Page 100 - 2022_01-Haematologica-web

T. Milan et al.
for sustained proliferation. Given the similarity of their DNA binding consensus sequences, defining the precise contribution of different CEBP family members in this process will require further focused studies.
Sequential changes in DNA methylation are not global- ly correlated with expression changes
Given the known role of KM3 as an epigenetic regula- tor and the demonstrated importance of DNA methyla- tion changes in other subtypes of AML,38,39 we next sought to investigate changes in DNA methylation in our model leukemias and genetically matched patients’ sam- ples. Using methyl-sequencing, capture-based, bisulfite sequencing,40 we assessed DNA methylation levels genome-wide at 84 Mbp of target sequence regions including CpG islands, promoters and published differen- tially methylated regions. Biological duplicates at each step of our model leukemia system were compared to bisulfite-treated DNA from three adult KM3-AML patients. In line with expression data from our model leukemias,15 the methylation data between replicates were highly consistent (Online Supplementary Figure S3) and were also consistent with published data regarding global methylation levels (Online Supplementary Figure S4).41
Hierarchal clustering of DMC (CpG coverage ≥8, q- value ≤0.1 and methylation differences ≥40%) demon- strated that relatively small numbers of changes occurred after introduction of the KM3 fusion gene and in vitro cul- ture (Figure 2A). The number of genes with DMC present varied from 7,645 (CD34+ vs. CD34+KM3) to 13,947 (CD34+ vs. model AML); however a large percentage of these genes (40% and 20%, respectively) contained only a single DMC. Interestingly, relative to the initial CD34+ cells with the KM3 fusion, there was a striking increase (>6 fold) in the number of DMC detected in the final model leukemias and patients’ samples. This suggests that the initial epigenetic changes induced by the fusion gene are relatively limited, with the majority of changes occurring only during in vivo growth. Along with the pre- vious stepwise changes in gene expression patterns we have reported,15 these results again support the impor- tance of microenvironmental cues received in vivo. With respect to the direction of change, in agreement with recent observations in patients with KMT2A-AML,41 DMC observed in both model and patient KM3-AML were profoundly biased (~89%) towards demethylation (Figure 2B). In opposition to this trend, model leukemias generated from the same donor cells but which devel- oped a lymphoid phenotype (B-cell ALL) had primarily hyper-methylated cytosines, and fewer DMC compared to the AML samples. The majority (>75%) of DMC were located in gene bodies and particularly in promoter regions and putative functional elements at various stages of the model AML and in patients (Online Supplementary Figures S5 and S6).
To clarify the importance of the localization of these changes and their potential impact, we next examined the correlation between DNA methylation changes and gene expression. Despite the general view regarding the impact of DNA methylation changes in the promoter regions and transcription levels, we observed only a weak global correlation between the two (Figure 2C). While we could find examples of expected changes (e.g., in previ- ously described KMT2A target genes such as GATA2,
HOXA7 and HOXA942,43) (Figure 2D), overall, there was a very poor global correlation. This observation mirrors those made in recent comprehensive single-cell studies of DNA methylation changes during normal hematopoiesis in humans44 as well as previous bulk studies.45 We were able to identify new examples of genes with evidence for epigenetic regulation such as the atypical adenylate cyclase gene ADCY9. The promoter region of ADCY9 exhibits hypomethylation specifically in the model AML compared to normal CD34+ cells and model B-cell ALL (Figure 2E) and the gene is only expressed in the KMT2A- rearranged model and patients’ leukemias (Figure 2F). ADCY9 is a poorly characterized and distantly related member of the adenylate cyclase gene family and no spe- cific inhibitors have yet been described, although it was recently identified as a fusion partner of the KMT2D gene in a case of ALL.46 We therefore performed shRNA knock- downs in several KM3-AML cell lines to determine whether ADCY9 expression was relevant to these leukemias. We found that all the KM3-AML lines showed impaired proliferation with multiple shRNA in contrast to no effect in a non-KM3-AML control line (Figure 2G), demonstrating that ADCY9 is required in these cells.
Beyond regulating specific genes, we next wanted to examine whether global methylation changes might have an impact on gene expression through their localization within the DNA binding sites of transcription factors. To do this, we selected all the DMC with coverage that exhibited consistently higher (≥40%) methylation levels in CD34+ cells than in either CD34+KM3 or model leukemias. Transcription factors in the ENCODE ChIP- sequencing dataset (covering 338 factors in 130 cell lines) that overlapped the positions of these cytosines were then used to generate a network (using GOnet47) based on shared gene ontology terms (Online Supplementary Figure S7). This network contains a large collection of factors that: (i) are highly expressed in normal bone marrow, (ii) exhibit a large loss of DNA methylation directly in their binding sites after KM3 addition, and (iii) are enriched for activity in regulating hematopoietic and myeloid differ- entiation. The results indicate that despite the poor over- all correlation between DNA methylation and the expres- sion of individual genes, the changes that occur are glob- ally relevant for leukemic transformation.
Epigenetic analysis reveals potential CCR1-CCL23 autocrine signaling in KM3-acute myeloid leukemia
Given that the KM3 fusion is a well-known epigenetic regulator, we next asked whether the presence of the gene fusion had a specific impact on the chromatin organ- ization in our model leukemias. To investigate this, we performed ChIP-sequencing on relevant histone marks (H3K4me3, H3K79me211) and ATAC-sequencing on sam- ples of each step of our leukemia model (CD34+, CD34+KM3, KM3-AML). Enrichment of the different histone marks correlated with gene expression levels (Online Supplementary Figure S7) and expected spatial dis- tribution, and the majority of peaks seen in the model leukemias were common to both the initial and trans- duced cells (Figure 3A). In addition, we analyzed the com- bined expression and epigenetic data to compare the ini- tial CD34+ cells and CD34+KM3 cells to highlight critical early events potentially implicated in the disease. Using the EaSeq package20 we identified a set of genes (n=331) (Online Supplementary Table S2) with coordinated epige-
92
haematologica | 2022; 107(1)