To the Editor:

Human cancer cell lines constitute useful models to study the primary disease and many relevant findings in tumor biology have been originated from them since the first immortalized cell line (HeLa) was obtained [1]. The easy experimental intervention, the purity of the transformed cells and their versatility to undergo high-throughput screenings represent advantageous features of the established cancer cell lines. In recent years, major efforts have characterized in detail the multiomics make-up of hundreds of cancer cell lines and studied their association with sensitivity to anticancer drugs [2,3,4,5]. Most of these studies have been genetic-centric and have not characterized in detail the epigenetic profiles underlying the characteristics of these cells or their impact on the efficacy of antitumoral compounds. In this regard, a past version of a DNA methylation microarray [3, 6, 7] or a more time-consuming readout such as reduced representation bisulfite sequencing [4, 8] have been used in those attempts to interrogate the epigenetic setting. Herein, we have obtained the DNA methylation profiles of 210 cell lines derived from hematological malignancies utilizing comprehensive DNA methylation microarrays that interrogates more than 850,000 and 285,000 CpG sites from human and mouse genomes, respectively [9, 10]. Importantly, we also provide a pharmacoepigenetic example of the potential use of this resource by showing how DNA methylation can predict response to nucleoside analogues.

In this regard, using the above-described platforms [9, 10], we analyzed the DNA methylation profile of 180 and 30 human and mouse hematological cell lines, respectively, as illustrated in Fig. 1A (Supplementary Methods). Overall, these 210 samples encompassed 80 lymphoma, 93 leukemia, 20 multiple myeloma and 17 non-malignant transformed cell types (Fig. 1A). The entire description of all the studied cell lines is shown in Dataset S1. The complete DNA methylation data are freely available at the GEO repository under accession number GSE270494. For both human and mouse, we found by using the whole DNA methylome that the hematological cell lines tend to cluster by disease (leukemia, lymphoma or multiple myeloma) as shown in the unsupervised hierarchical clustering (Fig. 1B) (Supplementary Methods). Further dimensionality reduction analysis by t-Distributed Stochastic Neighbor Embedding (t-SNE) (Supplementary Methods) yielded similar findings for each specie (Supplementary Fig. S1A). For the integration of human and mouse cells, we found that samples clustered by disease within species clusters (Supplementary Fig. S1A). Phylogenetic analysis (Supplementary Methods) of the different hematological subtypes and species according to the DNA methylation landscape provided additional evidence of the characteristic epigenetic blueprint of each category (Fig. 1C). The phylogenetic tree reinforced the differences between human and mouse DNA methylomes. Additionally, we found that in humans, hematological cell lines derived from B-cells shared the same node origin in the phylogenetic tree. We next performed a supervised differential methylation analysis (Supplementary Methods) between leukemia, lymphoma and multiple myeloma derived cell lines for human and mouse and we obtained a list of differentially methylated CpG sites (available as Dataset S2). We plotted the methylation β-values of these CpG sites in a heatmap where hematological cell lines were hierarchically clustered and nearly perfectly separated according to each hematological malignancy (Fig. 1D). These differentially methylated CpG sites between the three pathological entities were enriched in open sea regions and promoter regions. Pathway enrichment analyses (Supplementary Methods) unveiled an enrichment in pathways related to phosphorylation regulation and transcription processes (Supplementary Fig. S1B, C). The described clusters in Fig. 1B and D were also observed without data imputation (Supplementary Fig. S2). For both species, normal B and T-cells and myeloid cells clustered apart from their derived malignancies (Supplementary Fig. S3) and the corresponding differential CpG sites are available in Supplementary Dataset S3.

Fig. 1: DNA methylation landscape of human and mouse hematological cell lines.
figure 1

A Schematic representation of hierarchical subdivision of human (top) and mouse (bottom) cell line cohorts based on hematological malignancies. B Heatmap resulting from the unsupervised analysis of the human (left) and mouse (right) hematological cell lines. Dendogram clustering was performed using all methylation values, an 1% of randomly selected CpG probes were used for heatmap visualization. C Phylogenetic analysis of the different hematological subtypes and species. Each leaf represents the 1% most variable CpG sites within each hematologic subtype and specie. The inner circle colors represent if the leaf corresponds to leukemia, lymphoma, multiple myeloma or transformed cell lines, while the outer circle corresponds to the refined classification for human hematological cell lines. D Heatmap resulting from the differential methylation analysis within leukemia, lymphoma and multiple myeloma in human (left) and mouse (right) hematological cell lines. The bottom annotation indicates hematological disease, as described in the heatmap legend. Methylation β-values range from 0 (green) to 1 (red).

Full size image

Focusing on the human samples, we further examined the granularity of hematological cell lines by performing supervised differential methylation analyses and obtaining lists of CpG methylation sites characteristic of each specific disease in comparison to the others (shown in Supplementary Dataset S4). We used these CpG sites to perform a supervised hierarchical clustering that enabled an effective discrimination of acute myeloid leukemia (AML), chronic myeloid leukemia (CML), B-cell acute lymphoid leukemia (B-ALL), T-cell acute lymphoid leukemia (T-ALL), mantle cell lymphoma (MCL), Burkitt’s lymphoma (BL), diffuse large B-cell lymphoma (DLBCL), Hodgkin lymphoma (HL), T-cell lymphoma (TCL) and multiple myeloma (MM) in the heatmap representation (Fig. 2A). The genomic loci and CpG content of the DNA methylation sites that differentiate among the mentioned ten hematological malignancies are shown in Supplementary Fig. 4A. Additional dimensionality reduction analysis using t-SNE (Supplementary Fig. 4B) provided similar results than the supervised hierarchical clustering analysis. The pathway enrichment analyses that were significantly enriched in these ten pathological entities are shown in Supplementary Fig. 4C, mainly involved in T-cell receptor signaling, phosphorylation signaling regulation and transcription regulation. Most important, we translated these in vitro findings of CpG sites that are able to distinguish across different hematological malignancies to the primary setting (Supplementary Methods), using the same DNA methylation microarray platform. Using the above identified CpG sites (Supplementary Dataset S4), the primary cases of AML, B-ALL, T-ALL and DLBCL were clustered almost perfectly in the supervised hierarchical clustering analysis (Fig. 2B) (EGA repository: EGAS50000000627; https://ega-archive.org/studies/EGAS50000000627). Interestingly, unsupervised clustering analysis showed that cell lines and primary samples intermingled in many occasions, suggesting DNA methylation resemblance (Supplementary Fig. S5). Those CpG sites with distinct methylation content in vitro vs in vivo are available at Supplementary Dataset S5.

Fig. 2: The DNA methylome of human hematological cell lines can differentiate different hematological malignancies and is associated with drug sensitivity.
figure 2

A Heatmap resulting from the CpG sites differentially methylated within hematological diseases in human hematological cell lines. The bottom annotation indicates hematological disease, as described in the heatmap legend. Methylation β-values range from 0 (green) to 1 (red). B Heatmap resulting from the CpG sites differentially methylated within hematological diseases in human primary samples. The bottom annotation indicates hematological disease, as described in the heatmap legend. Methylation β-values range from 0 (green) to 1 (red). C Heatmap resulting from the common CpG sites associated with cytarabine, fludarabine and nelarabine drug sensitivity. Top annotation indicates whether the cells are in the sensitive cluster or resistant cluster. The bottom annotation indicates hematological disease and cytarabine, fludarabine and nelarabine sensitivity as described in the heatmap legend. Methylation β-values range from 0 (green) to 1 (red). D Boxplot representing cytarabine (top), fludarabine (middle) and nelarabine (bottom) IC50s of human hematological cell lines grouped according to whether they clustered in sensitive or resistant cluster. Two-sided Mann–Whitney–Wilcoxon test was performed.

Full size image

Finally, since expression profiles and IC50 values against hundreds of drugs are available for human hematological cell lines [3, 4, 6], we were able to perform an initial pharmacoepigenetics exploratory analysis. DNA methylation status of particular genes, such as the case of the DNA repair gene MGMT in gliomas, is currently used in precision cancer medicine [11]; and the epigenetic database for hematological malignancies, herein reported, could exhibit translational value. Using the experimental and bioinformatic pipeline described in Supplementary Methods for drugs that are used for hematological malignancies, we were able to identify CpG sites which methylation status and associated expression levels were linked to drug sensitivity (Supplementary Dataset S6). The example of enhanced response to the nucleoside analogs cytarabine, fludarabine and nelarabine [12] and to azacitidine according to the methylation levels sites is illustrated in Fig. 2C and Supplementary Fig. S6, respectively. The DNA methylation patterns of the CpG sites associated with the nucleoside analogs enabled a classification of hematological cell lines as sensitive or resistant through a hierarchical clustering analysis, as it is also shown in the IC50s boxplots in Fig. 2D. For azacitidine, a hypermethylated CpG site within the tumor necrosis factor-alpha-induced protein 3 gene (TNFAIP3) gene was associated with RNA downregulation and enhanced sensitivity (Supplementary Fig. S6). This CpG site was not present in the DNA methylation signature associated with sensitivity to nucleoside analogs, including cytarabine.

Overall, we have demonstrated how human and mouse cell lines derived from hematological malignancies exhibit unique DNA methylation profiles. These epigenetic fingerprints are so characteristic that allowed the obtention of a DNA methylation classifier for ten pathological categories. Most importantly, these data, now freely available to all researchers as a resource, can be further mined for various applications. This is illustrated by our preliminary results, which suggest a potential pharmacoepigenetic use that could pave the way for translation into clinical studies.