Demographics

Two largely overlapping placenta cohorts were used for first trimester sex differences analyses (Fig. 1A). Demographics analyses were performed to compare each cohort by fetal sex (Table 1, Additional file 1). The methylation cohort included 56 pregnancies (25 carrying a female fetus, 31 a male fetus). The methylation cohort subjects were mostly Caucasian non-Hispanic but had no differences in maternal/paternal age, race, ethnicity, BMI, maternal medical conditions, pregnancy outcomes, or pregnancy complications (Supplemental Table S1, Table 1). The RNA-seq cohort included 74 pregnancies (34 carrying a female fetus, 40 a male fetus). RNA-seq cohort subjects were mostly Caucasian, non-Hispanic. However paternal ethnicity was more often non-Hispanic (93%) in the male fetus cohort than in the female fetus cohort (85%) (p = 0.041). No differences were noted in maternal age, race, ethnicity, BMI, maternal medical conditions, pregnancy outcomes, or pregnancy complications (Supplemental Table S2, Table 1). Principal components analysis showed that when sex chromosomes were included, samples separated by fetal sex and not by other demographics variables (Additional file 2). When sex chromosomes were excluded, fetal Hispanic samples somewhat clustered in DNA methylation data, but were still located within the larger group (Additional file 2).

Fig. 1

Sample cohorts and quality control. A Summary of the first trimester placenta cohorts and analyses. F = females, M = males. B Principal components analysis for the DNA methylation cohort. The pre-filtered dataset confirms fetal sex. The filtered dataset no longer shows clustering by sex. C Quantile–Quantile plot for the differentially methylated probes (DMP) analysis. Points are individual methylation sites (probes). Red line at Pobserved = Pexpected

Table 1 Brief demographics and pregnancy outcomesDNA methylation analysis quality control

DNA methylation of first trimester human placentas was measured using the EPIC array. The fetal sex of processed placenta samples was verified experimentally using a principal components analysis starting with all sites on the EPIC array, including chromosomes X and Y which segregate samples by fetal sex (Fig. 1B, Additional file 2). However, sex chromosome DNA methylation cannot be accurately compared between female and male samples due to differences in copy numbers and phenomena such as X chromosome inactivation which involves wide spans of DNA and histone methylation across one copy of the female chromosome X, but not the other. Differentially methylated probes (DMPs) analysis was performed to identify sexually dimorphic sites using only autosomal chromosomes, without other sources of unreliability such as masked probes or probes at single nucleotide polymorphisms. Principal components analysis of the final sites shows that placental samples no longer cluster by fetal sex without the sex chromosomes, indicating that global placental DNA methylation is largely similar between sexes (Fig. 1B, Additional file 2). The Quantile–Quantile plot of P-values shows a lambda value of 1.1249 indicating low inflation (Fig. 1C).

Differentially methylated probes (DMPs) affected by fetal sex

We identified 151 autosomal differentially methylated probes (DMPs) significantly altered between female and male placentas at FDR < 0.05 (here equal to P < 1 × 10–5), with a slight majority of 59% (89 DMPs) hypermethylated in females, and 62 hypermethylated in males (Additional file 3, Additional file 4). Among these 151 DMPs, there were 29 CpG sites that reached genome-wide significance (P < 9 × 10–8), with 10 hypermethylated in females and 19 hypermethylated in males (Table 2). The most significantly female hypermethylated probe was cg00167275, located in the 5′ untranslated region (5′UTR) and first exon of glutamate dehydrogenase 1 (GLUD1), and before the transcription start site of double-strand break regulator FAM35A (also called SHLD2). The most significantly male-hypermethylated probe was cg17612569, located before the transcription start site of transcription factor subunit GABPA and at different locations of ATP synthase subunit ATP5J transcripts (5′UTR, first exon, and coding region). No CpH (CpA, CpC, or CpT) probes reached FDR < 0.05 significance.

Table 2 Differentially methylated probes (DMPs) in first trimester placenta that reach genome-wide significance (P < 9 × 10–8)

Manhattan plots were created to visualize site significance across the genome (Fig. 2A). Chromosome 5 showed a cluster of 13 probes mapped to the promoter region of zinc finger protein 300 (ZNF300) gene, 200 bp or 1500 bp upstream of the ZNF300 transcription start site (Additional file 4). Twelve sequential ZNF300 probes were significantly hypermethylated in male placentas with strong trends, including 9 in a CpG island and 4 in a south shore (downstream of the CpG island), and only the final 13th probe (further downstream from the rest) showed female-hypermethylation. Additional clusters of methylation differences were evident across the genome, though not all reached genome-wide significance (Fig. 2B).

Fig. 2

Manhattan plots of the top probes by P-values. Chromosomes alternate colors and magenta points highlight gene ZNF300. A The top 5000 probes by P-values separated by significance. Higher − log10(P) values means probes are more significantly sex different. Probes reach FDR < 0.05 (P < 1 × 10–5) significance above the solid green and genome-wide significance above the dashed green line at P = 9 × 10–8. B Only the significant probes (FDR < 0.05), plotted for direction of methylation: positive values indicate hypermethylation in female placentas, and negative values indicate hypermethylation in male placentas. C Top 10 enriched gene ontology terms for the DMPs, sorted by P-value

Gene ontology analysis was performed using the 151 DMPs, resulting in 248 enriched gene ontology terms at P < 0.05 (Additional file 5). The five most significant terms were “endothelial cell–cell adhesion” (P = 0.000123), “presynaptic active zone” (P = 0.000694), “intracellular non-membrane-bounded organelle” (P = 0.00204), “non-membrane-bounded organelle” (P = 0.002047), and “S100 protein binding” (P = 0.002195) (Fig. 2C).

Differentially methylated regions

A detailed analysis of differentially methylated regions (DMRs) was performed to confirm findings on the Manhattan plot. We identified eleven DMRs on autosomal chromosomes, eight regions hypermethylated in female placentas and three regions hypermethylated in male placentas (Table 3). The most significant DMR was composed of 14 probes in ZNF300, hypermethylated in males. The longest DMR between female and male placentas was male-hypermethylated, located in genes GABPA/ATP5J, and composed of 23 probes across 1543 base pairs. The most significant female-hypermethylated DMR was located in ZNF311 with 16 probes. Another zinc finger domain-containing transcription factor, ZNF175, appeared to be among the most male-hypermethylated genes (Fig. 2B) and was confirmed as a significant male-hypermethylated DMR. Chromosome 14 had three DMR, all female-hypermethylated, including two forkhead box transcription factor genes (FOXG1 and FOXA1). Chromosome 6 had two DMRs (ZNF311 and C6orf47/C6orf47-AS1), also both female-hypermethylated. Other DMRs were located on separate chromosomes.

Table 3 Differentially methylated regions (DMRs) between female and male placentas at first trimesterComparison to RNA-seq sex differences

We previously found ZNF300 gene expression significantly upregulated in female placenta, compared to male, in bulk total RNA-seq of n = 39 [9]. We combined additional bulk total RNA-seq data to perform a new sex differences analysis on n = 74 first trimester human placenta samples. Differential expression analysis identified 152 differentially expressed genes (DEGs) between females and males at FDR < 0.05, with 48 female-upregulated and 104 male-upregulated (Additional file 6). The most female-upregulated gene was XIST (167-fold higher in females, FDR = 0), an X-linked gene involved in X-chromosome inactivation. The most male-upregulated genes were Y-linked protein coding genes RPS4Y1, DDX3Y, UTY, EIF1AY, ZFY, USP9Y, KDM5D, and non-coding TTTY15 (Fig. 3A). By fold change, the most sexually dimorphic autosomal genes were chemokine ligand 9 (CXCL9, 6.3-fold higher in males, FDR = 0.0021) and insulin-like growth factor binding protein 1 (IGFBP1, 6.27-fold higher in females, FDR = 0.0090) (Fig. 3B). By significance, the most sexually dimorphic gene was ZNF300 (2.12-fold higher in females, FDR = 2.98 × 10–8).

Fig. 3

RNA-seq sex differences and correlation to DNA methylation. A, B Volcano plots for RNA-seq sex differences in placenta, with dashed lines at P = 0.05 and FDR = 0.05 (these lines overlap in A due to the y-axis range). C Violin plots comparing female and male gene expression: log2(baseMeans) adjusted for batch. Horizontal lines indicate the mean. Sex differences: *P < 0.05, **FDR < 0.05. D Manhattan plot of correlation results with the overlapping cohort (n = 51), comparing gene expression (“expr”) to DNA methylation (“methyl”). Each point is a probe. Probes located in the same gene are grouped with rectangles. Non-significant (“NS”, RNA-seq P ≥ 0.05) genes are semi-transparent. Magenta color highlights three genes with correlation FDR < 0.05 and RNA-seq FDR < 0.05 (FAM228A, ZNF300 and CSMD1)

DMPs at FDR < 0.05 were matched to the RNA-seq results to identify overlapping genes of interest (Additional file 7). Of the 151 DMPs, 42 were located in genes that showed at least nominally significant sex differences in RNA-seq (P < 0.05), and 17 DMPs were located in two genes significant after adjustment for multiple comparisons (FDR < 0.05 in the RNA-seq): ZNF300 which was consistently male-hypermethylated, and tumor suppressor gene CUB and Sushi multiple domains 1 (CSMD1) which had gene expression 1.78-fold higher in males but inconsistent CpG hypermethylation directions in the gene body (none in the promoter region). Probe cg05994094 was genome-wide significant and hypermethylated in females and its corresponding gene, coiled-coil domain-containing protein 178 (CCDC178), had 1.46-fold higher gene expression in males at P < 0.05 (Fig. 3C), consistent with expectations. Three genes from the DMR region analysis were associated with sexually dimorphic gene expression: ZNF300, ZNF311, and CCDC178.

To directly access which DNA methylation sex differences affect gene expression, we integrated the datasets. DNA methylation probes with sex differences at P < 0.01 (12,560 probes) were matched to overlapping or nearby genes, resulting in 35,862 gene-probe matches for a correlation analysis using the 51 overlapping placentas (Additional file 8). Of these, 3083 (5.7%) and 396 (0.72%) gene-probe matches reached correlation significances of P < 0.05 or FDR < 0.05, respectively, indicating that methylation sex differences alter expression of a small subset of placental genes. When limited to sexually dimorphic genes identified earlier in the full cohort RNA-seq, 152 gene-probe matches (146 unique probes) show significant methylation correlation to gene expression in first trimester placenta (P < 0.01 DNA methylation, P < 0.05 RNA-seq, and FDR < 0.05 correlation). Of these, three genes met FDR < 0.05 in the RNA-seq: FAM228A, ZNF300, and CSMD1 (Fig. 3C), indicating strong sex differences in the placental transcriptome. FAM228A showed a significant inverse (negative) correlation, indicating that female-hypermethylation causes male-upregulated expression (Fig. 3D). ZNF300 showed an inverse correlation as well. CSMD1 showed a positive correlation for the most significant 12 probes (correlating at FDR < 0.05), suggesting that hypermethylation at these sites increases gene expression. The correlation analysis also highlighted proteasome subunit alpha type 8 (PSMA8) which showed a significant inverse correlation in eleven FDR < 0.05 probes. ZNF175, a sexually dimorphic DMR, showed strong inverse correlation between placental DNA methylation and gene expression (FDR < 0.05), though the gene was not significantly sex different in this RNA-seq.

Methylation sex differences across studies

The 151 DMPs were cross-referenced across different studies to identify which are uniquely sex different in first trimester placenta, which show placental sex differences later in gestation, and which also show sex differences in the whole blood of neonates or adults (Figs. 4, 5). All studies were compared at significance threshold FDR < 0.05, and non-significant sites were examined for directional trends.

Fig. 4

Comparison with other sex differences studies, chromosomes 1–7. DMPs FDR < 0.05 in this study are compared to DMPs in Andrews et al., Santos et al., Inkster et al., and Grant [13, 14, 32, 33]. DMPs are sorted by chromosome position and labeled “genes_chromosome_probeID”. RNA-seq sex differences in this study are highlighted for significant (FDR < 0.05) and nominally significant (P < 0.05) genes using asterisks, with color bars for genes with > 2 DMPs

Fig. 5

Comparison with other sex differences studies, chromosomes 8–22. Continuation of Fig. 4, with similar labeling

Our DMPs showed good consistency with Andrews et al. [14] sex differences analysis of human placenta at different gestational ages, including an independent cohort of 22 first trimester placentas at 8–14 weeks. We found novel results for 60 DMPs (40%) due to the higher number of methylation sites in the EPIC array, compared to the older 450 k array. Of 91 (60%) DMPs remaining, methylation directions were consistent across gestation: 84, 83, 66, and 86 DMPs matched in the first, second, third trimester (preterm), and full term placenta cohorts, respectively. At FDR < 0.05, we identified 3, 16, 6, and 71 DMPs with shared significance, respectively. The 3 DMPs significant in both first trimester cohorts were cg26605406 (ZNF175), cg20696478 (ZNF175), cg17612569 (GABPA/ATP5J), all male-hypermethylated genes significant in our region analysis (Table 2). We identified more significant matches with Andrews’ second trimester and full term cohorts, possibly because our cohort is at late first trimester (10–14 weeks). Inkster et al.’s [13] full term analysis includes some overlap with Andrews’ full term cohort and showed similar results, with 79 DMPs in the same direction (61 DMPs significant) and 66 DMPs with no available data. We also examined a sex-stratified comparison of second versus third trimester placenta [31], but no DMPs overlapped.

We also compared our 151 DMPs to two EPIC array studies of preterm placenta and whole blood. In Santos et al.’s [32] sex differences analysis of the Extremely Low Gestational Age Newborn (ELGAN) cohort (23–27 weeks, n = 1171), we found FDR < 0.05 matches for 51 DMPs in preterm placenta (34%) and 22 DMPs in neonatal peripheral blood (15%). In Grant et al.’s [33] adult blood analysis, only 8 DMPs (5%) matched, the lowest percent of all compared. One CpG methylation site was significantly male-hypermethylated at FDR < 0.05 in all studies including adult blood: cg17612569 (GABPA/ATP5J, chr21) (Fig. 5). Another near-unanimous methylation site was female-hypermethylated cg06710937 (chr13) near genes RFESDP1, LINC00621, BASP1P1, and IPMKP1, none of which were sex different in our placental RNA-seq.

Placental female-hypermethylation resulted in higher gene expression for males at several genes including: GABA-receptor subunit GABRB1 (chr4), transmembrane glycoprotein embigin (EMB, chr5), transcription factors ZNF311 (chr6) and GTF3C6 (chr6), methionine sulfoxide reductase A (MSRA, chr8), the cell cycle regulating transcription factor TFDP1 (chr13), read-through product NEDD8-MDP1 (chr14), fibroblast growth factor 7 (FGF7, chr15), CCDC178 (chr18). Of these transcriptome-altering DMPs, MSRA female-hypermethylation is uniquely significantly at first trimester placenta. TFDP1 and EMB methylation may also be unique to first trimester placenta, but information is lacking across gestation. CSMD1 DMPs (chr8) had inconsistent methylation directions, mostly male-hypermethylated except for female-hypermethylated cg12269374 located in the gene body. Two DMPs showed female-hypermethylation and female-upregulated gene expression: cg16446577 at peptidase inhibitor PI16 (chr6) which was consistently female-hypermethylated across gestation, and cg18685455 (chr6) which corresponds to AHI1 (involved in vesicle trafficking) as well as one of several regions encoding Y_RNA and is thus difficult to interpret.

Placental male-hypermethylation rarely resulted in higher gene expression for female placentas, only seen for our most significant result ZNF300 (chr5), and for long non-coding gene AP001065.15 also called LINC02575 (chr21). ZNF300 was consistently male-hypermethylated across gestation in placenta, but mostly not blood except for cg08580836 which was significant in neonatal blood [32]. AP001065.15 (cg20877594) was also male-hypermethylated in preterm placenta and matching neonatal blood [32]. Three other genes were male-hypermethylated in placenta but showed nominally higher gene expression in males: RAS oncogene family member RAB7A (chr3), CSMD1 (chr8) discussed above, and nuclear lamin B2 (LMNB2, chr19). RAB7A was male-hypermethylated across gestation in placenta, whereas LMNB2 may be unique to first trimester placenta.