Sex-specific DNA methylation and associations with in utero tobacco smoke exposure at nuclear-encoded mitochondrial genes

ABSTRACT

Sex-linked differences in mitochondrial ATP production, enzyme activities, and reactive oxygen species generation have been reported in multiple tissue and cell types. While the effects of reproductive hormones underlie many of these differences, regulation of sexually dimorphic mitochondrial function has not been fully characterized. We hypothesized that sex-specific DNA methylation contributes to sex-specific expression of nuclear genes that influence mitochondrial function. Herein, we analysed DNA methylation data specifically focused on nuclear-encoded mitochondrial genes in 191 males and 190 females. We found 596 differentially methylated sites (DMSs) (FDR p < 0.05), corresponding to 324 genes, with at least a 1% difference in methylation between sexes. To investigate the potential functional significance, we utilized gene expression microarray data. Of the 324 genes containing DMSs, 17 showed differences in gene expression by sex. Particularly striking was that ATP5G2, encoding subunit C of ATP synthase, contains seven DMSs and exhibits a sex difference in expression (p = 0.04). Finally, we also found that alterations in DNA methylation associated with in utero tobacco smoke exposure were sex-specific in these nuclear-encoded mitochondrial genes. Interestingly, the level of sex differences in DNA methylation at nuclear-encoded mitochondrial genes and the level of methylation changes associated with smoke exposure were less prominent than that of other genes. This suggests more conservative regulation of DNA methylation at these nuclear-encoded mitochondrial genes as compared to others. Overall, our findings suggest that sex-specific DNA methylation may help establish sex differences in expression and function and that sex-specific alterations in DNA methylation in response to exposures could contribute to sex-variable toxicological responses.

Introduction

Epigenetic modifications are mitotically heritable and reversible changes that do not involve a change in DNA sequence but may result in altered gene expression. The epigenome, including histone modifications and DNA methylation, regulates gene expression in virtually all cell types and tissues. DNA methylation most often occurs at cytosine-guanine dinucleotides (CpG sites), where it can modulate gene expression. Generally, the presence of DNA methylation at CpG sites near promoters can repress gene expression by blocking transcription factors that would initiate gene expression upon binding to the DNA [1,2]. In contrast, DNA methylation within a gene body can positively influence transcription activity in some cases [3] and also affect splice site choice [4,5]. Sex differences in DNA methylation at birth have previously been described. It has been shown that CpG sites with the highest statistically significant sex differences correspond to genes involved in neurodevelopmental disorders [6–8]. Although there is evidence for sex differences in mitochondrial function, there are no studies to date investigating the role of sex differences in DNA methylation that might regulate mitochondrial function and metabolism.

Mitochondria are complex organelles that play important roles in energy metabolism, cell signalling, apoptosis, and ion homoeostasis. Mitochondria are organized and function quite differently in various cell types. Sex differences in mitochondrial function and metabolism have recently become a topic of interest within the field and have been identified across multiple species and tissue types [9–18]. In general, the pattern emerging is that female mitochondria are more efficient, have higher ATP production and elevated enzyme activities, and produce fewer reactive oxygen species (ROS) [19]. These subtle sex differences in mitochondrial function have been hypothesized to drive altered susceptibility to health and disease [19–21]. Evidence for this connection exists in the sex-specific prevalence of multiple types of diseases that have mitochondrial components [19,20,22]. We hypothesize that sex differences in mitochondrial capacity and function will also drive differences in susceptibility and health outcomes due to exposures to environmental toxicants, many of which target mitochondria [23–27].

Could sex differences in mitochondrial function result from sex differences in the epigenetic regulation of the nuclear genome? The mitochondrial genome encodes 13 proteins, all of which are essential subunits of the electron transport chain (ETC), along with two ribosomal RNAs, and 22 transfer RNAs. However, whether DNA methylation occurs on the mitochondrial genome, and if so, what functional role it plays, is debated [28–31]. Regardless of the presence of CpG methylation in the mitochondrial genome, DNA methylation has the potential to impact the expression of nuclear genes with mitochondrial function [31]. In fact, over 99% of the proteins present in mitochondria are nuclear-encoded. Identification of genes in the nuclear genome that localize to the mitochondria is an ongoing effort, currently catalogued in the database Mitocarta 2.0 [32], containing 1,158 genes. We hypothesized that some of the nuclear-encoded genes that drive mitochondrial function are regulated by DNA methylation that could, in turn, drive functional sex differences in the ‘differentiated’ mitochondria within different cell types.

In utero tobacco smoke exposure alters epigenetic profiles in newborns [33–39], with large (up to 15%) differences in methylation at CpG sites corresponding to genes including AHRR, CYP1A1, and GFI1 [33] that clearly relate to the mechanisms of action through which tobacco smoke disrupts biological function. In utero, exposure to tobacco smoke has also been associated with alterations in mtDNA content in both the placenta and cord blood, as well as negative outcomes for overall mitochondrial function within the placenta [40,41]. Recently, one group identified sex-specific associations in DNA methylation with pack-year in the lung tissue of COPD patients, and implicated mitophagy as one targeted pathway [42]. However, the mechanisms through which these outcomes occur have yet to be elucidated, and the possibility that tobacco smoke could alter mitochondrial function through the disruption of nuclear DNA methylation has not been examined.

In this study, we utilized Illumina’s Infinium HumanMethylation450 BeadChip (450K) DNA methylation data from the Newborn Epigenetics STudy (NEST) cohort to identify sex differences in nuclear genes obtained from Mitocarta 2.0 related to mitochondrial function. The observed sex differences in methylation of a selected set of CpG sites were validated using independent samples. Additionally, the effects of in utero exposure to tobacco smoke on DNA methylation at nuclear-encoded mitochondrial genes were analysed with respect to sex.

Material and methods

Study population

Participants in the Newborn Epigenetics STudy (NEST) were recruited in Durham, North Carolina at prenatal clinics between 2005 and 2011. The NEST cohort was established to identify early life exposures that occur in utero that alter epigenetic profiles of offspring and affect later health consequences. A detailed description of identification and enrolment procedures for this cohort has been described elsewhere [39,43]. Briefly, women were eligible if they were aged 18 years or older, pregnant, either English or Spanish speaking, and planning to use Duke or Durham Regional Hospital for the delivery. Because in utero smoke exposure is associated with negative health outcomes, women who smoked were intentionally over selected. For the purpose of this study, we used any self-reporting of cigarette use during pregnancy for our ‘tobacco smoke exposed’ category. Non-smokers were included only if they self-reported never having smoked cigarettes. The analytical sample for the 450K BeadChip analyses (n = 381) was restricted to participants with reported race/ethnicity and who had complete questionnaire and medical records data. The analytical sample used for validation via bisulphite pyrosequencing (n = 76) was restricted to those who were not included in the 450K BeadChip analyses and had available high-quality cord blood DNA as well as PAXgene cord blood RNA tubes for complementary gene expression analysis. Demographics of the individuals selected for the Illumina 450K analysis can be found in Table 1 and demographics of individuals selected for the validation cohort can be found in Table 2.

Table 1.

Demographics of individuals in NEST 450K beadchip cohort used in this study.

Variable	N (%)	450K BeadChip Cohort
Overall	381 (100%)
Sex Male Female	191 (50%) 190 (50%)	Female 190 (100%)	Male 191 (100%)
Race White Black	179 (47%) 202 (53%)	91 (48%) 99 (52%)	88 (46%) 103 (54%)	n.s.
Maternal Education High School or Less College and Beyond	129 (34%) 252 (66%)	68 (34%) 122 (64%)	61 (32%) 130 (68%)	n.s.
Maternal Smoking During Pregnancy Yes No	131 (34%) 250 (66%)	67 (35%) 123 (65%)	63 (34%) 127 (66%)	n.s.
Maternal Diabetes Status Normal Diabetes (gestational or Type II)	307 (88%) 41 (12%)	156 (88%) 21 (12%)	151 (88%) 20 (12%)	n.s.
Maternal Age Under 20 20–30 30-40 40–50	28 (7%) 182 (48%) 155 (41%) 16 (4%)	15 (8%) 91 (47%) 78 (41%) 6 (3%)	13 (6%) 91 (48%) 77 (40%) 10 (5%)	n.s.
Maternal BMI Underweight/Normal Overweight	176 (50%) 175 (40%)	97 (54%) 82 (46%)	79 (46%) 93 (54%)	n.s.
Gestational Age (Weeks) Preterm (<37) Normal (>37)	80 (21%) 301 (79%)	39 (21%) 151 (79%)	41 (21%) 150 (79%)	n.s.
Birth Weight (Grams) Low (less than 2500) Normal (2500–4000) High (over 4000)	66 (17%) 284 (75%) 29 (8%)	32 (17%) 148 (78%) 10 (5%)	34 (18%) 136 (72%) 19 (10%)	n.s.

Note: Comparisons were done between male and female participants. Chi-square was performed when multiple groups were present and Fisher’s Exact test was performed when only two groups were present.

Table 2.

Demographics of individuals in the NEST validation cohort used in this study.

Variable	N (%)	Validation Cohort
Overall	76 (100%)
Sex Male Female	39 (51%) 37 (49%)	Female 37 (100%)	Male 39 (100%)
Race White Hispanic Other	49 (64%) 18 (24%) 9 (12%)	26 (70%) 10 (27%) 1 (3%)	23 (59%) 8 (20.5%) 8 (20.5%)
Maternal Education High School or Less College and Beyond	33 (43%) 43 (57%)	16 (43%) 21 (57%)	17 (44%) 22 (58%)
Maternal Smoking During Pregnancy Yes No	39 (51%) 37 (49%)	20 (54%) 17 (46%)	19 (49%) 20 (51%)
Maternal Diabetes Status Normal Diabetes (gestational, Type I or Type II) No Data	62 (81.5%) 9 (12%) 5 (6.5%)	30 (81%) 5 (13.5%) 2 (5.5%)	32 (82%) 4 (10%) 3 (8%)
Maternal Age Under 20 20–30 30-40 40–50	6 (8%) 36 (47%) 32 (42%) 2 (3%)	0 (0%) 20 (54%) 16 (43%) 1 (3%)	6 (15%) 16 (41%) 16 (41%) 1 (3%)
Maternal BMI Normal Overweight No Data	32 (42%) 40 (53%) 4 (5%)	11 (30%) 24 (65%) 2 (5%)	21 (54%) 16 (41%) 2 (5%)
Gestational Age (Weeks) Preterm (<37) Normal (>37)	9 (12%) 67 (88%)	3 (8%) 34 (92%)	6 (15%) 33 (85%)
Birth Weight (Grams) Low (less than 2500) Normal (2500–4000) High (over 4000) No Data	6 (8%) 56 (74%) 12 (16%) 2 (2%)	3 (8%) 31 (84%) 3 (8%) 0 (0%)	3 (8%) 25 (64%) 9 (23%) 2 (5%)

Ethical approval

The NEST Cohort is approved by the institutional review board at Duke University (Pro00014548) and this research was performed in accordance with the 1964 Helsinki Declaration. Written informed consent was provided by all participants in this study.

Sample collection and Illumina Humanmethylation450 beadchip platform

Immediately following newborn delivery, cord blood was collected and stored in 10 mL ethylenediaminetetratacetic acid (EDTA) vacutainer tubes, inverted, and centrifuged. Genomic DNA from the leukocyte-containing buffy coat was extracted using Gentra Puregene Reagents (QIAGEN) according to manufacturer’s protocol and stored at −80°C until analysis. The concentration and purity of the resulting DNA was assessed using a Nanodrop 2000 Spectrophotometer (Thermo Scientific). Sodium bisulphite treatment was performed on 800 ng of the genomic DNA using the Zymo EZ DNA Methylation Kit (Zymo Research). Methylation levels for individual CpG sites were then measured using the Illumina Infinium HumanMethylation450 BeadChip at the Duke Molecular Genomics Core Facility. Processing of this data has been described elsewhere [39]. Briefly, probe intensity data (IDAT) files were processed using the RnBead (version 1.8.0) R package, the data were then normalized using wm.dasen and background correction was performed using methylumi.noob [44]. This resulted in a matrix of β values for each CpG site, which ranges from 0 (completely unmethylated) to 1 (completely methylated). Quality of each array was assessed by examining density plots of the values, bean plots of all β values, and β values for control beads. Probes containing SNPs at any position and low variability probes (< 0.05%) were removed based on best practices recommendations [44,45].

Mitocarta 2.0 CpG list

To curate the list of CpG sites present in the Mitocarta 2.0 gene database, gene names and aliases present in Mitocarta 2.0 were cross-referenced with the Illumina 450K manifest; only CpG sites that were annotated to any of the Mitocarta 2.0 gene names in the report were used. CpG sites that were annotated to genes included those in the 3'UTR, body, 5'UTR, 1^st exon, and 1500 bp upstream of the transcriptional start site. This resulted in a total of 19,916 CpG sites on the 450K platform associated with Mitocarta 2.0 genes.

Sex differences in DNA methylation at mitochondrial genes

Given the known sex differences in DNA methylation associated with the X-chromosome (Figure S1), CpG sites associated with sex chromosomes and probes that were cross reactive with sex chromosomes [46] were excluded, resulting in 18,709 CpG sites. Linear regression models for sex association with methylation at these 18,709 sites were adjusted for maternal race, smoke exposure status, maternal age, maternal education level, gestational age, Houseman-estimated cell proportions [47] (CD8T, CD4T, NK, B Cells, monocytes, granulocytes, and nucleated red blood cells), as well as the technical variables: plate, row, and column. Maternal education was adjusted for in the analysis as a proxy for social metrics [48]. We included race as a covariable as previous reports have identified differences in DNA methylation associated with race [49]. While race is not a biological variable, it is used as a proxy for social stress, which we do not have data for in this cohort. Moving forward, scientists should strive to collect data on stress levels and other social factors rather than relying on race as a proxy. The differentially methylated sites (DMSs) were then subjected to False Discovery Rate (FDR) corrections. After DMSs were identified, they were filtered to include only those with a > 1% difference in mean methylation between females and males. Genes that contained DMSs were entered into Ingenuity Pathway Analysis (IPA, QIAGEN) to determine enrichment for DMSs in pathways and biological processes that may have sex-specific DNA methylation patterns.

Microarray gene expression analysis

Genes containing DMSs from the methylation analysis were compared to an Affymetrix Gene Chip HG-U133A microarray dataset previously generated by our lab. This dataset included gene expression microarray data from the Affymetrix U133HTA platform for the cord blood total RNAs of 52 infants (26 females and 26 males) for a pilot study for the NEST Cohort [50]. The mothers were matched at recruitment for smoking status, age, county of residence and race. Umbilical cord total RNA was purified from PAXgene tubes and RNA quality and integrity was assessed using an Agilent Bioanalyzer prior to generation of the array data (average RNA Integrity Number = 8.3; SD = 1.4). The microarray data was filtered to specifically examine mitochondrial genes (1308 probes representing 865 of the Mitocarta 2.0 genes). All probes with a gene annotation to a Mitocarta 2.0 gene were included. Linear regression analysis was performed to identify probes with sex differences in expression. Genes with probes indicating Differentially Expressed Genes (DEGs) between the sexes (p < 0.05) were cross-referenced to genes with DMSs in the methylation dataset that passed FDR correction and met the threshold for a ≥ 1% difference in methylation. These overlapping genes were then analysed using STRING to probe for interactions between proteins of interest [51].

Bisulphite pyrosequencing

Pyrosequencing assay design was performed using the PyroMark CpG Assay Design Software (QIAGEN) and pyrosequencing was carried out on a Pyromark Q96 MD Pyrosequencing Instrument (QIAGEN). Bisulphite treatment of 800 ng of gDNA, PCR amplification of the bisulphite modified DNA using 20 ng per reaction assuming complete recovery, and pyrosequencing were performed as previously described [52]. Primers and PCR conditions can be found in Table S1. Pyrosequencing assay performance was tested in triplicate using Epitect human fully methylated and fully unmethylated standards (QIAGEN) to create mixtures containing 0%, 25%, 50%, 75%, and 100% methylated DNA. Results for each target are presented in Figure S2. Once assay performance was confirmed, NEST samples from the validation cohort were analysed. Statistical significance was determined using a two-tailed Student’s t-test for each CpG site.

Quantitative real time reverse transcription PCR

Quantitative real-time RT-PCR was performed for ATP5G2 expression in the validation cohort samples to examine the methylation-expression relationships. RNA was extracted from PAXgene tubes using PAXgene Blood miRNA Kit (QIAGEN) and stored at −80°C until analysis. qPCR was performed using the QuantStudio 6 Flex Real-Time PCR System (Thermo Fisher Scientific). ATP5G2 (Taqman probe ID: Hs01086654_g1), and the GAPDH endogenous control (Taqman probe ID: Hs02758991_g1) probes were multiplexed in a 1:1 ratio for each sample. PCR mastermix contained 200 ng RNA, 1 μL of each probe, 10 μl of Quantabio qScript one-step RT-qPCR ToughMix (QuantaBio) and was brought to a total reaction volume of 20 μL with nuclease-free water. Samples were run in duplicate and PCR cycling conditions were as follows: 50°C for 10 minutes, followed by 95°C for 1 minute, and finally followed by 40 cycles of 95°C for 10 seconds and 60°C for 1 minute. To examine the relationship between methylation and expression of ATP5G2, delta CT was calculated, and relative expression was determined using levels of GAPDH to normalize for RNA input. Pearson correlation was used to identify the relationship between the average level of DNA methylation at the four CpG sites measured by pyrosequencing and the relative expression for each sample, which are the reported R and P values. This analysis was done independently for males (n = 31) and females (n = 31).

Sex-specific responses to tobacco smoke exposure in DNA methylation at mitochondrial genes

The impact of smoke exposure on DNA methylation was evaluated by first subdividing the total group of 381 patients by sex. This resulted in a male-only group of n = 191 and a female-only group of n = 190. For each of these groups, regression models for smoke exposure association with methylation at all 19,916 CpG sites were adjusted for maternal race, maternal age, maternal education level, gestational age, Houseman-estimated cell proportions (CD8T, CD4T, NK, B Cells, monocytes, granulocytes, and nucleated red blood cells), as well as the technical variables: plate, row, and column. DMSs associated with smoke exposure were then subjected to False Discovery Rate (FDR) corrections. To compare the overlap between the differentially methylated CpG sites and genes containing differentially methylated CpG sites between sexes, a Fisher’s exact test was used to determine significance and odds ratios. P values <0.05 were considered significant and odds ratios > 1 suggests the association between lists is strong. Genes that contained DMSs were entered into Ingenuity Pathway Analysis (IPA, QIAGEN) to determine enrichment for DMSs in pathways and biological processes that may be disrupted by smoke exposure in each sex.

Random subset of genes

To determine if the number of DMSs associated with in utero smoke exposure or the level of sex differences in DNA methylation present in the Mitocarta 2.0 gene list was similar to that of the rest of the genome, we compared the results to those of three randomly selected samples of 19,916 CpG sites. To obtain the random CpG sites, Mitocarta 2.0 associated CpGs were excluded and then sampling without replacement was performed until the same number of CpG sites was achieved. We ran the same statistical analyses used on the Mitocarta 2.0 CpGs and compared the number of DMSs for each of these datasets using chi-square analysis.

Results

Sex differences

We found sex differences in DNA methylation at 596 CpG sites associated with 324 genes, defined as meeting the criteria of statistical significance, including an FDR p < 0.05 and >1% difference in mean methylation. When stratified by direction of association, 345 of these CpG sites, corresponding to 185 genes, had higher methylation in females; 251 of these CpG sites, corresponding to 157 genes, had higher methylation in males (Figure 1). The top 20 CpG sites, based on statistical significance, exhibited mean differences in methylation between 1% and 23% (Table 3).

Figure 1.

Sex differences in DNA methylation at mitochondrial genes in the nuclear genome. A) Manhattan plot for association between sex and DNA methylation at 18,709 CpG sites annotated to genes listed in the Mitocarta 2.0 database. In total, there were 1,324 CpGs that met FDR threshold (p < 0.05), indicated by the red line. B) Volcano plot of the differential methylation analysis with X and Y axes displaying, respectively, the difference in methylation and the -log10 FDR p-values for each CpG site. CpGs more methylated in females and males are on the left and right regions of the plot, respectively. Setting a threshold of a 1% difference in mean methylation, there were 596 FDR significant (p < 0.05) sites that correspond to 324 genes.

Table 3.

Top 20 CpG sites with sex differences in DNA methylation.

Gene Name	CpG Site	P-value (unadjusted)	FDR P-value	Mean Methylation Difference (%, Male-Female)
ATP5J	cg17612569	9.14E-111	2.74E-105	23.48
CCT7	cg06642617	3.41E-48	2.32E-44	5.25
YARS2	cg01225095	2.97E-40	7.80E-37	−3.74
FASTKD2	cg11065518	2.92E-35	3.47E-32	−1.56
GLUD1	cg22227586	2.77E-34	2.76E-31	−5.41
TOMM40	cg12266551	5.80E-31	3.45E-28	−3.34
CLYBL	cg27053299	1.29E-30	7.21E-28	−4.22
CLYBL	cg05833851	1.42E-28	5.83E-26	−3.45
NDUFA13	cg25274157	5.64E-27	1.78E-24	−4.04
LACTB2	cg19632760	3.29E-22	4.82E-20	−3.55
BOLA1	cg24364827	6.03E-22	8.44E-20	−3.65
NDUFA13	cg03233793	3.97E-21	4.83E-19	−3.87
MACROD1	cg01882150	2.40E-20	2.60E-18	−1.57
C15orf40	cg24717799	3.40E-19	3.07E-17	−4.39
PTGES2	cg22863798	8.27E-19	7.02E-17	1.55
NDUFA13	cg07624705	9.15E-19	7.72E-17	−3.46
SFXN5	cg07452306	1.28E-18	1.06E-16	−2.45
NDUFA13	cg08331981	1.61E-18	1.31E-16	−3.43
PKLR	cg03314840	6.96E-18	5.13E-16	−3.11
ATP5G2	cg02905198	3.31E-17	2.19E-15	−2.80

Ingenuity Pathway Analysis (IPA) was used to characterize the disease and function annotations associated with these genes and returned a list of 500 categories, each with specific disease or function annotations. We removed 57 cancer-associated category terms given their overrepresentation in the IPA knowledge base, leaving 443 categories. Annotations for the top 10 significant diseases or functions included ‘mitochondrial respiratory chain deficiency,’ ‘complex I deficiency,’ ‘termination of translation of proteins,’ ‘elongation of proteins’ and categories that include metabolism and synthesis of nucleotides and amino acids. Annotations for the top 10 significant canonical pathways included ‘oxidative phosphorylation,’ ‘sirtuin signaling pathways,’ ‘estrogen receptor signaling,’ and terms involving amino acid synthesis and degradation (Figure 2).

Figure 2.

Top Ingenuity Pathway Analysis disease and function annotations and canonical pathways for genes with sex differences in DNA methylation. A) Graph of the top ten most significant disease and function annotations associated with genes differentially methylated between males and females. B) Graph of the top ten most significant canonical pathways associated with genes differentially methylated between males and females. Y-axis represents the annotation, x-axis is the -log10 of the p-value.

To investigate the potential transcriptional regulatory significance of these findings, we utilized an unpublished microarray dataset collected in our lab. Cord blood from 52 infants (26 females and 26 males) was collected for gene expression analysis. We performed regression analysis specifically on genes annotated with the Mitocarta 2.0 database. Sex differences in mitochondrial gene expression in this dataset were found for 69 genes, and of these, 17 also contained a sex difference in DNA methylation identified in the 450K dataset (Figure 3a, Table 4). STRING analysis of these 17 genes indicated strong interactions among proteins encoded by these genes, particularly those involved in the electron transport chain (ETC) (Figure 3(b)). Six of the genes identified encode either subunits or assembly factors for complexes present in the ETC (Figure 3(c)). A striking result was the gene ATP5G2, which contained seven sexually dimorphic CpG sites. The DMSs associated with ATP5G2 were located upstream of the transcriptional start site, in the first exon, and in the body of the gene. These CpG sites were all within the north and south shores of a CpG Island that is spatially close to many regulatory elements. To test if methylation differences seen in the NEST 450K samples were also present in an independent sample set, methylation of four of the CpG sites were assessed in DNA samples from 76 additional independent NEST participant’s cord blood using bisulphite pyrosequencing. These four CpG sites were selected given their close proximity to one another and thus their ability to be captured in a single pyrosequencing assay. At all four of these CpG sites, we validated higher average levels of methylation in females, two of which were statistically significant (Figure 4(a,b)). Given the sex difference in expression of ATP5G2 identified in our microarray dataset (Figure 4(c)), we examined the relationship between expression and methylation in our independent validation cohort for these CpG sites. Pearson correlation analysis between the relative expression of ATP5G2 as determined using qPCR and the average level of methylation across the four CpG sites assessed via pyrosequencing revealed a significant positive correlation in males only (r = 0.39, p < 0.05), and a negative non-significant correlation in females (r = −0.24, p = 0.18).

Figure 3.

Sex differences in DNA methylation correspond to sex differences in gene expression in a subset of genes. A) Volcano plot of the differential expression analysis with X and Y axes displaying the delta-beta values and the -log p-values, respectively, for each microarray probe. Probes with higher expression in females and males are on the left and right regions of the plot, respectively. There were 69 genes meeting the defined threshold of p < 0.05 (red line). Of these genes, 17 also had sex differences in DNA methylation and are indicated as grey open circles. B) STRING schematic of protein interactions for the 17 genes that contained sex differences in DNA methylation and expression. The schematic shows that eight of these proteins interact with at least one other protein in this network. Six of these proteins are involved with the electron transport chain, five of them are directly involved with ATP synthase. The number of nodes represents the number of proteins analysed, and the number of edges represents the number of interactions present between the nodes. The significant Protein–Protein Interaction (PPI) enrichment p-value suggests that the interaction of these proteins is not random. C) Diagram of the electron transport chain showing the functional location within the ETC for the proteins encoded by the differentially methylated and expressed genes.

Table 4.

Genes with sex differences in expression and DNA methylation.

Gene Name	Gene Function	No. Sexually Dimorphic CpG Sites	Microarray Probe	P-value	Higher Expression
LONP2	Peroxisome homoeostasis	1	221834_at	0.003	Female
SLC25A40	Mitochondrial carrier protein	2	205716_at	0.009	Male
OXA1L	Insertion of proteins into inner mitochondrial membrane	2	208717_at	0.01	Female
GPX1	Glutathione peroxidase	1	200736_s_at	0.01	Female
CECR5/HDHD5	Haloacid Dehalogenase	3	218592_s_at	0.01	Female
CDC25C	Regulator of cell division	1	205167_s_at	0.01	Male
SPG7	ATP dependent zinc metalloprotease	1	214494_s_at	0.01	Female
FTH1	Iron storage	1	200748_s_at	0.02	Male
ATP5I/ATP5ME	ATP synthase subunit E	1	207335_x_at	0.02	Male
NDUFS1	Ubiquinone oxidoreductase in ETC	2	203039_s_at	0.02	Male
ATP5E/ATP5FE	ATP synthase subunit F1 epsilon	1	217801_at	0.03	Male
LETMD1	Mitochondrial outer membrane protein, role in tumorigenesis	1	207170_s_at	0.04	Female
ATP5G2/ATP5MC2	ATP synthase subunit C	7	208764_s_at	0.04	Female
ATPAF2	ATP synthase F1 assembly factor	1	214330_at	0.04	Male
ANGEL2	Deadenylase	1	217630_at	0.04	Female
P4HB	Protein disulphide-isomerase	3	200654_at	0.04	Female
MUTYH	Adenine DNA glycosylase	1	207727_s_at	0.05	Female

Figure 4.

Validation of sex differences at ATP5G2 in an independent set of samples. A) Schematic of pyrosequencing assay designed to assess DNA methylation at four CpG sites, with the corresponding 450 BeadChip probe ID indicated. Data for each of the four CpG sites assessed is shown below. The top row represents 450K BeadChip methylation levels at these sites in the 381 individuals, showing higher levels of methylation in females at each site. Y-axis represents the beta value and x-axis represents sex. B) The bottom row represents methylation data obtained at these four CpG sites via bisulphite pyrosequencing from a validation cohort of 76 additional NEST cord blood samples (38 males and 38 females), showing females have significantly higher levels of methylation at two of these sites. Y-axis represents the percent methylation and x-axis represents sex. C) ATP5G2 expression data from 52 additional NEST cord blood samples (24 males and 24 females) showing higher levels of expression in females. Y-axis represents the relative expression obtained from an Affymetrix Gene Chip HG-U133A microarray and x-axis represents sex. D) Relationship between ATP5G2 relative expression obtained via qPCR (Taqman Assay ID:Hs01086654_g1) and DNA methylation obtained via pyrosequencing from 72 of the NEST individuals used in the validation cohort in panel B, as determined by Pearson Correlation. (*: p < 0.05, **: p < 0.01, ***: p < 0.001, ****: p < 0.0001).

Smoke exposure

When analysed for response to in utero smoke exposure, we observed differences between males and females in the CpG sites with alterations in DNA methylation at the nuclear-encoded mitochondrial genes. In males, there were 219 CpG sites associated with exposure (FDR p < 0.05) corresponding to 197 genes, while in females there were 117 CpG sites corresponding to 106 genes (Figure 5(a)). Only one of these CpG sites was in common between the male and female groups (Figure 5(b)), and this site showed opposite directionality associated with exposure between the sexes. In females, the smoke exposed group had higher levels of methylation, while in males the smoke exposed group had lower levels of methylation. There was much more similarity in the amount of overlap between the genes that were impacted in each of the sexes rather than the individual CpG sites associated with these genes (Figure 5(b)). Even though only one CpG was found to be differentially methylated in both sexes, 23 of the same genes contained differentially methylated sites in both sexes, albeit different CpGs (Figure 5(b), p < 0.0001, OR = 31.3). IPA analysis was performed on the gene lists associated with altered methylation and smoke exposure in both male and female datasets (Figure S3). Both the top 10 significant disease and function annotations and the top 10 significant canonical pathways were remarkably similar in the males and females and contained terms such as ‘sirtuin signaling pathway,’ ‘oxidation of fatty acids,’ and terms involved with amino acid synthesis and degradation (Figure S3).

Figure 5.

Sex specific responses to in utero tobacco smoke exposure in DNA methylation at mitochondrial genes in the nuclear genome. A) Manhattan plot for association between smoke exposure and DNA methylation at 19,916 CpG sites annotated to genes listed in the Mitocarta 2.0 database for females on the left and males on the right. In total, there were 117 CpGs in females and 219 CpG sites in males that met the FDR threshold (p < 0.05). X axis represents chromosome and Y axis represents the -log10 FDR p-values for each site. B) Overlap of CpG sites and genes associated with smoke exposure between each sex. There was an insignificant level of overlap between the CpG sites identified in each sex, with only one overlapping. In contrast, there are 23 overlapping genes when comparing those containing CpG sites with changes in methylation associated with smoke exposure (p = 1.15e-24; Fisher’s Exact Test).

Discussion

Sex differences

Sex differences in mitochondrial function and metabolism have recently become a topic of interest in the field of mitochondrial biology and disease. A variety of diseases that involve mitochondrial dysfunction including cardiovascular, metabolic, and neurodegenerative diseases have been shown to have sex-specific risk [19,20,22]. The aetiologies of sex differences in mitochondrial metabolism are not yet completely understood. In this study, we found that 596 CpG sites belonging to 324 genes that contribute to mitochondrial function have sex differences in DNA methylation (FDR p < 0.05) with a mean difference in methylation of at least 1%. We were able to validate sex differences in methylation at the gene ATP5G2 in independent cord blood samples from the larger NEST cohort (Figure 4). Additionally, comparing the observed DMSs to genes with sex differences in expression revealed some potentially important genes whose expression may be influenced by sex-specific DNA methylation. Our results are the first that address the potential for DNA methylation to regulate sexually dimorphic mitochondrial function through the establishment of sex differences in expression of nuclear-encoded mitochondrial genes.

We identified 324 genes with levels of sex-specific differential CpG methylation that were statistically significant (Figure 1, Table 3). The significance of broad terms such as ‘mitochondrial dysfunction’ and ‘mitochondrial disorder’ from the IPA analysis terms is likely driven by the preselection of mitochondrial genes for this study (Figure 2). However, mitochondria carry out a wide range of functions [53–55], and we were able to identify sex differences in more specific pathways. For example, in both the top 10 significant annotations for the disease and function categories and the canonical pathways, many of the terms relate to amino acid and nucleotide synthesis and degradation. Consistent with these differences having functional significance, sex differences have been observed in amino acid metabolism and cellular concentrations of amino acids [56–58]. Additionally, the enrichment of the term ‘estrogen receptor signaling’ and ‘sirtuin signaling’ is particularly interesting given that there are known sex differences in these signalling pathways [59,60]. Furthermore, the enrichment for the term ‘oxidative phosphorylation’ is intriguing because previous studies have reported sex differences in ATP production and oxidative capacity [9,10,14,19]. Sex-specific patterning in DNA methylation could be regulating nuclear gene expression of some of the proteins required for oxidative phosphorylation. Of the 17 genes identified that have sex differences in expression and methylation, six of these play a direct or indirect role in oxidative phosphorylation: ATP5G2, ATP5I, ATP5E, ATPAF2, OXA1L, and NDUFS1. Of particular interest, five of these are either subunits or assembly factors required specifically for ATP synthase, or complex V, of the electron transport chain. However, whether these differences in methylation are causative for the sex differences in expression and functional differences in mitochondrial activity and output remains to be tested.

ATP5G2 appears to be an especially strong candidate for such investigation, as we observed both sex differences in DNA methylation at multiple CpG sites and a difference in gene expression. ATP5G2 encodes one of three identical proteins that make up subunit C of the ATP synthase, complex V of the ETC [61]. The c-ring, comprised of multiple c subunits, is the part of this complex that is embedded within the membrane and facilitates proton transport, which then drives the production of ATP. ATP5G1, ATP5G2, and ATP5G3 encode identical protein sequences except for their mitochondrial import sequences, which are cleaved after mitochondrial import. Previous studies have investigated the apparent redundancy of this protein and have provided evidence that once these proteins are transported to the mitochondria, their cleaved import sequences may have biologically functional roles in the maintenance of ETC complexes [62]. They specifically suggest that the ATP5G2 targeting peptide may play a role in the assembly of complex IV of the ETC [62]. If true, the sex difference in expression of this protein could be a mechanism through which sex differences in the efficiency of oxidative phosphorylation occurs. Future studies should confirm the role of the ATP5G2 targeting peptide in cytochrome c oxidase (COX) assembly and probe for sex differences in methylation and expression in other tissue types.

It is important to keep in mind that other pathways regulate mitochondrial function in a sex-specific manner that were not assessed here. Reproductive hormones, sex differences in signalling pathways, and other mechanisms of altering gene expression of these nuclear genes could help to establish these sexual dimorphisms [63–67]. Functional sex differences in mitochondrial metabolism are likely not a result of a dominant single mechanism acting alone, but rather the combination of hormonal interactions, double dosing of X chromosome genes that are not subject to X–inactivation, signalling cascades, and differential gene expression that could partly be due to regulation by epigenetic factors such as DNA methylation.

Smoke exposure

We were surprised by the small number of CpG sites associated with nuclear-encoded mitochondrial genes that were differentially methylated by smoke exposure, compared to the large number of CpG sites that had sex differences in methylation levels. We expected that smoke exposure would be a driving factor for differences in DNA methylation, as nicotine is known to interfere with DNMTs, and previous reports have identified changes in methylation across the genome with tobacco smoke exposure, including in cord blood [33,35,39,68,69].

We observed drastically different tobacco exposure response profiles in methylation between the sexes, in line with recent reports of sex-specific associations of DNA methylation with smoke exposure in adult COPD patients [42]. Taken together, this suggests that the sexes ought not to be grouped together when analysing DNA methylation in response to a stressor, as they may have very different responses. In this gene set, there was a clear lack of overlap in addition to divergent directionality of methylation changes between the sexes. In females, there were 117 CpG sites (FDR p < 0.05) corresponding to 106 genes that had altered methylation associated with smoke exposure. In males, there were 219 CpG sites (FDR p < 0.05) corresponding to 197 genes that had altered methylation associated with smoke exposure. However, while there was surprisingly only one CpG site that was common to both the male and female DMSs, the genes that were impacted were remarkably similar. Between both sexes, 23 of the genes that experienced methylation changes overlapped (Figure 5b, p < 0.0001).

This suggests that while different CpG sites are being altered between the sexes, they are within many of the same genes. This is quite interesting since the methylation changes at different CpG sites within the same genes may have very different impacts on gene regulation with the possibility of different functional outcomes. Additionally, IPA analysis revealed enrichment of similar pathways and cellular mechanisms, indicating that even though different CpG sites are being altered, these alterations are occurring in genes with similar functions (Figure S3). While these data cannot speak to the differences in health outcomes and the severity of response from tobacco smoke exposure in utero, it does suggest that females and males have different responses in DNA methylation to the same exposures.

Analysis comparing the Mitocarta 2.0 CpG sites to three randomly generated lists of CpG sites has shown that Mitocarta 2.0 contains a lower level of CpG sites with sex differences in methylation, fewer DMSs in response to smoke exposure in both sexes, and that males had higher numbers of DMSs associated with exposure in all gene lists (Figure S4, Table S2). It is not clear how these sex-specific responses in DNA methylation from tobacco smoke exposure are coordinated and how or why certain CpG sites are differentially methylated in response to exposure. Both the decreased level of sex differences in DNA methylation of the mitochondrial genes compared to that of the rest of the genome and the decreased susceptibility to exposure of these CpG sites compared to that of the rest of the genome, as seen in the chi-square analysis, suggest a conservative or more tightly controlled level of regulation on DNA methylation at these nuclear-encoded mitochondrial genes. Previously, our lab has shown that a handful of genes are divergently methylated at the same CpG sites in response to independent exposures to nicotine or THC [70]. It remains unclear what about these specific genes, which often experience differences in DNA methylation in response to a wide range of exposures, is driving this phenomenon.

Males appear to be more sensitive to DNA methylation changes in response to in utero tobacco smoke exposure in the sense that they have a greater number of CpG sites with statistically significant alterations following exposure (Figure S4, Table S2). These differences could be a result of sex differences in DNA methyltransferase activity [71] or metabolism of polycyclic aromatic hydrocarbons [72,73]. Additionally, mitochondria play a crucial role in the establishment of DNA methylation. Byproducts of mitochondrial metabolism serve as cofactors necessary for DNA methylation establishment and maintenance [31,74,75]. Certain components of tobacco smoke exposure, such as benzo[a]pyrene, are known mitochondrial toxicants and in utero tobacco smoke exposure has been associated with mitochondrial dysfunction [40,41]. The proposed increased resiliency seen in females may be due to their increased OXPHOS capabilities [19], which could allow them to better tolerate disruptive exposures like that to in utero tobacco smoke. Previous studies have observed sex differences in toxicological responses to cigarette components such as benzo[a]pyrene [76], though few studies have examined these differences in the context of mitochondrial function or the establishment of DNA methylation [77].

Strengths and limitations

The Illumina 450K platform is a well-established and highly reproducible methylation array that is typically analysed in its entirety, requiring more stringent statistical corrections. Previous studies assessing genome-wide sex differences in DNA methylation revealed that CpG sites with the highest degree of sex-specificity show enrichment for genes involved in neurodevelopment [6,7]. Here, we have used a hypothesis-driven approach to identify genes that may be involved in the regulation and establishment of sex differences in mitochondrial function. A limitation to this study is that DNA methylation profiles can vary across cell and tissue types. In this study, we used genomic DNA from the leukocyte fraction (buffy coat) of cord blood samples for our validation cohort studies. Although the 450K results were corrected for cell composition, blood contains multiplecell types, which could confound the validation cohort data results. We believe our findings of sex-specific patterns of DNA methylation at the nuclear-encoded mitochondrial genes will likely be generalizable, given the high level of conservation of mitochondrial function [78]. However, additional work will be required to determine the extent to which the sex differences we have observed in cord blood samples will be consistent across cell and tissue types.

Conclusions

In summary, this study has revealed sex differences in DNA methylation in genes corresponding to mitochondrial metabolism and oxidative phosphorylation, as well as sex-specific DNA methylation alterations in response to in utero tobacco smoke exposure. Future work should examine both the potential functional consequences to mitochondria of these sex-specific methylation differences, as well as effects of sex differences in mitochondrial function on the establishment of nuclear epigenetic patterning in the context of environmental exposures.

Funding Statement

This work was supported by the National Institute of Environmental Health Sciences awards P01ES022831, P42ES010356, P30ES011961, R01ES016772, R24ES028531 and T32ES021432; the National Institute of Diabetes and Digestive and Kidney Disease award R01DK085173; the United States Environmental Protection Agency award RD-83543701-0; and the National Science Foundation award GRFP DGE-1644868.

Data availability statement

Results of statistical analysis can be found in the supplemental data files. Raw gene expression microarray data used are available in the supplemental data files and the 450k methylation data used is available at the Duke Digital Research Data Repository (https://doi.org/10.7924/r44b36p48).

Author contributions

Dillon E. King: conceptualization, data curation, formal analysis, investigation, visualization, writing - original draft.

A. Clare Sparling: writing - original draft, visualization, formal analysis.

Dillon Lloyd: investigation, formal analysis.

Matthew J. Satusky: investigation, formal analysis, visualization.

Mackenzie Martinez: data curation, formal analysis.

Carole Grenier: data curation.

Christina M. Bergemann: formal analysis.

Rachel Maguire: data curation, resources.

Cathrine Hoyo: data curation, resources, review editing.

Joel N. Meyer: conceptualization, project supervision, writing - review editing.

Susan K. Murphy: conceptualization, resources, project supervision, writing - review editing.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplemental data for this article can be accessed here.