Extramedullary disease (EMD) in multiple myeloma (MM) affects up to 20% of patients throughout the disease course [1]. It is associated with poor outcomes due to aggressive disease kinetics and therapy resistance, but the biology remains relatively unknown [2, 3]. EMD is variably defined, but two major subtypes exist: extraosseous EMD arising from bone and extending into surrounding tissue, and true EMD arising by haematogenous spread, which is associated with the worst outcomes [2, 3]. To describe the genetic landscape of haematogenous-spread EMD, we performed whole genome sequencing of extramedullary tumours from 15 patients; three patients had multiple biopsies. Patient samples were collected after written consent was obtained. Tumour DNA was obtained from two cohorts: fresh biopsies (n = 8; QIAGEN DNeasy Blood & Tissue Kit) and formalin-fixed, paraffin-embedded (FFPE) historical samples (n = 8; Roche High Pure FFPET DNA isolation Kit). Germline DNA was obtained from buccal swabs or stored autologous stem cells. Library preparation was optimised for DNA from FFPE samples using IDT® xGen PRISM DNA library preparation kit. Sequencing was performed on an Illumina NovaSeq® by the Australian Genome Research Facility. Briefly, bioinformatics utilised BWA-MEM, the GATK Mutect2, CNVpytor, Manta, Accucopy, SigProfilerMatrixGenerator, mmSig and FFPEsig. Liquid biopsy of circulating tumour DNA (ctDNA) utilised QX200 droplet digital PCR system (Bio-Rad®). See supplementary material for detailed methods.

Patient data is shown in Table 1. The median age at MM diagnosis was 52 years. Two cases of primary EMD were included. There was no unifying genetic event in EMD. The median number of small nucleotide variants (SNV) was 18,819 mutations per sample across the whole genome, with no difference between fresh and FFPE samples, although this is impacted by two highly mutated fresh samples (Fig. 1A). The median number of SNV was increased compared to published data for MGUS, newly diagnosed MM (NDMM) and relapsed/refractory MM (RRMM) (3406, 5483 and 10,938, respectively [4, 5]; Fig. 1B). Three samples were hypermutated, defined as total SNV above the upper limit of the interquartile range. Of 137 published driver genes [6,7,8,9] in MM, a median of 3 were mutated per EMD tumour (Fig. 1C). This is compared to 2 in NDMM [7], but clear data for comparison is lacking in RRMM. Twenty-three genes were mutated in 3 or more patients, with oncogenicity assessment using intOGen [10] noting driver genes as NRAS, KRAS, BRAF and TP53 (Fig. 1D). We identified MAPK-activating mutations in NRAS, KRAS and BRAF in 80% of patients, and importantly all were present at clonal cancer fractions of >0.9. The RAS mutations almost exclusively impacted codon 61. Patients with no MAPK-activating mutation had hypermutated disease and translocations involving MAF or MAFB.

A Number of small nucleotide variants (SNV) per sample, demonstrating no impact of FFPE. Three samples with high SNV are noted. B Median number of SNV in EMD is higher than in MGUS, newly diagnosed and relapsed refractory MM. C Oncoplot demonstrating known MM driver genes mutated in EMD samples, with associated CNV. D Genes mutated in three or more EMD samples, with the intOGen-predicted driver genes highlighted. E Manhattan plots demonstrating the average copy number change (top) and proportion of patients with the loss or gain (bottom) across the genome. F Manhattan plots of CNV of three subsequent EMD biopsies of patient 3 at diagnosis, first relapse and third relapse, demonstrating ongoing genomic instability and increasing CNV with treatment. G Mutational signatures between hypermutated (left) and MAPK-driven (right) EMD demonstrating SBS9 in MAPK-driven lesions, and predominance of SBS2 and SBS13 in hypermutated disease. H Overall survival from EMD diagnosis in MAPK and non-MAPK EMD patients.

Known high-risk copy number variants (CNV) in MM were frequent in EMD lesions (Fig. 1E), including del(1p), gain/amp(1q), del(13/13q) and del(17p), seen in 10 (67%), 11 (73%), 9 (60%) and 5 (33%) samples. Additional focal copy number changes were also defined, including gains of MYC (6 patients, 40%), BRAF (7, 47%), RASA2 (12, 80%) and IL6R (9, 60%), and losses of TP53 (4, 27%), RB1 (3, 20%), and IGLL5 (9, 60%). There was a median of 44 structural variants (SV) per sample with complex rearrangements consistent with chromothripsis seen in 2 patients. Translocations involving known MM super-enhancers were seen in 7 patients, involving IGH, IGL, TXNDC5 partnered with MYC (n = 3), FGFR3 (n = 2), CCND2, MAF and MAFB (all n = 1). Forty percent of the total 1 153 SV identified occurred within 500 kb of published sites of recurrent SV in MM [11, 12] Most frequently this impacted IGH, IGK and IGL [n = 12 (80%), 6 (40%) and 5 (33%), respectively]. Less predictable SV targets were the tumour suppressor gene SP140, impacted by disruptive SV in 6 patients, and H1-10 and FGFR3 in 5 (33%) and 4 (27%) patients, respectively. When the impact of SNV, CNV and SV were combined, the most frequently affected genes were SP140 in 8 (53%) and BRAF, MYC and TENT5C, all in 7 patients (47%). Biallelic loss of TP53 was seen in 3 patients (20%). Interrogation of repeat biopsies provided information on disease progression and therapy resistance. One patient with primary EMD and a contemporaneous BM biopsy demonstrated t(8;22)(MYC::IGL) and TP53 mutation in the EMD tumour that were not present in the BM. These EMD ‘founder lesions’ were present at two subsequent EMD biopsies from alternative anatomical sites. In total 288 genes were mutated in the EMD tumour but not the BM biopsy; notable genes included NFKBIA, NFKB2 and NFRKB, suggesting a role for NF-κB signalling in disease progression.

Repeat biopsies of EMD tumours were characterised by persistence of key EMD drivers [NRAS p.Q61H, t(8;22)(MYC::IGL) and NRAS p.Q61K], remaining at a clonal level, and separated across anatomical sites (e.g. pancreas and bladder) and time, up to 7 years (Fig. 1F). Additionally, the repeat biopsies demonstrated increases in SNV, CNV and SV at subsequent timepoints, frequently tumour suppressor genes (e.g. TENT5C) reflecting a mechanism of therapy resistance and ongoing genomic instability. The mutational signatures differed between the MAPK mutated and hypermutated samples (Fig. 1G), with SBS2 and SBS13 (APOBEC-associated signatures, associated with MAF translocations in MM [13]) the most dominant in the hypermutated samples, while all patients with MAPK-activating mutations had SBS9, which was absent in the hypermutated samples, consistent with 2 potential and differing mechanisms of EMD development. Given these differences in mutational signatures, we sought to identify biological differences. We have previously demonstrated that 100% of EMD patients with MAPK-activating mutations in EM tumours have the mutation detectable in circulating tumour DNA (ctDNA) at the time of EMD diagnosis [14]. We thus expanded our EMD cohort to n = 30 by including EMD patients who had ctDNA analysis available but had not undergone WGS. Patients were then assigned to the MAPK group if ctDNA mutations were detected in either NRAS or KRAS at codon 61 or activating mutations of BRAF (see supplemental methods). All other patients were included in group 2. Subsequent analysis demonstrated that patients with MAPK-driven EMD had better overall survival from MM diagnosis (median 5.75 years compared to 2.9 years, Wilcoxon test p = 0.024, Fig. 1H) when compared to the non-MAPK cohort. There was no statistically significant difference in median OS from EMD diagnosis of 20 vs 7 months (p = 0.113), although this may reflect the limited numbers. Long-term survivors (>4 years from EMD diagnosis) were only seen in the MAPK group, with several patients remaining in remission at the time of publication (Supplementary Fig. 1).

In this study, the first to solely and comprehensively analyse the genome of a cohort of EMD, we have defined the genetic landscape of EMD and demonstrate that there is no single unifying factor underlying the development of EMD in MM. Frequently implicated oncogenes were BRAF, NRAS, KRAS and MYC, with the frequent loss of tumour suppressor genes SP140 and TENT5C. Mutational signature analysis suggests 2 mechanisms of mutagenesis in EMD, with clinically relevant differences between these subtypes. We show EMD is genetically highly complex, with SNV, CNV and SV co-operating to impact several driver genes, and with 80% of patients demonstrating clonal MAPK-activating mutations. The persistence of these lesions at subsequent biopsies across anatomical sites, suggest a central role of the MAPK-pathway in the development of EMD, particularly of mutations involving codon 61. Sequential biopsies reveal a complex pattern of disease evolution with SNV, CNV and SV all contributing to therapy resistance and disease progression.

Our results provide rationale for the use of targeted therapy in EMD. The clonal-level MAPK-activating mutations found in EMD at diagnosis and subsequent timepoints suggest a central role in pathogenesis but also a possible therapeutic vulnerability, with pan-RAS inhibitors and novel combinations of multilevel MAPK inhibition attractive therapeutic options. Furthermore, the tumours with high SNV may be susceptible to immunotherapy, given this is a biomarker for response to checkpoint inhibition in solid organ malignancies [15].

The main limitation of this study is the limited sample size and the limitations of WGS. While WGS provides a complete overview of genetic features, it comes at the cost of depth: clonal mutations can be confidently identified, but structural variants, particularly translocations, and small subclonal mutations may be underestimated or missed. Our main aim was to identify key drivers of EMD, and adequate depth for this was achieved, but deeper sequencing to explore the subclonal architecture of these tumours is required. An additional limitation is the lack of protein or mRNA expression data to confirm the impact of genomic changes; this is the subject of further work. Finally, in patients with secondary EMD, it is difficult to reliably determine what may represent truly EMD-specific genetic abnormalities versus abnormalities simply driving drug resistance mechanisms. Ultimately, however, these lesions likely impact both functions, evidenced by the persistence of EMD clones over time despite multiple therapeutic exposures, suggesting a degree of innate drug resistance, and thus remain relevant to this study. Regardless of these limitations, this study represents the first WGS-based study of haematogenous EMD in MM, identifies likely drivers of EMD development and persistence and importantly provides a potential rationale for targeted therapies in this difficult to treat population.