A Distinct Epigenetic Program Underlies the 1;7 Translocation in Myelodysplastic Syndromes

Abstract

The unbalanced translocation dic(1;7)(q10;p10) in myelodysplastic syndromes (MDS) is originated by centromeric juxtaposition resulting into 1q trisomy and 7q monosomy. More than half of cases arise after chemo/radio-therapy. To date, given the absence of genes within the centromeric regions, no specific molecular events have been identified in this cytogenetic subgroup.

We performed the first comprehensive genetic and epigenetic analysis of MDS with dic(1;7)(q10;p10) compared to normal controls and therapy-related Myeloid Neoplasms (t-MNs).

RNA-seq showed a unique downregulated signature in dic(1;7) cases, affecting more than 80% of differentially expressed genes. As revealed by pathway and gene ontology analyses, downregulation of ATP-binding cassette (ABC) transporters and lipid related genes, and upregulation of p53 signaling were the most relevant biological features of dic(1;7).

Epigenetic supervised analysis revealed hypermethylation at intronic enhancers in the dicentric subgroup, in which low expression levels of enhancer putative target genes accounted for around 35% of the downregulated signature. Enrichment of Kruppel-like transcription factor binding sites emerged at enhancers. Furthermore, a specific hypermethylated pattern on 1q was found to underlie the hypo-expression of more than 50% of 1q-deregulated genes, despite trisomy.

In summary, dic(1;7) in MDS establish a specific transcriptional program driven by a unique epigenomic signature.

Introduction

The dic(1;7)(q10;p10) is a rare cytogenetic alteration found in 1.5–6% of MDS, and less frequently in Acute Myeloid Leukemia (AML) and Myeloproliferative Neoplasms (1). Of note, dic(1;7)-associated MDS/AML are frequent after chemo/radio-therapy exposure. Interestingly, in both de novo and treatment-induced disorders this abnormality usually occurs as isolated or associated with one additional cytogenetic change (2).

Dic(1;7) differs from other typical chromosome translocations that act as leukemia drivers by originating fusion genes and chimeric transcripts (3), as it derives from the translocation of whole arms of chromosome 1 and 7 with juxtaposition at centromeric regions, without production of abnormal fusion genes.

Human centromeres are made by chromosome specific no-gene DNA encompassing transposable elements and repetitive alpha-satellite (α-sat) sequences. Their function is under the control of epigenetic marks that promote binding of protein complexes directing segregation during cellular divisions (Fig. 1a). Centromere instability, originating aberrant recombination of chromosomes 1, 9, and 16 at centromeric level characterizes a rare autosomal recessive disease with a constitutional methylation defect, immunodeficiency and facial anomalies (OMIM #614069, OMIM #616910, OMIM #616911) (4). In multiple myeloma cytogenetic instability of 1q originating jumping translocations prevalently involves pericentromeric heterochromatin decondensation rather than α-sat (5, 6).

Fig. 1.

Cytogenetics, FISH and SNPa results summary. a Human centromeric regions. Top: representative chromosome structure with position of repetitive sequences. Upper middle: small different colored circles represent monomers of alpha satellite DNA at centromeres; large red arrows indicate high-order repeat (HOR) units that form centromeric chromatin. Lower middle: specific histone modifications promote binding of centromeric (CENP-A, CENPB, CENP-C) and pericentromeric (HP-1) proteins. Bottom: the constitutive 16 proteins, centromere-associated network (CCAN), interact with centromeric proteins and promote microtubule attachment to the centromere. b Schematic representation with relative orientation of genomic FISH probes at centromere 1 [D1Z5 Spectrum Orange] and centromere 7 [D7Z1 Spectrum Green]. c Metaphase dual color FISH with D1Z5 (orange) and D7Z1 (green) showing a fusion signal on dic(1;7), indicate by a yellow arrow [DIC 4 case]. d Copy number alterations (CNA) detected by SNP arrays in all five dicentric cases.CN: Copy Number; chr: chromosome; cytostart: cytogenetic band start; cytoend: cytogenetic band end.

Due to an absence of genes within the centromeric regions, specific molecular consequences have not been identified in dic(1;7) cases and detailed genomic information is still lacking. Using a comprehensive genetic and epigenetic approach we delineated the specific (epi)genomic features underlying the typical 1;7 centromere-centromere recombination in MDS.

Materials and Methods

Samples:

Cases were retrospectively collected from the files of the Laboratory of Cytogenetics and Molecular Genetics at the Hematology Department of the University of Perugia, Italy. All patients gave their written informed consent to sample collection and biological analyses in accordance with the Declaration of Helsinki. The study was approved by the Institutional Bioethics Committee (Prot.1.X.2011). Hematological and cytogenetic features of patients are listed in Table 1. dic(1;7)(q10;p10) cases were compared with a series of controls and t-MNs. As controls we used bone marrow samples from non-neoplastic cytopenias, normal bone marrow morphology and normal karyotype. Mutational diagnostic screening of 14 MDS-related genes was performed on both controls and t-MNs (Supplementary Table 1).

Table 1.

Hematological and cytogenetic features of all patients analyzed. Previously primary tumor indicated between brackets. M: male, F: Female, DIC: dic(1;7) CTL: control, t-MN: Therapy related Myeloid Neoplasm, t-MDS: Therapy related Myelodysplastic Syndrome, AML: Acute Myeloid Leukemia, ET: Essential Thrombocythemia, JAK2+: Val617Phe detected at diagnosis of ET, NA: Not Available due to quality control. mERRBS was performed on all samples except on CTL 5, t-MN 3 and t-MN 5; RNA-seq analysis was performed on all samples except on CTL1, due to quality control. *Mutations: mutational results indicate WES for DIC cohort and DHPLC/Sanger screening of MDS-related genes for CTL and t-MN cohorts (detailed gene list and results are reported in Supplementary Table 1).

dic(1;7) cohort
Patient	Sex	Age	Diagnosis	Karyotype	mERRBS	RNAseq	Mutations*
DIC 1	M	67	MDS (ET, JAK2+)	47,XY,+1,dic(1;7)(p10;q10),+9[15]	yes	yes	Not Available
DIC 2	M	75	t-MDS (Lung Cancer)	46,XY,+1,dic(1;7)(p11;q11)[14]/46,XY[1]	yes	yes	GGPS1 (R74H); ANKS1B (T247I)
DIC 3	F	79	t-MDS (NHL)	46,XX,+1,dic(1;7)(p11;q11),del(13)(q13q31)[2]/47,idem,+8[8]/46,XX[2]	yes	yes	Not Available
DIC 4	M	42	t-MDS (Melanoma)	46,XY,+1,dic(1;7)(q10;p10)[7]/46,XY[8]	yes	yes	Not Available
DIC 5	M	45	MDS	47,XY,+1,dic(1;7)(q10;p10),+8[20]	yes	yes	CCDC8 (R429K); PSMF1 (P168L)
Control cohort
CTL 1	F	43	Wild Type	46,XX[20]	yes	NA	Not Found
CTL 2	M	39	Wild Type	46,XY[20]	yes	yes	Not Found
CTL 3	F	46	Wild Type	46,XX[20]	yes	yes	Not Found
CTL 4	M	46	Wild Type	46,XY[20]	yes	yes	Not Found
CTL 5	F	63	Wild Type	46,XX[20]	NA	yes	Not Found
t-MN cohort
t-MN 1	F	77	t-MDS (Lymphoma)	46,XX[20]	yes	yes	TET2 (P1115L*2); SF3B1 (K700E)
t-MN 2	F	46	t-MDS (NHL)	51,XXX,del(4)(q13),−5,+6,+8,+11,−17,−18,+19,+22,+2mar[11]/46,XX[7]	yes	yes	TP53 (R273H)
t-MN 3	M	70	t-MDS (Bowel carcinoma)	46,X,-Y,t(1;10)(q21;q21),−5,der(7)(?),del(7q)(q11),+8,+19,−22+mar[15]	NA	yes	TP53 (C135L*4)
t-MN 4	F	80	t-MDS (Ovarian cancer)	47,XX,+8[7]/46,XX[8]	yes	yes	Not Found
t-MN 5	M	77	t-MDS (Bladder Cancer)	43–45,XY,−5,del(7q)(q21q36),−9,add(12p)(p11),−13,+16,add(17)(p11),+19,del(21)(q?),+22[7]	NA	yes	TP53 (M246R)

Karyotype, Fluorescent In Situ Hybridization (FISH), Denaturing High-Performance Liquid Chromatography (DHPLC) and Sanger:

Cytogenetic analysis was performed following standard procedure and G-banded karyotypes (7). FISH experiments on metaphase chromosomes were performed as previously described (8) using the following probes: Vysis CEP 1 (D1Z5, SpectrumOrange Probe, Abbott) for alphoid sequences of chromosome 1 and Vysis CEP 7 (D7Z1, SpectrumGreen Probe, Abbott) for chromosome 7 α-sat DNA (Fig. 1b). Analyses were carried out using a fluorescence microscope Olympus BX61 equipped with a highly sensitive camera JAI and driven by CytoVision 4.5.4 software. At least 7 abnormal metaphases were analyzed in each experiment.

Nucleic acids were extracted from unsorted bone marrow cells of patients using All Prep DNA/RNA Mini Kit (Qiagen), quantified with Qubit fluorimeter using Quant-i-T dsDNA HS Assay Kit and RNA HS Assay Kit (Invitrogen) respectively and samples quality was evaluated using Tapestation visualization (Agilent 2100 Bioanalyzer). Mutational diagnostic screening of 14 MDS-related genes was performed using DHPLC (Wavemaker software, Wave System, MD Transgenomic Inc., USA) and Sanger sequencing (3500 Genetic Analyzer, Applied Biosystems) (Table 1; Supplementary Table 1).

Single Nucleotide Polymorphism Array (SNPa):

SNPa was performed on CytoScan HD Affymetrix platform (Affymetrix, Santa Clara, CA, USA), following the manufacturer’s instructions. 250 ng of high-quality genomic DNA obtained from bone marrow cells was used to study Copy Number Alterations (CNAs) and copy neutral loss of heterozygosity (cnLOH). Protocol was supported by Affymetrix GeneChip Command Console (AGCC) software and analyzed through Affymetrix Chromosome Analysis 3.1 (ChAS) software with filters set at 100 Kb dimension for CNAs and 10 Mb dimensions for cnLOH. Our data refer to NetAffx Build 32.3 (hg19) database. Polymorphic copy-number variations were excluded from the analysis (9).

Whole Exome Sequencing (WES):

WES was performed in DIC 2 and DIC 5 (Table 1) using germline DNA extracted from bone marrow fibroblasts with normal karyotype. All variations identified by WES were confirmed as somatic on matched tumor/normal samples and tested on the remaining three DIC cases by Sanger sequencing. Illumina paired-end libraries were generated according to manufacturer’s protocol (Illumina San Diego, CA). Image processing and basecall were performed using the Illumina Real Time Analysis Software. Paired whole-exome fastq data were aligned to the human reference genome (GRCh38/hg38) with the BWA-MEM algorithm (10). Duplicates were marked using Samblaster. Quality of the aligned reads, somatic variant calling and copy number analysis were performed through CEQer2 tool, as previously described (11). Variants were annotated using dbSNP142. Variants with Minor Allele Frequency < 0.01 or carrying a ‘Clinical’ dbSNP flag were further processed; the other variants were discarded from subsequent analyses. Filtered variants were exported as vcf files and used as input for Annovar (12) analysis/annotation.

RNA-seq and Quantitative Real-Time PCR (qRT-PCR):

All samples used for RNA-seq had RNA integrity number of 6 or above. Ribosomal RNA depletion and library preparation were obtained following manufacturer’s instructions using RiboGone Mammalian Kit (Clontech Laboratories Inc.) and NEBNext® Ultra™ RNA Library Prep Kit for Illumina (New England Biolabs, NEB), respectively. Reads were aligned against hg19 build of the human genome using TopHat (13), counts were generated using HTseq (14) and DESeq2 was used for differential expression analysis (FDR< 0.1 and log2FC ≥I1I) (15). To validate RNA-seq results, 1 μg of RNA was retrotranscribed using 100U of SuperscriptII and esarandom primers (Invitrogen, Carlsbad, CA, USA) and qRT-PCR was performed on ZMAT3 and NR1H2, using TaqMan assay probes Hs00536976_m1 and Hs01027215_g1 (Applied Biosystems, LifeTechnologies, Carlsbad, USA) respectively with endogenous ABL1 (Hs00245445 m1, Applied Biosystems, Life Technologies, Carlsbad, USA) and Universal Human Reference RNA (Stratagene, Cedar Creek, TX, USA) as reference controls. Real-time reactions were performed in triplicate using the Roche Light Cycler 480; fluorescence data were analyzed with the software version 1.5 and second derivative maximum method; gene expression was expressed as Cp (Crossing point) values. Statistical significance was tested by Mann–Whitney test (*p <0.05).

Multiplex Enhanced Reduced Representation Bisulfite Sequencing (mERRBS):

mERRBS libraries were prepared with a modified size selection fragments for multiplex application, as previously described (16). Only genomic regions with coverage ranging from 10 to 450 times were used for the downstream analysis. Reads were aligned against a bisulfite-converted human genome (hg19) using Bismark with Bowtie2 (17). Downstream analysis was performed with the MethylKit (18) and MethylSig (19) packages. We investigated cytosine methylation using beta-binomial regression (FDR ≤0.1) with a minimum cutoff of 25% methylation difference to identify Differential Methylated Cytosines (DMCs). Unsupervised clustering analysis with Principal Component Analysis was conducted using only high-variance CpGs tiles across all samples. Unsupervised analysis of DNA methylation using hierarchical clustering [distance= 1-Pearson Correlation, Ward’s agglomeration method] was performed using MethylKit. Differentially Methylated Regions (DMRs) were identified by first summarizing the methylation status of the genomic regions into 25-bp tiles and then identifying regions with an absolute methylation difference ≥ 25% and an FDR ≤0.1. DMRs were annotated to the RefSeq genes (NCBI) using the following criteria: (a) DMRs overlapping with a gene were annotated to that gene; (b) intergenic DMRs were annotated to all neighboring genes within a 50-kb window; (c) if no gene was detected within a 50-kb window, the DMR was annotated to the nearest transcription start site (20). For Chip-Enrich annotation we used the option locus.def=“nearest_tss” (21). Enhancer annotation was performed using chromatin immunoprecipitation sequencing (ChIP-seq) data sets generated in human-mobilized CD34⁺ cells. Enhancers were defined as regions with H3K4me1 and H3K27ac, with absence of the promoter marker H3K4me3 (22). Moreover, we used data sets generated in human-adult CD34⁺ cells based on histone repressive marks (H3K27me3 and H3K9me3) and histone active marks (H3K36me3, H3K4me1 and H3K4me3) to investigate if distorted DNA methylation in dic(1;7) was associated with a particular chromatin and/or structural microenvironment (22).

Sequencing:

All amplified libraries for RNA-seq and mERRBS underwent quality control steps as described for DNA and RNA extraction and were sequenced on Illumina HiSeq2500 using manufacturer’s recommended (Illumina San Diego, CA). The data were deposited in NCBI’s Gene Expression Omnibus (GEO) under the accession number GSE118648.

Integrative functional Analysis:

To analyze the functional pathway and Gene Ontology (GO) enrichment of the identified gene list, we used the Database for Annotation, Visualization, and Integrated Discovery 6.8 (DAVID) (23) and RNA-Enrich (24). Known and de novo DNA motif enrichment were identified with Hypergeometric Optimization of Motif EnRichment (HOMER) (25) using all tiled CpGs from our sequencing as background in the case of mERRBS analysis. In order to analyze the potential recruitment of transcriptional repressors that could be responsible of gene silencing, we extracted promoter position from the identified downregulated genes between dic(1;7) and controls and we intersected them with ChIP-seq data generated in CD34+ cells for CTCF (26), one of the most well characterized transcriptional repressor.

Results

The dic(1;7) breakpoints fell within α-satellite DNA

Double color FISH with Vysis D1Z5 and D7Z1 probes generated a fusion signal between centromeric regions in all our five dic(1;7) cases (Fig. 1c). Three cases bore one or two additional events as seen at both conventional karyotypes and SNPa, in particular, trisomy 8, trisomy 9 and del(13q) (Table 1 and Fig. 1d). In DIC 3, trisomy 8 characterized a subclone of a main line including dic(1;7) and del(13q) (Fig. 1d). No CNAs or cnLOH, with reduplication of a chromosomal allele alongside with loss of its respective homologous region, were detected by SNPa.

dic(1;7) bore no common somatic mutations

WES on DIC 2 (Table 1) identified a mutation in the ankyrin 7 domain of ANKS1B/EB1 gene (12q23.1), a promiscuous gene originating multiple fusion transcripts in solid tumors (27) and overexpressed in a subset of B-cell Acute Lymphoblastic Leukemia (B-ALL) (28). The same case harbored a variation in the polyprenyl synthetase domain of GGPS1 (1q42.3), a gene transcribing for a long non-coding RNA which is overexpressed in lung cancer and mutated in adult T-cell lymphoma cases (29, 30). In DIC 5 we identified two mutations, the first one within the alanine rich domain of CCDC8 gene (19q13), involved in maintaining microtubule integrity and mutated in the genetic 3M growth retardation syndrome (OMIM#614205) (31). The second mutation occurred in the proline rich domain of PSMF1 gene (20p13), a proteasome regulator previously described as mutated in a case of ALL (32). In silico analysis with three different softwares predicted the GGPS1 mutation to be disease causing, while the other 3 mutations showed conflicting results (Supplementary Figure 1). None of these variations were identified by Sanger sequencing in the other three patients (DIC 1, DIC 3 and DIC 4, Table 1).

dic(1;7) has a distinct expression signature

To characterize the transcriptomic differences between dic(1;7) cases and the other two groups, controls and t-MNs, we analyzed their gene expression profile. Unsupervised analysis using Principal Component Analysis of transformed RNA-Seq count data separated cases bearing dic(1;7) from controls and t-MNs along the first principal component (Fig. 2a and b). We identified a total of 4,860 Differentially Expressed Genes (DEGs) between dic(1;7) and controls, and 4,317 between dic(1;7) and t-MNs (Fig. 2c). Gene expression analysis showed a prevalent downregulation within the signature of dic(1;7) cases, affecting more than 80% of DEGs (Fig. 2c and d; Supplementary Figure 2). The CTCF analysis at level of promoters of silenced genes revealed that 920 out of 3,971 downregulated genes between dic(1;7) and controls (23%) were CTCF targets, suggesting a potential contribute of transcriptional repressors in mediating the observed downregulated gene expression profile (Supplementary Table 2).

Fig. 2.

Summary of RNA-seq results. a Scatter plot showing results of a Principal Component Analysis (PCA) of transformed count data from RNAseq using DESeq2 (varianceStabilizingtransformation). Samples are colored according to their disease status, observe that the DIC are separated from all other samples along dimension 1 (the x axis). b The scree plot line indicates the amount of the total variance preserved by a principal component. c Table summary of RNA-seq results analysis. Up refers to upregulated genes, down refers to downregulated genes. d Heatmap representing the top 50 differentially expressed genes (DEGs) across all three groups of samples. The heatmap colours reflect values representing the degree of expression from low to high as blue to red, respectively, as shown on the scale at the top-left hand side of the figure. Red= CTL; Blue= t-MN Green= dic(1;7). e Significance for ZMAT3 and NR1H2 expression were tested by Mann–Whitney test (*p < 0.05); values are expressed as means ± _SEM.

To capture biological differences, we analyzed functional pathways within DEGs. Dicentric cases showed 18 and 30 significantly enriched pathways when analyzed against controls and t-MNs, respectively (FDR ≤0.1) (Supplementary Table 2 and 3). The t-MN cohort compared to controls showed 102 enriched pathways (Supplementary Table 4). When compared to controls the two neoplastic subgroups included in this study, i.e., dic(1;7) and t-MNs, showed aberrant downregulation of biologically relevant pathways, specifically: MAPK signaling, spliceosome, ribosome, GnRH signaling, IL12/Stat4 dependent signaling and TNF/stress related signaling (Supplementary Tables 2–4). In addition, six pathways emerged as specifically deregulated (4 down- and 2 up-) in dic(1;7) cases (Supplementary Tables 2–4). Notably, the signaling of ATP-binding cassette (ABC) transporters, a superfamily of proteins involved in the transport of metabolic products and drugs across biological membranes, emerged as specifically downregulated (Supplementary Tables 2–5). Of the ABC genes detected as enriched by our RNA pathway analysis, seven were found to be significantly deregulated in dic(1;7) compared to the other two subgroups (Supplementary Table 5). In particular, 5/7 were downregulated; ABCA6 and ABCA10, known as phospholipid and cholesterol transporters; ABCB4 and ABCG2 involved in phosphatidylcholine and drug transport; and ABCB6, which plays a crucial role in heme synthesis by mediating porphyrin uptake into mitochondria (33). Two other ABC genes, the phospholipid and drug transporters ABCA3 and ABCC3, were upregulated (Supplementary Table 5). Moreover, the cholesterol transporter ABCA9 was significantly downregulated in dic(1;7) only when compared to t-MNs; while the peptide transporter ABCB2 was significantly upregulated only against controls (Supplementary Table 5).

GO analysis of differentially expressed genes showed significantly enrichment in lipid metabolism terms involving twelve genes, ten of which were downregulated only in dic(1;7) (Supplementary Tables 2–5). Among them, we found one group of lipid metabolism genes related to oxidative/inflammatory response: the apolipoprotein APOM and the cholesterol esterifying enzyme LCAT that have been reported to decrease upon oxidative and/or inflammatory response (34, 35); the cytidine deaminase APOBEC2, found to be regulated by the inflammatory TGF-beta signaling (36) and the nuclear receptor NR1H2 that was involved in lipid homeostasis and inflammation (37). A second group of genes is encoding for lipids related to cancer; LPL, lipoprotein lipase hypo-expression that was shown to predict evolution in CLL patients (38); PLA2G7, a phospholipase which expression in the peripheral blood was predictive of survival in melanoma patients (39); apolipoprotein APOC1 that was associated with poor prognosis in pancreas cancer patients; moreover, when inhibited, it induced apoptosis in pancreatic cancer cell lines (40).

Only two specifically upregulated pathways emerged in dic(1;7); most important of which was the TP53 signaling pathway, which included 9 of the DEGs (Supplementary Tables 2–5). In particular, the four TP53 target genes upregulated in dicentrics, compared to both t-MNs and controls, were: ZMAT3, STAT1, SUMO1 and CCNG1 (Supplementary Table 5). Moreover, SHISA5 and MYC were upregulated only against controls and SERPINE1, FAS and ATR only against t-MNs. By contrast, t-MNs versus controls showed TP53 signaling downregulation (Supplementary Table 4). The TP53 gene by itself was not significantly deregulated in any subgroups although there was a trend towards RNA hyper-expression in dic(1,7) and hypo-expression in t-MNs (data not shown). We validated the upregulation of the TP53 target gene ZMAT3 and the downregulation of NR1H2 gene, involved in lipid homeostasis and inflammation, by qRT-PCR (Fig. 2e).

Both gene dosage effect and DNA methylation profile contributed to the dic(1;7) signature

We investigated gene dosage effect of 1q trisomy and 7q monosomy from RNA-seq. Compared with controls and t-MNs, 97.5% (117/120) and 95.9% (118/123), respectively, of 7q DEGs were downregulated in dic(1;7), in keeping with 7q loss (Fig. 3a). We also found downregulation of EZH2 (41) and MLL3 (42), previously identified as haploinsufficient putative suppressor genes on chromosome 7. Surprisingly, among 198 1q DEGs included in the signature of dicentrics against controls, 90 were upregulated and 108 downregulated (Fig. 3b). Also, when dic(1;7) were compared to t-MNs the 157 1q DEGs resulted to be 52 up- and 105 down-regulated (Fig. 3b). These findings showed that, despite of trisomy, more than 50% of 1q genes where always hypo-expressed in the dic(1;7). Consistent with this, 85/119 1q DMRs were hypermethylated in dicentrics compared with controls (67/109 compared with t-MNs) (Fig. 3c) and they were annotated to 83 genes, 31 of which were downregulated. Interestingly, 61/85 hypermethylated 1q DMRs (71.8%) were annotated to enhancers, thus linking 1q hypermethylation to functional DNA region (Supplementary Table 6).

Fig. 3.

Gene dosage effect and DNA methylation pattern on 7q and 1q regions of dic(1;7). a Chromosome view of regions significantly enriched in genes differentially expressed using PGE on 7q (left). Percentage of DEGs distribution within down- and up- regulated genes in dic(1;7) compared to controls (up-right) and to t-MNs (bottom right). Compared to controls, 117/120 of 7q DEGs were downregulated in dic(1;7), while compared to t-MNs 118/123 7q DEGs were downregulated. b Chromosome view of regions significantly enriched in genes differentially expressed using PGE on 1q (left). Percentage of DEGs distribution within down- and up- regulated genes in dic(1;7) compare to controls (up-right) and to t-MNs (bottom right). The signature of dic(1;7) contained 90 up- and 108 (54.5%) down-regulated genes on 1q compared to controls; 52 up- and 105 (66.8%) down-regulated genes on 1q compared to t-MNs. c DMRs distribution along chromosome 1q involved in the translocation. Dicentric cases showed hypermethylation of DMRs on 1q compared to both, controls (left) and t-MN (middle). By contrast, t-MNs showed a prevalent hypomethylated pattern on 1q (right).

Global DNA methylation profiling captured 3.2M CpGs across the genome. Principal component analysis of high-variance CpGs identified dic(1;7) and t-MN cases from controls (Fig. 4a and b). Unsupervised analysis using hierarchical clustering segregated dicentrics with additional anomalies in a separate cluster with the t-MN cases with complex karyotype (Fig. 4c; Table 1). Supervised analysis showed both global hyper- and hypo-methylation in dic(1;7), compared with the other subgroups, while t-MNs showed global hypomethylation (Fig. 4b and c). We identified a total of 3,261 DMRs between dic(1;7) and controls, and 4,568 DMRs between dicentric cases and t-MNs (Fig. 5a). Genome-wide distribution of DMRs and annotation to genomic regions showed that, compared to controls, dicentrics were significantly depleted at promoter regions (15% vs. background 30%, binomial test p <2.2e⁻¹⁶) and significantly enriched in non-promoter regions (Fig. 5a and b). These results were maintained also in the comparison against the t-MN group.

Fig. 4.

Summary of mERRBS analysis. a Unsupervised multivariate analysis using principal component analysis (PCA) was able to separate controls from all tumor samples. The scree plot line indicates the amount of the total variance preserved by a principal component (middle). Unsupervised analysis of DNA methylation using hierarchical clustering (left). Red= Control; Blue= t-MNs; Green= dic(1;7). b Circos plot showing the hyper- and hypo-methylated DMRs found for each group comparison: DIC/CTL (left), DIC/t-MN (middle) and t-MN/CTL (right). The outer circle indicates the G-banded human chromosomes, red and blue colours refers to hyper- and hypo-methylated DMRs distribution, respectively. c Volcano plots of all DMRs distribution between each group. The plot shows the DNA methylation difference on percentage (x-axis) with respect to the −log₁₀ of the adjusted p-value for each DMR. Thresholds are shown as red lines: dot vertical lines indicate DNA methylation difference cut-off (−25% and +25%), dot horizontal line indicates statistically significant cut-off (q-value ≤ 0.1). Significantly different hypo- and hyper-DMRs are highlighted as blue and red dots, respectively. DIC vs. CTL (left), DIC vs t-MN (middle) and t-MN vs. CTL (right).

Fig. 5.

Summary of annotation analysis. a Table summarizing the distribution of annotated DMRs on genomic regions for each group. We used Methylkit package on R to annotate previously identified DMRs on Methylsig to hg19. Statistical results for Binomial enrichment/depletion test were conduct on R. For all tests, we used 95 percent confidence interval and percentage of background annotation for each category (promoters, exons, introns, intergenic) as true probability of success. b Pie charts illustrate DMRs annotated to promoter (red), exonic (green), intronic (blue) and intergenic regions (purple) in DIC vs CTL (left) and vs t-MN (right). As background, we used total identified DMRs without filters (top left). c Distribution of hypo- and hyper-methylated DMRs on promoter regions for each comparison (top) and relative number of hypo- and hyper- promoter DMRs for each group (bottom). d Venn diagram showed the overlap between genes annotated to promoter DMRs and all DEGs identified on RNA-seq analysis in DIC vs. CTL (i) and vs. t-MN (ii).

Focusing on promoter regions, t-MNs showed gene-specific promoter hypermethylation (Fig. 5), confirming previous studies (43). dic(1;7) cases compared with controls showed hypermethylation of 356/502 (68%) promoters (Fig. 5c). However, promoter hyper-methylation correlated with significant changes in expression level only for 31 genes, suggesting that its contribution to dic(1;7) expression profile was limited (Fig. 5d).

Enhancer hypermethylation enriched on TFs binding sites emerged in dic(1;7)

Analysis of the non-promoter regions showed that 43% of the dic(1;7) DMRs were enriched for enhancers (background 34%, binomial test p< 2.2e⁻¹⁶) when compared with controls, while t-MNs had 38% enrichment (background 34%, binomial test p= 0.046) (Fig. 6a). In silico analysis using ChIP-seq data sets for histone repressive marks (H3K27me3 and H3K9me3) and histone active marks (H3K36me3, H3K4me1 and H3K4me3) for CD34+ primary cells from public database (22), confirmed that distorted DNA methylation in dic(1;7) is mainly associated with changing in active histone marks specifically linked to enhancer regions (72% of DMRs overlaps with H3K4me1). Searching for super-enhancers using the Hnisz D data set of 234 super-enhancers identified in CD34+ adult cells (44), we found five super-enhancers out of the 744 in our series (Supplementary Table 6). In both dicentric and t-MNs, enhancer DMRs were predominantly found in intronic regions. A direct enriched test confirmed that the enrichment at enhancers was stronger in dicentric cases than in t-MNs, with a prevalent hypermethylation (Fig. 6a). Since DNA methylation negatively correlates with enhancers’ activity and transcription of target genes (45), we used the Predicting Enhancer Target by modules tool to identify putative enhancer-target genes (46). Indeed, compared to controls, dic(1;7) RNA-seq revealed that 1,747 out of 4,860 DEGs were target genes of hypermethylated enhancers (Fig. 6b). Moreover, 1,377/1,747 DEGs had decreased gene expression, suggesting that enhancer hypermethylation directly accounts for 34.7% of the downregulated signature in dicentric cases (Fig. 6b). These results were maintained by dic(1;7) against t-MNs, with enhancer hypermethylation accounting for 40% of the downregulated profile (Fig. 6b). By contrast, the t-MN group showed 128 out of 192 enhancer DMRs (66%) to be hypomethylated (Fig. 6a). Since DNA methylation has been shown to prevent binding of transcription factors (TFs) to their target sites (47), we analyzed DMRs with HOMER. This analysis revealed that DMRs were enriched for TF binding sites (Supplementary Figures 3–6), with those regions differentially methylated in dic(1;7) compared to t-MNs showing enrichment for the key hematopoietic TF factors, GATA1 (p-value= 1e⁻⁴), GATA2 (p-value= 1e⁻³), GATA3 (p-value= 1e⁻²), ETV1 (p-value= 1e⁻²), STAT3 (p-value= 1e⁻²) and RUNX1 (p-value= 1e⁻²) (Supplementary Figure 4). Specifically, GATA and RUNX1 transcription factors binding sites were hypermethylated while ETV1 and STAT3 binding sites were hypomethylated. With respect to hypermethylated enhancers, dic(1;7) showed enrichment of 28 and 27 TFs, compared to controls and t-MNs, respectively (Supplementary Figures 5 and 6). Interestingly, dic(1;7) hypermethylated enhancers were strongly enriched in motifs for Krüppel-like factor (KLFs) protein family transcription factors. Specifically, KLF5, KLF6 and KLF 4 were included among the top enriched transcription factors in the analysis against controls (Supplementary Table 5), while KLF3, KLF5 and KLF4 emerged in the analysis against t-MNs (Supplementary Table 6). By contrast, none of these KLF transcription factors emerged as enriched in t-MNs enhancers when compared to controls (Supplementary Figures 7).

Fig. 6.

Summary of enhancers analysis. a Pie chart illustrates DMRs enrichment on enhancer regions. Red= feat1, enhancers; light blue= others, non-enhancers. Compared with controls, dic(1;7) DMRs showed an enrichment on enhancers with p-value< 2.2e⁻¹⁶ (left), compared with t-MN showed an enrichment on enhancers with p-value of 6.617e⁻⁰⁶ (middle) and t-MNs compared with controls showed an enrichment on enhancers with p= 0.046 (right). For statistical analysis, we used binomial test with 95 percent confidence interval and percentage of background annotation as true probability of success. b Venn diagrams illustrated the overlap between hypermethylated enhancer target genes and downregulated signature on DIC/CTL (left) and DIC/t-MN (right) groups.

Discussion

This is the first in-depth genetic and epigenetic characterization of dic(1;7) which provided us with new insights into the biological features of a recurrent leukemic chromosome rearrangement involving repetitive DNA at centromeric regions.

In our patients breakpoints of dic(1;7) fell within α-satellite DNA at chromosome 1 and chromosome 7, in agreement with previously reported data (48). As a result, the affected cells had genomic imbalances involving both euchromatic regions, due to entire 1q trisomy and 7q monosomy, and heterochromatic regions, with an excess of α-satellite and pericentromeric sequences of chromosome 1. Interestingly, centromeric juxtaposition in dic(1;7) did not alter centromere action on chromosome segregation and the der(1;7) behaved as a stable clonal marker along mitotic divisions of many cell generations, even during long lasting cytogenetic monitoring, for up to five years (data not shown). Moreover, the dic(1;7) chromosome behaved as a cytogenetically primary abnormality since additional karyotypic changes, when present, were individually different, in both de novo and therapy-induced MDS. Accordingly, data from 258 cases of dic(1;7) collected from Mitelman database showed 82.5% of cases with the dic(1;7) as sole cytogenetic aberration or plus only one/two additional abnormalities (2). In addition, here we excluded the presence of cytogenetically cryptic changes by using SNP-arrays. Altogether these data highlight dic(1;7) as the cytogenetic hallmark of a biological entity including both de novo and t-MDS strikingly different from other t-MNs typically associated with complex karyotypes (49).

Common somatic mutations did not emerge in our cases with dic(1;7) in which we found only private variations affecting GGPS1; CCDC8; and PSMF1 genes, still not reported in MDS/AML, and ANKS1B/EB1 gene, recently annotated in two AML cases (50).

Interestingly, when compared to both controls and t-MNs, the dic(1;7) subgroup was characterized by a signature with prevalence of gene downregulation that was supported by a specific epigenetic program. Notably, promoter hypermethylation contributed to downregulation to a lower extent than hypermethylation at level of non-promoter regions, particularly at intronic enhancers which target genes represented around 35% of the entire downregulated expression profile. Both enhancers and super-enhancers aberrant DNA methylation have been identified in malignancies (51, 52). Interestingly, in multiple myeloma, Agirre X. et al. found hypermethylated intronic enhancers to be associated with binding sites for transcription factors related to B-cell differentiation (53). Similarly, the hypermethylated enhancers in dic(1;7) included binding sites for important transcription factors in normal hematopoiesis, such as GATA1, GATA2, GATA3, and RUNX1. In addition, in dic(1;7) the Krüppel-like factor (KLFs) protein family specifically emerged with four top enriched members involved in regulation of: stem/progenitor proliferation and localization in the bone marrow (KLF5) (54); erythroid differentiation (KLF5, KLF6, KLF3) (55), stem cell pluripotency (KLF4) (56), and tumor suppression (KLF4) (57). Intriguingly, a checkpoint role following DNA damage was attributed to KLF4 loss of function in a murine model, in which its inactivation was associated with high level of apoptosis and genetic aberrations, including dicentric chromosomes (58). Altogether these data first emphasize a pathogenetic role of the KLFs family in a distinct cytogenetic subgroup of MDS/AML. Future studies are required to investigate potentialities of their modulation for new therapeutic strategies.

Dosage effect due to chromosome number variations is a well-known mechanism altering gene expression in leukemic cells with a strong relationship between gene hypo-expression and chromosome loss (59). Accordingly, in the dicentric signature we found the expected gene dosage effect due to 7q monosomy, as more than 95% of differentially expressed genes belonging to the long arm of chromosome 7 were downregulated. In contrast, a gene dosage effect by 1q trisomy acted only partially, given that more than 50% of 1q differentially expressed genes were downregulated in the cases with dicentric chromosome. Indeed, DNA methylation profile clarified that discrepancies between 1q trisomy and gene hypo-expression were related to high level of 1q hypermethylation in dic(1;7), compared to both controls and t-MNs. Whether gain of the 1q heterochromatin is related to this epigenetic effect is still unknown.

Pathway analysis of the signature showed that ABC transporters were significantly down-regulated in dic(1;7) when compared to both controls and t-MNs. ABC transporters, a large family of proteins with localization at mitochondria, endoplasmic reticulum, Golgi, endosomes, and cytoplasmic membranes, are involved in multiple cellular functions, including protection against genetic damage caused by xenobiotic and chemotherapeutical compounds, and transmembrane transport of amino acids, ions, polysaccharides and lipids (60). In particular, ABCB1, alias MDR1/p-glycoprotein, and ABCG2, alias BCRP/MXR, have been extensively studied in drug resistance in AML (61). The ABC transporters specifically involved in the dic(1,7) signature are known to play a role in the transport of phospholipids, cholesterol and phosphatidylcholine. Accordingly, only in the dic(1;7) subgroup, our Gene Ontology analysis showed a significant enrichment for genes regulating the lipid metabolism involved in stress and inflammatory responses (37,58). Among them the downregulated gene NR1H2 is.an ubiquitous isoform of the liver X receptor (LXR), a targetable candidate in leukemic cells with deregulation of cholesterol homeostasis (62, 63). Whether the ABC involvement impacts on the chemo-resistance/sensitivity of dic(1;7) affected cells remains to be elucidated.

The p53 signaling pathway emerged as a specifically up-regulated pathway in dic(1;7) versus both controls and t-MNs. In the last subgroup loss of function p53 mutations accompanied complex karyotypes, as expected (64). Interestingly, up-regulation of p53 target genes has been previously found in AML with wild type p53 gene, suggesting that, in tumor cells, downstream targets may mediate p53 dysfunction regardless of gene mutations (65). In line with this hypothesis both our series of dicentrics and the published AML series (65), sharing wild type p53, showed up-regulation of ZMAT3, a p53 target gene encoding for an RNA-binding zinc-finger protein regulating cell cycle arrest (66). Altogether these data suggest that dic(1;7) MDS, belong to a subgroup of myeloid malignancies in which, in the absence of mutations at the gene, p53 is involved through deregulation of its targets.

In conclusion, in this study we showed that MDS with dic(1;7) are a distinct cytogenetic-epigenetic entity characterized by a downregulated expression profile closely connected to gene dosage effect and site-specific hypermethylation, mostly at intronic enhancers enriched for hematopoietic transcription factors. Downregulation of pathways and genes involved in lipid homeostasis and up-regulation of TP53 signaling emerged as specific biological features, and separated dic(1;7)-positive MDS from t-MN.