Germline Lysine-Specific Demethylase 1 (LSD1/KDM1A) Mutations Confer Susceptibility to Multiple Myeloma

In multiple myeloma, abnormal plasma cells (PC) derived from post-germinal center B cells abnormally proliferate and produce high amounts of immunoglobulin or paraprotein, which leads to lytic bone lesions, anemia, and renal failure (1). Multiple myeloma is preceded by a premalignant condition, monoclonal gammopathy of undetermined significance (MGUS). Consistent with genetic predisposition, having a first-degree relative with multiple myeloma elevates multiple myeloma risk 2- to 5.5-fold (2).

Identifying multiple myeloma predisposition genes can provide mechanistic insights into MGUS, myeloma pathogenesis, and PC differentiation. Genome-wide association studies have identified 16 common variants at loci significantly associated with multiple myeloma risk (3–8). However, the role of rare multiple myeloma high-penetrance predisposing genes is poorly understood. To identify multiple myeloma predisposition genes, we performed germline whole-exome sequencing (WES) for familial multiple myeloma probands and identified kindreds that carry germline truncating mutations in lysine (K)-specific demethylase 1 (KDM1A, also called LSD1; ref. 9). KDM1A is an epigenetic transcriptional repressor that primarily demethylates monomethylated and dimethylated histone H3 on lysine 4 (H3K4me1/me2) to repress target gene promoters and enhancers (10–12).

We used CRISPR to introduce a “second hit” mutation in lymphoblastoid B cells from a germline KDM1A mutation carrier, which increased H3K4me1 levels. MGUS and multiple myeloma cells have significantly lower KDM1A transcript levels compared with normal PCs, and may be particularly sensitive to KDM1A mutations causing loss of function or haploinsufficiency. We also performed mutation burden test analysis of patients with multiple myeloma unselected for family history and controls, which showed higher rates of germline KDM1A mutations in patients with multiple myeloma. Mice treated with a KDM1A small-molecule inhibitor, GSK-LSD1, have enhanced secondary immune response with expansion of PCs, increased immunoglobulin production, and appearance of serum paraprotein. RNA sequencing (RNA-seq) analysis of these abnormal mouse PCs shows enrichment of MYC oncogene transcriptional targets. Transcriptomic analysis of multiple myeloma cells from KDM1A mutation carriers shows upregulation of the MYC target oncogene Cyclin D2 and enrichment of pathways associated with both intrinsic multiple myeloma pathogenesis and extrinsic multiple myeloma–bone marrow microenvironment interactions. Our findings show that KDM1A is a novel germline predisposition gene for multiple myeloma and provide new insights into its mechanistic roles as a tumor suppressor in B cells.

All patient studies were conducted in accordance with the U.S. Common Rule, after approval by an Institutional Review Board (IRB) at the respective recruiting institution. Informed written consent was obtained from all subjects. Familial multiple myeloma probands (n = 50; Supplementary Table S1) analyzed by exome sequencing met inclusion criteria: (i) confirmed diagnosis meeting revised criteria of the International Myeloma Working Group, (ii) IgG heavy/light chain analyzed, and (iii) ≥1 first-degree or ≥2 second-degree relatives diagnosed with multiple myeloma. KDM1A-Sanger sequencing European ancestry (EA) validation cohort (n = 400) inclusion criteria were: (i–iii; N = 200) or (i), (ii), and (iv) multiple myeloma onset younger than 60 years (n = 200).

Germline DNA extracted from peripheral blood was used for whole exome capture using Agilent SureSelect 38 Mb paired-end sequencing and ran on Illumina HiSeq 2000s/2500s. FASTQ files were aligned to human reference genome (GRCh37) to generate BAM files using BWA v0.7.12. Picard tools were used for quality metric calculation and marking duplicate reads. GATK version 3.5-0-g36282e4 was used for variant calling using the haplotype caller algorithm. Variant quality score recalibrated (VQSR) data were used for filtering variants. Variant-level and interval-level annotations used SNPEff, ANNOVAR, and CAVA programs. Downstream analysis consisted of filtering out low-quality variant calls and common variants. Average coverage depth was 80×–100×. Variants with read depth (DP) of 10 or greater and a genotype quality (GQ) score of 20 or greater were included in analyses. Variant-, exon-, and gene-level data were obtained using information from the 1000 Genomes Project, NHBLI GO Exome Sequencing Project Exome Variant Server (EVS), Exome Aggregation Consortium (ExAC), and the combined annotation dependent depletion (CADD) server (13). Deleterious variants were defined as loss-of-function (frameshift insertion or deletion, stop-gain, splice-site change) or missense variants with CADD score >15. We performed segregation analysis using either exomes from family members or targeted Sanger sequencing. Cosegregating qualifying variants in Family 1 (Fig. 1A) shared by exomes are listed in Supplementary Table S2. Exome sequencing data can be accessed at NCBI (NCBI SRR5641111).

For the KDM1A-Sanger sequencing EA validation cohort, all coding regions of KDM1A (NM_015013) were amplified using primers available upon request. PCR amplicons were sequenced in the Applied Biosystems 3730 DNA (Applied Biosystems) Analyzer and analyzed using Mutation Surveyor (SoftGenetics). KDM1A mutations in probands and indicated family members (Fig. 1A) were verified by Sanger sequencing.

Cell culture methods used are described in Supplementary Data S1.

To generate biallelic KDM1A truncating mutations in lymphoblastoid (LCL) cells that carry the KDM1A c.805_806delAG (p.Arg269Aspfs*7) heterozygous mutation, CRISPR/Cas9 gene editing sgRNA sequences were designed to target genomic sites near KDM1A c.805_806delAG (p.Arg269Aspfs*7). pX458_sgRNA (5′- CACCGTCAACTTC-GGCATCTATAAG-3′, 5′- AAACCTTATAGATGCCGAAGTTGAC-3′) was transfected into LCLs (Amax Nucleofector II device). GFP-expressing clones were seeded as single-cell clones in 96-well plates using flow cytometry and screened for mutations by PCR amplification, target sequencing, and Western blot analysis.

LCL cells were harvested in lysis buffer containing complete Protease inhibitors (Roche, 4693159001). Whole-cell extracts were subjected to SDS-PAGE. Antibodies used include anti-H3K4me1 (Abcam ab8895, 1:500 dilution), anti-H3K4me2 (Abcam ab32356, 1:500 dilution), anti-H3K9me1 (Abcam ab9045, 1:500 dilution), anti-H3K9me2 (Abcam ab1220, 1:500 dilution), anti-Histone H3 (Abcam ab1791, 1:1000 dilution), anti-γ-tubulin (Sigma T5192), and anti-KDM1A (Santa Cruz Biotechnology sc-398794, 1:200 dilution).

A Bio-Rad assay specific for detection of the KDM1A frameshift mutation was designed (forward primer: 5′ CGTCATGGTCTTATCAACT3′; reverse primer: 5′ AGGAGG-TCCTTACTTGGT3′; Wt probe: 5′CGGCATCTATAagAGGATAAAACC3′-; mutation probe: 5′ CGGCATCTATAAGGATAAAACC3′. Cycling conditions were optimized for annealing/extension temperature and separation of positive from empty droplets on a QX200 ddPCR system (Bio-Rad). Technical duplicates were run. PCR reactions contain primers/probes, DNA, and digital PCR Supermix for probes (no dUTP). Reactions were partitioned into median approximately 16,000 droplets/well using QX200 droplet generator. Emulsified PCRs were run using cycling conditions (95°C 10 minutes; 40 cycles of 94°C 30 minutes 55°C 1′, 98°C 10 minutes, 4°C hold). Plates were read and analyzed with QuantaSoft software to assess the number of droplets positive for mutant DNA, wild-type (wt) DNA, both, or neither. Assay threshold sensitivity was set at 2 mutant droplets.

Exome data from 879 patients with sporadic multiple myeloma (dbGaP phs000748 and phs000348) and 2389 control subjects (phs000179, phs000276, phs000403, phs000687, and phs000806) were simultaneously processed and variants jointly called. Exomes from cases/controls were required to have ≥75% of the 33.27 Mbps of CCDS with >10-fold coverage and ≤3% contamination using VerifyBamID (14). Exomes from cryptically related individuals were removed using KING algorithm (15). Principal component analysis (PCA) was performed using SMARTPCA (16) to identify EA. EA exomes meeting these quality control metrics were selected (733 patients with sporadic multiple myeloma and 1,480 controls). We compared the sum of (i) predicted deleterious mutation or (ii) neutral synonymous coding variant qualifying carriers for each gene in multiple myeloma cases and controls using Fisher exact test (statistical function scipy.stats.fisher_exact in SciPy v0.14.0.). Qualifying variant criteria include (i) a quality score (QUAL) ≥50, (ii) a genotype quality (GQ) score ≥20, (iii) a quality by depth (QD) score ≥2, (iv) a mapping quality (MQ) score ≥40, (v) a read position rank sum (RPRS) score ≥-3, (vi) mapping quality rank sum (MQRS) score ≥-10, and (vii) coding variants with mean allele frequency (MAF) ≤ 0.05% in all three databases of the 1000 Genomes Project, ESP, and ExAC. For rare synonymous variant burden testing, synonymous coding variants meeting criteria (i)–(vii) were used. For mutation burden testing, loss-of-function (frameshift insertion or deletion, stop-gain, splice-site change) and missense variants with Phred CADD score ≥15 meeting criteria (i)–(vii) were used. For association testing, common variants were required to meet criteria (i)–(vi), have MAF ≥5%, and be in Hardy–Weinberg equilibrium in cases, controls, and the combined group (P > 0.001 for each). PCA, rare synonymous, predicted mutation gene burden, and common variant association lambda statistics were used to confirm case–control matching and lack of notable population stratification (Fig. 2A–D; Supplementary Fig. S1).

Peg-Mal (polyethylene glycol-malate; AA peptides) hydrogels with REDV (AA peptides) peptidic ligands were impregnated with 30,000 U266 myeloma cells and 5,000 HS5 nonproliferating (mitomycin-treated) stromal cells, or 30,000 mononucleated cells from a primary bone marrow biopsy sample as previously described (17) with the addition of VLA-4/Integrin α4β1 ligand fibronectin (18). After gelation, hydrogel 3D organoids were rehydrated with media and cultured as described previously (17, 19). A detailed protocol for multiple myeloma organoid culture is available upon request. U266 organoids untreated or treated with 1 μmol/L LSD1 inhibitor (GSK-LSD1) were followed by microscopy. Cells were harvested and counted by measuring ATP content/well (CellTiter Glo, Promega) at days 0, 1, 3, 5, and 7. Three to five replications were used per organoid datapoint. For the primary bone marrow biopsy sample, after 6 days in culture, 2 organoids per experimental condition were mechanically digested and single cells assessed for PC markers CD138 and p63 using flow cytometry, and growth assessed at the same time point for 3 organoid datapoints per treatment condition using CellTiter Glo. To assess proliferation in a bone marrow sample of a patient with multiple myeloma in remission, mononuclear cells were preloaded with CellTrace Violet (Thermo Fisher Scientific).

For mouse PCs, polyadenylated single-read RNA-seq was performed using standard Illumina Truseq protocols and run at 3/lane on HiSeq 2500. Reads were trimmed of adapters with Flexbar then aligned to mm10 mouse genome (20). Uniquely mapped reads were used for read counts. Raw read counts were processed in R with DESeq2 (21) and accessed at NCBI (NCBI GEO GSE85596). Gene set enrichment analysis (GSEA) scores were calculated as the negative log₁₀ of the false discovery rate (FDR) multiplied by the sign of the log₂ fold change (log FC). Scores were run against preranked GSEA Hallmarks (weighted score type, 1,000 permutations; ref. 22) or Boylan multiple myeloma gene sets (23). Differentially expressed genes (DEG) were graphed as a volcano plot in R.

For analysis of MMRF CoMMpass RNA-seq multiple myeloma transcriptomes from KDM1A mutation carriers and patients with wt multiple myeloma, raw counts (19) were analyzed in R with DESeq2 (20). Multiple myeloma RNA-seq data were available from 4 KDM1A mutation carriers (MMRF_1201, 1730, 2293, and 2068). RNA-seq data from these 4 tumors were compared against 4 randomly picked multiple myeloma tumors from the 799 nonmutation carriers and run for 1,000 iterations. Significantly up- or downregulated genes (P_adj < 0.05) were compared against mouse GSK-LSD1–treated PCs. Outputs with >5% frequency were input for pathway enrichment in ClueGO (23), using GO Biological Process, Kyoto Encyclopedia of Genes and Genomes, REACTOME, and Wikipathways ontologies. Two-sided hypergeometric test was used to calculate P values, followed by Bonferroni correction. Pathways with P_adj < 0.05 were considered significant. The complete lists of genes and pathways are given in Supplementary Table S3.

For analysis of hypodiploid acute lymphoblastic leukemia (ALL) gene expression data (24), we compared samples with truncating KDM1A mutations (SJHYPO021 and SJHYPO032) to 50 KDM1A wt hypodiploid ALL tumors and used GSEA (n = 1,000 permutations) to test for enrichment of HoxA9 and LSK gene sets (25).

C57BL/6 male mice aged 11 weeks (n = 18) were immunized intraperitoneally with 100 μg NP-CGG in alum on day 0 and day 21. On day 1, intraperitoneal treatment with either vehicle (PBS) or GSK-LSD1 (Cayman Chemical #16439; 0.5 mg/kg//day) started and continued until day 45. On days 10 and 35, mice were bled. On day 45, mice were sacrificed, and heart blood, femurs, and spleen were collected and analyzed.

Mouse spleens were processed into single-cell suspensions, pooled in groups of 3–4, and washed before resuspending Ficoll-isolated mononuclear cells in PBS containing 0.5% BSA and antibodies B220-PECy7 (eBioscience, #25-0452-82) and CD138-APC (BD Pharmigen #558626). After one wash, 20 million cells/mL were resuspended and DAPI (1 μg/mL) added to exclude dead cells. Cells collected were B220⁺ CD138⁻ (B cells) and CD138⁺ (PCs), and were >95% pure. Bone marrow samples extracted from mouse femurs and spleen mononuclear cells were also analyzed by FACS with B220, CD138, CD3, CD11b, and Gr1 antibodies.

Day 35 retro-orbital blood sera were serially diluted, incubated overnight in 96-well plates precoated with NP (4)-CGG or NP (27)-CGG (Biosearch, #N-5055A and N-5055C), and preblocked with PBS with 1% BSA. After three washes with 0.05% Tween-20 PBS, horseradish peroxidase–conjugated antibodies against IgG1, IgG2a, IgG2b, IgG3, IgM, or IgA (SouthernBiotech, #5300-05) were used to detect NP-specific antibodies. ABTS (SouthernBiotech, #0404-01) was used as peroxidase substrate and the reaction stopped with 1% SDS. Absorbance of the product was measured at 410 nm.

Serum electrophoresis (Helena Laboratories) in agarose was performed with day 45 heart blood sera. Densitometry-based quantification of serum protein bands was expressed as percentage of total protein in samples quantified in a Roche/Hitachi Modular P Chemistry Analyzer.

Patients were consented under IRB-approved protocols in accord with CIOMS at Weill Cornell Medicine (New York, NY), Memorial Sloan-Kettering (New York, NY), Creighton University (Omaha, NE), University of Utah (Salt Lake City, UT), Mayo Clinic (Rochester, MN), Dana-Farber Cancer Institute (Boston, MA), and University of Lyon (Lyon, France). All mouse experiments were performed under Institutional Animal Care and Use Committee–approved protocols at Weill Cornell Medicine (New York, NY).

Statistical tests used are described in Supplementary Data S1.

Internet resources cited are listed in Supplementary Data S1.

Germline exome sequencing of 50 familial multiple myeloma kindreds (Supplementary Table S1) revealed EA family carrying a N-terminal truncating KDM1A mutation (c.805_806delAG [p.Arg269Aspfs*7]; Fig. 1A) absent in ClinVar or Exome Aggregation Consortium (ExAc) databases. This was the only kindred with a qualifying truncating mutation (Supplementary Table S2). Therefore, we focused follow-up on this family. Sanger sequencing identified five relatives carrying this KDM1A mutation. Three were multiple myeloma affected (II:4, III:6, and III:7), one (III:3) with both acute myelogenous leukemia (AML) and MGUS, and one unaffected at age 48 (III:1; Fig. 1A). Three additional family members were diagnosed with pancreatic (II:2, III:2) and colon cancers (II:1).

KDM1A is composed of three major domains: SWIRM and Tower, which mediate protein–protein interactions, and a C-terminal amine oxidase (AO) domain (26). The truncating KDM1A c.805_806delAG (p.Arg269Aspfs*7) mutation eliminates the AO and Tower domains. Western blot analysis of LCL derived from patient III:6 showed normal levels of full-length KDM1A and no truncated protein. RT-PCR and Sanger sequencing of cDNA revealed the absence of KDM1A-mutant transcripts.

We next sequenced KDM1A exons in 400 additional EA multiple myeloma early-onset (<age 60) or familial myeloma probands. This identified another EA patient with early-onset multiple myeloma (age 59) and a truncating KDM1A mutation (KDM1A c.707delA [p.Gln236Hisfs*3]) in the SWIRM domain (Fig. 1B). No other family members were diagnosed with multiple myeloma/MGUS. One sibling was affected with colorectal cancer, a half-sibling with renal cell cancer, and one uncle with prostate cancer.

CADD scoring is a widely used computational tool to predict whether germline variants are benign or deleterious (13). The higher a CADD score, the higher the likelihood that a genetic variant is deleterious. Using CADD ≥15 threshold we identified two predicted rare deleterious KDM1A missense mutations in the Tower and AO domains, (KDM1A c.1424T>C [p.Leu475Pro] and c.2003G>C [p.Arg692Pro]) in patients with multiple myeloma with the first-degree relatives also affected with multiple myeloma/MGUS. Collectively, these data confirm the presence of rare truncating and predicted deleterious missense KDM1A mutations in familial and patients with early-onset multiple myeloma.

To understand the prevalence of KDM1A truncating mutations in controls, we examined the ExAC database. Including all ethnicities only 9 predicted truncating and splice-site loss-of-function (LoF) variants were identified, carried by 10 individuals (10/60,694; LoF constraint metric pLI = 0.99). Seven of these are of EA. All KDM1A truncating mutations had mean allele frequencies <1.6 × 10⁻⁵ and no homozygotes were identified. Joint analysis of LoF mutations in familial and early-onset multiple myeloma cases (2/450) and ExAC EA participants as controls (7/36,664), showed enrichment of KDM1A LoF mutations among patients with familial and early-onset myeloma [P = 0.005, Fisher exact test; RR, 23.28; 95% confidence interval (CI), 4.85–111.75]. Overall, despite possible small mismatch in mean age and the precise definitions of our multiple myeloma probands and ExAC database assignments of EA, our data provide evidence of KDM1A LoF mutation enrichment in familial and early-onset multiple myeloma.

Myeloma cells from bone marrow biopsies were available for two members (II:4 and III:6) of the kindred carrying the KDM1A c.805_806delAG (p.Arg269Aspfs*7) mutation. Digital droplet PCR (ddPCR) was performed on DNA isolated from multiple myeloma cells and normal bone marrow. The KDM1A mutation was enriched in multiple myeloma cells from deletion of the wt allele compared with germline bone marrow cells carrying the heterozygous germline mutation in both family members tested [both P < 0.01, Mann–Whitney test (Fig. 1C)]. Next, we generated LCLs from members of this kindred carrying KDM1A heterozygous or wt genotypes. We used CRISPR/Cas9 gene editing technology to introduce a second hit on the wt allele and an LCL cell line was derived carrying biallelic KDM1A truncating mutations [(c.805_806delAG (p.Arg269Aspfs*7) and c.801_802insT (p.Lys268*)]. Western blot analysis of KDM1A biallelic (−/−), heterozygous (+/−), and wt (+/+) LCLs revealed that heterozygous and wt LCLs had similar KDM1A protein levels, but that biallelic mutant LCLs had no detectable KDM1A protein (Fig. 1D). Accordingly, KDM1A^−/− LCLs have significantly higher overall cellular levels of H3K4me1, whereas total H3K4me2, H3K9me1, and H3K9me2 levels were not significantly changed (Fig. 1D and E).

We performed gene burden tests using jointly called germline exomes from EA patients with multiple myeloma unselected for family history and EA controls (Fig. 2A). Exomes were matched for quality metrics as before (27, 28), and KING algorithm used to exclude individuals with cryptic relatedness (15). PCA showed no apparent population stratification between EA multiple myeloma cases/controls (Supplementary Fig. S1). The distribution of neutral synonymous coding variants showed no significant difference in the number of genes carrying neutral synonymous coding variants in multiple myeloma cases (N = 733; yellow, average 17.55 ± 0.26/gene) and controls (N = 1480; blue, average 17.52 ± 0.18/gene; Mann–Whitney U test, P = 0.91; Supplementary Fig. S1). We used Fisher exact test (28, 29) to generate quantile–quantile plots for neutral synonymous coding variant and predicted deleterious coding mutation gene burden tests. We also used Fisher exact test to generate a quantile–quantile plot comparing multiple myeloma case and control common variants (MAF ≥ 5%). The genomic inflation factor (λ) statistic confirmed appropriate matching between multiple myeloma cases and controls (λ = 1.005, 1.028, and 1.06, respectively; Fig. 2B–D). This showed significant enrichment of KDM1A predicted deleterious missense and LoF/truncating mutations (P = 1.27 × 10⁻³) in patients with multiple myeloma compared with controls (Table 1).

Among patients with multiple myeloma, we detected another patient (56 years old) with early-onset multiple myeloma with an exon 2 splice-donor LoF/truncating mutation KDM1A c.517+1G>A, (p.Glu119Trpfs*15), but no LoF/truncating mutation in controls. The myeloma tumor from this patient carried a somatic deletion encompassing the entire KDM1A locus (chr1:830,800-120,666,600). In addition, screening for predicted deleterious KDM1A missense mutations with CADD scores ≥15 identified 8 patients with multiple myeloma with heterozygous-predicted deleterious missense mutations and 2 non-multiple myeloma control subjects (1.23% LoF/missense mutations in patients with multiple myeloma and 0.14% LoF/missense mutations in controls; RR, 9.08; 95% CI, 1.97–41.95; Table 1). The rate of predicted KDM1A LoF and missense mutations in EA controls is similar to the ExAc database (0.22%; 81/36,664). However, none of the 7 tumors from patients with multiple myeloma carrying KDM1A missense mutations whose tumor somatic mutation data were available demonstrated loss of heterozygosity (LOH; ref. 30).

We analyzed published transcriptome databases containing gene expression from normal PCs, MGUS, and multiple myeloma cells, and found that KDM1A mRNA levels are significantly lower in MGUS or multiple myeloma cells versus PC (7.45 × 10⁻⁶ and 1.7 × 10⁻⁰⁵, respectively, moderated t test; Fig. 2E; ref. 31). Overall, these data support a broader role for KDM1A inactivation in sporadic MGUS and multiple myeloma, in addition to its role in familial and early-onset multiple myeloma. Because they have significantly lower KDM1A levels compared with normal PCs, MGUS and multiple myeloma cells may be particularly vulnerable to epigenetic dysregulation from KDM1A haploinsufficiency or loss-of-function mutations.

To test a role of KDM1A in post-germinal center B-cell differentiation, we conducted antigen-driven B-cell differentiation in vivo experiments. To induce differentiation of B cells into PCs, we challenged immunologically mature C57BL/6 mice with T-cell–dependent antigens on day 1 and boosted at day 21. In one group of mice, we used sheep red blood cells (SRBC) as antigen and in another group NP-chicken gamma globulin (NP-CGG), which allows quantification of total and high-affinity immunoglobulins. KDM1A was inhibited with GSK-LSD1, an irreversible KDM1A inhibitor (Fig. 3A) with IC₅₀ of 16 nmol/L that shows >1,000× selectivity for KDM1A over closely related FAD utilizing enzymes including LSD2, MAOA, and B. Mice were randomized to receive vehicle (PBS) or GSK-LSD1 0.5 mg/kg daily by intraperitoneal injection (Fig. 3B). Mice were monitored for primary (day 10) and secondary (day 35) antibody immune responses, and sacrificed (day 45).

Phenocopying the Kdm1a knockout mouse models (25, 32), GSK-LSD1 significantly decreased neutrophils and increased monocytes in bone marrow (Fig. 3C). Antigen-challenged mice treated with GSK-LSD1 showed increased splenic PCs (Fig. 3C). Analysis of serum total and high-affinity anti-NP antibody production revealed that KDM1A inhibition promoted secondary immune response to the T-cell–dependent antigen NP-CGG (Fig. 3D). KDM1A inhibition also promoted IgM to IgG2b isotype class-switching, a process where B cells activated upon antigen encounter are costimulated by Th cells (Fig. 3D). The increase in immunoglobulin production could be detected by serum electrophoresis as an increased gamma globulin fraction (Fig. 3E). These findings are consonant with a model whereby KDM1A activity is required to control abnormal generation and/or expansion of PC.

To characterize the transcriptional consequences of KDM1A inhibition in PCs, we performed RNA-sequencing of DEG in sorted PC (CD138⁺) from GSK-LSD1 treated and control mice. To functionally categorize these genes, we conducted pathway analysis using GSEA (33) and identified “MYC targets (v1)” as a significant pathway associated with upregulated genes (Fig. 4A, P = 0.0, FDR = 0.0).

To evaluate mouse genes dysregulated by Kdm1a inhibition in PCs with development of multiple myeloma, we compared DEG in PCs from GSK-LSD1 treated versus control mice with Bcl-xL/Myc transgenic mouse multiple myeloma cells (23). GSEA showed enrichment of genes upregulated in Bcl-xL/Myc transgenic mouse multiple myeloma cells in GSK-LSD1–treated PC DEG (normalized enrichment score 1.64, P = 0.0, FDR 0.008; Fig. 4A), including Myc-driven multiple myeloma oncogenes Ccnd1 and Ccnd2 (Supplementary Table S3; refs. 34–36). Thus, Kdm1a pharmacologic inhibition enriched for upregulated MYC transcriptional targets including Ccnd1 and Ccnd2 as potential drivers of PC proliferation.

To determine the impact of germline KDM1A mutations on patient multiple myeloma tumors, we compared multiple myeloma tumor transcriptomes from patients with KDM1A-predicted deleterious mutations to those from KDM1A wt patients. First, we evaluated shared DEG common to both (i) GSK-LSD1–treated PCs and (ii) multiple myeloma cells from KDM1A mutation carriers versus wt patients. Upregulated shared DEG included CCND2 (Fig. 4B). Among downregulated shared DEG were the transcription factor KLF4 that has tumor suppressor functions in multiple myeloma (37) and PRDM11, a tumor suppressor in MYC-driven B lymphomagenesis (Fig. 4B; ref. 38). Thus, CCND2, KLF4, and PRDM11 are potential transcriptional targets downstream of KDM1A whose dysregulation promotes MGUS and multiple myeloma in patients.

We then analyzed DEG in multiple myeloma tumors from KDM1A mutation carriers compared with wt patients by ClueGO pathway analysis (Fig. 4C; ref. 39). Pathways related to myelomagenesis enriched in upregulated genes include Immunoglobulin Production, Antigen Processing, and Presentation (40), IL10 signaling (41), ERK Signaling Cascade (42), Vasculature Development (43), and Extracellular Matrix Organization (44), while in downregulated genes, the TGFβ signaling pathway was enriched (Fig. 4C). This included SMAD1, which suppresses MYC-dependent transcription in multiple myeloma cells (45). Altogether, these data reveal enrichment in pathways associated with both intrinsic multiple myeloma pathogenesis and multiple myeloma–bone marrow microenvironment interactions in KDM1A mutation carriers.

To directly test whether KDM1A affects myeloma cell proliferation, we treated human myeloma cells with GSK-LSD1 in 3D organoids with extracellular matrix–like biophysical properties that resemble the microenvironment of lymphoid tissues (17, 19). When U266 multiple myeloma cell line (Fig. 5A) or primary sporadic multiple myeloma patient cell (Fig. 5B and C) organoids were treated with GSK-LSD1, KDM1A inhibition significantly stimulated multiple myeloma cell number (P < 0.001 and <0.01, t test). In contrast, GSK-LSD1 treatment did not affect proliferation of nonmultiple myeloma CD138⁻ bone marrow cells as controls (Fig. 5D and E). These data support roles for KDM1A regulating PC differentiation to prevent MGUS and multiple myeloma development, and also to promote multiple myeloma proliferation.

MGUS and multiple myeloma risk is increased in relatives of affected individuals (2). We used whole-exome sequencing of familial myeloma kindreds, mutation cosegregation in family members, somatic tumor studies, CRISPR-induced somatic second hit mutations, targeted sequencing of additional patients with familial and early-onset multiple myeloma, mutation burden analysis in patients with multiple myeloma unselected for family history, multiple myeloma tumor transcriptome studies and functional studies of a KDM1A inhibitor on human multiple myeloma cells and mouse PC that promoted PC expansion, and increased immunoglobulin production and multiple myeloma cell proliferation to nominate KDM1A as the first autosomal-dominant multiple myeloma predisposition gene. Our data are consistent with germline KDM1A mutations predisposing to familial multiple myeloma, having a prevalence of approximately 1.23% in patients with multiple myeloma unselected for family history and conferring approximately 9-fold increased risk of developing multiple myeloma.

Whole body knockout or shRNA knockdown of mouse Kdm1a is embryonic lethal (25, 32). In hematopoietic cells, Kdm1a knockout/knockdown causes immature HSPC blasts, LSK progenitors, and monocytes to proliferate while LT-HSC, granulocytes, megakaryocytes, and erythrocytes are reduced (25, 32). The early lethality of Kdm1a-mutant mouse models has limited studies of B lymphocytes, which continue to be produced despite absent Kdm1a (25, 32). Therefore, the roles of KDM1A in B lymphocytes are poorly understood.

Previously, chronic exposure to lysolipid and dioxin antigens was proposed to drive the transition from PC to MGUS (46, 47). In mice, Kdm1A pharmacologic inhibition upregulates PC MYC proto-oncogene targets including Ccnd1 and Ccnd2, increases PC numbers, high-affinity secondary immune response to T-cell–dependent antigens, immunoglobulin production and causes the appearance of serum paraprotein (Fig. 3C). On the basis of similarities between these PC and Bcl-xL/Myc–driven mouse multiple myeloma cells, this transcriptional dysregulation likely predisposes to additional myelomagenesis-driving events. Furthermore, analysis of tumor transcriptomes from KDM1A-mutant patients is consonant with selected alterations persisting in multiple myeloma cells, such as increased CCND2 and reduced KLF4 and PRDM11 (37, 38). In line with multiple myeloma organoid findings, tumor transcriptomes from KDM1A-mutant patients are enriched in pathways associated with both intrinsic multiple myeloma pathogenesis and extrinsic multiple myeloma–bone marrow microenvironment interactions. These data are consistent with KDM1A inhibition/mutation acting at distinct steps to promote both PC immune response and multiple myeloma proliferation by dysregulating both cell-intrinsic (including MYC driven) and -extrinsic microenvironment signaling pathways.

We identified both KDM1A germline truncating and missense mutation carriers (Fig. 2). Somatic multiple myeloma tumor analyses performed with two members of our index family carrying a KDM1A truncating mutation (Fig. 1) showed evidence of possible LOH. However, this finding may be by chance, as chromosome 1p deletion is frequent in myeloma (1, 42). Of seven additional KDM1A predicted germline mutation carriers with somatic multiple myeloma mutation data, only one patient with multiple myeloma tumor showed somatic LOH. The other 6 patients with multiple myeloma without tumor LOH carried germline missense mutations. KDM1A is a component of the NuRD and CoRest chromatin remodeling complexes, and participates in multiple protein–protein interactions with distinct histone demethylase activities (10–12). It is possible that KDM1A missense mutations act through a dominant negative mechanism in PC NuRD or CoRest complexes to promote MGUS/multiple myeloma predisposition. Alternatively, because MGUS and multiple myeloma cells both have significantly lower KDM1A levels compared with normal PCs (Fig. 2E), they may be particularly vulnerable to epigenetic dysregulation from KDM1A hypomorphic missense mutations and haploinsufficiency.

Scales and colleagues recently reported a multiple myeloma exome mutation burden test study in 513 United Kingdom northern EA multiple myeloma cases from two clinical trials (48). While no individual gene reached genome-wide significance in this study, the authors nominated KIF18A as a promising multiple myeloma predisposition candidate. However, in our EA multiple myeloma mutation burden study, we did not see an association between KIF18A and multiple myeloma (Fisher exact test, P = 0.787). Furthermore, the study of Scales and colleagues did not nominate KDM1A as multiple myeloma predisposition gene. A likely reason for these different findings is that Scales and colleagues used variant MAF filter <1% while our study used MAF ≤0.05% to define rare candidate mutations. Thus, each study used different inputs for mutation burden testing. Additional potential reasons include methodologic differences in variant annotation and mutation burden testing pipelines, possible differences in EA population substructure (American vs. UK patients with multiple myeloma), and different rates of clinical covariates for patients with multiple myeloma and their tumor molecular subtypes.

Bolli and colleagues performed exome sequencing of a large, multiplex MM/MGUS kindred (49). The authors nominated EP300 and PDPK1, but not KDM1A, as candidate MGUS/multiple myeloma predisposition genes. However, in contrast to the KDM1A mutation carrying kindreds we identified (Fig. 1A and B), this family did not have non-multiple myeloma/MGUS cancers. Thus, additional MGUS/multiple myeloma specific high-risk predisposition genes may exist.

Recently, recurrent somatic KDM1A mutations were identified in chronic myeloid leukemia (50). We also identified one patient with a germline truncating KDM1A mutation diagnosed with MGUS and AML (Fig. 1A). In mice, Kdm1a knockout causes monocytosis (25, 32). However, KDM1A inhibition was found to induce terminal differentiation of AML cell lines and several primary AML cultures (51). On the basis of this, KDM1A small-molecule inhibitors are used in clinical trials for AML and other conditions. Our study of germline KDM1A mutations supports individuals being treated with KDM1A inhibitors being monitored for MGUS/multiple myeloma.

We also surveyed germline data from patients with pediatric hypodiploid ALL (24). This revealed a germline KDM1A truncating mutation c.1368_1371delAGAA (p.Lys456Asnfs*12) in a patient with near-haploid hypodiploid ALL whose tumor demonstrated LOH and another patient with near-haploid hypodiploid ALL whose tumor carried a somatic KDM1A truncating mutation c.1249C>T (p.Gln417*, MAF 83%). Mouse Kdm1a^−/− LSK hematopoietic stem-progenitor cells have upregulated HOXA9 and LSK leukemia stem cell signatures (25). Reanalysis of KDM1A-mutant and wt hypodiploid ALL tumor transcriptomes revealed that KDM1A mutations enriched for these same leukemia stem cell signatures (Supplementary Fig. S2; ref. 25). In addition, query of germline sequence data from The Cancer Genome Atlas revealed germline truncating and splice-site KDM1A mutations in 3 patients with glioblastoma multiforme (KDM1A c.2371C>T [p.Gln791*]), glioma, and ovarian serous cystadenocarcinoma (both KDM1A c.1476+1_1476+2delGT [p.Met448_Lys492del]). Thus, our data are consistent with KDM1A mutations contributing to predisposition of other tumors in addition to multiple myeloma.

K.E. Nichols reports receiving commercial research grants from Incyte and Alpine Biosciences. L.A. Godley has provided expert testimony for UpToDate, Inc. J.E. Garber reports receiving commercial research support from Novartis Pharmaceuticals, is a consultant/advisory board member for Novartis Pharmaceuticals, GTC Pharmaceuticals, and Helix Genetics, and has provided expert testimony for Oric Pharmacetuicals. C.G. Mullighan reports receiving commercial research grants from Loxo Oncology, Pfizer, and Abbvie, has received speakers bureau honoraria from Amgen, and is a consultant/advisory board member for Genome Quebec and Cancer Science Institute. No potential conflicts of interest were disclosed by the other authors.

Conception and design: X. Wei, M.N. Calvo-Vidal, V. Joseph, C.M. Vachon, S. Chen-Kiang, M. Salama, N.J. Camp, R.J. Klein, L. Cerchietti, S.M. Lipkin

Development of methodology: X. Wei, M.N. Calvo-Vidal, J. Zhang, M.F. Walsh, K. Curtin, R.J. Klein, H. Yu, L. Cerchietti

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): X. Wei, M.N. Calvo-Vidal, M.F. Walsh, K.E. Nichols, V. Joseph, C. Snyder, C.M. Vachon, J.D. McKay, D.S. Jayabalan, D. Becirovic, A. Viale, J.M. Phillip, D. Atanackovic, R. Niesvizky, O. Landgren, S.L. Slager, L.A. Godley, J. Churpek, J.E. Garber, C.Dumontet, H.T. Lynch, C.G. Mullighan, N.J. Camp, S.M. Lipkin

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): X. Wei, M.N. Calvo-Vidal, S. Chen, G. Wu, M.V. Revuelta, J. Sun, J. Zhang, M.F. Walsh, V. Joseph, J.D. McKay, M. Artomov, J.M. Phillip, K. Curtin, R. Niesvizky, O. Landgren, K.C. Anderson, M.J. Daly, C.G. Mullighan, N.J. Camp, R.J. Klein, H. Yu, S.M. Lipkin

Writing, review, and/or revision of the manuscript: X. Wei, M.N. Calvo-Vidal, S. Chen, M.V. Revuelta, K.E. Nichols, V. Joseph, C. Snyder, C.M. Vachon, D.S. Jayabalan, R.G. Waller, A. Viale, K. Curtin, M. Salama, R. Niesvizky, O. Landgren, S.L. Slager, L.A. Godley, J.E. Garber, K.C. Anderson, C. Dumontet, H.T. Lynch, C.G. Mullighan, N.J. Camp, K. Offit, R.J. Klein, L. Cerchietti, S.M. Lipkin

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): X. Wei, S.-P. Wang, D.S. Jayabalan, L.M. Jacobs, R.G. Waller, D. Atanackovic, O. Landgren, R.G. Roeder, N.J. Camp

Study supervision: X. Wei, N.J. Camp, H. Yu, L. Cerchietti, S.M. Lipkin

We acknowledge funding from R01 CA167824, R01 CA13464, R01 CA178765, T15 LM007124, R21 CA152336, LLS 6067-09, NCI P30 CA42014, HHSN261201000026C, the V Foundation (V2015-003), the Weill-Cornell Program in Mendelian Genetics, the Utah Genome Project, Huntsman Cancer Institute, Utah Population Database (UPDB), the Utah Cancer Registry (UCR), Icahn School of Medicine at Mount Sinai Office of Research Infrastructure of the NIH award number S10OD018522, and a generous donation from Matthew Bell.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.