Monosomy X in isogenic human iPSC-derived trophoblast model impacts expression modules preserved in human placenta

Significance

Monosomy X causes Turner syndrome (TS), but the most likely developmental outcome is miscarriage. In contrast to miscarried monosomy X, live-born TS individuals are often mosaic due to embryonic loss of Y or inactive X, suggesting that some euploid cells, even when confined to the placenta, may be required for term pregnancy. We reprogrammed X-monosomic alongside isogenic euploid pluripotent stem cells from mosaic fibroblasts to test how monosomy X impacts an in vitro model (“trophoblast-like,” TBL) for early human placenta. We find moderately impaired secretion of placental growth factor and chorionic gonadotropin in X-monosomic TBLs. Correlated expression changes also vary in euploid first trimester and term placenta and implicate genes common to X/Y. Our study identifies candidate pathways that may impair X-monosomic trophoblast function.

Abstract

Mammalian sex chromosomes encode homologous X/Y gene pairs that were retained on the Y chromosome in males and escape X chromosome inactivation (XCI) in females. Inferred to reflect X/Y pair dosage sensitivity, monosomy X is a leading cause of miscarriage in humans with near full penetrance. This phenotype is shared with many other mammals but not the mouse, which offers sophisticated genetic tools to generate sex chromosomal aneuploidy but also tolerates its developmental impact. To address this critical gap, we generated X-monosomic human induced pluripotent stem cells (hiPSCs) alongside otherwise isogenic euploid controls from male and female mosaic samples. Phased genomic variants in these hiPSC panels enable systematic investigation of X/Y dosage-sensitive features using in vitro models of human development. Here, we demonstrate the utility of these validated hiPSC lines to test how X/Y-linked gene dosage impacts a widely used model for human syncytiotrophoblast development. While these isogenic panels trigger a GATA2/3- and TFAP2A/C-driven trophoblast gene circuit irrespective of karyotype, differential expression implicates monosomy X in altered levels of placental genes and in secretion of placental growth factor (PlGF) and human chorionic gonadotropin (hCG). Remarkably, weighted gene coexpression network modules that significantly reflect these changes are also preserved in first-trimester chorionic villi and term placenta. Our results suggest monosomy X may skew trophoblast cell type composition and function, and that the combined haploinsufficiency of the pseudoautosomal region likely plays a key role in these changes.

Placental female mammals maintain dosage parity of most X-linked genes with males via X chromosome inactivation (XCI), a process that independently evolved long noncoding RNAs (eutherian XIST, metatherian Rsx) to silence one of two X chromosomes (1). This dosage-compensation strategy was likely necessitated by attrition of the proto Y in the heterogametic male germline of ancestral mammals following acquisition of the male-determining factor (SRY) (2). Yet, because some sex-chromosomal genes were selected to resist Y attrition and escape XCI, thereby maintaining expression from two active copies in males and females alike (3), mammalian development is likely sensitive to proper dosage of these X/Y gene pairs (4). This multigenic dosage sensitivity is perhaps best reflected in the pronounced rarity of live-born monosomy X in mammals that feature a long pseudoautosomal region (PAR). Present on X and Y, the PAR is maintained via meiotic recombination in the male germline and likewise escapes XCI in females (5). In contrast, mice feature a short PAR, as well as fewer XCI “escapee” genes overall (6, 7) and tolerate monosomy X with very little developmental impact (8, 9).

Human monosomy X (45,X) causes Turner syndrome (TS, ∼1:2,500 live births), which ranges from full penetrance of short stature and early, often prepubertal ovarian failure to other skeletal and craniofacial changes, lymphedema of hands and feet, cardiovascular defects, and to impaired hearing in about half of TS patients (10, 11). Yet, most monosomy X pregnancies result in miscarriage, estimated to account for 6 to 11% of all spontaneous terminations (12–15). Because the rate of detectable mosaicism for euploid cells in live-born TS is very high (∼50%), but very low (∼0.5%) in karyotypic follow-up of miscarried monosomy X, Hook and Warburton compellingly hypothesized zygotic monosomy X to be near-invariably lethal in utero, and that live-born TS results from mitotic sex chromosome loss in early embryos, which gave rise to either detectably mosaic TS, or TS with cryptic (e.g., confined placental) mosaicism (15).

Conversely, this suggests that in the absence of placental mosaicism, homogenously 45,X extraembryonic tissues would be defective in supporting conceptuses to term. Because this phenotype is not shared with the mouse, new mammalian and human in vitro models are needed to test this important prediction. While two prior reports of 45,X human embryonic and induced pluripotent stem cell (hiPSC) lines pointed to lower expression of some placental genes in generic (embryoid body) differentiation (16, 17), the impact of monosomy X on relevant in vitro models of human extraembryonic development has not been assessed.

To address this important question in a widely used hiPSC-based model of the primitive syncytiotrophoblast (STB) (18, 19), we derived 45,X hiPSCs alongside isogenic euploid control lines from reportedly mosaic samples. This enables us to largely exclude the impact of autosomal variation and leverage phased variants to quantify allele-specific dosage contributions from X and Y. While these hiPSCs trigger a GATA2/3- and TFAP2A/C-mediated gene circuit (20) irrespective of karyotype, differential expression indicates monosomy X may alter the balance between cytotrophoblast (CTB), STB, and extravillous trophoblast (EVT) markers. These changes are also reflected in secretion of human chorionic gonadotropin (hCG) and placental growth factor (PlGF) and correlate with gene modules that are preserved in late first-trimester and term placentas. Together, our study represents the largest single source of cytogenomically validated 45,X hiPSC and isogenic euploid control lines to date and provides an assessment for how monosomy X may impact human trophoblast–relevant gene networks.

Results

Because the low rate of term pregnancy with 45,X conceptuses may reflect a selective bottleneck, we turned to aging as a postdevelopmental model of mosaic sex chromosome loss, which by age 75 is detectable in 19 to 44% of males and ∼0.45% of females (21–23). In total, we reprogrammed mosaic fibroblasts with reported mosaic 45,X karyotype (SI Appendix, SI Methods) from two male and three female donors and validated the resulting clones in systematic fashion (Fig. 1 A and B). Two of the female samples resulted in exclusively X-monosomic or euploid hiPSC lines, and were not pursued further, as our analysis aimed to compare 45,X to isogenic euploid control lines of the same donor. In contrast, one female (AG05278) and two male (AG09270 and AG04381) samples gave rise to both euploid and 45,X hiPSC clones (male1 XO, male2 XO7/15, and female-derived XO2/8/9) from each donor (aged 65, 67, and 82), as determined by short tandem repeat PCR and Y-linked amplicons (SI Appendix, Fig. S1B). Karyotyping (SI Appendix, Fig. S1C) further ruled out three male1-derived clones with chromosome (chr) 12 trisomy. Finally, all remaining male- and female-derived lines passed high-resolution cytogenetic testing (CytoSNP-850k) to rule out any other genomic copy-number variation (CNV). No other CNV calls (Materials and Methods) were made in AG09270- and AG04381-derived lines, while all AG05278-derived (euploid and 45,X alike) hiPSC lines carried an intronic 440-kb duplication in neuron-specific OPCML that was likely present in the donor fibroblasts. Altogether, a final set of 12 hiPSC clones passed this quality-control workflow with validated karyotypes (Fig. 1B).

Fig. 1.

Cytogenomic validation of monosomy X and isogenic euploid hiPSC panels. (A) Schematic of reprogramming and hiPSC characterization. (B) International System for Human Cytogenomic Nomenclature (ISCN) call from the CytoSNP-850k (Illumina) platform of hiPSC clones used in this study. (C) XIST expression by RT-qPCR as a percentage of GAPDH (Mann–Whitney U test P values).

We next characterized XCI in female-derived 46,XX clones (XX6, XX19, and XX23), because loss of XIST expression is common in standard hiPSC culture and can lead to progressive reactivation of the inactive X (Xi), referred to as Xi erosion (24, 25). We recently reported a prevalent and largely contiguous Xi DNA hypomethylation trajectory after loss of XIST expression, which we validated in two of our 46,XX (XX19/23) lines over 6 months of continuous passage (26). However, in early-passage and differentiation experiments described herein, all three 46,XX clones expressed XIST at or above female fibroblast WI-38–equivalent levels by qPCR (Fig. 1C). As expected, these early passages reflect heterogeneity for XIST by fluorescence in situ hybridization (FISH) and associated H3K27me3 deposition on the Xi (SI Appendix, Fig. S2 A and B).

We next confirmed that X-linked CpGs that best reflect Xi erosion in our earlier report (26) remain hypermethylated in all three hiPSC lines relative to male control hiPSC data (27). This methylationEPIC array analysis indicated XCI maintenance was largely intact across all early hiPSC passages used herein (XX6 < p21, XX19 < p9, and XX23 < p11), as >89 to 97% of probes remain methylated in these samples (SI Appendix, Fig. S2C). Interestingly, XX6 hiPSC featured a hypermethylated X relative to XX19/23 lines (SI Appendix, Fig. S2D), as >10% of X-linked promoter CpGs were differentially methylated in the XX6 line (SI Appendix, Fig. S2E, p.adj ≤ 0.05, median +15% in CpG methylation; Dataset S1). Importantly, the distribution of hypermethylated probes is significantly different (Kolmogorov–Smirnov [KS] test SI Appendix, Fig. S2F) from the erosion-specific order of demethylation we reported during Xi erosion (26), and instead most closely matches overall probe density on the methylationEPIC array. These data indicate that relative to XX19/C23, the XX6 Xi is evenly hypermethylated across its entire length, including over differential escapees (Dataset S1). This may suggest that variance in XIST levels (Fig. 1C) may contribute to variable escape, which we assess below by allele-specific RNA sequencing (RNA-seq).

To this end, we first performed linked-read whole-genome sequencing on the 10× Genomics platform (phased WGS) to ∼30× coverage (SI Appendix, Fig. S2G), yielding a catalog of ≥76,000 heterozygous variants that distinguish the female donor’s X chromosomes. In the mRNA-seq experiments below, we leverage these phased variants (≥1,000 in exons, 3′ and 5′ UTRs) to determine that all female-derived 45,X clones maintained the same parental X chromosome and to quantify escape from XCI. We also confirmed high expression of pluripotency markers by immunofluorescence (IF) for OCT4 and SSEA4 and low expression of differentiation markers by 3′ mRNA sequencing (SI Appendix, Fig. S3 A and B). All 45,X hiPSC lines expressed high levels of pluripotency genes SOX2 and DNMT3B and low levels of lineage-specific genes, equivalent to their respective euploid control lines, in line with reported 45,X hiPSCs (17, 28, 29).

Several methods for derivation of trophoblast-like (TBL) cell fates from hiPSCs have been reported in recent years (reviewed in ref. 19), starting from either preimplantation blastocyst-like (naïve or ground state) (30, 31), postimplantation epiblast-like (or primed) (18, 32, 33), or intermediate (extended/expanded) (34, 35) human pluripotency states. We chose a well-characterized TBL-induction method compatible with primed hiPSCs (18), to ensure our validated karyotypes and DNA methylation were stably maintained. These requirements ruled out TBL induction from naïve hiPSCs, which can suffer increased genome instability relative to primed hiPSCs (36, 37), and a genome-wide drop in DNA methylation (38, 39) that removes gametic imprints (40), to many of which extraembryonic development is highly sensitive (41–43).

Primed hiPSCs were differentiated to TBL cell fates by exposure to BMP4 and inhibition of TGFβ1/activin/nodal and fibroblast growth factor (FGF) signaling over 8 d. This widely used “BAP” (BMP4, A83-01, PD173074) model (18, 32, 33, 35, 44–46) is thought to reflect the primitive STB (18, 46), as BMP4 activates a conserved trophectodermal transcription factor (TF) circuit via GATA2/3 and TFAP2A/C (20, 47). All BAP-treated lines formed flat epithelial sheets that progressively fused into large syncytiated cells, seen as clustered nuclei inside Na⁺/K⁺ ATPase-marked membranes. These syncytia featured high levels of the hCG beta-subunit (SI Appendix, Fig. S4), produced only by the fused STB upon implantation (48). Secretion of hCG showed a moderate but statistically significant decrease (P ≤ 0.1, Mann–Whitney U test) in 45,X lines (female-derived XO2/8/9, male1 XO, and male2-derived XO7/15) relative to their isogenic euploid controls in each panel, and in comparing all 45,X to euploid samples (Fig. 2A, P = 8.4e-05). Another important protein secreted from the STB, placental growth factor (PlGF), a member of the vascular endothelial growth factor (VEGF) family, is critical for proper placental angiogenesis (49). Secreted PlGF levels were also lower in 45,X relative to their isogenic euploid controls, a decrease that was statistically significant in each of the three individual panels, and in comparing all 45,X to all euploid samples (Fig. 2B, P = 9.2e-06, Mann–Whitney U test).

Fig. 2.

Moderately impaired hCG and PlGF secretion and TB markers in monosomy X. (A) Secreted hCG ELISA in mIU/mL normalized to total of RNA (per microgram) harvested from the same well. (B) Secreted PlGF ELISA in picograms per milliliter (pg/mL) per microgram of RNA, as in A. Mann–Whitney U test P values in A and B compare 45,X samples to otherwise isogenic euploid controls (denoted by brackets). (C) TBL RNA-seq vst counts of genes relevant to BAP-induced cell fates by line for female and male1 samples. (D) Expression of genes in C by RT-qPCR as a percentage of GAPDH (Mann–Whitney U test P values) for male2 samples. Boxes denote interquartile range (in C and D).

Female euploid XX6 TBLs, however, secreted less hCG and PlGF (Fig. 2 A and B) and fused at a lower rate than XX19/23 lines (SI Appendix, Fig. S5 A and B), pointing to differences among 46,XX euploid lines addressed below. Fusion indices were determined in unbiased fashion via computational analysis of DNA content (Hoechst staining) and revealed no other differences that rose to statistical significance. Likewise, differences in Transwell migration rates were significant only in the female panel, not the male panel (SI Appendix, Fig. S5C). All BAP-treated lines also gave rise to migratory cells that expressed the EVT marker HLA-G with similar, albeit variable frequency (SI Appendix, Fig. S5D).

To develop a more comprehensive understanding of how monosomy X may impact TBL cell fates, we performed mRNA-seq in four independent rounds of BAP differentiation. We first assessed the BMP4-induced TF circuit triggered via GATA2/3 and TFAP2A/C (20), all four of which were robustly expressed (>13 variance-stabilizing counts (vst) roughly approaching log2-scaled counts). XX19/23 and XY lines expressed moderately higher levels (log2FC −0.2 to −0.9, p.adj ≤ 0.05, Dataset S2) of TFAP2A, PGF, and HLA-G than their isogenic 45,X counterparts, which was not true for XX6, however (Fig. 2C). We replicated significantly lower TFAP2A and PGF expression in the male2 panel (XO7/15), alongside decreased GATA3 levels (Fig. 2D), which was likewise significantly lower in the female panel. Of the TFs responding to the GATA2/3 and TFAP2A/C quartet, matching decreases in the male1 and female-derived panels were confined to a handful of weakly expressed TFs, except for MEIS1 and EPAS1, which were modestly decreased across 45,X lines (SI Appendix, Fig. S6A).

Next, we rigorously compared expression levels of lineage markers identified in single-cell RNA-seq studies of early human and macaque embryos (50–52). A subset of human pregastrulation lineage markers (51) were reclassified recently based on single-cell RNA-seq from postgastrulation macaque embryos (52) to resolve human trophectoderm (TE), epiblast (EPI), and early amniotic (E-AM) lineages (53). As expected, we found that levels of both TE-associated gene sets (50, 53) significantly exceed levels of all other human lineage-associated gene sets in our BAP-treated cells (SI Appendix, Fig. S6 B and C), with a median differential of +1.5 to 2 vst counts (approximately fourfold difference, see Materials and Methods). We also assessed all original gene sets (50–52) against the reclassified (TE, E-AM, and EPI) markers (53) and again found that both distinct TE gene sets far exceed levels of genes associated with all other lineages, with early STB markers (51) representing the next-highly expressed gene set (SI Appendix, Fig. S6C). This is important in regards to a number of purported markers of the primate amnion, which have led to the suggestion that BAP treatment of primed hiPSCs induces an amniotic rather than TE cell fate (54, 55). Yet, recent publications acknowledge that both naïve (+AP without BMP4) (54) and primed hiPSCs treated with BMP4 (35, 53, 56, 57) adopt TBL cell fates, and new work suggests that human TBL may differentiate through a transient amnion-like state in vitro (53, 58). Our data across the full breadth of these lineage-associated markers demonstrate that our BAP-treated lines reflect a predominantly TE and early STB-like expression profile (SI Appendix, Fig. S6 B and C).

Transcriptome-wide principal component analysis (PCA) indicates that 45,X and euploid BAP-treated samples are clearly distinct, segregating along PC1 and PC2 by karyotype and donor, respectively (Fig. 3A). Surprisingly, the XX6 euploid samples cluster with otherwise isogenic 45,X samples (XO2/8/9), mirroring their relative decrease in hCG and PlGF levels (Fig. 2 A and B). Because the predominant TS hypothesis posits a haploinsufficiency of genes that escape XCI and were maintained on the Y, we next assessed allele-resolved and overall expression of PAR genes, interspersed X-Y pair genes (“X-pair”), and X-specific escapees without the Y homolog (Fig. 3B). Overall, our phased variants covered 451 X-linked genes previously assessed in a large human GTEx (Genotype-Tissue Expression) study (59), with sufficient allelic read depth to unambiguously call XCI status for up to 226 genes. A total of 166 genes were called (inactive/escape) across all three isogenic 46,XX lines, revealing excellent agreement overall (SI Appendix, Fig. S7A). Of the 60 escapees we identified by allelic expression (lesser allele fraction, LAF ≥ 0.1, binomial P ≤ 0.05), 38 had previously been reported to escape XCI (59), and another nine variable escapee genes (“VarOther,” Fig. 3B) were previously identified in separate studies (60–64), including placenta-specific escapees IRAK1, MBTPS2, and SMS. The remaining 13 escapees that were overexpressed relative to 45,X samples comprised of four noncoding RNAs of unknown XCI status, and respectively, five and four genes that escape XCI in at least two 46,XX lines (POLA1, TMEM164, MBNL3, AMMECR1, and AMOT) or only the C19XX line (MID1, FAM199X, ELF4, and MIR503HG). Overall, these data indicate that aside from partial reactivation of a handful of genes in the C19XX line, and a few potentially TBL-specific novel escapees, all three 46,XX lines faithfully maintained XCI with at least 47/60 biallelic genes representing bona fide escapees.

Fig. 3.

X-chromosomal gene dosage in female- and male-derived TBL cells. (A) PCA of 45,X and isogenic euploid control TBL RNA-seq data. Respective symbols and colors indicate karyotype and cell line. (B) Median-vst normalized expression values for PAR, X-pair, and X-specific escapee genes, which are grouped by the independent escapee calls in all 46,XX lines, 2/3 or a single female euploid line (all, 2/3, 1/3). Escapee annotation panel indicates reported XCI states across multiple studies (“reported,” from ref. 59), as in SI Appendix, Fig. S7, and from refs. 60–64 for “VarOther”). Heatmap column annotation denotes hiPSC lines (XX6 values bordered in black; names of escapees silenced in XX6 bolded). Left-most barplot panels report LAF and three Centered barplot panels report log2-fold change in expression (Log2FC) in 45,X samples relative to isogenic euploid controls (XX19/23 or XX6, or male1XY). (C) Left: Autosomal-median normalized vst ratio of X-linked genes subject to XCI (“silenced”), relative to autosomal, X-specific escapees, PAR, and X-linked pair genes. Significant differences to female 45,X denoted by Mann–Whitney U test P value (≤0.1). Right: X:autosome ratios normalized by gene length (fpkm) compares each class of genes within each condition (see text), with median X:A ratio denoted next to each boxplot.

These allele-resolved mRNA-seq data also reveal that the active X (Xa) in XX19 and XX23 is the same X retained in all female-derived 45,X lines, whereas this copy was chosen as the Xi in the XX6 line (SI Appendix, Fig. S7 B and C). Curiously, seven escapees common to XX19 and XX23 were silenced on the Xi of the XX6 line, including bona fide escapees MXRA5 (65), PUDP (66, 67), STS (68), SMS (60), and STK26 (63), as well as AMMECR1 and AMOT (69). In standard differential expression, each of the XX19/23-specific escapees was also significantly lower in the XX6 line (p.adj ≤ 0.05, abs(log2FC) ≥ 0.3), whereas only XX6-specific escapee SMC1A was significantly higher (SI Appendix, Fig. S8A). We also found many PAR and X/Y pair genes to be significantly decreased in XX6 relative to XX19 and XX23 euploid lines. Because the relative difference in these groups of escapees between XX6 (log2FC panel and boxed differential vst samples) and 45,X samples (XO2/8/9) was less pronounced than between XX19/23 and 45,X lines (Fig. 3B and SI Appendix, Fig. S8A), it is plausible that escapees across the XX6 Xi were repressed as a whole. This is consistent with X hypermethylation in XX6 relative to XX19/23 lines (SI Appendix, Fig. S2), and higher XIST levels in XX6 hiPSCs that persist in TBLs (Fig. 1C and SI Appendix, Fig. S8B). In summary, while differential and allelic expression analyses demonstrate that XCI remained largely intact across all three female euploid TBL sets, XX6 TBLs revealed excessive repression of escapees across the Xi, relative to XX19 and XX23.

The observation that XX6 TBL clustered with 45,X cells and phenocopied their significant decrease in hCG and PlGF secretion (Figs. 2 and 3A) may reflect the consequences of their reduced escape from XCI, consistent with the haploinsufficient monosomy X hypothesis. We therefore grouped samples by karyotype and Xa identity throughout this analysis, in effect separating female euploid XX19/23 lines from XX6. Indeed, median expression of PAR, Pair, and X-specific escapees in XX19 and XX23 (purple), but not XX6 (lavender) is significantly higher than in female 45,X lines, which is also true for PAR genes in the male panel (Fig. 3C, normalized to autosomal median). Comparing X:autosome (X:A) ratios across gene categories using the fragments-per-kilobase-exon-per-million (fpkm) measure, we also found that genes subject to XCI are fully dosage compensated across male and female samples (X:A fpkm ratio ∼1). This result is consistent with the Xa hyperactivation hypothesis (70), which posits that single-copy X-linked genes evolved to match mean transcription of autosomal genes that are expressed from two alleles. Interestingly, PAR, Pair, and X-specific escapees are expressed at significantly higher levels (X:A fpkm ratio >1), even when present in single-copy in 45,X samples. This result suggests that genes that evolved to escape XCI also tend to be expressed from the Xa at well above median autosomal gene levels.

We then performed a systematic assessment of differentially expressed genes (DEGs), comparing: 1) isogenic XX19/23 female euploid and 45,X samples (“fXO”), 2) isogenic male1 euploid and 45,X (“m1XO”), and 3) male1 euploid to female XO samples (“fXO-m1XY”). Altogether over 5,000 genes were found to be differentially expressed (p.adj ≤ 0.05, abs(log2FC) ≥ 0.3) in at least two of these comparisons (Dataset S2). As expected, these DEGs clustered samples by karyotype, but also featured highly significant overlap and concordance in direction (Fig. 4A), including 936 concordant DEGs out of 1,283 common to all three comparisons (73% concordant, sign test P = 8.2 × 10⁻²⁸⁵). We next performed gene set enrichment analysis (GSEA), ranking genes by individual fXO, m1XO, or fXO-m1XY DESeq2 Wald test statistic, or an equally weighted mean Wald score (“aveXO”). As an additional control, we also ranked genes by comparing karyotypically identical 45,X samples across donors (“XOXO”). Expectedly, chrY and chrXp22 were recovered as significantly reduced in, respectively, male1 XY-relative and all comparisons, alongside other chromosomal region-specific enrichments (SI Appendix, Fig. S9). Among the computational Molecular Signatures Database (MSigDB) modules, the placental gene module38 was the top gene set reduced in all 45,X comparisons, and from a large human fetal single-cell RNA-seq dataset (71), three trophoblast-related gene sets are among the 15 most commonly reduced lineage-associated sets.

Fig. 4.

Transcriptome-wide impact of monosomy X by differential expression. (A) Median-vst normalized expression (“vst diff.”) of all differentially expressed genes (DEGs, DESeq2 p.adj ≤ 0.05, abs(log2FC) ≥ 0.3) in 45,X lines relative to their isogenic euploid controls (“fXO,” “m1XO”), and female-derived 45,X relative to male1 46,XY (“fXO-m1XY”). Number of overlapping and concordant DEGs identified in all three (Top) or any two comparisons listed as a fraction alongside sign-test P value, Left of DESeq2 Wald barplot panels. DEGs concordant across comparisons are annotated in dark gray, discordant in light gray. (B) GSEA against the Wikipathway collection for each comparison, as well as gene list reranked by the average Wald statistic from all three equally weighted sets (“aveXO”), and control comparison of male- and female-derived 45,X samples (“XOXO”). Bubble position, color, and size denote the signed log10-scaled GSEA p.adjust, normalized enrichment score, and number of core genes, respectively, and are plotted opposite of abbreviated Wikipathway titles. (C) Significant GO terms (beige) relevant to cilia, lysosomes, or autophagy and genes colored by m1XO Wald statistic (red for higher, blue for lower expression). D as in C for fXO Wald score-based GO terms.

Interestingly, the Wikipathway (Fig. 4B), Reactome (SI Appendix, Fig. S9), Hallmark, and other MSigDB collections (Dataset S3) point to impaired NRF2, cholesterol metabolism, and estrogen signaling, all of which are important for placental function (68, 72, 73). These terms were also significantly enriched in a recent transcriptome analysis of primary EVT and CTB (74), alongside gene sets relating to the cell cycle. Indeed, several of the significant terms reflecting increased expression in monosomy X related to the primary cilium (Fig. 4B: “ciliopathies,” “Joubert syndrome”; SI Appendix, Fig. S9: “anchoring of the basal body,” “BBSome-mediated cargo-targeting to cilium,” among others). The scale of this enrichment is clearly appreciable in the biological process (BP) and cellular component (CC) gene ontologies (GO), where proliferation-, splicing-, and translation-related categories are generally up-regulated in 45,X TBL cells, but the two top terms (BP: “cilium assembly,” “cilium organization,” and CC: “centriole” and “ciliary basal body”) represent the primary cilium by some margin (Fig. 4C and SI Appendix, Fig. S10). Among down-regulated gene sets, we find terms relating to the lysosome, autophagy, transmembrane transport, immunity, and lipid and steroid metabolic processes (SI Appendix, Figs. S9 and S10). The contrast between increased ciliary genes and decreased lysosomal genes in both male- and female-derived 45,X is particularly striking (Fig. 4 C and D). Recent reports show the primary cilium is important for proper human trophoblast invasion and migration (75, 76). The primary cilium also has an inverse relationship with NRF2 (77–79), NRF2 can activate PlGF expression directly (80), and NRF2 has been linked with birth outcomes in humans (72, 73) and mice (81, 82). Likewise, impaired autophagy has been implicated in recurrent and early miscarriage (83, 84).

To further clarify the relationships between these biological processes, we performed standard expression-trait correlation, and weighted coexpression network analyses (WGCNA) (85). First, we assessed how the levels of cell cycle (86) and cell type markers (87), as well as PAR, Pair, X-specific, and all combined escapees (“All”), correlated with hCG and PlGF secretion in matched RNA-seq and enzyme-linked immunosorbent assay (ELISA) experiments. “All” escapee, and PAR genes were highly and significantly correlated with hCG and PlGF secretion, as well as STB and EVT cell fates (Fig. 5A), whereas markers of cell cycle, CTB, and other proliferative cell fates correlated with each other. Unsurprisingly, STB and EVT cell fates anticorrelate with cell cycle and CTB markers, as STB and EVT arise from proliferative CTB but exit the cell cycle upon fusion (87, 88) and maturation (74, 89), respectively.

Fig. 5.

Correlation of gene modules sensitive to monosomy X and their preservation in primary placenta. (A) Significant (P ≤ 0.05) Spearman ρ coefficients relating escapee gene classes (All, PAR, X-pair, X-specific) to secreted PlGF and hCG levels, and to cycling (86), and cell type markers from early human embryonic studies (50, 51, 53, 87), suffixed by first author’s last initial (W/X/Z/C). Cell fates abbreviated for cyto-TB (CTB), early amnion (E-AM), epiblast (EPI), extravillous TB (EVT), mixed (MIX), primordial endoderm (PE), and syncytio-TB (STB). (B) Signed network from WGCNA plotted over module assignments, DESeq2 Wald stats (fXO, m1XO, fXO-m1XY, and aveXO; red >0, blue <0), and Pearson correlation with PlGF and hCG levels, and each median-normalized marker set (from A). (C) Z summary statistic for preservation of modules (from B) in another BAP dataset (90), (sex-stratified) first trimester chorionic villi sampling (66), and two studies of term placenta (91, 92), the latter further stratified by sex, birthweight, or randomized (“rand”) to control datasets of similar size. (D) Correlation coefficient (kME) of each escapee with its assigned module eigengene, and enrichment analysis (−log10-scaled Fisher P value, above heatmap) for module assignment of escapee classes (same genes and reported XCI annotation as in Fig. 3B).

Next, we performed WGCNA across all 32 BAP samples, which segregated by karyotype, except for XX6 samples that expectedly clustered with female-derived 45,X samples (SI Appendix, Fig. S11A). Plotted over Wald scores and gene-specific Pearson coefficients with hCG and PlGF levels (Fig. 5B), we observed a signed network of 17 modules that splits the transcriptome. Group1 genes (Right side) largely correlate with DEGs increasing in 45,X samples and anticorrelate with hCG and PlGF secretion, whereas group2 genes (Left side) correlate with DEGs decreasing in 45,X and increasing hCG and PlGF levels. Strikingly, group1 genes correlate with cell cycle and CTB markers and anticorrelate with STB and EVT markers, whereas the inverse is true for group2 genes. These data suggest that the network may be shaped by mutually exclusive cell fates (Fig. 5B and SI Appendix, Fig. S11B). To determine which specific modules were significantly driven by the contrast between euploid vs. X-monosomic expression, we quantified the degree and significance of their preservation in a subset of exclusively 45,X samples and a mixed karyotype control dataset of equal size. Preservation (Z) scores of modules representing over two-thirds of all genes show a decrease by 20 to 80 SDs in the 45,X-only dataset relative to the mixed karyotype set (SI Appendix, Fig. S11C). Correlating each module to traits of interest (Fig. 5A), we find the same set of modules (black, blue, brown, turquoise, and yellow) to be strongly anti- or co-correlated with each other and with euploidy, PAR expression, hCG, and PlGF secretion (SI Appendix, Fig. S11 B and D).

To test whether these modules are recovered in independent BAP and primary placental samples, we performed module preservation analysis against RNA-seq datasets from another hiPSC-based BAP study (90), first trimester chorionic villi samples (CVSs) (66), and two WGCNA studies on placental samples at term (91, 92). Remarkably, the same 45,X–euploid contrasting modules (SI Appendix, Fig. S11 C and D) are also moderately to highly preserved in primary placental samples, largely irrespective of fetal sex or birth weight categories (Fig. 5C). Because higher CTB and cycling markers correlated with monosomy X in our WGCNA modules (SI Appendix, Fig. S11D) and standard correlation analysis (Fig. 5A), we interpret this high level of preservation to reflect variable cell type composition (e.g., CTB vs. STB) that is inherent in the sampling of first trimester CVSs, and term placenta, and is commonly captured in WGCNA for any primary tissue (93). Here, in the context of our BAP model, gene modules preserved in CVS and placental samples may suggest that monosomy X hinders commitment to postmitotic STB and EVT cell fates or their survival, thereby increasing CTB marker representation and continued expression of cycling markers. Indeed, our WGCNA modules are most strongly preserved in the first-term trimester CVSs, when proliferation and cell fate commitment are likely even more variable than in term placenta (Fig. 5).

We also tested whether modules were overrepresented for gene sets assessed in the differential expression analysis. We recovered many similar terms (SI Appendix, Fig. S12 and Dataset S4) relating to cell cycle and primary cilium (turquoise, green), translation and regulation of autophagy (yellow), membrane-anchored signaling pathways (brown), as well as adipogenesis and cholesterol metabolism (blue). This indicates a strong overlap with enriched terms from differential expression overall (Fig. 4 and SI Appendix, Figs. S9 and S10), while providing module-level resolution of cellular functions (SI Appendix, Fig. S11) and lineage markers identified in single-nuclei BAP RNA-seq (94) and early embryonic studies (SI Appendix, Fig. S12A).

Finally, to implicate the dosage of specific X/Y-linked gene classes in TBL differentiation, we first tested which modules featured an overrepresentation of PAR, Pair, or X-specific escapees. The brown module was significantly enriched (P ≤ 0.05) for all escapees as one class (“all escape”), escapees without Y homolog (“X-specific”), and X-linked X/Y pair genes, whereas PAR genes were most overrepresented in the blue module. This latter module was of particular interest because it was also highly enriched for MSigDB’s placental gene module38, and human fetal EVT markers (Dataset S4), as well as for TE, STB, and EVT markers identified across early human embryonic studies (SI Appendix, Fig. S13A, p.adj < 0.1). In addition, the blue module correlated with hCG and PlGF levels (Fig. 5B), best reflected euploidy and escape from XCI (SI Appendix, Fig. S11D), and was most highly preserved in first trimester CVS and term placental RNA-seq samples, tied with the yellow and turquoise modules (Fig. 5C).

To identify potential X-linked drivers strongly associated with specific modules, we determined the correlation of each gene with the assigned module expression average (“eigengene”) across samples. We find that PAR gene ZBED1 is the top X/Y-linked hub gene in the blue module, ranking 90th of 1,979 genes (eigengene connectivity measure, kME = 0.94) and highly connected in the module overall (Fig. 5D and SI Appendix, Fig. S13B). Additionally, other PAR genes (GTPBP6, AKAP17A, PPP2R3B, and CD99 > 0.85 kME) ranked highly in this module, while escapees repressed in XX6 associated with the correlated black module (kME 0.93 to 0.84 for MXRA5, PUDP, SMS, and AMOT), and PRKX topped the X-pair genes in the brown module (0.95 kME, ranked 22nd/1,909 genes). In summary, our WGCNA analysis indicates that PAR expression most strongly reflects placental gene expression in this TBL model, which may help to prioritize a core set of X/Y-linked genes for follow-up in complementary in vitro models of human extraembryonic development.

Discussion

As a leading cause of spontaneous termination in humans (15), monosomy X can serve as a penetrant genetic model for miscarriage. Few genome-wide association studies on spontaneous or recurrent miscarriage have been published to date, in part due to the challenge of accounting for and excluding embryonic/fetal karyotypic changes (95–97). Characterizing the impact of monosomy X using in vitro human cell models may therefore provide a complementary approach toward implicating cellular functions and pathways in miscarriage.

The hiPSC BAP model (18) used herein has previously revealed sex-divergent expression patterns (44), addressing another important question in trophoblast biology (98, 99). In contrast, we applied this model to identify TBL cellular phenotypes and expression signatures that are common to the absence of the Xi or Y (rather than sex divergent), to better understand the consequences of this multigenic haploinsufficiency. We also rigorously validated the BAP model by comparing expression of markers identified across several independent early human and primate embryonic studies (50–53, 87), and found TE and STB markers to be the predominantly expressed lineage-associated gene sets in our experiments (SI Appendix, Fig. S6).

Importantly, we find significantly decreased secretion of hCG and PlGF from 45,X TBL cells compared to isogenic euploid controls across three independent isogenic panels (Fig. 2). Although differences in STB fusion index or the fraction of HLA-G⁺ cells did not rise to statistical significance (SI Appendix, Fig. S5), PGF and HLA-G transcript levels, respective markers of STB and EVT, were also reduced significantly in 45,X samples (Fig. 2). This is relevant because misregulation of HLA-G alone can result in miscarriage (100), and significantly lower PlGF levels have previously been reported in X-monosomic first trimester pregnancies (101, 102).

Two related insights emerged from global 45,X-associated expression changes: 1) Among significantly enriched gene sets undergoing concordant changes in male- and female-derived 45,X samples (Fig. 4), we find increased proliferation-associated terms (primary cilium, DNA replication, splicing, translation) and decreased terms related to maturing STB and EVT cellular functions (transmembrane transport, immune regulation, and metabolism), which included lysosomal processes like autophagy. 2) Likewise, standard correlation and WGCNA reveals cell cycle and CTB markers correlate with genes that increase in 45,X samples, whereas STB and EVT markers correlate and cluster with genes that decrease in 45,X samples (Fig. 5). Because both STB and EVT cells derive from CTB, but must exit the cell cycle upon fusion or maturation, the correlation between monosomy X and cell cycle/CTB markers (Fig. 5 A and B and SI Appendix, Fig. S11D) suggests 45,X TBL cells are still skewing toward actively cycling CTB at the end of the 8-d BAP differentiation, which may explain their lower secretion of hCG and PlGF (Fig. 2).

While the molecular basis of this delay in committing to STB or EVT cell fates remains unclear, our WGCNA indicates that PAR genes in general, and ZBED1 specifically, are strongly positively correlated and well connected inside the blue placental gene module (Fig. 5 and SI Appendix, Fig. S13). This module was also among the most preserved gene modules in RNA-seq studies of term placenta and especially first trimester CVSs (Fig. 5C), which are likewise heterogenous in CTB vs. STB/EVT contributions due to sampling. Intriguingly, ZBED1 does regulate proliferation (103) and is expressed in human placenta, with higher levels in postmitotic STB expression than CTB (104). Recently, partial knockdown of ZBED1 in the BeWo trophoblastic cell model was reported to increase apoptosis of syncytiated cells (105), which is consistent with the STB defect reported herein.

Our study highlights promising areas for follow-up in future in vitro work and study of primary samples. For example, it is unclear whether higher expression of primary cilia components merely reflects the ciliary resynthesis in actively cycling 45,X cells or altered function of this important signaling organelle. In contrast to mouse trophoblasts, human trophoblasts carry cilia (75, 76), and cilia regulate autophagy and NRF2 (79). Curiously, lysosomal genes were widely down-regulated in 45,X BAP samples (Fig. 4), and impaired autophagy has previously been implicated in recurrent miscarriage (83, 84, 106). Whether placental autophagy is deregulated in miscarried 45,X conceptuses specifically, and whether cell fate proportions are altered in such 45,X-associated CVS or placental samples, would therefore be of particular interest toward understanding why human monosomy X terminates early.

Materials and Methods

Reprogramming, iPSC Culture, and Validation.

Sex chromosomal mosaicism in starting fibroblasts (partial 45,X recorded in Coriell Biorepository) was confirmed by interphase DNA FISH using sex centromere (DXZ1/DYZ3) probes (WiCell). Fibroblasts were reprogrammed to hiPSCs using the CytoTune iPSC 2.0 Sendai Reprogramming kit (Thermo Fisher Scientific) and maintained by weekly mechanical passaging, as described (26). Prior to BAP differentiation, hiPSCs were transitioned to mTeSR media (Stem Cell Technologies), grown on extracellular matrix (Geltrex, Thermo Fisher), and passaged weekly with ethylenediaminetetraacetic acid. Standard IF was performed for pluripotency markers and H3K27me3, as well as FISH and qRT-PCR for XIST (SI Appendix, SI Methods include details).

Cytogenomic Analysis and DNA Methylation Profiling.

Cytogenomic analysis was performed at the UConn Center for Genome Innovation on the CytoSNP-850k v1.2 (Illumina) platform, at a CNV call resolution of 400 kb. DNA methylation analysis was performed as described (26) to query previously reported probes differentially methylated (DMP) during Xi erosion, alongside new XX6-specific DMP calls (P ≤ 0.05) using minfi (107). KS distances and KS test significance were calculated comparing the Gaussian probe densities across the X chromosome for DMP or background probe set). Change in DNA methylation (β-value) was calculated as the difference between the mean XX6 probe β- and the mean XX19/23 probe β-value.

DNA Sequencing and Phasing.

High molecular weight (HMW) genomic DNA was prepared by sarkosyl/proteinase-K/RNAseA digestion and phenol-chloroform extraction, followed by ethanol precipitation (SI Appendix, SI Methods). HMW-gDNA was sequenced to ∼30× coverage, following library preparation on the 10× Genomics linked-read platform at the UConn Center for Genome Innovation. LongRanger (v2.2.2, 10× Genomics) was used for read alignment and phasing of variants genome-wide (hg38). X-linked phased variants were supplied alongside RNA-seq data to obtain A and B allele counts for X chromosome genes using phASER (108).

Trophoblast Differentiation.

TBLs were differentiated from hiPSCs as described (18) with minor modifications. Briefly, confluent hiPSC clones cultured in mTESR were dissociated with Accutase and plated at 50,000 cells/well of a six-well plate in mTESR with 10 μM Y-27632 (Tocris Bioscience). After 1 d, media were changed to mouse embryonic fibroblast conditioned media (MEF-CM), supplemented with 8 ng/μL human basic FGF (Thermo Fisher Scientific) and 10 μM Y-27632. The following day, media were switched to BAP differentiation media, which consisted of MEF-CM with 10 ng/mL BMP4 (Peprotech), 1 μM A83-01 (Stem Cell Technologies), and 0.1 μM PD173074 (Stem Cell Technologies). Media were changed daily until day 8 when cells and supernatant were harvested for RNA collection, IF, or ELISA, respectively. For ELISA, supernatants were diluted 1:1,000 for the hCG ELISA (GenWay Biotech) and 1:50 for the PIGF ELISA (R&D Systems).

RNA Sequencing and Analysis.

RNA was extracted from hiPSCs using the PureLink RNA Mini Kit (Thermo Fisher Scientific). For 3′ mRNA-seq, libraries were prepared using the Quant-seq 3′ mRNA Library Prep Kit FWD (Lexogen) and single-end 75-bp reads were sequenced on the NextSeq 500 (Illumina). Genes queried for hiPSC characterization were listed on the TaqMan hPSC Scorecard Assay (Thermo Fisher Scientific). For standard mRNA-seq, libraries were prepared at the UConn Center for Genome Innovation using the Illumina Strand mRNA Kit and 100-bp paired-end reads were sequenced to an average depth of 40 million reads/replicate on the NovaSeq (Illumina).

Read pairs were trimmed, aligned to the human genome (hg38), and quantified against GENCODE version (v36). For escapee analysis based on phased linked read variants, A and B allele counts from phASER were tabulated and calls made by binomial test (LAF > 0.1, P ≤ 0.05) for all X-linked genes.

For differential expression using DESeq2 (109), count tables were filtered for genes with sufficient expression. Surrogate variables were estimated using the sva package (110), and added to the DESeq2 design. GSEA using clusterProfiler (111) was performed on all genes ranked by DESeq2’s Wald statistic in three separate conditions, as well as the average of their quantile-normalized Wald scores to ensure equal weighting. Weighted gene coexpression network analysis (WGCNA) was performed on vst counts, using the WGCNA package (85), as a signed hybrid network using the biweight midcorrelation raised to a soft thresholding power of 17 (scale-free topology fit ≥0.85). Modules were correlated to normalized hCG and PlGF ELISA values and to averaged early embryonic lineage marker sets, which were median vst normalized to ensure equal weights across all sets. Module preservation analysis was performed against published BAP (90), CVS (66), and term placenta (91, 92) RNA-seq datasets. Enrichment analysis of gene sets across WGCNA modules was performed with clusterProfiler (SI Appendix, SI Methods for details).

Data, Materials, and Software Availability

RNA-seq, CytoSNP850k, and methylEPIC array data are deposited under superseries accession no. GSE207114 in the Gene Expression Omnibus (GEO) (112).