promote cancer cell survival, metastasis and drug resistance [35][36][37] was notable for the highest H3K27ac signal gain across its own preserved TAD ( Figure 1E and Supplemental Figure S1D and S1H). These features comprise a unique case of amplification of an entire regulatory domain linked to epigenetic activation, gene overexpression and adverse prognosis. Further analysis using the MMRF dataset confirmed PBX1 as a marker of high-risk MM disease, with its amplification significantly correlating with its overexpression (Supplemental Figure S2A and S2B), while PBX1 overexpression was associated with high-risk clinical features, high myeloma plasma cell proliferative index, progressive/relapsed disease and worse overall survival (Supplemental Figure S2C-S2J).

The pro-proliferative role of PBX1 in chr1q-amp MM
We explored further the functional role of PBX1 in chr1q-amplified MM cells, by assessing its mRNA and protein expression levels across healthy and tumour cells. Based on RNA-seq data, we found that in normal hematopoiesis, PBX1 is expressed in bone marrow hematopoietic stem and progenitor cells as well as megakaryocytes, but not in B cells or plasma cells (Supplemental Figure  S3A). In MM, we confirmed ectopic expression of PBX1 in four chr1q-amp MMCL by RT-qPCR ( Figure  2A) and in 9/11 patient myeloma PC samples with FISH-verified chr1q-amp by RT-qPCR and immunohistochemistry, using breast cancer as a positive control ( Figure 2B Table S2). Transcriptome profiling of PBX1-depleted cells showed similar numbers of genes de-regulated in the two MMCL, while Gene Set Enrichment Analysis revealed significant enrichment for cell cycle-related pathways amongst down-regulated and interferon response pathways in up-regulated genes ( Figure 2H). This is consistent with the reported enrichment of interferon response pathways in early-stage, non-proliferative MM and of cell cycle-related pathways in advanced disease and MMCL 38,39 . Accordingly, flow-cytometric analysis showed significant G1-phase cell cycle arrest in PBX1-depleted MMCL ( Figure 2I, Supplemental Figure S3H).

Defining the epigenetic and regulatory programme of PBX1 in chr1q-amp cells
ChIP-seq analysis against PBX1 in MM.1S and U266 cells identified 30,000-40,000 binding sites ( Figure 3A and Supplemental Table S2). Further annotation using chromHMM maps (built upon ENCODE/Blueprint Consortium data) showed that 60-80% of PBX1 recruitment occurs in activechromatin promoter and enhancer areas, while motif enrichment analysis identified the PBX1 motif among the top hits ( Figure 3A and Supplemental Figure S4A-S4D). Additional analysis of H3K27acseq profiles from eight primary myeloma PC and nine MMCL 34 identified 2,400 super-enhancers (SEs), 70% of which are PBX1-bound ( Figure 3B). Sample stratification based on chr1q-amp status showed significantly higher H3K27ac signal in PBX1-bound SEs in chr1q-amplified versus nonamplified cells, suggesting extensive epigenetic reprogramming associated with PBX1 binding in chr1q-amplified myeloma cells ( Figure 3C and Supplemental Figure S4E-S4F). Interestingly, the PBX1-bound SEs in chr1q-amplified cells are predicted to regulate critical cellular pathways, including cell cycle ( Figure 3D).
Next, we integrated the PBX1 cistrome with the PBX1-depleted transcriptomes to generate the gene regulatory network of PBX1 in chr1q-amplified cells (Figure 3E and Supplemental Figure S4G-S4I and Supplemental Table S3). We identified approximately 700 and 300 genes to be directly activated and repressed, respectively, by PBX1 in both MM.1S and U266 MMCL. Again, among other prominent oncogenic pathways, the former were primarily enriched in cell cycle-related biological processes and the latter in interferon response pathways ( Figure 3E).

The PBX1-FOXM1 axis regulates cell proliferation in chr1q-amp MM
Amongst the PBX1-dependent targets, we detected significant enrichment of the pro-proliferative FOXM1 and E2F transcription factors and their corresponding targets (Figure 3F), such as the FOXM1-dependent NEK2 that regulates drug resistance in MM 40,41 , ( Figure 4A). Further, we identified PBX1 binding on active PBX1, E2F1/2, NEK2 promoters and PBX1, FOXM1, E2F2, NEK2 enhancers ( Figure 4B), while FOXM1 was found to bind to the same FOXM1 and NEK2 regions as PBX1 (Supplemental Figure Fig S5A). To better explore the regulatory interplay among those factors (Figure 4A), we characterized further the role of FOXM1 in chr1q-amp cells. Knockdown of FOXM1 using two validated shRNAs was toxic to MM.1S cells (Figure 4C), as previously shown 40 . In addition, depletion of FOXM1 mRNA was associated with downregulation of NEK2 but not of PBX1 ( Figure  4D), suggesting that FOXM1 acts downstream of PBX1 ( Figure 4A). Moreover, RNA-seq analysis revealed approximately 800 differentially expressed genes after FOXM1 knockdown in MM.1S cells (Figure 4E), with cell cycle-related pathways found to be significantly enriched amongst downregulated genes ( Figure 4F). Cell cycle arrest at G2/M was corroborated by flow-cytometry, thus confirming the pro-proliferative role of FOXM1 in chr1q-amplified MMCL (Supplemental Figure  S5B).
For further validation of the PBX1-FOXM1 axis, we forced expression of exogenous PBX1 into MM.1S and NCU.MM1 chr1q-amplified MM cells (Figure 4G). This led to modest but significant increase in FOXM1, NEK2 and E2F2 mRNA levels ( Figure 4H) and significantly reduced sensitivity of the MMCL to thiostrepton, an inhibitor of FOXM1 transcription 40,42 (Figure 4I and Supplemental Figure S5C). Rescue of PBX1 depletion by shRNA-resistant PBX1 cDNA resulted in a significantly lower MMCL toxicity, ameliorated cell cycle arrest and dampened downregulation of FOXM1, NEK2 and E2F2 (Supplemental Figure S5D-5G), thus validating the genetic and functional interactions in the PBX1-FOXM1 axis ( Figure 4A) and its role in orchestrating an oncogenic, proliferative process in chr1q-amp MM cells. Next, we sought to validate activity of the PBX1-FOXM1 axis in primary myeloma plasma cells ( Figure  5A). For this purpose, we combined RNA-seq with ATAC-seq profiling of highly purified chr1qamplified (n=6) and non-amplified (n=6) primary myeloma PC, and explored differences in chromatin accessibility, gene expression and predicted TF connectivity (Figure 5A and Supplemental Table S4). In addition to previously established gene-markers (CKS1B, IL6R, ARNT, PDKZ1, ADAR), we also found overexpression of all main PBX1-FOXM1 module components (PBX1, FOXM1, E2F1/2, NEK2) in chr1amp cells (Figure 5B). Moreover, there was significant enrichment of proliferative pathways and FOXM1-dependent targets in genes overexpressed in chr1q-amp cells (Figure 5C). Comparative ATAC-seq analysis revealed enhanced chromatin accessibility in the regulatory regions of genes over-expressed in the same cells (Figure 5D). Differential TF footprinting analysis revealed a higher number of TFs with increased connectivity (measured as differential regulatory potential, ΔP) in chr1q-amp versus non-amplified cells (Figure 5E). By combining transcriptional and regulation profiles, we identified 34 TFs with increased expression and connectivity in chr1q-amplified cells, including all four TFs involved in the PBX1-FOXM1 module (PBX1, FOXM1, E2F1, E2F2; Figure 4A and 5F). Notably, as compared to non-amplified cells (n=3), chr1q-amplified primary myeloma cells (n=3) were selectively sensitive to thiostrepton treatment, while expression of FOXM1 and NEK2, but not PBX1, decreased in response to treatment (Figure 5G and 5H).
In addition, we validated functional activation of the PBX1 and shared PBX1-FOXM1 transcriptional programmes in a large cohort of patients (MMRF, n=813) and confirmed significant co-expression of PBX1 and FOXM1 with almost all of their gene targets across patients in two different cohorts (MMRF, Arkansas; Supplemental Table S5 and Figure S6A). Importantly, the majority of genes previously shown to comprise high-risk disease signatures in MM 17,41,43,44 were found to be directly regulated by PBX1 (Supplemental Table S5 and Figure S6B). Together, these findings strongly support the critical role of PBX1-FOXM1 axis in promoting proliferative regulatory circuitries determining adverse prognosis and high-risk disease in chr1q-amp MM patients.

Targeted therapy against chr1q-amp in cancer using a novel, selective PBX1 inhibitor
As the PBX1-FOXM1 axis acts as a central regulatory hub for chr1q-amp MM cells, we next sought to explore the prognostic impact and therapeutic potential of selective PBX1 targeting in chr1-amp cells across several types of cancer. For this purpose, we first analysed transcriptomic data from multiple cancer patient cohorts and found that activation of the PBX1-dependent regulatory signature (n=320 genes) predicts adverse prognosis in multiple myeloma and 12 solid tumour patient cohorts, including breast, ovarian, lung and brain cancer, in which chr1q-amp is a frequent CAN 2,3 ( Figure 6A and Supplemental Figure S7A). Next, we tested the impact of our novel, recently reported smallmolecule compound T417, which specifically inhibits PBX1 binding to its cognate DNA motif 45 in chr1q-amp ovarian cancer cells. We screened four myeloma (MM.1S, U266, NCU.MM1, OPM2), two breast (MCF-7, LTED), two ovarian (OVCAR3, A2780), two lung (A549, H69AR) and one brain (SNB-75) cancer cell lines harboring at least one additional chr1q copy (Supplemental Figure S7B). Cell viability assays revealed sensitivity of all cell lines to T417 at low M concentrations (4-28μΜ), while no significant toxicity was detected upon treatment with the inactive analogue/pro-drug compound DHP52 in two myeloma and two ovarian cancer cell lines (Figure 6B and Supplemental Figure S7C). In addition, cell cycle analysis revealed significant depletion of the G2/M phase along with G0/1 phase arrest upon T417 treatment in all tested cell lines ( Figure 6C). RT-qPCR-assessed mRNA levels of the PBX1-regulated FOXM1, NEK2 and E2F2 genes showed their significant decrease upon treatment with T417 in almost all 11 cell lines ( Figure 6D). Interestingly, a significant decrease of PBX1 mRNA itself was also detected in 8 out of 11 cell lines. This, in conjunction with the binding of PBX1 to its own promoter and putative enhancer, are consistent with a potential mechanism of PBX1 transcriptional autoregulation ( Figure 4B) which would potentiate activity of T417 in chr1qamp cells. Next, using a subcutaneous xenograft myeloma model, we also validated the antimyeloma activity of T417 in vivo. We observed significantly reduced tumour size and weight in the T417-treated versus control mice, while in explanted myeloma cells we detected cell cycle arrest and mRNA depletion of the PBX1-regulated genes (Figure 6E-6G and Supplemental Figure S7D-S7H). In addition, selective cytotoxicity of T417 was detected against PBX1-expressing primary chr1qamplified myeloma cells (X1-X3; n=3), but not against non-amplified MM (X4, X5; n=2) or normal donor peripheral blood B cells (PBBC; n=1) with undetectable PBX1 mRNA levels ( Figure 6G and 6H and Supplemental Figure S7I and S7J). Overall, these findings highlight the efficacy of T417 against chr1q-amplified cancer cells and its clinical potential as an adjuvant approach against chr1q-amp high-risk myeloma and other solid tumours.

Discussion
Recurrent, high frequency CNA such as chr1q-amp are major oncogenic drivers shared across different types of cancer 1-3 . However, delineating the prognostic and functional role of hundreds to thousands of genes and downstream oncogenic pathways associated with specific CNA for development of targeted therapies remains an unmet challenge. In this study, we focused on chr1qamp, the most frequent CNA linked to high-risk MM [4][5][6]10 .
First, by combining WGS and 3D genome data we found that genetic amplification disrupts a large proportion of the chromatin structure throughout the chr1q arm. This level of disruption likely reflects contribution of multiple mechanisms to structural changes in chr1q 46 , including isochromosome formation 47 , hypoxia-driven tandem duplications 48 , jumping translocations 4 , chromothripsis 49 , chromoplexy 50 , and combination of the above 51 . Nevertheless, we detected four main blocks of co-amplification (hyper-domains) which are the product of distinct amplification patterns and retain their overall chromatin structure across MM patients. Of those, only two hyperdomains (B1, B4) contribute to adverse prognosis, and therefore have potential implications for the chr1q-amp biology.
In contrast to previous studies which traditionally focused on 1q21 band alone [13][14][15][16][17] , here we employed a large-scale, integrative analysis of clinical and multi-omics datasets to identify adverse prognosis driver genes across the whole chr1q arm. This analysis validated previously reported highrisk markers in 1q21 locus [13][14][15][16] , but also linked novel genes to adverse prognosis and highlighted the biological and prognostic significance of two other new areas, namely the 1q22 and 1q23.3 bands. Collectively, the adverse prognosis genes identified across chr1q are predicted to promote cell cycle and proliferation, suggesting their direct involvement in the well-characterized proliferative phenotype associated with chr1q-amp in MM 17,43 .
Identification of PBX1, located in 1q23.3, as a prominent candidate driver of high-risk disease in chr1-amp MM, also exemplifies the potential of our approach for biological discovery. Indeed, the role of PBX1 in promoting cancer cell survival, metastasis and drug resistance has been previously reported in other types of cancer [35][36][37]52 . In addition, previous in silico work predicted a role of PBX1 in the biology of t(4;14) myeloma 53 which, interestingly, is highly associated with chr1q-amp 6 . Here we found ectopic expression of PBX1 in chr1q-amp myeloma cells associated with genetic amplification and strong epigenetic activation of its entire TAD (including proximal and distal DNA elements), suggesting a selective process acting on a whole regulatory domain rather than the gene alone, as previously suggested in oncogenesis 54 . Moreover, our combined genetic, epigenetic and pharmacological approaches establish the mechanisms and regulatory networks through which PBX1 regulates the activity of FOXM1, a master TF promoting cell cycle progression 40,42 . The proliferative circuitries regulated by PBX1 and the PBX1-FOXM1 axis are of wider importance in cancer, as they exert a powerful prognostic impact in several cancers. Pertinently, chr1-amp is one of the most frequent CNA not only in MM but also other cancers, including breast and ovarian cancer [35][36][37] .
The finding that pharmacological abrogation of the PBX1-FOXM1 axis selectively impacts survival of chr1q-amp myeloma cells is one of the most notable findings of this work. As well as providing proof-of-principle for developing CNA-specific therapeutic approaches, our data strongly support the central role of PBX1 and FOXM1 in regulation of the transcriptional programme driving the proliferative phenotype and adverse prognosis in chr1q-amp MM. In addition, these findings support our recent efforts for development of T417, a small molecule inhibitor of PBX1 binding to its cognate DNA motif 45 , and suggest the potential benefit of its use in MM and other cancers with chr1-amp and PBX1 over-expression. Indeed, along with our previously reported pre-clinical activity of T417 against ovarian cancer 45 , the data herein demonstrate selective targeting against MM, breast, lung, liver and brain cancer cells with chr1-amp. These findings not only validate the presence of a common, PBX1-FOXM1 axis underlying chr1q-amp that is active in many types of cancer, but also provide the basis for clinical development of T417 as a chr1q-amp-targeting therapy.
In summary, we showed that our systems medicine dissection of CNA in cancer, which includes integration of genetic, epigenetic, transcriptional and 3D-chromatin profiles, is a powerful strategy for discovery of genes and cellular oncogenic pathways of biological significance and clinical impact. Through this process, we show that the ectopically expressed PBX1, in co-operation with FOXM1, is a critical driver of the proliferative phenotype in chr1q-amp MM and several other cancers, and we provide proof-of-principle for selective therapeutic targeting of chr1q-amp, the most prevalent CNA in MM and in cancer in general.
(D) Analysis overview of candidate driver genes (103) across chr1q bands. Distribution of WGS multivariate analysis scores (-log 10 P-value; top) and percentage (%) of candidate genes (relative to band gene density) per cytogenetic band. The highest candidate genes density was detected in 1q22 and 1q23.3 bands (highlighted here), with 1q23.3 also displaying the highest survival significance scores. (E) The PBX1 gene as a prominent candidate occupying alone a single TAD, displays strong epigenetic activation across PBX1 body and putative enhancers in chr1amp myeloma PC.    (E) Differential ATAC-seq footprinting analysis of expressed TFs in chr1q-amp(+) versus chr1q-amp(-) cells (ΔP: differential regulatory potential). TFs of interest are indicated here. (F) Scatter plot representation of differential expression (x-axis) and differential regulatory potential (y-axis) of 63 TFs displaying significant differences in both dimensions. Green quartile: TFs with increased expression and ΔP in chr1q-amp(+) cells; orange quartile: TFs with decreased expression and ΔP in chr1q-amp(+) cells. Key transcription factors are also highlighted here. (G) Selective sensitivity of chr1q-amp(+) (n=3, green) versus chr1q-amp(-) (n=3, orange) primary myeloma plasma cells to thiostrepton at 48h after treatment. IC 50 values were calculated for each patient sample using a non-linear fitting model (fitting line represented here). ****, P<0.0001. (H) Transcriptional profiling (RT-qPCR) of FOXM1 and NEK2 mRNA levels in chr1q-amp(+) (green) and chr1q-amp(-) (orange) primary samples 24h upon thiostretpon (1μM) or mock (0nM) treatment. The (%) decrease in FOXM1 and NEK2 mRNA levels is also indicated here.