Female iPSC X-chromosome inactivation (XCI) erosion and its transcriptomic effects during CRISPR gene editing and neural differentiation

SUMMARY

Human induced pluripotent stem cells (hiPSC) and iPSC-differentiated neural cells, in combination with CRISPR editing, are commonly used for studying neurodevelopmental and other brain disorders. Female iPSCs undergo random X-chromosome inactivation (XCI) via epigenetic silencing by noncoding X inactive specific transcript (XIST). It is known that female iPSCs may lose XIST expression, leading to XCI erosion that affects both X-linked and autosomal gene expression. However, the effects of CRSIPR editing and neural differentiation on XCI erosion in iPSC-derived neurons and how this may confound a real-world transcriptomic analysis of differentially expressed genes (DEGs) are poorly understood. Here, leveraging bulk RNA-seq of hundreds of CRISPR-edited female iPSC lines from four donor lines for 66 genes and single-cell RNA-seq of iPSC-derived neurons of a subset of 42 edited genes, we investigated the effects of XCI erosion during CRISPR editing and in iPSC-derived neurons. We found that XCI erosion was variable in CRISPR-edited female iPSCs and largely preserved in iPSC-derived neurons. Like in iPSCs, XIST in neurons predominately influenced the expression of X-linked genes; however, its effect on autosomal genes was more pronounced in single neurons. Mechanistically, XIST epigenetically causes allelic imbalance of both X-linked and autosomal genes, with the former showing stronger allele-specific expression (ASE) bias. Notably, XIST-induced ASE bias exhibited a conserved positional pattern at loci affecting neurodevelopmental genes across different female lines and cell types. Finally, we demonstrated a confounding effect of XCI erosion on DEG analyses in iPSC-derived neurons. These results have significant implications in hiPSC modeling of neurodevelopmental and other brain disorders.

INTRODUCTION

Female cells (X/X) are well known to undergo random X-chromosome inactivation (XCI), resulting in dosage compensation between the sexes. The key regulator of XCI is a long non-coding RNA called X-chromosome inactive specific transcript (XIST). It coats one of the X-chromosomes and induces transcriptional silencing by recruitment of various RNA-binding proteins and chromatin modifiers^1–3. This permanent epigenetic change occurs during the blastocyst stage at approximately 4–6 days after fertilization in humans and is maintained in all somatic cells. Although XCI in females is permanent in vivo, XIST expression in cultured human embryonic stem cells (hESCs) and induced pluripotent stem cells (hiPSCs) has been reported to be highly dynamic. Studies have shown that hiPSCs and hESCs often gradually lose XIST expression after a certain number of cell passages or under specific conditions, leading to XCI erosion^4–11. Recent studies in naive human pluripotent cells and in hiPSCs have shown that XIST not only represses X-linked genes as expected, but also spreads a repressive effect to some autosomal regions^12,13. XCI erosion in females thus has profound effects on iPSC reprogramming, cell differentiation, and disease modeling in the context of widely used CRISPR/Cas9 gene editing in iPSCs.

While XCI erosion is well known in iPSCs, the effects of genome editing and cellular differentiation on XCI erosion dynamics remain poorly understood. In fact, CRISPR-mediated gene editing of iPSCs is a lengthy process spanning multiple cellular passages, and differentiation of iPSC cells into different cell types may further alter XIST expression, potentially influencing the regulation of X-linked and autosomal genes. Emerging evidence suggests that gene reactivation associated with XCI erosion in hiPSCs can persist through trilineage and cardiomyocytes differentiation¹⁴. Whether similar stabilization occurs during neuronal differentiation, particularly in the context of genome editing, remains unclear. Given that hiPSC-differentiated neural and glial models are increasingly used for studying neurodevelopmental and neurodegenerative disorders^15–21, it is imperative to understand XCI erosion dynamics in hiPSCs during CRISPR gene editing and in neuron/glia differentiation and to determine which X-linked and/or autosomal genes expression are affected, as both the occurrence of erosion and the degree to which genes are reactivated or silenced can vary across cells and genes.

While global effects of XCI erosion on X-linked and autosomal genes are commonly ascertained by examining average expression levels of X-linked or autosomal genes¹², allele-specific expression (ASE; i.e., allelic imbalance of gene expression) analysis²² has been used to identify specific genes that may be influenced by XIST^13,14. ASE analysis allows for the detection of imbalanced expression between the two alleles at heterozygous single nucleotide polymorphism (SNP) sites, providing insights into cis-regulatory mechanisms. Recent studies have shown that a modest decrease in XIST expression increases biallelic expression (i.e., less ASE bias) of X-linked genes^13,23. However, the effect of XCI erosion on ASE bias of X-linked and autosomal genes in iPSC-derived neurons has not been systematically investigated. Moreover, whether the genomic loci showing XIST-induced ASE biases are conserved across different genetic backgrounds remains unknown.

Recently, as part of the Scalable and Systematic Neurobiology of Psychiatric and Neurodevelopmental Disorder Risk Genes (SSPsyGene) consortium, we have been utilizing CRISPR-based cytosine base editors (CBEs) to establish loss-of-function (LoF) mutagenesis by introducing a premature stop codon (changing “C” to “T” in DNAs) for a large number of neurodevelopmental and psychiatric disorder (NPD)-associated genes in iPSCs from multiple donor lines^19,24,25. The edited isogenic lines and the non-targeting control (NTC) lines (unedited) are differentiated into both excitatory and inhibitory neurons¹⁹. These CRISPR-edited iPSC lines for many LoF genes and the respective iPSC-derived neurons provide a unique opportunity for assessing XCI erosion dynamics during CRISPR editing of iPSCs and neuronal differentiation. Here, we investigated the effects of XCI erosion in CRISPR-edited female iPSCs for 66 genes across six donor lines (including 4 female lines) and in iPSC-derived excitatory and inhibitory neurons¹⁹ for 42 genes along with their matched unedited control lines. We assayed the transcriptomic profiles of iPSC lines and their derived neurons co-cultured with mouse astrocytes by using bulk RNA-seq and single-cell RNA-seq (scRNA-seq)²⁶ (Figure 1A, Supplementary Table 1, see methods), respectively. We found that XCI erosion was widespread in CRISPR-edited female iPSC lines and the erosion status of XIST expression was largely maintained in iPSC-derived neurons. Like in iPSCs, XIST predominantly influenced the expression of X-linked genes in iPSC-derived neurons; however, the repressive transcriptional effect of XIST on autosomal genes was more pronounced in single neurons and in the allelic imbalance of expression. We also showed that XIST-induced ASE bias on various loci was conserved across different female donor lines and between iPSCs and neurons, likely affecting certain neurodevelopmental genes. Finally, we demonstrated the non-negligible confounding effects of XCI erosion on a real-world differential gene expression analysis in iPSC-derived neurons.

Figure 1: RNA-seq reveals that XCI erosion in female iPSC lines is preserved during iN differentiation.

(A) Schematic of experimental design. Six donor lines were used for CRISPR-based DNA editing of 66 target genes to introduce a premature stop codon (iSTOP) in batches (~23 genes and one donor line per batch). The isogenic iPSC lines for 42 edited genes (both edited and unedited) were differentiated into excitatory and inhibitory neurons and co-cultured with mouse astrocytes. The iPSC and iPSC-derived neuron/glia co-cultures were subject to bulk RNA-seq and scRNA-seq, respectively. (B) Representative immunofluorescence staining of iPSC mutant lines for pluripotent stem cell markers (NANOG, SSEA4 and OCT-4). DAPI, nuclei. Scale bar: 100μm. (C) Box plot showing expression of XIST (counts per million reads, CPM) for different editing batches. Each point represents an RNA-seq sample of each line for each batch. Dot color indicates different donor line. The percentage of XIST⁺ samples per donor line is indicated in parenthesis. (D) PCA plot of RNA samples of different isogenic iPSC. Color scheme of each dot is the same as in (C). (E) XIST expression in different isogenic iPSC lines in CW20107 cell lines of the same batch. Non-targeting control (NTC; line V4043) is shown in red. (F) Immunofluorescence staining of the iPSC-differentiated neurons of day 35. MAP2 staining is for neural dendrites. tdTomato staining is for excitatory neurons (infected by AAV-tdTomato before co-cultured with inhibitory neurons). Scale bar: 50μm. (G) Violin Plot showing XIST expression in excitatory (Ex) and Inhibitory (Inh) neurons of different isogenic lines for one example batch using CW20217 as a donor iPSC line for iSTOP base editing. Non-targeting control (NTC; line V4043 and its repeat sample V4043R) is shown in red.

RESULTS

XCI erosion is widespread in CRISPR-edited female iPSC lines and largely preserved during neuron differentiation

We first examined the XIST expression changes in female iPSC lines after CRISPR DNA base editing. We analyzed bulk RNA-seq data of 297 iPSC lines derived from 6 donor lines (4 female and 2 male), including both the edited and unedited control lines. Immunostaining of pluripotency markers (NANOG, SSEA4, and OCT4) of the CRISPR-edited isogenic lines confirmed their pluripotency following gene editing (Figure 1B). The edited iPSCs from different donor lines exhibited a donor-line-specific pattern of XIST expression, with more CD14 isogenic lines showing XIST expression compared to the other female lines (Figure 1C). Principal component analysis (PCA) showed no separation of samples based on source cell line, indicating that they remain transcriptome-wide similar to each other despite variable XCI erosion across all female lines (Figure 1D). We observed varying levels of XCI erosion across CRISPR-edited isogenic lines of the same donors (Figure 1E, Supplementary Figure 1A). These iPSC lines had cell passage numbers ranging from 19 to 26; however, there was no significant difference in XIST expression between different passages, even for the same donor line (e.g., CW20107) (Supplementary Figure 2A). From comparing NTCs (i.e., iPSCs which went through the same editing process as other LoF mutant lines but with no base editing reagents added) of the same donor line, we observed that editing did not significantly affect XIST expression (Supplementary Figure 2B). Thus, the variable XCI erosion across the CRISPR-edited lines from the same female donor line can likely be attributed to heterogeneity of XIST expression in the source iPSC population.

We next conducted XIST erosion analysis in excitatory and inhibitory neurons differentiated from a subset of these iPSC lines (n=234 lines from 6 donor lines; two were males for comparison purpose) (Figure 1F). Successful differentiation of neurons was confirmed by immunostaining for MAP2, a pan-neuronal marker (Figure 1F). scRNA-seq of these neurons confirmed that in most cases, expression of XIST in female iPSCs was preserved in neurons differentiated from these cells; conversely, when XIST was not expressed in iPSCs, the derived neurons also lacked XIST expression (Figure 1E, G; Supplementary Figure 1). However, in some instances, we did observe that neurons differentiated from the iPSC lines without XIST expression (e.g., V4502 in batch 3) regained XIST expression (Supplementary Figure 1), suggesting a stochastic nature of XIST expression regulation during neuron differentiation from iPSCs.

To examine whether XIST expression levels may influence iPSC differentiation into neurons, we calculated pluripotency scores of these iPSC lines from their RNA-seq data and examined their neuronal differentiation capability. Although we found that higher XIST expression appeared to be significantly associated with lower pluripotency scores, the differences of pluripotency scores were very subtle (from 0.95 to 0.94) (Supplementary Figure 2C). Likewise, we observed no correlation between XIST expression and arbitrarily defined “difficulty” scores of neuron differentiation (Supplementary Figure 2D).

XCI erosion elevates X-linked and mitochondrial gene expression in CRISPR-edited iPSCs

In human pluripotent stem cells, XIST has been shown to repress the expression of X-linked genes and induce gene expression dampening across specific autosomal regions¹². Here, to test for the effect of XCI erosion on X-linked genes and a broader transcriptional effect in iPSC lines during CRISPR editing, we first compared the mean expression of X-linked and autosomal genes among female iPSC lines (with or without XIST expression, XIST⁺ or XIST^-) compared to that of male iPSC lines. As expected, we found that the XIST^- female iPSC lines had higher expression of X-linked genes compared to XIST⁺ female lines and male lines, confirming the effect of XCI erosion for the XIST^- group (Figure 2A, Supplementary Figure 2E, 2G). Interestingly, we also observed a subtle increase of X-linked gene expression in XIST⁺ female lines vs. male lines, suggesting a possibly incomplete XCI in those XIST⁺ female lines (Figure 2A). However, for autosomal genes, in contrast to prior study of hiPSCs¹², we found lower expression of autosomal genes in XIST^- female iPSC lines compared to the other groups, though, it was a subtle difference of expression (Figure 2B, Supplementary Figure 2F, H).

Figure 2: Transcriptomic effects of XIST erosion in iPSCs.

(A-B) Box-and-whisker plots comparing the expression of (A) X-linked genes and (B) autosomal genes between male, XIST⁺ females, and XIST^- females. The box represents the interquartile range (IQR) with the median shown as the central line. Whiskers extend to the furthest data points within 1.5 × IQR from the quartiles. Each dot represents an isogenic cell line. (C-D) Scatter plots showing the correlation of XIST expression in a cell line to average expression of (C) X-linked genes and (D) autosomal genes. CPM, counts per million reads. (E) Heatmap showing the expression levels (normalized to male) of the reported XCI-escape genes (N = 48) in female isogenic iPSC lines with different XCI status. PAR, pseudoautosomal regions of X-chromosome. (F) Box-and-whisker plot comparing the expression of mitochondrial (MT) genes (encoded by MT genome) between male, XIST⁺ females, and XIST^- females. Box represents interquartile range and whisker represents 1.5 × IQR from the quartiles. Each dot represents an isogenic cell line.

To further corroborate the above-described effects of XIST expression on X-linked and autosomal genes, we performed linear regression analysis on XIST⁺ isogenic lines using XIST expression as a predictor and average gene expression as the outcome. We found a modest negative correlation between XIST expression and X-linked gene expression, with XIST expression explaining ~20% of the variance in X-linked gene expression (R² = 0.21, p < 0.01) (Figure 2C). In contrast, no significant association was found between XIST and autosomal gene expression (R² = 0.04, p > 0.09) (Figure 2D). Together, our data shows that the regulatory effect of XIST is largely specific to X-chromosome genes in CRISPR-edited iPSCs.

Studies have shown that over 15% of X-linked genes escape XIST-mediated inactivation and remain transcriptionally activated even on the inactive X-chromosome across human tissues^23,27,28. We hypothesized that the absence of XIST may also lead to increased expression of these XCI-escaping genes in our iPSC lines. To investigate this, we examined the expression pattern of known XCI-escaping genes across different XIST⁺ and XIST^- isogenic lines (Figure 2E). Sex chromosomes include PARs (pseudoautosomal regions), which are terminal regions of the chromosomes that are autosome-like in that these homologous sequences at the tips of the X- and Y- chromosomes are recombined during meiosis²⁹. Consistent with a previous study, we found that genes in the PARs showed higher expression in males compared to females, suggesting that the presence or absence of XCI is unable to overcome male-biased expression²³. In non-PARs, we see that most of the known XCI-escaping genes indeed show robust expression even in XIST⁺ iPSC lines, which is consistent with what is reported across tissues in humans. Furthermore, for many of these XCI-escaping genes, XCI erosion in XIST^- samples further increased their expression (Figure 2E). Together, these results suggest that the effects of XCI erosion are primarily detectable in non-PARs, with limited impact on sex-biased expression in the PARs, further supporting the predominant effect of XCI erosion on the expression of X-linked genes rather than autosomal genes in our CRISPR-edited iPSCs.

Emerging evidence also suggests that XIST may lead to mitochondrial (MT) dysfunctions^30,31. We thus examined whether the loss of XIST is associated with transcriptional changes in MT gene expression (n = 33 MT genes). We observed that MT gene expression was lowest in XIST⁺ female lines, and significantly increased in XIST^- female lines. Notably, MT gene expression levels in XIST^- females were comparable to those in males (Figure 2F). This pattern is consistent with a recent study reporting a sex specific bias for heart MT gene expression³², where protein coding MT genes were expressed in higher levels in males compared to females in mice. These findings suggest that the loss of XIST could lead to de-repression of MT genes in females, resulting in expression levels comparable to those observed in males.

XCI erosion in iPSCs has a stronger effect on allelic imbalance of expression of X-linked genes than autosomal genes

Our observed lack of a significant effect of XCI erosion on average expression level of autosomal genes may be due to the insensitivity of using average gene expression in detecting more subtle effects of XCI erosion. ASE analysis²², by directly comparing the RNA-seq counts of the two alleles of an expressed heterozygous SNP within the same sample, provides a sensitive way to identify target genes regulated by cis-regulatory genetic variants or epigenetic factors such as XIST (Figure 3A). XCI erosion is expected to make an XIST-affected gene to shift from ASE towards biallelic expression (Figure 3A). We thus analyzed SNP ASE patterns across both X-linked and autosomal genes in XIST^- and XIST⁺ female lines. To ensure robust estimation of allelic expression, we restricted the analysis to highly expressed heterozygous SNPs (i.e., total read count ≥ 5 times the number of samples per group; see Methods) for X-linked and autosomal chromosomes. For each donor line, we matched the number of XIST⁺ and XIST^- samples, prioritizing keeping the total RNA-seq read depth balanced between groups to ensure comparability. To assess how XIST expression influences allelic expression, we categorized the SNPs as biallelic or exhibiting allelic ASE based on the results of a binomial test. We observed substantial overlap in ASE SNPs between XIST⁺ and XIST^- samples, regardless of chromosome type (Figure 3B). This suggests that most SNPs with allelic bias are not uniquely associated with XIST status, although the magnitude of allelic imbalance at shared SNPs may differ between XIST⁺ and XIST^- lines. However, when focusing specifically on SNPs classified as biallelic in XIST^- samples, we found that a large proportion of these X-linked SNPs became allelically biased in XIST⁺ samples (Figure 3C). This shift toward allelically biased expression in the presence of XIST is consistent with re-establishment of XCI, reflecting the fact that XIST⁺ cells maintain a partially inactive X-chromosome that still contributes low-level transcription. In contrast, only a small proportion of autosomal biallelic SNPs in XIST^- samples transitioned to ASE in XIST⁺ samples (Figure 3C), which was consistent with our observed predominant effect of XIST expression on X-linked genes (Figure 2).

Figure 3: Allelic imbalance of gene expression caused by XCI erosion in female iPSC lines.

(A) Schematic representation of biallelic (paternal and maternal alleles) and allele-specific expression (ASE) at a SNP site. (B) Bar plot showing overlap of ASE SNPs in XIST+ and XIST-samples for different female iPSC donor lines. (C) Bar plot showing proportion of biallelic SNPs that change to ASE in the presence of XIST expression in different cell lines. (D) Density plot showing the distribution of ASE ratio for X-linked and autosomal SNPs across XIST- and XIST+ in iPSC line CD14. (E) Scatter plot to compare the degree of allelic imbalance between XIST+ and XIST- samples in iPSC line CD14. Each dot represents one SNP. ASE ratio, the proportion of reference allele counts of all sequencing reads at a SNP site. (F) Pearson correlation heatmap comparing ASE ratios (ref Count / total Count) for each SNP across different XCI statuses of isogenic cell lines derived from donor line CD14. (G) Positional scatter plot showing the distribution of XIST+/XIST-ASE ratio for SNPs across the X-chromosome of iPSC line CD14. Horizontal red dashed line highlights SNPs with XIST+/XIST-ASE ratio < 0.75 or > 1.25 whereas vertical red dashed box highlights the 70–80Mb hot spot of allelic imbalance. (H) Venn diagram showing overlap of genes containing SNPs with ASE ratio (XIST+/XIST-) < 0.75 or > 1.25 across cell lines. (I) Gene ontology (GO) enrichment (biological processes) for genes with XIST+ ASE SNPs overlapping between all cell lines. Top 10 enriched GO terms are listed for overlapping X- linked genes (n = 25) and autosomal genes (n = 339) from (H).

To more quantitatively ascertain ASE differences of X-linked and autosomal genes between XIST^- and XIST⁺ groups, we first examined the distribution of the ASE ratios (sum of reference allele count / total allele count) within each donor line (Figure 3D, Supplementary Figure 3A). We found that for the X-chromosome, XIST⁺ groups revealed a higher density of monoallelic SNPs (ASE ratios closer to 0) compared to the XIST^- groups, consistent with XCI. Conversely, XIST^- groups displayed a shift towards biallelic SNPs (minor allelic ratios closer to 0.5), reflecting XCI erosion. Across all four female lines, XCI erosion showed a consistent pattern of having a much weaker but detectable effect on ASE bias of autosomal SNPs compared to SNPs in X-linked genes (Figure 3D, Supplementary Figure 3A). We next examined the ASE ratio of individual SNPs to compare the degree of allelic imbalance between XIST⁺ and XIST^- samples. For X-linked SNPs, although allelic ratios between XIST⁺ and XIST^- samples remained moderately correlated (Pearson’s R² = 0.73), there was increased deviation from the diagonal (y = x) in both directions, indicating heterogeneous allelic shifts consistent with XCI erosion across the X-chromosome in XIST⁺ samples. In contrast, the ASE ratios of autosomal SNPs exhibited stronger correlation (Pearson’s R² = 0.83) between XIST⁺ and XIST^- groups, reflecting the smaller impact of XCI erosion on autosomal gene expression (Figure 3E, Supplementary Figure 3B).

To evaluate global differences in allelic expression between XIST⁺ and XIST^- samples, we computed the pairwise Pearson correlation of ASE ratios of highly expressed SNPs between different iPSC lines derived from the same donor line. Higher correlations were consistently observed within the same XIST status group (XIST⁺ or XIST^-) for X-linked SNPs compared to those between XIST status groups (Figure 3F, Supplementary Figure 3C), consistent with the above-mentioned effect of XIST erosion on ASE bias. The clear separation of the XIST⁺ and XIST^- groups also indicated that the iPSC donor line used for CRISPR editing was clonal (i.e., the same paternal or maternal allele showing bias across all derivative iPSC lines), which otherwise would not show strong correlation of SNP ASE bias across different isogenic lines of the same donor line. Compared to X-linked SNPs, correlations of autosomal SNP ASE ratios between XIST⁺ and XIST^- groups did not show clear separation (Figure 3F, Supplementary Figure 3D), again indicating smaller global effect of XCI erosion on autosomal genes. Together, these results illustrate that XCI erosion in XIST^- CRISPR-edited iPSC lines alter ASE of both X-linked and autosomal genes, but with a much stronger effect on X-linked genes.

Leveraging our SNP ASE data on multiple donor lines, we next investigated whether the SNPs with ASE bias influenced by XIST erosion were restricted to specific loci across different genetic backgrounds. We plotted the ratio of ASE values between XIST⁺ and XIST^- samples for each SNP along the X-chromosome. A ratio of 1 indicates equal ASE between the two cell groups, whereas deviations reflect differential allelic expression potentially due to XCI. We found that most loci had an ASE ratio close to 1, indicating balanced allelic expression (Figure 3G, Supplementary Figure 3D). Notably, although loci showing strong ASE bias appeared to be evenly distributed across the X-chromosome, the pattern of ASE bias seems to be similar across different donor lines, suggesting that specific genomic regions of the X-chromosome were influenced by XCI erosion (Figure 3G, Supplementary Figure 3D). To further identify genes in those regions that are consistently affected by XIST-associated allelic regulation across different donor lines, we selected SNPs showing at least a 25% difference in allelic expression between XIST⁺ and XIST^- samples (XIST⁺/XIST^- ASE ratio < 0.75 or >1.25) to their corresponding genes for all source cell lines (Supplementary Tables 2, 3). We found that among genes retaining those strong ASE SNPs, 25 (out of 202) X-linked genes were shared across all female cell lines and 339 (out of 3,875) autosomal genes were shared across all female cell lines (Figure 3H). The gene ontology (GO) term enrichment analysis revealed that the XIST-influenced X-linked genes were enriched for biological processes related to TORC2 signaling, positive regulation of phospholipid biosynthesis, and acetyl-CoA metabolism (Figure 3I), indicating possible effects on lipid homeostasis, membrane dynamics, and stress response. For XIST-influenced autosomal genes, the enriched GO-terms included interleukin-23-mediated signaling, leukocyte proliferation, and cytoplasmic translation (Figure 3I), suggesting possible alterations in immune-related signaling and the machinery involved in protein synthesis.

XCI erosion in single neurons affects the expression of both X-linked and autosomal genes

There has been a lack of systematic investigation of the functional effects of XIST erosion in iPSC-derived neurons, especially at single cell resolution. As part of the SSPsyGene effort¹⁹, we analyzed scRNA-seq data of 121,317 excitatory (SLC17A6⁺) and 69,022 inhibitory (GAD1⁺) neurons derived from iPSC lines of 42 CRISPR-edited LoF alleles and unedited (i.e., NTC) controls for 4 female and 2 male donor lines (Figure 4A, B). We found that most female cell lines showed variable neuronal expression of XIST, with the majority of neurons lacking detectable XIST (XIST⁺: 79,788 cells, XIST^-: 110,551 cells) (Figure 4C, Figure 1G, Supplementary Figure 1B).

Figure 4: Transcriptomic effects of XCI erosion in female iPSC-derived neurons.

(A) UMAP projection of snRNA-seq data of excitatory and inhibitory neurons of 225 isogenic iPSC lines derived from four female donor lines. (B) Feature plots for single cell gene expression of cell type specific markers (SLC17A6 for excitatory and GAD1 for inhibitory). (C) Feature plots for single cell XIST gene expression. (D) Pearson’s correlation of single-cell XIST log-normalized expression and average log-normalized expression of X-linked genes (left) and autosomal genes (right) in excitatory neurons. Each dot is a single cell. Dot color indicates different cell lines. (E) Average expression levels of X-linked in excitatory (left) and inhibitory (right) neurons between XIST- female, XIST+ female, and male lines. Box represents the IQR and whiskers represent 1.5*IQR from the quartiles. Wilcoxson p-value is given. (F) Heatmap showing the expression levels (normalized to male) of the reported XCI-escape genes (N = 74) in excitatory neurons of female isogenic iPSC lines with different XCI status. PAR, pseudoautosomal regions of X-chromosome.

Leveraging our scRNA-seq data, we first examined at single-cell resolution whether XCI erosion affects the expression of X-linked and autosomal genes. By analyzing single cells with detectable XIST expression, we found that XIST expression showed a moderate effect (R² = 0.31, p < 0.01) on X-linked gene expression and a lesser effect (R² = 0.17, p < 0.01) on autosomal gene expression in excitatory neurons (Figure 4D). XIST expression in inhibitory neurons showed a weaker correlation with X-linked genes (R² = 0.20, p < 0.01) but a similar level of correlation with autosomal genes (R² = 0.18, p < 0.01) (Supplementary Figure 4A), suggesting a possible neuron subtype-specific effect of XCI erosion in single cells.

We further analyzed the transcriptomic effects of XCI erosion by comparing pseudobulk RNA-expression levels of X-linked and autosomal genes across XIST⁺ and XIST^- female lines as well as male lines. Like in iPSCs, for both excitatory and inhibitory neurons, we observed higher expression of X-linked genes, but not autosomal genes, in XIST^- female lines (vs. XIST⁺ lines) (Figure 4E, Supplementary Figure 4B). As opposed to a significant effect of XIST erosion on MT gene expression in iPSCs (Figure 2F), we did not observe an effect of XIST expression on MT genes in neurons (Supplementary Figure 4C). Lastly, we examined the expression of known XCI-escaping genes on the X-chromosome in both the inhibitory and excitatory neurons (Figure 4F, Supplementary Figure 4D). We found that unlike in iPSCs (Figure 2F), most known XCI-escaping genes did not “escape” from XCI in XIST⁺ female lines, showing relatively low expression (normalized by the expression in male lines) in induced neurons (Figure 4F, Supplementary Figure 4D). However, like in iPSC, XIST^- neurons did exhibit increased expression of these XCI-escaping genes in the nonPAR, suggesting that XCI erosion still contributes to the de-repression of XCI-escaping genes in neurons (Figure 4F, Supplementary Figure 4D).

Taken together, these results suggest that although XCI erosion does not seem to affect average autosomal gene expression, it is associated with modest transcriptional changes of both X-linked and autosomal genes in single neurons with a stronger effect on X-linked genes.

XCI erosion in iPSC-derived neurons affects ASE bias of neurodevelopmental genes

Given the observed effects of XCI erosion on ASE bias in iPSC (Figure 3), we also examined whether similar ASE bias can be seen in iPSC-derived excitatory and inhibitory neurons. We first categorized individual cells from each gene’s LoF mutant line into XIST⁺ and XIST^- groups based on detectable levels of XIST transcripts. This classification was performed at the single-cell level to ensure accurate representation of XIST status across heterogeneous neuronal populations. To minimize the effect of confounding variables, comparable sequencing depths and an equal number of samples were selected for the XIST⁺ and XIST^- groups. Next, to assess the influence of XIST expression on allelic bias, we categorized the expressed heterozygous SNPs into biallelic and ASE groups. Like in iPSC, we found that the neuronal ASE SNPs largely overlapped between the XIST⁺ and XIST- groups across all cell lines and both neuronal subtypes (Figure 5A), and a large proportion of X-linked biallelic SNPs in XIST-samples became allelically biased in XIST⁺ samples (Figure 5B). In contrast, a smaller proportion of autosomal biallelic SNPs in XIST^- samples transitioned to ASE in XIST⁺ samples (Figure 5B). Also like in iPSCs (Figure 3D), density plots of allelic ratios for highly expressed SNPs on the X-chromosome in both types of neurons showed a higher density of monoallelic SNPs (ASE ratios closer to 0) for XIST⁺ groups compared XIST^- groups (Figure 5C, Supplementary Figure 5A). In contrast, for autosomal SNPs, although XIST⁺ neurons also exhibited a more pronounced monoallelic expression pattern (centered around allelic ratio 0.1), both the XIST⁺ and XIST^- groups had a bimodal distribution of allelic ratios, with the prominent peak around an allele ratio of 0.5 (i.e., biallelic expression) (Figure 5C, Supplementary Figure 5A). Altogether, these results suggest much stronger transcriptional effects of XCI erosion on X-linked genes compared to autosomal genes in iPSC-derived neurons.

Figure 5: Allelic imbalance of gene expression caused by XCI erosion in female iPSC-derived neurons.

(A) Bar plot showing overlap of ASE SNPs in XIST⁺ and XIST^- excitatory (left) and inhibitory (right) neurons for different female iPSC donor lines. (B) Bar plot showing proportion of biallelic SNPs that change to ASE in the presence of XIST expression in excitatory (left) and inhibitory (right) neurons for different female iPSC donor lines. (C) Density plot showing the distribution of the ASE ratio for X-linked (top) and autosomal (bottom) SNPs across XIST^- and XIST⁺ in excitatory (left) and inhibitory (right) neurons derived from donor line CD14. (D) Heatmap showing Pearson’s correlation of SNP ASE ratios (ref Count / total Count) of X-linked (top) and autosomal (bottom) genes in excitatory (left) and inhibitory (right) neurons between isogenic lines with different XIST status (⁺ or ^-) derived from the female donor line CD14. (E) Positional scatter plot showing the distribution of XIST⁺/XIST^- ASE ratio for SNPs across X-chromosome of the female donor line CD14. Horizontal red dashed line highlights SNPs with XIST⁺/XIST^- ASE ratio < 0.75 or > 1.25 whereas vertical red dashed box highlights the 70–80Mb hot spot of allelic imbalance. (F) Venn diagram showing overlap of genes containing SNPs with ASE ratio (XIST⁺/XIST^-) < 0.75 or > 1.25 for X-linked (top) and autosomal (bottom) genes in excitatory (left) and inhibitory (right) neurons across cell lines. NDD genes overlapping in three or more cell lines are indicated by arrows. (G) Gene ontology (GO) enrichment (biological processes) for genes containing neuronal XIST⁺ ASE SNPs overlapping between at least two donor lines. Top 10 enriched GO terms are listed for the overlapping X-linked genes (not available for inhibitory neurons) and autosomal genes from (G).

We also computed the pairwise correlation of ASE ratios to evaluate the global differences in ASE bias between XIST⁺ and XIST^- samples within each cell line and neuronal subtype. In contrast to iPSC data, we did not find evidence of clustering between the same XIST status group even for the X-linked genes (Figure 5D, Supplementary Figure 6), suggesting a weaker impact of XCI erosion in neurons than in iPSC.

To investigate positional patterns of XCI-affected SNP ASE bias in neurons across different donors, we plotted the ratio of XIST⁺ and XIST^- ASE values for X-chromosome SNPs. Consistent with iPSC cell lines (Figure 3G), most SNPs were clustered around a ratio of 1, indicating comparable allelic expression between groups (Figure 5E, Supplementary Figure 5B). Despite a smaller number of called SNPs from scRNA-seq (10× Genomics’ 3’-capture method) compared to bulk RNA-seq with iPSCs, we were able to identify a putative “hot spot” where the expressed SNPs with allelic imbalance were clustered (70–80 Mb) on the X-chromosome in excitatory neurons across all donor lines (Figure 5E, Supplementary Figure 5B, Supplementary Table 6). Interestingly, this hot spot region also encompassed XIST. Looking at the same region in iPSC cells, we identified SNPs associated with 12 genes (DLG3, SLC7A3, ZMYM3, PIN4, RPS4X, FTX, JPX, TTC3P1, NONO, TAF1, SEPHS1P4, RPS6P26) (Figure 3G, Supplementary Figure 3D, Supplementary Figure 5C, Supplementary Table 6). This suggests that XIST-associated silencing may be confined to specific regions of the X-chromosome.

To systematically identify genes consistently influenced by XIST across different donor lines in iPSC-derived neurons, we mapped ASE SNPs (XIST⁺/XIST^- ASE < 0.75 or > 1.25) to their corresponding genes. Among X-linked genes, 13 genes for excitatory neurons and 2 genes (FTX, NAP1L3) for inhibitory neurons were identified in at least 2 cell lines of which 1 gene (TTC3P1) was consistently found across all cell lines in excitatory neurons (Figure 5F, Supplementary Table 5). TTC3P1 was among 8 genes (AL135749.6, HDAC8, MIR325HG, ATRX, NLGN3, FTX, JPX, TTC3P1) that were present in the hot spot region in the X-chromosome of excitatory neurons. For autosomal genes, 313 genes for excitatory neurons and 63 genes for inhibitory neurons were present in at least 2 cell lines, where 25 genes for excitatory neurons and 3 genes (BX470102.3, CRABP1, S100A6) for inhibitory neurons were shared across all cell lines (Figure 5F). When we combine the neuronal types, 22 of 332 autosomal genes and 5 of 14 X-linked genes have been implicated in neurodevelopmental disorders (Supplementary Tables 4, 5). Furthermore, these genes were overrepresented in biological processes related to postsynaptic signaling, brain morphogenesis, and CNS neuronal development (Figure 5G). These observations raise the possibility that XIST erosion is associated with altered regulation of neurodevelopmental genes, although direct functional effects remain to be determined.

XCI erosion in iPSC-derived neurons has an effect on common transcriptomic differential expression analysis

Given our observed effects of XCI erosion on ASE bias of both X-linked and autosomal genes in iPSC-derived neurons, we attempted to examine how variable XIST expression across samples may affect an empirical transcriptomic-wide differential expression (DE) analysis. Using the pseudobulk RNA expression values in our scRNA-data of iPSC-derived neurons from all four female-donor isogenic lines, we performed a linear regression analysis to identify XIST-associated gene expression, regressing out the effects of the LoF allele and cell line (see methods). We only identified 52 and 12 X-linked genes affected by XIST expression in excitatory and inhibitory neurons, respectively, where 9 genes were shared between the two neuronal subtypes (Figure 6A, 6B, Supplementary Table 7). The lack of XIST-associated autosomal genes was consistent with our above-described much weaker effect of XCI erosion on autosomal genes. Among the XIST-associated genes, we found a subset of genes implicated in neurodevelopmental disorders (NDD): 7 genes (ALG13, CASK, GDI1, HNRNPH2, PAK3, EIF2S3, ZC4H2) in excitatory neurons and 3 genes (PAK3, DCX, ALG13) in inhibitory neurons. Consistent with the gene set enrichment for iPSCs (Figure 3I), the enriched biological processes for these neuronal genes perturbed by XIST included TORC2 signaling, pyrimidine biosynthesis, heat acclimation, and regulation of acyl-CoA biosynthetic processes (Figure 6C).

Figure 6: Impact of neuronal XIST expression on transcriptome-wide differential expression analysis.

(A) Volcano plot showing X-linked genes with expression impacted by XIST expression based on standard linear regression. Genes with adjusted p-value < 0.05 are highlighted in red (upregulated) or blue (downregulated), of which NDD genes are labeled. (B) Upset plot showing XIST-associated genes from (A) in both cell types. (C) GO enrichment (biological process) for X-linked genes affected by XIST expression. (D) Pearson correlation of log₂FC of gene expression, calculated using DESEq2 differential expression analysis for each iSTOP edited gene with and without XIST expression as a covariate. (E) Jaccard similarity index calculated using intersection of differentially expressed genes (p-value < 0.05 and log₂FC < or > 1) identified using DESeq2 associated with each iSTOP edited gene with and without XIST expression as a covariate. (F) Comparison of autosomal and X-linked iSTOP edited genes based on the absolute log₂FC differences (with XIST vs without XIST expression as a covariate) of non-overlapping DEGs (p-value < 0.05 and log₂FC < −1 or > 1) with Jaccard similarity index < 0.8. Box represents IQR and whisker represents 1.5 * IQR. (G) Scatter plot showing correlation between log₂FC (with XIST vs without XIST expression as a covariate) for non-overlapping DEGs (p-value < 0.05 and log₂FC < −1 or > 1) with Jaccard similarity index < 0.8.

We next examined to what extent XIST erosion may confound any differentially expressed genes (DEGs) associated with LoF alleles. Using DESeq2³³ to perform pseudobulk DEG analysis between neurons of isogenic lines for each LoF mutation with and without including XIST expression as a covariate. Because of the small number of DEGs that have an adjusted p value < 0.05, for comparison purpose we used a relaxed DEG cut-off (nominal p value < 0.05 and FC > 2). We found that the log₂FC of the DEGs for each LoF line, with or without XIST expression as a covariate, highly correlated (Pearson’s R > 0.85) (Figure 6D). The list of DEGs with or without XIST expression as a covariate was also mostly overlapping; however, there were still about 20% of LoF genes showing a lower Jaccard similarity index between 0.5 to 0.75 (Figure 6E). As expected from our observed stronger effect of XIST on X-linked genes, the non-overlapping X-linked DEGs (with XIST vs. without XIST as a covariate) exhibited a larger FC difference compared to those non-overlapping autosomal DEGs (Figure 6F, Supplementary Table 8). Notably, despite a sizable number of non-overlapping DEGs, their log₂FCs across all assayed LoF genes with or without using XIST expression as a covariate were highly correlated (Pearson’s R = 0.98) (Figure 6G). Altogether, these results demonstrate a subtle but significant confounding effect of XCI erosion on common DEG analysis, highlighting the need of accounting for variable XIST expression across samples for data interpretation.

DISCUSSION

XIST is a central regulator of X-chromosome dosage compensation in females, ensuring balanced gene expression between sexes. However, the effects of CRISPR editing on XIST expression in iPSCs and iPSC-differentiated neurons are not well characterized, limiting our understanding of X-linked regulation in these cellular models for brain disorders. Here, we systematically investigated a large number of CRISPR-edited iPSC lines of multiple donors and iPSC-derived neurons, providing an in-depth view of how gene editing and neuron differentiation influence XCI erosion in female cells. We also examined the XIST-induced ASE bias on specific X-linked and autosomal genes in both iPSCs and neurons, highlighting an epigenetic regulatory mechanism that may be conserved across different genetic backgrounds and during neuron differentiation from iPSCs.

Our study demonstrates that XCI erosion observed in iPSCs largely persists in differentiated neurons. Consistent with these findings, Raposo et al. (2025) reported that XCI erosion in hiPSCs is maintained following differentiation into multiple lineages, including trilineage commitment and cardiomyocytes¹⁴. However, we did observe a small number of instances in which isogenic lines that expressed XIST in the iPSC state lost expression in induced neurons, or conversely gained XIST expression following neuronal differentiation, suggesting a possible stochastic nature of XCI in iPSCs that may be influenced by CRISPR editing and neural differentiation. Nonetheless, the overall persistence of XCI erosion in CRISPR-edited iPSCs and iPSC-derived neurons has important implications for disease modeling using iPSCs, as abnormal XIST expression has been associated with multiple diseases including autoimmune disorders and cancer and may influence the expression of NDD genes^34–36. Supporting this, Topa et al. (2024) specifically focused on NDD genes and identified 17 X-linked genes affected by XIST in iPSCs and 10 in iPSC-derived sensory neurons, of which three genes overlapped between the two contexts¹³. In our induced neurons, four of these X-linked NDD genes (PAK3, DCX, EIF2S3, ALG13) were affected, with EIF2S3 and DCX showing specificity to excitatory and inhibitory neurons, respectively.

XCI erosion has been reported to occur spontaneously with increasing passage number^2,10,37. Contrary to these studies, we did not observe XCI erosion due to increased passage number, likely due to the narrow range (19–26) of passage numbers in our study. Alternatively, as Briggs et al. (2015) reported, cells at late passage may be more resilient to XIST loss¹⁰. Thus, differences in XIST expression and ASE observed in our study are unlikely to be influenced by passage number, allowing us to focus on XIST-dependent transcriptional effects. As XIST coats the X-chromosome to enforce transcriptional repression, XIST^- females showed the expected elevation of X-linked expression in both iPSCs and neurons. Notably, this increase was more pronounced in iPSCs, suggesting X-linked gene activation is partially constrained in differentiated neurons. Our analyses of known XCI-escaping gene expression and ASE bias in neurons also supported a stronger effect of XCI erosion in iPSCs than in neurons. However, despite weaker effects of XCI erosion in neurons, we observed a localized “hot spot” (70–80 Mb) on the X chromosome in excitatory neurons and iPSCs that exhibited allelic imbalance (Figure 3G, 5E, Supplementary Figures 3D, 5B). This region encompassed the non-coding genes FTX and JPX ^35,38,39, which are known regulators of XIST activation, and overlapped the X-chromosome inactivation center region spanning Xq13-q21. Together, these findings support a model in which XIST loss leads to modest, gene-specific relaxation of X-linked repression in neurons.

Compared to the known effect of XCI erosion on X-linked genes in female iPSCs, its influence on autosomal genes has not been well established, especially in iPSC-derived neurons. Previous studies have shown XIST in female hESCs and hiPSCs spreads its repressive effect to some autosomal regions^12,13. On the contrary, we did not observe significant repressive effects of XIST on autosomal gene expression by comparing bulk RNA levels between XIST⁺ and XIST^- iPSC lines or neurons. However, we did find modest negative expression correlation of XIST and autosomal genes in single neurons expressing variable levels of XIST. Effects of XCI erosion were also observed for autosomal genes in both iPSC and neurons from our ASE bias analysis by comparing XIST⁺ and XIST^- samples. Despite a much weaker effect of XIST expression on autosomal genes than X-linked genes, our DEG analysis using commonly used pipelines in iPSC-derived neurons supported a potential confounding effect of XIST expression on autosomal genes, highlighting the need to consider female XIST expression as a covariate in DEG analysis.

Our study has some limitations. Although we had large cohort of CRISPR-edited iPSC lines and their matching neuronal samples, only 4 female donor lines were included. Additional female donor lines would allow us to gain a more comprehensive view of the genetic loci exhibiting XIST-induced allelic imbalance of expression of both X-linked and autosomal genes. Moreover, our iPSC lines had a relatively narrow range of late cell passages (19–26); including unedited NTC lines of early passages and assaying more time points of neural differentiation would help reveal more of the dynamics of XCI erosion and its stage-specific effects on X-linked and autosomal genes. Nevertheless, our study represents the first extensive evaluation of XCI dynamics and its transcriptional impacts on X-linked and autosomal genes in the same set of CRISPR-edited iPSC lines and neurons. The observed persistent XCI erosion in iPSC-derived neurons and its confounding effect on commonly used DEG analyses have significant implications in improving the design and interpretation of hiPSC-based disease modeling of neurodevelopmental and other brain disorders.

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Jubao Duan (jduan@uchicago.edu).

Materials availability

The hiPSC lines will be made available as part of the SSPsyGene Consortium to fulfill the NIMH (National Institute of Mental Health) material/data-sharing commitment.

Data and code availability

All the reported data and code used during analysis, including hiPSC eSNP-Karyotyping results are deposited at https://doi.10.5281/zenodo.18776733. The RNA-seq data’s GEO accession numbers are GSE262442 and GSE322198.

Supplementary Material

LIST OF SUPPLEMENTARY MATERIALS

Methods

Supplementary Text

Figures: S1 to S6

Tables S1 to S8

References (1-28)