Genome-wide Analysis of Histone Lysine Methylation Variations Caused by Diabetic Conditions in Human Monocytes*

Aberrant histone lysine methylation patterns that change chromatin structure can promote dysregulated gene transcription and disease progression. Diabetic conditions such as high glucose (HG) are known to alter key pathologic pathways. However, their impact on cellular histone lysine methylation is unknown. We hypothesized that chronic HG can induce aberrant changes in histone H3 lysine 4 and lysine 9 dimethylation (H3K4me2 and H3K9me2) within target cells. Chromatin immunoprecipitation linked to microarrays (ChIP-on-chip) is currently a widely used approach for acquiring genome-wide information on histone modifications. We adopted this approach to profile and compare the variations in H3K4me2 and H3K9me2 in human gene coding and CpG island regions in THP-1 monocytes cultured in normal glucose and HG. Subsequently, we identified key relevant candidate genes displaying differential changes in H3K4me2 and H3K9me2 in HG versus normal glucose and also validated them with follow-up conventional ChIPs. Relevance to human diabetes was demonstrated by noting that H3K9me2 at the coding and promoter regions of two candidate genes was significantly greater in blood monocytes of diabetic patients relative to normal controls similar to the THP-1 data. In addition, regular mRNA profiling with cDNA arrays revealed correlations between mRNA and H3K9me2 levels. These novel results show histone methylation variations, for the first time, under diabetic conditions at a genome-wide level.

Aberrant histone lysine methylation patterns that change chromatin structure can promote dysregulated gene transcription and disease progression. Diabetic conditions such as high glucose (HG) are known to alter key pathologic pathways. However, their impact on cellular histone lysine methylation is unknown. We hypothesized that chronic HG can induce aberrant changes in histone H3 lysine 4 and lysine 9 dimethylation (H3K4me2 and H3K9me2) within target cells. Chromatin immunoprecipitation linked to microarrays (ChIP-on-chip) is currently a widely used approach for acquiring genome-wide information on histone modifications. We adopted this approach to profile and compare the variations in H3K4me2 and H3K9me2 in human gene coding and CpG island regions in THP-1 monocytes cultured in normal glucose and HG. Subsequently, we identified key relevant candidate genes displaying differential changes in H3K4me2 and H3K9me2 in HG versus normal glucose and also validated them with follow-up conventional ChIPs. Relevance to human diabetes was demonstrated by noting that H3K9me2 at the coding and promoter regions of two candidate genes was significantly greater in blood monocytes of diabetic patients relative to normal controls similar to the THP-1 data. In addition, regular mRNA profiling with cDNA arrays revealed correlations between mRNA and H3K9me2 levels. These novel results show histone methylation variations, for the first time, under diabetic conditions at a genome-wide level.
Histone modifications in chromatin, particularly histone lysine methylation, play key roles in gene expression and are emerging as a visible new layer of gene transcription regulation (1)(2)(3). Depending upon the timing and chromosomal location, histone methylation cannot only undergo dynamic changes during gene transcription and cell division, but also remain semi-stable, well maintained, and somatically inheritable. Along with DNA methylation, histone methylation can con-tribute to epigenetic heritable changes in gene function that do not involve a local change in DNA sequence. It has been shown that changes in histone methylation follow specific patterns and encode information during cell cycle changes and development (4). Therefore, aberrant alterations in histone lysine methylation patterns that change chromatin structure could lead to dysregulated gene transcription and disease progression (5). Elucidating the biological and functional relevance of these post-translational histone modifications is crucial to our understanding of the role of chromatin in gene expression. To date, compared with the extensive literature available on gene mutations, there is very little data linking histone modification variations to human disease, largely due to the lack of effective identification methods.
DNA microarray technology has made it possible to profile and quantify the expression of thousands of genes simultaneously (6). Although the major use of DNA microarrays has been for mRNA expression profiling, there are other applications (7). Various types of profiling of functional elements have been designed to acquire cellular genome information such as DNA copy number (8), mapping DNA-binding proteins (9, 10), mapping transcriptional networks (11), DNA methylation (12), and histone modifications (13,14). Currently, chromatin immunoprecipitation combined with DNA array analysis (ChIP-on-chip) 2 is the best approach for profiling and acquiring genome-wide information on histone modifications. Whereas several reports of histone methylation profiling in higher eukaryotes have been recently reported (15)(16)(17), applying this method for the reliable identification of the full scope of human epigenetic variations still poses an enormous challenge.
To examine whether histone methylation is altered by diabetic conditions, in this paper, we used the ChIP-on-chip approach to analyze histone lysine methylation variations caused by high glucose (HG) in THP-1 cells. Our understanding of the pathological role of chronic HG in diabetes and the mechanisms of glucotoxicity has increased substantially (18 -25). Hyperglycemia leads to several adverse effects in tissues and cells including pancreatic ␤-cells, endothelial cells, and monocytes, and contributes to several microvascular and macrovascular complications in both type 1 and type 2 diabetes (18) by inducing key factors including oxidant stress (21), advanced glycation end products (20,22), protein kinase C, and inflammatory gene activation (21,24). Production of mitochondrial superoxide has been implicated as a common mediator (21,25). Overall, chronic HG can program cells to undergo damage by affecting key cellular targets. Despite intensive studies, many of these target genes or in vivo nuclear mechanisms regulating them at the level of chromatin remain unclear. Most importantly, we know very little about how HG affects histone modifications of chromatin in target cells.
In the current study, we mimicked diabetic conditions by culturing THP-1 monocytes in HG (25 mM) relative to normal glucose (NG, 5.5 mM). We recently reported that HG treatment of human monocytes leads to dynamic changes in histone H3 lysine acetylation (26). However, it is not clear whether methylation undergoes such variations. Thus, we examined how HG affects the status of chromatin histone lysine methylation genome-wide by profiling histone H3 lysine 4 dimethylation (H3K4me2) and lysine 9 dimethylation (H3K9me2) in THP-1 monocytes cultured in NG and HG. Rationale for the choice of K4me2 and K9me2 is that these modifications are generally associated with gene activation and repression, respectively (2,3). Subsequently, we identified genes or chromatin regions in which histone methylation is susceptible to chronic HG. In addition, we tested the in vivo relevance of the THP-1 data by examining whether key target genes were similarly methylated at the coding or promoter regions in peripheral blood monocytes obtained from diabetic patients. Our new results could provide key links between these susceptible genes and the diabetic state, and also uncover a potential mechanistic basis for metabolic memory, diabetes, and its complications.
Microarrays-Human 12K cDNA arrays were from the University of Pennsylvania Functional Genomics Core. Human 12K CpG island arrays were from the Universal Health Network Microarray Center (Toronto, Canada). The CpG array contains 12192 CpG island clones originally from the Sanger Center, United Kingdom (27). The sequences of CpG islands on the array and alignment data are available on-line (28).
Cell Culture-Human THP-1 cells were obtained from ATCC and maintained in RPMI 1640 medium containing 5.5 mM glucose (NG) as described earlier (22,32). Treatments with HG (25 mM) or mannitol (19.5 mM) were for 72 h.
Isolation of Human Peripheral Blood Monocytes-50 ml of blood from adult volunteers with established type 1 (T1D, n ϭ 7) and type II diabetes (T2D, n ϭ 6) and from normal healthy donors (n ϭ 6) were collected in the presence of anticoagulant in accordance with an approved Institutional Review Board protocol (number 00123). The blood was diluted with equal volumes of phosphate-buffered saline. An equal volume of diluted blood was overlaid on Ficoll-Paque Plus in a 1:1 ratio and centrifuged at 1200 ϫ g for 20 min at 18 -20°C. The leukocyte population was collected from the interface and washed with phosphate-buffered saline several times to remove plasma and Ficoll. About 50 million washed cells in 10 ml of RPMI medium containing 10% fetal calf serum were plated in 100-mm culture dishes to allow monocytes to adhere on the surface of the dish for 2-3 h. The non-adherent cells (mainly lymphocyte population) were removed, washed with fresh medium, and cultured in RPMI medium. Attached monocytes were washed twice with warm RPMI medium containing 10% fetal calf serum and allowed to remain in the dish overnight at 37°C in 5% CO 2 . During this period the monocytes detach from the dish. They were collected in fresh RPMI medium and were used for the next ChIP experiment.
Microarray Data Collection and Statistical Analyses-After washing, the hybridized microarray slides were scanned using a GenePix 4000B scanner (Axon Instruments, Foster City, CA). Acquired microarray images were analyzed with GENEPIX version 6 software. Preprocessing of raw data and statistical analyses were performed using Bioconductor packages in the R programming environment (29). Spots marked as "bad" or "not found" by GENEPIX or with intensity less than the negative control spots were excluded. Background correction was performed using the "normexp" method implemented in the Bioconductor LIMMA package to adjust local median background estimates (30). Background corrected intensity data were normalized using the Print-tip group Lowess method to remove the bias within each array and the median absolute deviation normalization to remove the bias between arrays. H3K4 and H3K9 methylation profiles (enrichment factors) were determined by ratios between enriched and un-enriched samples for THP-1 cells growing in NG and HG. Ratios of the enrichment factors between NG and HG (difference factor) measures the methylation difference between these two conditions. To determine the K4 and K9 targets and those that are differentially methylated between NG and HG, methylation profiles were compared using a statistical linear model in LIMMA. For identification of K4 and K9 targets in both conditions, p values were adjusted for multiple testing using the method of Ben-jamini, Hochberg and co-workers (31) to control the false discovery rate at a level of 0.05, and the enrichment factor was set at Ͼ1.75. For identification of K4 and K9 targets that are differentially methylated between HG and NG conditions, p values were not adjusted for multiple testing due to the limited number of replicates, and difference factor was set at Ͼ1.5.

RESULTS
Profiling Histone Lysine Methylation of THP-1 Cells under NG and HG Conditions-Based on the principle of genomewide location analysis (9) and the use of cDNA microarrays in comparative genome hybridization analysis (8), we recently use a ChIP-cDNA array approach to profile and analyze histone lysine methylation patterns in the coding regions of human genes (17). Although the resolution is lower than high density oligonucleotide tiling arrays, the ChIP-cDNA array approach can identify histone variations at the gene level, is extremely sensitive and needs as little as 100 ng of antibody enriched or ChIP DNA. We now adapted this procedure to detect differences in histone K9 and K4 dimethylation between NG-(5.5 mM) and HG (25 mM)-treated THP-1 cells. The approach is illustrated in Fig. 1. The microarray experiments are summarized in supplemental data Table 3. All experiments were performed at least in duplicate. Within the vast human genome, the most interesting portions are in the gene coding and promoter regions. We used human 12K cDNA and 12K CpG island arrays to cover these regions. The 12K CpG array contains a significant percentage of the CpG islands found in the human genome with ϳ68% located near a transcription start site, although not fully representative of promoter regions (28). Together, the cDNA and CpG arrays used in this study cover a relatively small but significant and meaningful portion of the human genome.
Early studies show that H3K4me2 and H3K9me2 are present in partitioned chromosome regions that are structurally and functionally distinct in eukaryotes (32). H3K4me2 is closely associated with H3K4me3 (16,17,33,34), which mainly occurs concomitantly on active loci, especially in the first exon. H3K4me2 can be present on active as well as inactive genes. Its distribution can extend to promoter regions and gene coding regions (32,33). Histone H3K9me2 is also widespread in chromatin and could be either a repressive mark in euchromatin or a hallmark for heterochromatin (3,35). Increases or decreases in H3K9 methylation can change chromatin structure and affect gene expression (3). More recently, we found that, in the coding regions of human genes, H3K4me2 marks are associated with H3K4me3, H3K9Ac, H3K36me2, H3K79me2, but not with H3K27me2, H3K9me2, and H3K9me3, whereas H3K9me2 appeared not to be closely associated with any of the  marks we tested (17). Overall, compared with their trimethylated states, H3K4me2 and H3K9me2 marks cover broader regions of the genome. Based on these, we selected H3K4me2 and H3K9me2 marks for evaluation by ChIP-on-chip in this study.
To obtain a global overview of histone methylation profiles under HG and NG conditions, the ChIP-on-chip data were first corrected for background and normalized to remove system-atic bias. Methylation profiles were then determined by the ratios between normalized Cy5 and Cy3 intensities. We applied a criterion of 1.75-fold change under at least one condition (K4-HG, K4-NG, K9-HG, or K9-NG). A total of 809 probes (6.7%) on the cDNA array and 3632 probes (29.8%) on the CpG island array showed at least a 1.75-fold enrichment by histone antibodies under HG or NG conditions. Fig. 2 shows the hier- archical clustering diagram of H3K4me2 and H3K9me2 candidates under NG and HG conditions. We observed very little overlap between K4me2 and K9me2 candidates on this 12K cDNA array ( Fig. 2A), which is consistent with our previous human 1.7K cDNA array results (17), i.e. majority of the K4me2 candidate genes are not associated with K9me2 candidate genes and vice versa, but a very small fraction of genes can in fact be both K4me2 and K9me2 candidates. From Fig. 2B, similarly we noted that H3K4me2 and H3K9me2 profiles are also almost mutually exclusive on CpG islands. Another interesting observation is that the number of the H3K9me2 candidates is almost 10 times greater than H3K4me2 on CpG islands array, indicating that H3K9me2 is enriched within CpG islands. This set of results provides, for the first time, a snapshot view of H3K4me2 and K9me2 status at gene coding and promoter regions in monocytes separately under NG and HG conditions.
We applied statistical analysis using the LIMMA package in bioconductor tools to filter H3K4me2 and H3K9me2 targets under NG and HG conditions. With a false discovery rate of 5% and ratio Ͼ1.75, we identified H3K4me2 targets (299 and 275 for NG and HG, respectively) and H3K9me2 targets (228 and 286 for NG and HG, respectively) using cDNA arrays, and H3K4me2 targets (235 and 244 for NG and HG, respectively) and H3K9me2 targets (2570 and 2606 for NG and HG, respectively) using CpG arrays. There were only 12 common targets between H3K4me2 and H3K9me2 on the cDNA array, and 34 common targets on the CpG array. Detailed lists of candidate genes are provided in supplemental data Tables 4 -7 (cDNA array) and Tables 8 -11 (CpG array).

Differential Histone Lysine Methylation in THP-1 Cells under HG
Conditions-Next, we analyzed H3K4me2 and H3K9me2 candidate gene variations occurring under NG versus HG conditions. Our aim was to identify genes that underwent changes in their methylation status during HG treatment. This includes two scenarios: first, genes that are H3K4me2 or H3K9me2 candidates that are preferentially enriched by the respective antibodies (normalized ratio Ͼ1.75, FDR Ͻ5%) only under NG or HG; second, genes that are candidates under both conditions, but show significant differences (un-adjusted p value Ͻ0.05 and normalized ratio Ͼ1.5). Therefore, genes that show significant methylation differences between HG and NG, but are not identified as H3K4me2 or H3K9me2 candidates, were excluded in the target list. Whereas these stringent restrictions might have caused some false negatives (i.e. genes that underwent changes in methylation but not identified by our analysis), it reduced the possibility of introducing too many false positives.
Genes displaying significant differential methylation status were identified by applying the above criteria to the cDNA array data, and the results are illustrated in Table 1. Nine genes, including ICAM3 and FOS, displayed increased H3K4me2 in HG compared with NG. On the other hand, 26 genes, including GSTA-4, IL-8, and BCL-9, showed decreased H3K4me2. For H3K9me2, 39 genes displayed increased methylation under HG compared with NG and 11 genes showed decreased methylation (Table 1).
To validate the microarray results, selected genes that displayed HG-mediated decreases in H3K4me2 (IL-8, GSTA4, and BCL-9), and increases in H3K9me2 (PTEN, MLL3, and UBB) were then verified by conventional ChIP assays. We also included ChIP DNA from cells treated with mannitol (20 mM, control for osmolality) to confirm that changes seen are specific to HG. As shown in Fig. 3 (A and B), the conventional ChIP results of these chosen K4me2 and K9me2 candidates, respectively, are consistent with the ChIP array analyses. The differences in histone methylation noted by these conventional ChIPs were also clearly visible in the original DNA microarray images (Fig. 3, lower color panels). Taken together, these ChIP validations support the accuracy of the array data.
The same analysis procedure was also applied to the 12K CpG array. 34 probes showed differential H3K4me2, and 6 probes showed differential H3K9me2. The detailed list is available in Table 2. To identify the associated genes, each of these 40 probe sequences was then aligned to the UCSC human genome (28), and 34 of these probes could be mapped to the human genome. The annotation of these "altered " CpG island targets can be summarized as follows: 1) within promoter; 2) within gene exons or introns; 3) within the genome but not close to any genes; 4) blank or unblastable against the data base. Thus, although several CpG islands showed variations in his-tone K4me2 and K9me2, only a fraction of these mapped to their promoter or coding regions.
Among the few candidates identified in the CpG array, interestingly, JMJD2A, a newly discovered histone lysine demethylase belonging to the jumonjC domain containing family of pro-  MAY 4, 2007 • VOLUME 282 • NUMBER 18 teins (36, 37), showed a significant increase in histone H3K4me2, whereas IL-1A showed an increase in histone H3K9me2 at their respective promoter regions under HG conditions (Table 2). Follow-up conventional ChIPs further confirmed that H3K4me2 at the JMJD2A promoter and H3K9me2 at the IL-1A promoter are indeed increased under HG conditions (Fig. 3, C and D). Of interest is that JMJD2A is a K9me demethylase (33, 34) that can potentially augment gene expression by decreasing K9 methylation. Because K4me is usually a gene activation mark, our novel new observation that HG could increase K4 methylation of the K9 demethylase JMJD2A suggests a new mechanism by which HG could trigger the expression of genes previously silenced by K9me. On the other hand, IL-1A is a mediator of inflammation and also has regulatory functions in the immune and endocrine systems (38). To verify the relevance of our observed HG-mediated K9me at the IL-1A promoter, we examined the mRNA expression of IL-1A in THP-1 cell under NG versus HG conditions (Fig. 4). The basal expression level of IL-1A was very low in THP-1 cells, but was dramatically increased by the inflammatory stimulus, TNF-␣. We noted that, under HG conditions, the stimulatory effects of TNF-␣ were markedly reduced (by over 35%) relative to NG conditions (Fig. 4). Altered K9me2 at the promoter region of IL-1A might be responsible for this impaired response to TNF-␣ in THP-1 cells under HG conditions. This suggests an adaptive response to the effects of HG and implicates a finetuning role for histone H3K9me2 in IL-1A expression that further underscores the significance of cellular histone modification status in response to environmental stimuli.

Histone Lysine Methylation Variations in Diabetes
Parallel Methylation Changes in Peripheral Blood Monocytes Obtained from Patients with Diabetes-Circulating monocytes in diabetic individuals are exposed to hyperglycemic conditions and thus our studies with THP-1 cells can potentially be extrapolated to human blood cells to determine disease and clinical relevance. To evaluate the direct relevance to diabetes, we compared histone H3K9me2 levels in two chosen candidate genes from the THP-1 cell data. Thus we compared H3K9me2 in the PTEN coding and IL-1A promoter regions by ChIP assays in blood monocytes obtained from a group of diabetic patients relative to normal controls. ChIP DNA samples were prepared and analyzed as described earlier (26) from primary monocyte fractions isolated from the peripheral blood of patients with T1D, T2D, and normal healthy control volunteers with an approved IRB protocol. The ChIP DNA sample from the monocytes of each volunteer was separately prepared and analyzed using the conventional ChIP method. The ChIP PCR data in Fig. 5A and bar graph quantification in Fig. 5B below show that diabetic patients (both T1D and T2D) have statistically higher levels of histone H3K9me2 around the IL-1A promoter and PTEN coding regions relative to those in the normal control group. This is similar to the results seen with HG-stimulated THP-1 cell (Fig. 3). No bands were observed in "no antibody" controls (not shown).
Correlation between mRNA Expression Levels and H3K4me2 and H3K9me2-An advantage of using cDNA arrays for mapping histone methylation is that we can integrate histone methylation and mRNA expression profiles to explore possible associations. We therefore prepared RNA from NG and HG cells and performed traditional mRNA profiling (HG versus NG) by hybridizing RNAs to 12K cDNA arrays to compare changes in mRNA levels among genes displaying altered histone methylation. The correlation between HG induced variations of histone methylation and the ratio of mRNA expression under HG/NG conditions are also illustrated in Table 1. Among 38 genes that displayed a HG-induced increase in H3K9me2, the mRNA expression of 24 was down-regulated (63%), 9 remained unchanged (24%), and only 5 were up-regulated (13%). Among the 11 candidate genes depicting the HG-induced decrease in H3K9me2, the mRNA expression of 5 were upregulated, 4 remained unchanged, and 2 were down-regulated. On the other hand, the columns on the right show that the correlation of H3K4me2 with mRNA expression is much less clear. In conclusion, HG-induced variations in H3K9me2, but not H3K4me2, have an inverse correlation with gene expression.  In an effort to examine potential causes for observed exceptions to the rule repressive marks, we chose four candidates (PTEN, MLL3, UBB, and SBF1) for further analysis of their histone modification status including H3K9me3 and H3K14Ac (acetylation). As shown in Fig. 6, no signals were detected at the level of histone H3K9me3 in all four candidates. But they did show changes in H3K14Ac in response to HG. Notably, MLL3, whose gene expression is up-regulated in HG despite increased H3K9me2 (Table 2), showed increased H3K14Ac (which is usually associated with active genes). However, SBF1, whose gene expression was lower despite decreased H3K9me2, also showed increased H3K14 acetylation in response to HG. PTEN and UBB did not show significant changes in K3K14Ac. These results suggest that histone modifications can be correlated to the expression of many but not all genes, as also observed by others (3,39).
Finally, we performed cluster analysis of the histone K4me2 and K9me2 candidate lists (supplemental data Tables 4 and 5) from methylation profiling and the related expression data from mRNA profiling to examine the overall correlations between methylation and mRNA expression levels. Statistically, genes with higher expression are expected to have higher intensity on the mRNA profiling array. With this assumption, cluster analyses for H3K4me2 and H3K9me2 candidate genes with the mRNA intensities were obtained from combining two data sets (Fig. 7). As anticipated, a portion of H3K4me2 candidate genes show high intensity (expression, yellow) on the cDNA array. But surprisingly, some H3K9me2 candidate genes also showed high intensity (expression) on the cDNA array (Fig. 7). This demonstrates that not all histone H3K9me2 candidates are silenced in gene expression. These results are not fully anticipated because H3K4 and H3K9 methylation are generally linked to transcription activation and repression, respectively. However, our data supports the argument that histone methylation profiling provides data sets that reflect the methylation status of the chromatin, whereas mRNA profiling provides data sets of mRNA levels that reflect the outcome of transcription activities and other related mRNA metabolism events. This is further supported by several recent reports showing unexpected outcomes of K9 and K4 methylation (40,41).

DISCUSSION
Unlike traditional microarray profiling of gene expression, histone methylation profiling generates data sets that reflect chro-mosome histone methylation status under a specific condition, and provides a snapshot view of the cell at the layer of histone methylation. It is flexible and scalable, and variable parameters such as antibody specificity and the types of DNA microarrays, including tiling arrays, can considerably increase the power of this approach. Data sets from the profiling provide information about the distribution of histone methylations in the genome, the corre-FIGURE 6. ChIP analysis of histone modifications in selected candidate genes. Conventional ChIP assays were carried out on four selected candidate genes to compare changes in histone H3K9me2, H3K9me3, and H3K14Ac levels in their coding regions in response to 72 h HG treatments. N represents normal glucose (5 mM) and H represents high glucose (25 mM). One-tenth of lysate was used as Input control. lation between different methylation marks and also their variations under different states. Our strategy in this study was to detect dynamic alterations in histone lysine methylation by taking these snapshots separately under NG and HG conditions and then analyze and compare these individual snapshots. We also tried a direct comparison approach, i.e. profiling with NG and HG ChIP DNA on the same DNA array without a no-antibody control, and found that it has a much higher false positive rate. 3 We therefore adopted the indirect comparison approach.
The cDNA and CpG arrays that we used here cover about one-third of human genes. The use of these arrays may have limitations in that they may prevent detection of important histone methylation changes at certain areas, and their resolution and accuracy are lower than high density oligonucleotide tiling arrays. However, they still have key advantages in that they cover a significant portion of human genes at an affordable price. Furthermore, an important aspect is that the cDNA arrays are highly sensitive, needing as little as 0.1 g of ChIP DNA compared with 5 g needed for tiling arrays. Thus cDNA arrays would be particularly useful especially when cell number is a limiting factor.
By performing parallel mRNA profiling on cDNA arrays, we uncovered both anticipated and unanticipated correlations between H3K4me2, H3K9me2, and gene expression. One interesting finding was that not all histone H3K9me2 candidates were silenced in gene expression. It is possible that these H3K9me2 candidate genes are actually highly transcribed but inhibited by a feedback mechanism through H3K9 methylation. Thus, mRNAs of some H3K9me2 candidates are still present in the cell that is in line with our observation that changes in H3K9me2 correlated with changes in gene expression ( Table 1).
The biological functions of histone methylation at gene promoter and coding regions are still not fully understood. Although early studies suggest that transcriptionally active genes can be methylated at H3K4, H3K36, and H3K79, whereas silenced or repressed genes are marked by methylation at H3K9, H3K27, and H4K20 (3), deviations from this rule are being reported and also noted in our current studies. Notably, a recent report showed that H3-K9 trimethylation was present in the coding regions of active transcribed genes (40). On the other hand, although K4me2 is associated with active genes, it is not clear whether increased K4 methylation will boost or hinder the next round transcription. Our results show the absence of such a correlation (Table 1). Increased K4me2 could also serve as a feedback counterregulatory response to slow down on-going transcription. Consistent with this, two recent studies link H3K4me3 to gene repression where the H3K4me3 mark is read by ING2 (41,42), a component of repressor complexes. Collectively, these imply that a single chromosomal mark has versatile functions and the "histone code " (1) is more complicated than we anticipated. Furthermore, each nucleosome has two copies of H2A, H2B, H3, and H4. Currently, we do not know whether modifications occur in one or two copies of histone proteins, and whether this is functionally important.
Despite the knowledge that HG can affect multiple cellular pathways and genes, several unsolved issues remain, including the impact of the long term memory of chronic HG on cellular chromatin. Importantly, the effects of HG on histone methylation have not previously been assessed in a genomewide scale. Here, for the first time, we systematically evaluated the effects of HG on H3K4me2 and K9me2 in THP-1 cells and gained new insights into the links between key genes and diabetes in the context of histone methylation. Noticeably in this study, HG could alter the methylation of some diabetes-related genes such as IL-1A and IL-8 that are associated with inflammation and also stimulated by chronic HG in monocytes (24). Furthermore, we uncovered HG-induced K4 methylation at histone K9 demethylase JMJD2A (36,37), which could indicate a novel mechanism by which HG de-represses silenced genes. HG also altered the methylation of genes such as PTEN, ABCB6, IGF1R, GSTA4, BCL-9, FOS, ICAM3, NOTCH1, TCF3, MAPK10, ESR1, and ING5 ( Table 1) that could be relevant to diabetes because these genes are associated with signal transduction, transporter, inflammation, and oxidant stress pathways. Importantly, in parallel, we noted a significant increase in K9me2 at the PTEN coding and IL-1A promoter regions in monocytes obtained from T1D and T2D patients, thus providing direct in vivo relevance, at least with these genes.
Interestingly, in unpublished studies, we used Ingenuity Pathway Analysis, a web-based application and noted potential diabetes and HG-related biological networks among our methylated genes listed in Table 1. This placed our candidate genes in canonical pathways, for e.g. c-FOS, IL-8, and insulin-growth factor receptor related to IL-6 and MAPK signaling, PTEN and insulin-growth factor receptor related to phosphatidylinositol 3-kinase signaling, TCF related to Wnt/␤-catenin signaling, and USP33, PSMA2 in the ubiquitination pathway. Future studies will explore such interactive pathways as well as the functional and biological relevance of methylated targets.
The connection between HG, diabetes, and histone methylation variations is clearly worthy of investigation because it could provide clues to the unexplained hyperglycemic or metabolic memory phenomenon observed in the Epidemiology of Diabetes Interventions Complications trial (43). Changes in H3K4me2 and H3K9me2 marks may act as this "memory" of transcription history initiated by chronic HG and sustained thereafter. Several key questions, however, remain unanswered, including whether these changes in histone methylation are reversible, and what are the "thresholds " in time or the degree hyperglycemia needed to mediate these changes. These issues are further complicated by the fact that glycemia fluctuates in diabetic subjects.