Genome-wide promoter methylation of hairy cell leukemia

Classic hairy cell leukemia (HCL) is a rare mature B-cell tumor that is characterized by the accumulation of leukemic cells in the bone marrow, spleen, and peripheral blood.¹ 

The universal genetic fingerprint of HCL is the acquisition of the BRAF V600E mutation in all individual hairy cells.^1-5  The mutation leads to constitutive BRAF–MEK–ERK pathway activation^1,2  and represents an effective therapeutic target in patients.^3,6  KLF2 and CDKN1B (p27) mutations may cooperate with BRAF V600E in the tumor cells of some patients.⁷  However, HCL typically has a highly stable genomic profile,^8,9  and the inability of BRAF inhibitors to completely eradicate HCL in patients suggests that factors other than genetics may contribute to disease pathogenesis and behavior.² 

Expression of multiple functional immunoglobulin isotypes is another unique feature of HCL.^10,11  Its association with low levels of intraclonal variations of the immunoglobulin gene heavy chain variable (IGHV) region and ongoing isotype-switch events prior to deletional recombination are suggestive of ongoing environmental interactions promoting or maintaining the tumor clone.^12-15  However, the behavior of mature B-cell tumors is also influenced by the DNA methylation status of the transformed cell.^16-18  DNA methylation is involved in controlling cellular differentiation and cell type specification during hematopoietic development.^17,19  In the most common form of adult leukemia, chronic lymphocytic leukemia (CLL), the methylation profile is clearly different between the 2 main subsets with unmutated (U-CLL) or mutated IGHV (M-CLL) and is stable over the course of the disease, likely reflecting the maturation of the cell of origin.^17,20-22  Methylation profiling also helps to better define specific disease subentities, like IGHV3-21⁺ CLL, and it can contribute to defining of disease prognosis.^17,23,24 

The DNA methylation profile of HCL has not been extensively investigated. Here, we investigated the DNA methylation profiles of a series of HCL using the Illumina HumanMethylation27 array and compared them with other B-cell tumor entities and with normal peripheral blood B cells at different stages of differentiation.

Peripheral blood mononucleated cells were obtained at diagnosis or prior to any treatment from 41 mature B-cell tumors, including 11 HCLs, 7 splenic marginal zone (MGZ) lymphomas (SMZLs), 7 U-CLLs, and 6 M-CLLs. The CLL cohort also included 10 IGHV3-21⁺ CLLs (CLL–VH3-21, all mutated for IGHV), which was analyzed as a separate subentity. Diagnosis was made according to the World Health Organization 2018 Classification of Tumors of Hematopoietic and Lymphoid Tissues.²⁵  Differential diagnosis of HCL and SMZL was verified by allele-specific oligonucleotide polymerase chain reaction and sequencing.²⁶  HCL samples were confirmed BRAF V600E mutated, whereas all SMZLs were confirmed BRAF V600E unmutated. Use and mutational status of the expressed tumor IGHV gene were determined using our previously reported procedures.¹⁵  Purity of tumor B cells was ≥70% in all samples, as measured by immunophenotyping.⁸  The characteristics of the 11 HCL samples are shown in supplemental Table 1. Patients provided informed consent in accordance with the local institutional review board requirements and the Declaration of Helsinki.

DNA extraction and quality control were performed as previously described.⁸  Methylation profiling was performed with the Infinium HumanMethylation27 array (Illumina, San Diego, CA), as previously described.²⁷ 

Probes inside or outside cytosine guanine dinucleotide islands (CGIs)²⁸  were analyzed separately, as previously reported.²⁷  The methylation profiles of the CLL cases were derived from previous publications.^23,27  To identify the normal counterpart of HCL, defined as the nontumor B-cell subset with the closest methylation profile to HCL cells, we studied a series of B-cell subpopulations obtained with the Infinium HumanMethylation450 BeadChip and acquired from the European Genome-phenome Archive dataset EGAS00001000534.¹⁸  Methylation data were preprocessed, including background correction and quantile normalization, using the “minfi” package in R environment.²⁹  Batch effect was corrected for 27k and 450k platforms using the ComBat algorithm,³⁰  and only probes commonly annotated in both 27k and 450k platforms were considered in the study. To stabilize the variance of methylation range across the series, β values were transformed into M-values for the identification of differentially methylated probes. A moderated Student t test (limma) was used to identify the differentially methylated regions (q value < 0.05; absolute difference of average M-values > 1.0).³¹  To further identify probes with changes in methylation, the Fisher's exact test was performed on the β values (P < .05; not methylated for β from 0 to 0.33, hemimethylated for β from 0.33 to 0.66, and methylated for β > 0.66).

Gene expression profiling (GEP) data, obtained with Affymetrix U95A and U95Av2 arrays, were extracted from Gene Expression dataset GSE2350³²  and analyzed by limma to characterize the differentially expressed genes (q value < 0.05 and absolute log₂ fold change >1.0). The methylation–GEP correlation (Pearson) was calculated only for genes with probes common to the GSE2350 dataset.³²  We considered inversely correlated those genes significantly hypomethylated and overexpressed, or hypermethylated and underexpressed, in ≥1 of the following independent comparisons (computed separately on methylation and gene expression profiles using the moderated Student t test): HCL vs post–germinal center (GC) B cells, HCL vs SMZL, HCL vs M-CLL, and HCL vs U-CLL. For functional annotation, >6100 gene sets belonging to The Molecular Signatures Database v5.1 gene set enrichment analysis (GSEA) collection,³³  were grouped into 100 major biological themes (“concepts,” supplemental Table 2), reducing the number and the redundancy of gene sets. Single-sample GSEA ([ssGSEA], “GSVA” package in R environment)³⁴  was then performed on the methylation and GEP data. The ssGSEA output was subsequently analyzed by limma at the concepts level.³¹  Analyses were performed using R environment (R Studio console; RStudio, Boston, MA).

We studied the genome-wide promoter methylation profile of 11 HCLs (all harboring BRAF V600E mutation) vs defined normal B-cell subsets^18,35  or vs SMZL (n = 7), U-CLL (n = 7), or M-CLL (n = 6).

Unsupervised analysis of all probes provided insight into the potential normal B-cell counterpart of HCL. Unsupervised clustering (Figure 1A) and multidimensional analysis (Figure 1B) highlighted a marked distance of HCL from naive B-cell and GC founder B-cell subsets, whereas the closest subsets to HCL were the post-GC subsets, including splenic MGZ, low-maturity memory, and intermediate-maturity memory B-cell subsets.

When the other B-cell tumors were compared, HCL samples clustered very much closer to SMZL than to CLL. Although the latter distributed largely according to its IGHV mutational status (U-CLL, M-CLL, IGHV3-21 CLL), as expected,^16,18,24  SMZL cases clustered with (2/7) or very close to (5/7) HCL samples. Indeed, 1 of the 2 cases, initially classified as CD103⁺DBA44⁺ HCL, lacked the BRAF V600E mutation and was reclassified as an SMZL only after central immunohistochemical revision due to ANXA-1 negativity.

Two additional unsupervised analyses were performed to better understand the epigenetic program of HCL cells: 1 for the CGI probes only (supplemental Figure 1A) and another 1 with probes outside of the CGIs (supplemental Figure 1B). Both analyses generally overlapped with the clustering on the global genome-wide methylation profile, although it is worth mentioning that the CGI-only methylation profile clustered all HCL samples in an independent branch, separately from post-GC B cells but still together with the 2 SMZL cases mentioned above.

Hence, this analysis documented that HCL has a global methylation profile close to post-GC B cells and SMZLs.

HCL has a gene expression profile similar to that of post-GC B cells but with a specific signature that is distinct from normal and other neoplastic B cells.³²  The published HCL gene-expression signature³²  was then integrated with promoter methylation profiles by independent comparisons of HCL with post-GC B cells and with other B-cell tumors (HCL vs SMZL, HCL vs M-CLL, and HCL vs U-CLL). Enrichment analysis by GSEA revealed hypermethylation of the underexpressed HCL gene signature, whereas the overexpressed HCL signature was hypomethylated compared with post-GC B cells (supplemental Figure 2A). An integrated analysis using the moderated Student t test showed that 47% (36/76) of the differentially transcribed genes were inversely correlated with their methylation status (supplemental Figure 2B; supplemental Table 3). The top 10 hypomethylated and overexpressed transcripts were EPB41L2, DST, RIN2, EMP1, PDE4DIP, ENG, RCBTB2, AIF1, FLT3, and PLOD2, whereas only 4 transcripts were hypermethylated and underexpressed: CXCR5, TRAF5, PAWR, and TNFAIP8.

The observed inverse correlation between gene expression and methylation (Pearson correlation ρ = −0.375, P < .001) points to DNA promoter methylation as a mechanism involved in the regulation of the specific gene expression signature of HCL.

Independent supervised analyses were performed to compare HCL methylation patterns with those of post-GC B cells and with the methylation profiles of SMZL, M-CLL, U-CLL, and CLL–VH3-21.

Commonly hypermethylated genes in HCL, compared with post-GC B cells or with the other B-cell tumors, included those involved in regulation of B-cell proliferation, cell motility, and immune system process. The top hypermethylated genes in HCL included TNFRSF13B, MCM5, VHL, ENTPD1 (CD39), CCL22, MNDA, SFTPB, FOXN1, FCER1G, and RUNX3. Commonly hypomethylated genes included those involved in the RAS signaling cascade. The top 10 hypomethylated genes in HCL were RIN2, ACTA1, EMP1, SPARCL1, COL11A2, IL10, LRRC32, CMKLR1, RAD50, and FOXD1 (Figure 2; supplemental Tables 4-10).

We then searched methylation differences between HCL and post-GC B cells for genes that were not differentially methylated from other B-cell tumors. Hypermethylated genes in HCL were annotated for transmembrane transport and genes downregulated by overexpression of an oncogenic form of KRAS; among the top genes we found RNF126, BLCAP, CDH12, ELF2, and PTENP1. The top hypomethylated genes in HCL were LAPTM5, TRAF1, MAP2K1, IL10RA, and IL5RA; and GSEA provided enrichment in antigen-dependent B-cell activation and RAS, interleukin-10 (IL-10), IL-2, and MAPK signaling pathways (Figure 2; supplemental Tables 4-10).

DNA promoter methylome of HCL was further interrogated to identify genes differentially methylated from other B-cell tumors (SMZL, M-CLL, U-CLL, and CLL–VH3-21) but not from post-GC B cells. Among the top hypermethylated genes we found CHI3L2, MMP11, LILRB4, PRDM2 (RIZ1), and BRD4, whereas TJP1, SIX1, ADAM12, BTBD3, and GATA4 were the top hypomethylated genes. GSEA showed that hypermethylated probes were enriched in genes modulated by AKT-mTOR, IL-2, and IL-15 signals and by JNK inhibitors; meanwhile, polycomb repressive complex 2 targets, cell–cell interactions, and G protein–coupled receptor signaling were enriched in the hypomethylated genes in HCL (Figure 2; supplemental Tables 4-10).

Methylation and gene expression profiles were further integrated to compare HCL with post-GC B-cell subsets. The unsupervised methylation profiling analysis identified low-maturity memory and intermediate-maturity memory B cells and MGZ B cells as the normal populations closest to HCL; unfortunately, gene expression data for these populations were not publicly available.³²  Thus, we integrated the methylation profile of post-GC B cells (nonclass switched and class switched) with the gene expression profile of the closest available B-cell subset, CD27⁺ memory B cells, although these were primarily represented by class-switched memory B cells.

Despite this shortcoming, the comparison identified 119 hypermethylated or hypomethylated genes in HCL. TNFRSF13B, ENTPD1, KLK1, CCL22, TRPM2, CXCR5, LY86, CD86, RUNX3, and APOBEC2 were the 10 most hypermethylated and underexpressed genes, whereas the top 10 genes with hypomethylation and high expression included SPARCL1, RIN2, EMP1, LRRC32, AIF1, FLT3, CMKLR1, DST, RCBTB2, and ITGAD (Figure 3; supplemental Figure 3; supplemental Table 11). Interestingly, GEP signatures and several genes associated with BCR–TLR–NF-κB and BRAF-MAPK signaling pathways and cell adhesion were hypomethylated and overexpressed. Conversely, cell-differentiation markers and methylated genes in cancer appeared hypermethylated and underexpressed (Figure 4; Table 1; supplemental Table 12).

We then integrated gene expression and methylation data to compare HCL with SMZL and CLL. This analysis showed a marked overlap with the deregulated pathways identified in the comparison of HCL with post-GC B cells, suggesting a specific methylation-driven gene expression signature of HCL cells. A set of 245 genes was differentially methylated and expressed in HCL compared with SMZL and CLL. The top 10 hypermethylated and underexpressed genes were LILRA4, SFTPB, TNFRSF13B, KCNN4, PMAIP1, CXCR5, TRAF5, TGIF1, FAM65B, and LILRB4, whereas RIN2, BTBD3, SPARCL1, SLITRK5, EMP1, CXADR, CAMK1, PLOD2, LRRC32, and CHN2 were hypomethylated and overexpressed (Figure 5; supplemental Figure 4; supplemental Table 13). Finally, when HCL methylation profiles were compared with those of CLL–VH3-21, 85 differentially methylated genes were differentially identified, because there were no publicly available GEP data for this subset of CLL. Among the top low-methylated genes we found RAD50, FOXD1, RORA, CCNC, and WT1, whereas FCER1G, CCL2, NFKBIE, CD86, and MGMT appeared hypermethylated (supplemental Table 13).

Enrichment analysis revealed high expression of signaling pathways that are key to B-cell survival among the hypomethylated and overexpressed transcripts. These included integrin and CD40 signaling, MAPK (ERK, JNK) and RAS–RAF–MEK–ERK pathways, NF-κB activation upon TLR signaling, and B-cell activation mediated by the IKK complex. In accordance with the postulated HCL cell of origin, GC and pre-GC programs were methylated and repressed in HCL. Similarly, genes involved in transcription and the respiratory cascade, as well as genes harboring methylation marks, appeared more methylated and less expressed in HCL than in the other entities (Figure 6; supplemental Table 14).

We have characterized the genome-wide DNA promoter methylation pattern of HCL and compared it with normal B-cell subsets and with other mature B-cell tumor entities (U-CLL, M-CLL, IGHV3-21 CLL, and SMZL). HCL had a distinct methylation pattern. The closest normal counterpart was post-GC B cells (low-maturity memory and intermediate-maturity memory B cell), along with MGZ B cells. The closest neoplastic entity was SMZL, which is in agreement with HCL and SMZL being morphologically and phenotypically similar, albeit distinct, tumor entities.

HCL is characterized by the presence of the BRAF V600E somatic mutation, which leads to constitutive BRAF–MEK–ERK pathway activation and represents an effective therapeutic target in patients.^1-5  However, BRAF inhibitors are unable to completely eradicate HCL,^2,3,6  suggesting that factors other than genetics may contribute to disease pathogenesis and behavior. The platform used in this study had a limited number of genes compared with the arrays currently available and was not associated with the gene-expression profile of the individual cases. However, we were still able to define the contribution of promoter methylation to the previously reported HCL gene expression signature,³²  identifying a set of hypomethylated and highly expressed genes associated with the BCR and the BRAF signaling pathways. This result supported the notion that promoter methylation is generally associated with repression of transcription and vice versa.³⁶ 

Different methylation changes that we observed in HCL were consistent with the constitutive activation of the RAS–RAF–MEK–ERK pathway,^1-3  pointing to a permissive epigenetic landscape as a player in the upregulation of BRAF-related genes and, thus, participating in the specific biology of HCL. RIN2, encoding for a Ras effector that plays a role in the stimulation of GTPase activity,³⁷  was hypomethylated and overexpressed. Transcriptional targets of the Ras cascade, such as EMP1, encoding an integral tetraspan membrane protein,³⁸  were hypomethylated and overexpressed. There was hypomethylation and overexpression of IGFR1, which initiates a cascade of downstream signaling events leading to activation of the RAS/MEK/ERK pathway,³⁹  IGF-related signatures, and its direct target gene SPARCL1.⁴⁰ CMKLR1 appeared hypomethylated and overexpressed in HCL. The protein product of CMKLR1, RARRES2, which induces cell proliferation by increasing phosphorylation of ERK1/2,⁴¹  was also overexpressed. Hypermethylation of CD151 may also cooperate with BRAF V600E somatic mutation in the RAS constitutive signaling of HCL patients, because the gene acts as a negative regulator of adhesion-dependent activation of Ras.⁴² CHI3L2, the most hypermethylated gene in HCL compared with other B-cell tumors, encodes for a component of the chitinase family that has been reported to be hypermethylated in RAS-activated cancer subtypes.⁴³ VHL is a tumor-suppressor gene that mediates degradation of the hypoxia-inducible factor, contributing to the activation of a series of pathways, including the RAS–RAF–MEK–ERK pathway,⁴⁴  which is constitutively active in HCL. Last, and consistent with this constitutive activation, a number of hypomethylated genes in HCL are likely to contribute to RAS signaling, including MAP2K1, IL10RA, LAPTM5, and TRAF1. Among these, MAP2K1 and IL10RA also had a high expression in HCL. MAP2K1 encodes for the BRAF-downstream kinase MEK1, and it is activated by somatic mutation in 50% of HCL-variant patients.^5 ,45 IL10RA is reported to promote survival upon ERK phosphorylation.⁴⁶ 

Our current work also identified methylation patterns that may affect homing/migration and survival pathways of tumor cells. HCL is characterized by a gene expression signature³²  that appears to be inversely correlated with DNA promoter methylation, indicating the importance of methylation in gene expression control. Interestingly, the CXCR5 promoter was methylated in our HCL samples. CXCR5 is a chemokine receptor for B cells, and its absence on HCL cells⁴⁷  might explain the lack of tropism of the tumor cells to the white pulp and to lymph nodes.

In this study, we found that antiapoptotic genes, including TRAF5 and TNFAIP8, were hypermethylated. TRAF5 and TNFAIP8 expression is known to be downregulated in HCL.³² TRAF5 is a gene involved in the signal transduction of tumor necrosis factor–type receptors, including CD27, which is reduced by gene expression profile and not detectable by phenotype.¹² TNFAIP8 (TIPE) is a member of the tumor necrosis factor-α–induced protein family. Its repression protected hematopoietic cells from apoptosis,⁴⁸  and TNFAIP8-deficient mice showed increased leukocyte infiltration.⁴⁹  Other genes, including cell-differentiation markers and known tumor-suppressor genes, appeared highly methylated in HCL. O-6-methylguanine–DNA methyltransferase, coded by MGMT, is involved in DNA repair and is often methylated in cancers, including lymphoid tumors.⁵⁰ PRDM2 is also hypermethylated in leukemia and solid tumors, and its silencing induces high-grade B-cell lymphoma in mice.^51-53 ENTPD1 encodes for the plasma membrane protein CD39 and may have diagnostic value, because it was hypermethylated in HCL compared with all CLL subsets (M-CLL, U-CLL, and CLL–VH3-21). Unsupervised and supervised analyses documented a highly methylated TNFRSF13B (TACI) and MNDA in HCL; indeed, these genes were proposed for the differential diagnosis of MZL vs B-cell tumor entities other than HCL.^54,55 

Among the genes that appeared hypomethylated in HCL, we found FGF2 (fully nonmethylated across the series) and FLT3, which are known to be upregulated and might contribute to the bone marrow fibrosis that is characteristic of HCL.^32,56  Indeed, autocrine secretion of FGF2 by tumor cells is responsible for fibronectin production.⁵⁷  The ligand of FLT3, a potential therapeutic target in other leukemias,⁵⁸  is responsible for B-cell adhesion to fibronectin.³² RBBP4 and SUZ12, coding for components of the polycomb repressive complex 2, were hypomethylated, whereas genes silenced by methylation and transcripts harboring the trimethylation marks were methylated and repressed, pointing to the acquisition of a polycomb repression-associated methylator phenotype,⁵⁹  potentially linked to BRAF V600E mutation, at least in solid cancers.⁶⁰ 

In conclusion, our data reveal that HCL differs from SMZL, CLL, and normal B cells and indicate that the HCL-specific methylation pattern affects pathways involved in the homing, migration, and survival of HCL cells.

This work was supported by the Hairy Cell Leukemia Foundation (New York, NY); the Nelia and Amedeo Barletta Foundation (Lausanne, Switzerland); Cancer Research UK (CRUK Centre grant C34999/A18087); the Southampton Cancer Research UK and NIHR Experimental Cancer Medicine Centres; the University of Southampton; Bloodwise (grants 16003, 14037, and 12021); the Keanu Eyles Haematology Fellowship for the Cancer Immunology Centre, University of Southampton; and the Gilead UK & Ireland Oncology Fellowship Programme 2016.

Contribution: A.J.A. performed statistical analyses, interpreted data, and wrote the manuscript; A.R. performed DNA methylation profiling; G.C. reviewed and collected sample materials and diagnoses and interpreted data; I.K. and L.C. performed statistical analyses; D.R. and G.G. collected well-characterized tumor samples; M.K. and R.R. provided CLL methylation profiles; A.A.M., E.Z., and P.W.J. provided advice and interpreted data; C.C.O. analyzed and provided the profiles from the normal B-cell subsets; F.B. designed research, performed statistical analyses, interpreted data, and wrote the manuscript; F.F. designed research, collected and characterized tumor samples, interpreted data, and wrote the manuscript; and all authors approved the final manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Francesco Bertoni, Institute of Oncology Research, via Vincenzo Vela 6, 6500 Bellinzona, Switzerland; e-mail: francesco.bertoni@ior.usi.ch; and Francesco Forconi, Cancer Sciences Unit, University of Southampton, Cancer Research UK Centre, Somers Cancer Research Building, MP824, Southampton General Hospital, Southampton SO16 6YD, United Kingdom; e-mail: f.forconi@soton.ac.uk.