Female naïve human pluripotent stem cells carry X chromosomes with Xa-like and Xi-like folding conformations

Abstract

Three-dimensional (3D) genomics shows immense promise for studying X chromosome inactivation (XCI) by interrogating changes to the X chromosomes’ 3D states. Here, we sought to characterize the 3D state of the X chromosome in naïve and primed human pluripotent stem cells (hPSCs). Using chromatin tracing, we analyzed X chromosome folding conformations in these cells with megabase genomic resolution. X chromosomes in female naïve hPSCs exhibit folding conformations similar to the active X chromosome (Xa) and the inactive X chromosome (Xi) in somatic cells. However, naïve X chromosomes do not exhibit the chromatin compaction typically associated with these somatic X chromosome states. In H7 naïve human embryonic stem cells, XIST accumulation observed on damaged X chromosomes demonstrates the potential for naïve hPSCs to activate XCI-related mechanisms. Overall, our findings provide insight into the X chromosome status of naïve hPSCs with a single-chromosome resolution and are critical in understanding the unique epigenetic regulation in early embryonic cells.

Chromatin tracing was used at single-chromosome resolution to define the 3D state of X chromosomes in human stem cells.

INTRODUCTION

Embryonic stem cells (ESCs) are derived from the inner cell mass (ICM) of the blastocyst and present key features of early embryonic cells (1). In particular, female murine ESCs (mESCs) have both X chromosomes in an active (Xa) state, one of which undergoes the canonically described process of X chromosome inactivation (XCI) upon differentiation (1). Thus, female mESCs have provided a robust model system to investigate the mechanisms of XCI. In human ESCs (hESCs), X chromosome status varies according to the cell lines and culture conditions examined (2–4). Primed hESCs are derived from the epiblast and exhibit “primed” features, pluripotent features which closely resemble those of mESCs derived from postimplantation epiblasts (1). In the past decade, hESCs and human induced pluripotent stem cells (hiPSCs) have been derived into the naïve state, which represents the preimplantation epiblast. More optimized medium conditions that better support the naïve identity have been reported. These conditions use human leukemia inhibitory factor; extracellular signal–regulated kinase inhibition; glycogen synthase kinase 3β (GSK-3β) inhibition, in combination with protein kinase C inhibitor (t2iLGö) (5, 6); or BRAF and SRC inhibitors (5iLA) (7). Naïve human pluripotent stem cells (hPSCs) cultured under these conditions demonstrate a high transcriptional similarity to murine naïve PSCs and the human early embryo, as well as demonstrate similar patterns of global DNA hypomethylation (1, 8). Critically, many studies report that naïve X chromosomes are largely biallelically active under these 5iLA or t2iLGö conditions (6, 8–13).

There has been a great deal of interest and success in defining the X chromosome status in naïve human hPSCs (6, 8–10, 12–17). However, our incomplete understanding of human XCI has stymied attempts to resolve X chromosome status in pluripotent cells. In human preimplantation embryos, XIST is transcribed from both X chromosomes and can coat one, both, or neither X chromosome in a given cell (10, 11, 18). In naïve hPSCs, XIST status appears to be similarly heterogeneous, which may explain the inconsistency observed in reports of this topic (6, 8, 10). Cells with two XIST clouds, similar to those in early-stage embryos (11, 18), are observed in naïve hPSCs at proportions as different as 0 to 95% of a population (6, 9, 10, 19). Furthermore, different researchers have also observed naïve female hPSCs that show no XIST clouds (6, 9, 10). As is the case with XIST, the in vitro enrichment of H3K27me3, another canonical marker of XCI (15), on the naïve X exhibits heterogeneity in the literature (6, 9, 10, 13, 19), despite evidence from preimplantation human embryos suggesting an H3K27me3-null landscape (10, 18). In light of the heterogeneous landscape of XCI-related readouts reported in the literature, determining the exact state of naïve female X chromosomes and what heterogeneity is normally present within that state becomes quite difficult with traditional methods of assessing XCI status.

Recent studies on human XCI regulation have leveraged single-cell transcriptome methodologies to assess the allelic output of the human transcriptome during the earliest stages of development (10, 11, 20). However, most current single-cell RNA sequencing datasets experience high levels of gene dropout and a high level of noise because of large-scale mRNA amplification (21), which can ultimately result in data that may be more open to interpretation. In the context of XCI, changes to the analytical pipeline can shift most X-linked single-nucleotide polymorphism calls from biallelic expression to monoallelic expression, thereby affecting researchers’ interpretation of X status (9, 11, 20, 22, 23).

To avoid these aforementioned complications in transcriptomically assessing the naïve female X chromosome, we leveraged recently developed orthogonal methods in three-dimensional (3D) genomics and spatial biology (24–32). We performed chromatin tracing (25) in naïve and primed cells, targeting the center 100 kb region of 40 selected topologically associated domains (TADs) across the whole X chromosome. We determined that, although naïve cells typically exhibit X chromosomes with a spectrum of Xa-like and Xi-like 3D structures, these structured chromosomes lack differences in compactness that are canonically associated with these same structural changes in somatic X chromosomes. Naïve hPSCs, when cultured under the tested conditions, appear to have initiated some level of 3D conformational change to their X chromosome structure. However, naïve hPSCs have not initiated the transcriptional silencing and compaction processes that are canonically associated with these 3D changes in the context of XCI.

RESULTS

5iLA naïve hPSC lines are derived from the primed source lines and characterized for naïve pluripotency

To derive naïve cells, we converted primed hPSCs in a manner used previously (7–10). Naïve hPSCs were derived from primed hPSCs (H9 hESC, H7 hESC, HES3 hESC, and IMR90 iPSC lines) over the course of 4 weeks with an additional 2 to 3 weeks of culture to expand cells for downstream experimentation (Fig. 1A). Primed hPSCs cultured in mTeSR1 (33, 34) were passaged and cultured on mouse embryonic fibroblasts (MEFs) before replacing the primed culture medium with naïve culture medium; in this study, we used the 5iLA medium developed previously (7, 8). The cells were successively passaged as single cells and small clumps of 2 to 10 cells until a dome-shaped morphology typical for naïve hPSCs was achieved (Fig. 1A). In most cell lines, an early wave of cell death was observed as primed cells adjusted under naïve medium conditions, as noted by others (7–9).

Fig. 1. The 3D structures of X chromosomes of naïve cells derived under 5iLA conditions are different from those in primed cells.

(A) Schematic of deriving naïve hPSCs from primed hPSCs. Primed hPSCs were adjusted to MEF culture before a 3- to 6-week derivation and stabilization period. (B) Schematic of assaying X chromosome (ChrX) 3D structure using chromatin tracing. A representative nucleus with two copies of X chromosome is shown for primary probe hybridization. Scale bar, 10 μm. We bleached the dye-labeled primary probes and introduced dye-labeled secondary probes targeting two TADs, imaged with 647 (cyan) and 561 nm (magenta) illuminations and then bleached. This process was repeated for all 20 rounds of sequential hybridization and imaging. Representative images of TAD foci and the reconstructed chromatin trace were shown for the chromosome copy in the yellow dashed box. Scale bar, 5 μm. Illustration was drawn using BioRender. (C) Mean spatial distance matrix and domain boundary analysis of the X chromosome folding conformations for all tested primed and naïve 5iLA hPSC lines as well as IMR90 somatic Xa and Xi chromosomes. The matrices display the population average of the distance between each pair of TADs, with red representing smaller distances and blue representing further distances. Genomic positions of centromere (yellow), XIST (green), DXZ4 macrosatellite (blue), and FIRRE (pink) are pointed by arrowheads above the matrix. The yellow line on each matrix depicts the superdomains with boundaries determined by the boundary scores. Positions of the centromere and DXZ4 macrosatellite are indicated (black arrowheads) on the boundary score graphs. The matrices for IMR90 somatic Xa and Xi were reproduced with permission from the previous work (25). A.U., arbitrary units.

To confirm the acquisition of naïve pluripotency, we characterized hPSCs via immunofluorescence and quantitative reverse transcription polymerase chain reaction (qRT-PCR). Immunofluorescence was performed for the canonical pluripotency markers OCT4 (fig. S1), NANOG (fig. S2), and TRA-1-60 (fig. S3), as well as the naïve-specific cell surface marker Sushi Domain Containing 2 (SUSD2) (fig. S4) (35–37). As previously noted in the literature, naïve hPSCs derived under 5iLA conditions exhibit low (38) to generally positive (39) levels of TRA-1-60 expression on the cell surface, while primed hPSCs display heterogeneous yet generally positive expression of this mark (fig. S3), quite similar to previous studies (37, 40–49). Conversely, primed hPSCs displayed negligible expression of SUSD2 compared to naïve hPSCs (fig. S4). Naïve and primed hPSCs were further characterized via qRT-PCR for canonical pluripotency marks OCT4 and NANOG, as well as naïve-specific markers TFCP2L1 and REX1 (fig. S5A). 5iLA naïve cells expressed significantly higher NANOG, OCT4, TFCP2L1, and REX1 than the primed cells across most of the lines tested. We also performed karyotyping on these lines to evaluate the presence of genomic aberrations. Whereas primed hPSCs showed normal karyotypes, aberrant karyotypes were detected in a part of naïve hPSCs as reported previously (fig. S5B) (50).

Chromatin tracing reconstructs the 3D X chromosome structure

Previous work has shown that the X chromosome epigenetic state is prone to disruption under conventional primed hPSC culture conditions (51, 52). The progressive loss of silencing marks and chromosome compaction associated with XCI is known as XCI erosion. This ultimately leads to the abnormal transcription of Xi genes from decompacted and derepressed regions (14, 15, 53). To characterize the status of XCI erosion in the X chromosomes of our cells, we stained primed and 5iLA naïve hPSCs for H3K27me3, a core repressive histone mark deposited during XCI that is highly sensitive to XCI erosion (fig. S6) (15). The hPSCs in the primed state were devoid of H3K27me3 nuclear foci, suggesting the eroded state of their Xi (15, 54, 55). On the contrary, 5iLA naïve hPSCs all displayed robust subpopulations of cells with H3K27me3 foci when compared to their primed counterparts (fig. S6). As reported before (6, 9, 13), H3K27me3 accumulation has been observed in naïve hPSCs derived from primed cells and blastocysts. However, this repressive histone mark is not sufficient to impose the suppressive effects exhibited on XCI (56). Neither the presence of XIST RNA clouds nor the presence of the H3K27me3 mark on the X chromosome seem sufficient to prevent biallelic X-linked expression in naïve hPSCs (6, 8–10, 13).

As the canonical diagnostic marks of XCI do not correlate with the transcriptional activity in the naïve state at the global level, we sought alternative (nontranscriptomic) methods to interrogate the X chromosome state. Recently, there has been great success in characterizing large-scale elements of the 3D structure of the X chromosome in mammals through the application of 3D genomics approaches (24, 26, 30–32). We applied a highly multiplexed DNA fluorescence in situ hybridization (FISH) method, termed chromatin tracing (25, 57, 58), and defined the 3D folding conformation of X chromosomes at a TAD-to-chromosome length scale. We selected a total of 40 TADs spanning the genomic length of the entire X chromosome at roughly equal intervals (i.e., about 4 Mb genomic resolution) and sequentially mapped them in 3D space. We then generated matrices of the mean spatial distances between pairs of TADs for each different cell state and cell line (Fig. 1B). The spatial distance matrices confirmed that X chromosomes are folded into multimegabase superdomains (Fig. 1C).

Chromatin tracing revealed that naïve and primed hPSCs exhibited major differences in their 3D conformation near the centromere and the DXZ4 macrosatellite. All 5iLA naïve cell lines showed a superdomain boundary at the centromere (Fig. 1C, left black arrowheads at boundary score). Conversely, all primed hPSC lines lacked this feature but exhibited local minimums in superdomain boundary score at the centromere (Fig. 1C, left black arrowheads). Furthermore, all 5iLA cell lines showed relatively more robust boundary signals for a superdomain boundary at the DXZ4 macrosatellite, while this boundary is attenuated in the primed cells (Fig. 1C, right black arrowheads). Previous studies (25, 26, 30–32) identified a DXZ4-located macrodomain boundary as a marker of the Xi (Fig. 1C). A previous publication using this same chromatin tracing strategy on the X chromosome in female somatic cells identified a macrodomain boundary on the Xa near the centromere (Fig. 1C) (25), quite similar to the one observed in our 5iLA cells. Both features, which are more strongly observed in naïve cells, appear to be attenuated in the primed cells (Fig. 1C), perhaps because of XCI erosion.

Recently, multiple groups have tested modifications to the 5iLA naïve hPSC culture to improve the resulting naïve cell quality. Fibroblast growth factor 2 (FGF2) is a common component of naïve medium formulations (7, 59), yet it appears dispensable for the stable maintenance of naïve pluripotency (8). To ameliorate the genomic instability noted in hPSCs, one of the modified 5iLA medium conditions, called 4iLAF + tMEKi in this paper, leverages FGF2 supplementation to counterbalance a titrated reduction in mitogen-activated protein kinase (MAPK) kinase (MEK) inhibition (50). Ideally, this would titrate the MAPK activity to achieve a karyotypically stable culture. However, a recent publication reported that the FGF2 addition, such as in the so-called 5iLAF medium (7), may perturb the X chromosome status of naïve hPSCs (19). To determine whether adding FGF2 or reducing MEK inhibition have any effect on the X chromosome conformation of naïve hPSCs, we derived cells under these conditions. Derivation of naïve cells under both modified medium conditions was attempted in H9, HES3, and IMR90 primed iPSCs. 5iLAF naïve hPSCs were successfully derived in all three cell lines. The 4iLAF + tMEKi state was similarly successful in H9 and HES3 using 0.5 μM PD0325901 (50). In IMR90 iPSCs, however, low MEK inhibitor concentration was insufficient to derive stable naïve iPSC line in four independent derivation trials.

We performed the previously described immunofluorescence and qRT-PCR experiments to confirm that these cells had acquired naïve pluripotency (figs. S1 to S5A) and epigenetic status (fig. S6), similar to 5iLA naïve hPSCs. We then performed chromatin tracing and downstream analysis of the X chromosome folding conformation (fig. S7). Naïve hPSCs cultured under these FGF2-inclusive conditions showed X chromosome 3D structures highly similar to those under the previously characterized 5iLA conditions, including the superdomain boundary peaks at the centromere and DXZ4. Along with the findings from previously characterized 5iLA naïve hPSCs, these data suggest that naïve cells derived with or without exogenous FGF2 supplementation exhibit canonical 3D chromatin conformation features of both Xa and Xi chromosomes.

Population analysis of 3D chromosome structure finds Xa-like and Xi-like subpopulations of X chromosomes in both naïve and primed hPSCs

From chromatin tracing data, we identified two separate superdomain boundary markers in our naïve stem cell X chromosomes that are suggestive of a mixture of distinct 3D conformations (Fig. 1C). Hence, we set out to characterize the potential heterogeneity of X chromosome structures in naïve and primed hPSCs (Fig. 2). We first tried to combine XIST RNA FISH with chromatin tracing for the X chromosome. In our preliminary experiments, however, H9 near-euploid naïve cell line exhibited a relatively low expression of XIST (fig. S8A). Thus, XIST RNA FISH could not be used as an effective method to compare inferred X status. In primed cells, XIST was not expressed sufficiently to be meaningfully observed in this study, most likely because of the erosion of XCI (10, 15). Therefore, to distinguish Xa-like and Xi-like conformations in hPSCs, we compared these conformations to known Xa and Xi folding conformations from IMR90 somatic cells measured in a previous chromatin tracing study (25). We first performed Louvain-Jaccard clustering (60) of the folding conformations from IMR90 somatic cells by using each X chromosome trace as a high-dimensional data entry and the distances between pairs of TADs in each chromosome trace as variables. The clustering results and known Xa and Xi identities are displayed onto a 2D t-distributed stochastic neighbor embedding (t-SNE) plot (fig. S8, B and C) (61). We found that X chromosome traces from IMR90 somatic cells formed two major clusters, cluster 1 and cluster 2 (fig. S8B), which consisted of mainly Xa or Xi traces, respectively (fig. S8, C and D). Therefore, this unsupervised clustering can distinguish and separate Xa and Xi traces on the basis of their 3D conformations alone.

Fig. 2. Primed and naïve hPSCs have Xa-like and Xi-like chromosome conformations.

(A and B) Louvain-Jaccard clustering results of individual X chromosome traces of IMR90 primed iPSCs (A) and naïve 5iLA iPSCs (B) displayed with t-SNE plots. Chromosome trace clusters were identified, represented by different pseudocolors. (C and D) Mean spatial distance matrix and compartmentalization analysis for the two major chromosome trace clusters, cluster 1 (C) and cluster 2 (D), identified in IMR90 primed iPSCs in (A). The compartment scores determined the assignment of TADs into macrodomains (red or blue bars). The positions of the DXZ4 macrosatellite and centromere are pointed by black arrows. (E) Mean spatial distance matrix and compartmentalization analysis for the three major chromosome trace clusters identified in IMR90 naïve iPSCs in (B). Assignment of TADs to macrodomains (red or blue bars), positions of the DXZ4 macrosatellite and centromere (black arrows), and the p and q arms (black lines) are indicated. (F) Compartment score for IMR90 somatic Xa or Xi chromosomes. The assignment of TADs to macrodomains (red or blue bars), positions of the DXZ4 macrosatellite and centromere (black arrows), and the p and q arm (black lines) are indicated. The graphs were reproduced with permission from the previous work (25). (G to I) Radius of gyration of all X chromosome traces of IMR90 primed iPSCs and the two major chromosome trace clusters (G), IMR90 naïve iPSCs and the three chromosome trace clusters (H), and IMR90 primed Xa-like and IMR90 naïve Xa-like clusters (I). P values from unpaired t test are shown between clusters.

Next, to segregate hPSC chromosomes into the Xa-like and Xi-like 3D conformations in an unbiased manner, we performed Louvain-Jaccard clustering with the primed or 5iLA naïve chromosomes of IMR90 iPSCs (Fig. 2, A and B). Chromosome traces in different clusters were analyzed separately. We generated mean spatial distance matrices of different clusters and applied a previously established compartmentalization analysis pipeline (25) to identify Xa- or Xi-specific compartmentalization signatures or macrodomain organizations (Fig. 2, C to E). Critically, we found that different clusters corresponded to either Xa-like or Xi-like conformations. The primed cells analyzed here were found to contain two major clusters (cluster 1 and cluster 2) of X chromosomes (Fig. 2A). Primed cluster 1 chromosomes displayed two macrodomains with a boundary close to the DXZ4 sequence in their traces (Fig. 2C), which is highly similar to the Xi conformations identified in previous studies (Fig. 2F) (25, 26, 30–32). Primed cluster 2 chromosomes displayed a macrodomain boundary close to the centromere in their traces (Fig. 2D), which is highly similar to previously reported Xa organizations (Fig. 2F) (25). However, note that unlike the macrodomain boundaries that directly coincide with DXZ4 and the centromere in somatic cells (Fig. 2F) (25, 26), the macrodomain boundaries in the primed cells are shifted. This shift is consistent with the analysis presented beforehand, as the superdomain boundary scores at the centromere and DXZ4 positions are relatively low in primed cells (Fig. 1C). In 5iLA naïve cells, which contained three clusters based on the unsupervised clustering (Fig. 2B), two of the clusters, cluster 1 and cluster 3, showed the Xa-like macrodomains with a centromeric boundary, while the cluster 2 had a boundary near DXZ4 to separate the Xi-like macrodomains (Fig. 2E). These analyses further support that both primed and naïve X chromosomes contain traces with either Xa-like or Xi-like conformations.

Although the boundary position is defined by population-averaged analysis, we set out to quantify the variation of the macrodomain organization in individual chromosome copies. To do so, we calculated the polarization index between the two macrodomains as previously described (25). A higher polarization index of a chromosome copy indicates that the two macrodomains are more spatially segregated and polarized. From the distribution of the polarization index in each cluster (fig. S9), we see that, compared to the primed cells in matching Xa-like or Xi-like clusters, the naïve X chromosomes tend to have higher variations in their macrodomain polarization index. This indicates that the level of macrodomain segregation varies more in naïve X chromosomes.

We noted that the boundary near DXZ4 could move around nearby TADs (Fig. 2, C and E). As DXZ4 is a large tandem repeat with a number of CTCF binding sites, we searched the nearby TADs associated with the shifted boundary positions for genomic elements that fit these criteria. We identified a region composed of ~4 kilo–base pair units, including microsatellite [(AAAT)n, (TCTTTC)n, and (CCT)n], Alu (AluSc8 and AluSx3), and MER33 elements; this region also constitutes the CT47A gene cluster (CT47A1 to CT47A12) (fig. S10A). We found that this region was bound by CTCF in chromatin immunoprecipitation data from our H7 5iLA cells. In addition, this region is more conserved in primates than in other vertebrate lineages (fig. S10B). These findings raise questions about the potential roles other conserved, CTCF-enriched, repetitive elements could play in orchestrating the 3D structure of the X chromosome.

Given that we have determined Xa-like or Xi-like conformation for each chromosome on the basis of the clustering results, we then mapped each chromosome back to their corresponding individual cells to interrogate the X chromosome state at the single-cell level in IMR90 iPSCs (fig. S11, A and B). A representative image showed that single cells can have pairs of X chromosomes from all three combinations of ChrX states: Xa-like/Xa-like, Xa-like/Xi-like, and Xi-like/Xi-like (fig. S11A). We quantified the percentage of cells that carried each combination (fig. S11B). We found that more than half of IMR90 primed iPSCs (54%) carry Xa-like/Xi-like chromosomes, about one-third carry Xi-like/Xi-like chromosomes, and only a small portion (14%) carries Xa-like/Xa-like chromosomes. When looking at the IMR90 5iLA iPSCs, 40% of the cells carry Xa-like/Xa-like chromosomes, 40% carry Xa-like/Xi-like chromosomes, and only a small portion of cells (19%) carries Xi-like/Xi-like chromosomes. This single-cell analysis was also performed in all the other primed and naïve cell lines tested in this study (fig. S11C). We observed all three combinations of ChrX states in each of the primed and naïve cell lines. The Xa-like/Xa-like proportion of cells was generally expanded in naïve hPSCs compared to their primed counterparts; primed hPSCs exhibited between 14 and 28% Xa-like/Xa-like cells, while naïve hPSCs exhibited 30 and 69% Xa-like/Xa-like cells. Furthermore, the percentage of the cell population with Xi-like/Xi-like chromosomes was generally reduced in naïve hPSCs compared to primed hPSCs. These results would indicate a shift toward more Xa-like X chromosome structures in naïve hPSCs compared to primed cells.

In primed hPSCs, the identification of Xa-like and Xi-like chromosomes correlates well with the current understanding of X chromosome regulation in this pluripotent state (10, 14). The 3D chromosome states identified in naïve X chromosomes, however, have not been characterized well in the literature. Therefore, it was necessary to determine whether there were XCI-related differences between the two 3D conformational states we observed in the X chromosomes of these naïve cells. The canonical silencing and compaction of the X chromosome in XCI should generate marked differences in compactness between the Xa and an established Xi; these compaction differences have been quantified via imaging methods before (10, 15, 25, 62). Thus, we compared the compaction between the Xa-like and Xi-like chromosomes by calculating the radius of gyration of each individual chromosome (Fig. 2, G to I). The radius of gyration is a metric that approximates the radial size of the chromosome territory (27), thereby allowing us to compare the approximate volume of individual chromosome copies. We found that the Xi-like chromosomes are significantly larger than the Xa-like chromosomes in IMR90 primed iPSCs (Fig. 2G), possibly because of severe XCI erosion. In 5iLA naïve cells, the Xi-like chromosomes have a significantly larger radius of gyration than the Xa-like chromosomes as well (Fig. 2, H and I).

As these analyses showed significant differences in approximate size distribution between Xa-like and Xi-like chromosomes in primed and naïve cells, we wanted to further characterize the differences between the X chromosome populations in different pluripotent states. To achieve the goal, we compared the radii of gyration of all X chromosomes from different primed and naïve cell lines to the radii of gyration derived from IMR90 somatic chromosomes collected in a previous study (25). These somatic chromosomes served as a relative metric for Xa and Xi compaction (Fig. 3A). The statistical analysis showed that primed X chromosomes have similar compactness to the IMR90 somatic Xa. For all naïve hPSC cell lines and conditions examined, except the H9 5iLA cell line, the X chromosomes are more compacted than IMR90 somatic Xa, yet less compact than the IMR90 somatic Xi (Fig. 3B). We noted that the naïve cell lines showed smaller radii of gyration compared to primed cells in most cases (Fig. 3C).

Fig. 3. Naïve hPSCs have less-compact X chromosomes than primed hPSCs.

(A) Violin plots of the radius of gyration of every chromosome copy analyzed in every cell line and culture condition used in this study, with IMR90 somatic cells included as a comparison. The red and green dashed lines indicate the median value of the radius of gyration in the Xa and Xi chromosomes in IMR90 somatic cells, respectively. (B) Statistical analysis of the changes in the radius of gyration in all hPSCs compared to somatic Xa and Xi. The log₂ fold change (log₂FC) is represented by color. The level of significance calculated as −log₁₀(P value) from unpaired t test is represented by the size of the dots. The cutoff of −log₁₀(P value) for nonsignificant comparison is set to 1.30, which corresponds to a P value of 0.05. (C) Statistical analysis of the changes in the radius of gyration of naïve hPSCs, compared to the primed state, for each cell line. (D) Violin plots of the radius of gyration of Xa-like and Xi-like chromosomes in IMR90 iPSCs and that of the Xa and Xi chromosomes in IMR90 somatic cells (left). Statistical analysis of the changes in the radius of gyration in Xa-like and Xi-like chromosomes in IMR90 iPSCs compared to somatic Xa and Xi (right). n.s., not significant.

In addition, we compared the radii of gyration of Xa-like and Xi-like chromosomes in IMR90 iPSCs with the Xa and Xi chromosomes from IMR90 somatic cells (Fig. 3D, left). On the basis of the statistical analysis results (Fig. 3D, right), all IMR90 pluripotent chromosome groups are significantly less compact than the IMR90 somatic Xi. IMR90 5iLA naïve Xa-like chromosomes are significantly more compacted than the IMR90 somatic Xa, while naïve Xi-like chromosomes are comparable with the somatic Xa. The IMR90 primed Xa-like chromosomes have a comparable level of compaction to somatic Xa chromosomes, while the primed Xi-like chromosomes are slightly larger (Fig. 3D).

Together, these data suggest that both primed and naïve hPSCs contain Xa-like and Xi-like chromosomes that differ in 3D conformation. Naïve and primed Xi-like chromosomes are more expanded than Xa-like chromosomes in their respective cell state. Naïve X chromosomes are more compacted than the IMR90 somatic Xa, yet still larger than the IMR90 somatic Xi. Hence, we propose a model where the X chromosomes of naïve hPSCs acquire some traits of the 3D folding organization of somatic X chromosomes, but lack the XCI-driven compaction observed in the somatic Xi.

In light of this proposed model, we wanted to ascertain whether the Xa-like and Xi-like folding conformations could correlate with the transcriptional status of the X chromosomes and XIST expression. To investigate this, we combined the XIST RNA FISH with chromatin tracing on H9 naïve cells under all three naïve culture conditions, 5iLA, 5iLAF, and 4iLAF + tMEKi (fig. S12). The XIST RNA FISH identified the chromosomes with or without XIST coating, shown as XIST⁺ or XIST⁻ traces (fig. S12A). Louvain clustering facilitated the identification of these chromosomes by their Xa-like or Xi-like folding conformation (fig. S12B). We then correlated the chromosome conformations with the XIST expression state and quantified the percentages of Xa-like and Xi-like conformations assigned to XIST⁺ or XIST⁻ chromosomes (fig. S12C). No significant difference was identified between these two groups (P = 0.92, Fisher’s exact test), suggesting no correlation between XIST expression and Xa-like or Xi-like conformations.

To investigate whether the Xa-like and Xi-like folding conformation could correlate with the transcription status of other X-linked genes, we combined ATRX intron RNA FISH and chromatin tracing on our H9 5iLA cells (fig. S13). The RNA FISH targeting the first intron region of the X-linked gene ATRX measures the transcription activity of the gene at its genomic site (fig. S13A). We then performed Louvain clustering to identify Xa-like or Xi-like chromosomes (fig. S13B) and linked the conformations to the ATRX transcription state (fig. S13C). The percentages of Xa-like and Xi-like chromosomes are not significantly different between the ATRX⁺ (with transcription activity) and ATRX⁻ (without transcription activity) groups (P = 0.38, Fisher’s exact test). This finding suggests that ATRX transcription is independent of the Xa-like/Xi-like conformations, similar to the XIST coating state determined above.

Characterization of H7 hESCs with Hi-C and FISH demonstrates how severe X-linked DNA damage can modulate XCI-related phenomena in naïve cells

In our initial study, the H7 line was included as a comparative cell line. H7 5iLA naïve hESCs successfully acquired naïve pluripotency (fig. S14). However, chromatin tracing with combined XIST RNA FISH revealed that H7 5iLA naïve hESCs had lost substantial portions of the X chromosome p arm, especially among XIST⁺ X chromosomes (fig. S15A and Fig. 4A). Karyotyping demonstrated that an unequal translocation had effectively removed an entire copy of the X chromosome p arm from the H7 5iLA genome (fig. S15B), including the pseudoautosomal region (fig. S15C). This loss the short arm of one copy of the X chromosomes was observed twice in two independently derived H7 5iLA naïve hESC cultures. The short arm of the X chromosome was replaced by most of the q arm of chromosome 18, as shown by a deeper analysis of the Hi-C data from H7 naïve hESCs (Fig. 4, B and C). Hi-C data analysis reveals an obvious and strong association between the 18q and Xq regions in the 5iLA naïve cells that was completely absent under the starting primed condition (Fig. 4C). From this, we can infer that the X chromosome lost its p arm to an unequal translocation with the q arm of chromosome 18; the derivative X chromosome produced from this event will be referred to as der(X).

Fig. 4. Abnormal accumulation of XIST in H7 naïve hESCs due to der(X) deletion.

(A) Mean spatial distance matrix for time course RNA/DNA co-FISH in H7 5iLA naïve hESCs. Chromosomes were segregated by XIST RNA FISH, and the matrices were generated from DNA FISH data. Data were acquired over the course of a normal passage on days 2, 4, and 6 (D2, D4, and D6). Gray areas indicate a detection efficiency of <20% for the targeted TADs in D2 and D6, as well as <36% in D4. The plots of detection efficiency of each TAD are shown in the two bottom panels. The number of traces analyzed is the following: 292 (D2, XIST⁻), 123 (D2, XIST⁺), 101 (D4, XIST⁻), 125 (D4, XIST⁺), 171 (D6, XIST⁻), and 104 (D6, XIST⁺). (B) Hi-C genome-wide DNA depletion maps demonstrating naïve ChrX p-arm loss. Deletions are shown as blue lines; Hi-C contact enrichment is shown as red regions. Chromosomes span from Chr1 at the bottom left corner to ChrX at the top right corner. (C) Hi-C DNA contact enrichment map between ChrX and Chr18 in H7 primed and 5iLA naïve cells. An illustration of the translocation of Chr18q to replace X p arm is shown. (D) XIST RNA cloud counts over the course of a normal passage on D2, D4, and D6. XIST^−/− cells show no XIST expression, XIST puncta cells show minor levels of XIST expression, XIST^+/− cells show one X chromosome coated in XIST, and XIST^+/+ cells show two XIST RNA clouds. Bars represent the relative proportion of the population occupied by each class of chromosomes, averaged over two technical replicates. Error bars represent the SD. There are n > 190 chromosomes for every time point for each replicate.

To further characterize the extent of ChrX p-arm chromatin loss across the H7 naïve population and how this related to XIST coating status, we quantified the percentages of X and der(X) chromosomes that are coated in XIST in our H7 5iLA data directly. The der(X) constitutes at least ~65% of all XIST⁺ chromosomes across all time points; at days 2 and 6 (D2 and D6), this percentage is closer to 85 to 95% (fig. S15D). We also performed the opposite analysis to demonstrate the proportion of total der(X) or intact X chromosomes that were coated with XIST or uncoated. From these data, we can see that the der(X) is much more likely to be coated by XIST over passage time (fig. S15D). In addition, we looked at the chromosome intactness at the single-cell level in the H7 naïve 5iLA cells using our FISH data combined from all time points and quantified the percentage of cells carrying pairs of intact X/X, der(X)/X, or der(X)/der(X) chromosomes (fig. S15E). On the basis of the results, 75% of the cells carry one copy of der(X) and one copy of the intact X, 21% of them carry two copies of der(X), and only 4% of the cells retain two copies of the intact X chromosomes. Therefore, the der(X) clone is prevalent in the H7 naïve 5iLA cell line, based on FISH data. After filtering out these der(X) chromosomes, we proceeded to cluster the intact X chromosomes as in other cell lines to determine whether the H7 5iLA cell line exhibits Xa-like and Xi-like conformations of the normal X chromosome (fig. S16, A and B). We found the intact X chromosome from H7 5iLA cells generate Xa-like and Xi-like conformations (fig. S16B), quite similar to our other naïve hPSCs. Unlike the other cell lines, however, cluster 3 in our H7 5iLA cells could not be classified as Xa-like or Xi-like. Next, we performed the same clustering analysis with the q arm of the aberrant der(X) chromosome in naïve H7 (fig. S16, C and D). The results showed two major clusters (clusters 1 and 2), both of which partially resembled the Xi-like conformation with a boundary near DXZ4.

Despite the karyotypic issues observed, H7 5iLA cells showed an interesting behavior related to XIST expression (Fig. 4D). On D2 postpassaging, over 70% of H7 5iLA naïve cells either had one robust XIST cloud (XIST^+/−) or punctate expression of XIST around a single chromosome (XIST punctate/−). The other 30% had no XIST expression (XIST^−/−). By D6, almost all naïve cells displayed either one XIST cloud (XIST^+/−) or two robust XIST clouds (XIST^+/+) per cell. Very few cells lacked XIST expression by this time. Quantification of the individual X chromosome copies in these cells demonstrated that der(X) chromosomes were highly enriched for XIST coating over the time span of a single passage (fig. S15D). Similar massive losses of chromatin from the short arm of the X chromosome have been previously reported to cause nonrandom XCI bias toward inactivating the afflicted X chromosome (63). Our results demonstrate a potential link between large-scale DNA damage to the X chromosome and XIST accumulation in naïve hPSCs.

DISCUSSION

There have been extensive and successful efforts to characterize the X chromosomes of naïve hPSCs or human ICM cells (6–12, 19). However, defining X chromosome states in these cells has been challenging (9, 22), because of, in part, the lack of consensus on this subject within the scope of the literature. Some groups have identified subpopulations (12, 19) or culture conditions (23, 64) in which only partial reactivation of the X chromosome may occur. However, many studies have demonstrated that female naïve hPSCs generally maintain two Xa chromosomes (6, 8–13). Furthermore, common readouts of XCI function, such as XIST expression (6, 8–10, 19) or H3K27me3 foci (6, 9, 10, 13, 19, 64), exhibit similar inconsistency within the literature. Hence, it becomes quite difficult to determine a consensus within the field on the exact state of the naïve female X chromosome. However, recent developments in 3D genomics and spatial biology have made it feasible to profoundly assess the spatial conformation of chromatin (24–26, 28, 30). Thus, we sought to characterize the X chromosomes in human female pluripotent cells using 3D genomic methods such as chromatin tracing and Hi-C.

We found that naïve and primed hPSCs generally exhibit Xa-like and Xi-like chromosomes (Fig. 5). These two different conformations were readily identified from the macrodomain boundaries that they exhibit, a boundary near the centromere for Xa-like chromosomes (25) and one near DXZ4 for Xi-like chromosomes (Fig. 5, A to D) (25, 26, 30–32). The Xi-like chromosomes in naïve hPSCs do not exhibit the Xi compaction identified in somatic female cells (Fig. 5C). The canonical compaction difference evident between the larger somatic Xa and the smaller somatic Xi (Fig. 3D) is not observed in naïve and primed X chromosomes; rather, the Xi-like chromosomes are more expanded than the Xa-like chromosomes in these states (Fig. 2, G and I). Female naïve and primed X chromosomes therefore present unconventional 3D states when compared to somatic X chromosomes. From a population distribution perspective, primed hPSCs are more likely to exhibit X chromosomes in Xa-like/Xi-like and Xi-like/Xi-like configurations than naïve hPSCs. Naïve hPSCs exhibit an obvious shift toward Xa-like/Xa-like configurations of their X chromosomes across the various cell lines and culture conditions assessed in this study.

Fig. 5. X chromosome conformations in naïve and primed hPSCs and the implications in XCI.

(A) Scheme of compartmentalized macrodomains of Xa-like and Xi-like chromosomes in naïve and primed hPSCs as well as somatic Xa and Xi. Macrodomains are represented by red and blue rectangles. The positions of centromere and DXZ4 are indicated by green lines on the X chromosome. (B) Schematic illustration of Xa and Xi chromosome conformations in somatic cells. The red or blue dashed circles represent the macrodomains. Positions of centromere and DXZ4 macrosatellite are indicated by black arrows. (C) Schematic illustration of Xa-like and Xi-like chromosome conformations in naïve hPSCs. (D) Schematic illustration of Xa-like and Xi-like chromosome conformations in primed hPSCs.

This shift toward Xa-like 3D conformation in the naïve state complements findings from previous studies, wherein naïve hPSCs exhibited biallelically Xa chromosomes in the 5iLAF-derived and T2iLGö states (6, 8–10, 12, 13). However, other research groups reported contrasting results that naïve hPSCs exhibit an Xa and a partially reactivated Xi in these culture contexts (6, 12, 20, 23). In a larger picture, we suggest that naïve hPSCs may be slightly more developmentally advanced in culture than expected. This advancement is reflected in their consolidation of canonical Xa-like and Xi-like 3D chromosomal structures (Fig. 5, A to D) and in their apparent ability to initiate certain levels of XCI-like processes and 3D genomic changes canonically associated with large-scale lesions on the X chromosome (figs. S14 and S15) (63). However, these naïve hPSCs retain the ability to transcribe XCI-sensitive genes such as ATRX from both Xi-like and Xa-like chromosomes. As ATRX transcription apparently occurs from both Xa-like and Xi-like chromosomes (fig. S13), this suggests that transcriptional activity is not tied to 3D conformation under 5iLA naïve conditions. This result orthogonally complements findings from previous studies, where transcription from XIST⁺ X chromosomes was observed in naïve hPSCs (7, 9, 10).

In primed cells, the apparent lack of Xi compaction is somewhat expected; previous literature describing XCI erosion under this condition (6, 15) demonstrated that this process can completely reverse H3K27me3 deposition within 20 passages for many primed hPSC lines. Our data showed that the Xi-like chromosome in primed cells is even more expanded in 3D structure than their Xa-like chromosome (Fig. 5D). Primed cells included in this study have been cultured for over 20 passages and may have, thus, attained an even more pronounced XCI erosion state. A recent study has found that a component in mTeSR1, lithium chloride, can drive the loss of XIST expression in primed cells by GSK-3 inhibition (65). As the primed cells in this study were extensively cultured in mTeSR1, they have likely undergone XCI erosion. This also has important implications for naïve culture conditions such as 5iLA, which uses GSK-3 inhibition to help stabilize naïve pluripotency (1, 7, 65).

In combination with chromatin tracing for the X chromosome, several cell lines characterized in this study were examined for their XIST expression via RNA FISH. Unlike other naïve cell lines, which may have limited XIST expression because of the previously mentioned GSK-3 inhibition (1, 65), 5iLA H7 naïve hESCs displayed a unique accumulation of XIST during normal passage time. XIST preferentially coated the X chromosomes exhibiting der(X) and accumulated rapidly over one passage period (Fig. 4D). This aberrant and rapid XIST increase may be linked to the massive haploinsufficiency issues that would be caused by a near-complete loss of one copy of ChrX p-arm chromatin via unequal translocation. Previously, it was shown that DNA lesions in this chromosomal arm, specifically in the pseudoautosomal region, can result in severely skewed XCI (63). However, the observed der(X)/X clonal dominance in H7 5iLA culture suggests that in vitro selection pressure may have played an important role in producing this specific outcome. Previous studies on human XCI in hPSCs has found that clonal selection can result in highly variable outcomes for XCI readouts (10) and that some of these clonal selections are driven to success as a result of X chromosome aberrations (66). Although deletions that confer survival advantage are less commonly encountered (67), the X chromosome may be a more common location for such deletions to occur because of the low number of haploinsufficiency-sensitive genes contained in this chromosome (68). The loss of an entire X chromosome copy in 5iLAF-cultered hESCs has been briefly reported previously (50). Considering that chromosomal abnormalities commonly occur during the course of hPSC culture (6, 50, 69–73), it is possible that de novo chromosomal aberrations and subsequent selective pressures may work in tandem to amplify the incidence of genomically aberrant cells in naïve hPSC culture. In naïve hPSCs, genetic lesions that affect the X chromosome may contribute to marked changes in global XIST expression as a result of clonal dominance by selection. As demonstrated in this study, the resulting naïve hPSCs may exhibit XCI outputs contrary to expectations as a result of these in vitro pressures.

MATERIALS AND METHODS

Cell culture

hPSCs in the primed state were grown feeder-free on Matrigel-coated plates under the mTeSR1 medium condition. Primed hPSCs were passaged every 6 to 7 days with dispase. The medium was replenished every 24 hours. Transition of primed hPSCs to naïve hPSC was performed as previously described (7–10). Briefly, primed hPSCs were dissociated with Accutase and 10 nM Y-27632 for 2.5 min at 37°C and resuspended in 2 ml of naïve medium. Cells were pipetted approximately five times. The largest clumps were allowed to settle before the remainder was replated on 100% confluent MEFs. Depending on cell volume and density, cells were plated on 10 cm plates (weeks 1 to 2), 4 wells of a 6-well plate (week 3), 1 well of a 6-well plate (weeks 4 to 6), or 1 well of a 12-well plate (weeks 4 to 6 for certain lines). Cells were then amplified and stabilized in culture over two to three passages until sufficient cells could be aliquoted, frozen down for liquid nitrogen storage, and sent for karyotyping. Naïve hPSCs were cultured in 5iLA medium on MEFs, as reported previously (7).

Immunofluorescence

Cells were fixed in 4% paraformaldehyde in 1× Dulbecco’s phosphate-buffered saline (DPBS) and permeabilized and blocked with 5% (w/v) bovine serum albumin (BSA) and 0.3% (v/v) Triton X-100 in 1× DPBS. Next, cells were incubated with primary antibodies for NANOG (1:800) or OCT4 (1:400) in 1% (w/v) BSA in 1× DBPS at 4°C overnight. Primary antibodies were aspirated, and cells were incubated with secondary antibodies conjugated with Alexa Fluor 488 and a costain of SUS2D*PE (2 μl per sample) with NANOG and TRA-1-60*PE (2 μl per sample) with OCT4. Secondary antibodies were incubated at room temperature for 1 hour, after which the antibodies were aspirated and 4′,6-diamidino-2-phenylindole (DAPI) was added. DAPI was incubated for 10 min at room temperature before the cells were triple-washed in 1× DPBS. Cells were imaged at 20× on a Leica DMI6000 B inverted fluorescence microscope with adaptive focus control for phase contrast.

For H3K27me3 immunofluorescent staining, we followed the method described previously (74). After fixation, permeabilization, and blocking, cells were incubated with primary antibodies for H3K27me3 (1:450) overnight at 4°C in blocking buffer. The primary buffer was aspirated, and cells were incubated in secondary antibodies conjugated with Alexa Fluor 488 for 1 hour at room temperature (1:1000). Cells were then stained with DAPI in 1× PBS (1:1000) before triple-washing in 1× PBS. Imaging was performed at 40× on a Leica DMI6000 B inverted fluorescence microscope with adaptive focus control for phase contrast. Z-stack images were acquired, with 0.69 μm of z-thickness per stack. 2D images were generated from at least 10 in-focus frames through z-projection with maximum intensity using FIJI.

RNA extraction and PCR

RNAs were isolated using the RNeasy Mini Kit according to the manufacturer’s protocol. Next, cDNA was synthesized using the amfiRivert cDNA Synthesis Master Mix. qRT-PCR was performed using iQ SYBR Green Supermix. All primers used in this study are listed in table S1.

DNA FISH for X chromosome PAR

DNA FISH for two genes of the X chromosome pseudoautosomal region, STS and KAL/KAL1, was performed using a standardized protocol (75) by the Yale Cytogenetics Lab in both intact nuclei and metaphase spreads.

Hi-C

Hi-C 2.0 was performed as described previously (28). Briefly, 10 cm plates of naïve and primed hPSCs were cross-linked and aliquoted into 5.5 × 10⁶ cell aliquots. Each aliquot was separately treated as a Hi-C sample. Each sample underwent plasma membrane lysis and removal of the cytoplasm before nuclear semipermeabilization. Next, semipermeable cross-linked nuclei were digested with the Dpn II restriction enzyme. The sticky ends were filled in with deoxynucleotide triphosphates (dNTPs) and biotinylated 2′-deoxycytidine 5′-triphosphate before blunt-end ligation. After ligation, cross-linking was reversed, nuclei were broken down, and the correctly sized biotinylated DNA was purified out of the larger DNA pool. Illumina sequencing primers were added to these biotinylated fragments, which were subsequently amplified. Before sequencing, libraries were TOPO-cloned and ~10 clones were sequenced and analyzed to ensure that reads with the proper structure were present in the library. Sequencing was performed at the Yale Genomics Core.

Bioinformatics analysis of Hi-C

Hi-C data were processed as described previously (76). Briefly, mapping, quality control, and detection of valid interaction pairs of HiC reads were performed by HiC-Pro software (v2.9.0) with Dpn II restriction sites in hg19 genome and “GATCGATC” ligation site (77). Subsequent analyses were implemented using HOMER suite (v4.11.1) (78). Briefly, the mapped reads were converted into HOMER format by the makeTagDirectory command. Then, inter- and intrachromosomal interactions were visualized by the analyzeHiC command with “-res 1000000 -norm” or “-res 5000000 -norm” options.

Probe design for chromatin tracing, XIST RNA FISH, and ATRX intron RNA FISH

We performed chromatin tracing according to our published protocol (25). Briefly, the probes were designed to target the center 100 kb region of each TAD, and in total, 40 TADs across the whole X chromosome were targeted (table S2). A total of 1000 different primary probes were used to target each TAD, and the probe sequences were reported in our previous work (25). Each probe contains a 30 nt genome-targeting sequence, a 30 nt secondary probe binding sequence, and two flanking 20 nt priming sequences for probe synthesis.

The XIST RNA FISH probes that we used were the same as published before (27). Briefly, 50 probes were designed to target the XIST transcript. Each probe contains a 30 nt transcript-targeting sequence and two flanking 15 nt priming sequences for library amplification. Each probe has a 5′ acrydite modification incorporated during probe synthesis, which is used for cross-linking the oligos into a polyacrylamide gel matrix (see the “Combined XIST RNA FISH or ATRX intron RNA FISH with chromatin-tracing primary probe hybridization” section).

We designed 60 probes for ATRX intron RNA FISH using ProbeDealer (79). The sequences of the first intron of human ATRX were downloaded from Ensembl (https://useast.ensembl.org/Homo_sapiens/Info/Index). Each probe contains a 30 nt intron-targeting region, two identical 20 nt secondary binding regions, and two flanking 20 nt primer binding regions for library amplification. Each probe incorporates the Cy3 dye and acrydite modification at the 5′ end during probe synthesis for polyacrylamide gel embedding (see the “Combined XIST RNA FISH or ATRX intron RNA FISH with chromatin-tracing primary probe hybridization” section). The template oligo sequences for ATRX intron 1 are listed in table S3 as a separate file.

Probe synthesis

The template oligo pool for ChrX chromatin tracing primary probes was purchased from CustomArray, and the template oligo pools for XIST RNA FISH probes and ATRX intron RNA FISH probes were purchased from Integrated DNA Technologies. The protocol for probe synthesis was adopted from previous works (25, 80, 81). Briefly, the template oligo pool was amplified via limited-cycle PCR using PCR primers targeting the priming regions, where an additional T7 promoter sequence was added to the PCR product through the forward primer. The PCR product was column-purified and subjected to T7 in vitro transcription to produce RNA products. The RNA products were subjected to reverse transcription to produce DNA oligo probes, where the Alexa Fluor 647 dye was added to the ChrX chromatin tracing primary probes or the Cy3 dye and acrydite modification were added to the XIST RNA FISH probes or ATRX intron RNA FISH probes through the reverse transcription primer. The DNA oligo probes were column purified after removal of RNA products by alkaline hydrolysis. Limited-cycle PCR primers and reverse transcription primer for ChrX chromatin tracing probe synthesis were reported before (25). Limited-cycle PCR primers for XIST RNA FISH and ATRX intron RNA FISH probe synthesis were reported before (27) and are relisted in table S4. Reverse transcription primer for XIST RNA FISH and ATRX intron RNA FISH probe synthesis is reported in table S4.

Chromatin tracing primary probe hybridization

The cells grown on a coverslip were fixed in freshly made 4% (w/v) paraformaldehyde in 1× DPBS for 10 min at room temperature, followed by three DPBS washes. Cells were incubated with freshly made sodium borohydride (1 mg/ml) in 1× DPBS for 10 min at room temperature, followed by three DPBS washes. Then, the cells were permeabilized with 0.5% (v/v) Triton X-100 in 1× DPBS for 15 min at room temperature, followed by three DPBS washes. Cells were treated with 0.1 M HCl for 5 min at room temperature and washed twice with DPBS, followed by a 45 min digestion with ribonuclease A (RNase A) (0.1 mg/ml) in 1× DPBS at 37°C and two DPBS washes. Cells were then incubated with a prehybridization buffer composed of 50% (v/v) formamide and 0.1% (v/v) Tween 20 in 2× saline sodium citrate (SSC) for 30 min at room temperature. The coverslip was gently dried to the side on a tissue paper to remove the excess liquid and then flipped onto a 25 μl hybridization buffer composed of 50% (v/v) formamide and 20% (w/v) dextran sulfate in 2× SSC containing 6 to 20 μM DNA FISH probes on a glass slide, with cells in direct contact with the hybridization buffer. The coverslip slide hybridization assembly was heat-denatured on an 86°C dry bath for 3 min so that the sample reached 80°C and incubated overnight for 15 to 18 hours at 37°C in a humid chamber. The cells were then washed with 0.1% (v/v) Tween 20 in 2× SSC at 60°C water bath twice for 15 min each and once at room temperature for 15 min. Last, the coverslip was attached with fiducial beads by incubation with 0.1 μm yellow-green beads at 1:100,000 dilution in 2× SSC for 2 min at room temperature. The excess beads were removed by two washes with 2× SSC.

Combined XIST RNA FISH or ATRX intron RNA FISH with chromatin-tracing primary probe hybridization

The cells grown on a silanized coverslip (see the “Silanization of coverslips” section) were fixed and permeabilized as previously described in the “Chromatin tracing primary probe hybridization” section. The cells were incubated with prehybridization buffer composed of 50% (v/v) formamide in 2× SSC for 5 min at room temperature and then hybridized with 1 to 2 μM XIST RNA FISH probes or ATRX intron RNA FISH probes in hybridization buffer composed of 50% (v/v) formamide, 10% (w/v) dextran sulfate, yeast tRNA (1 mg/ml), and 1% (v/v) murine RNase inhibitor in 2× SSC and incubated in a humid chamber/petri dish at 37°C overnight. The cells were washed with 0.1% (v/v) Tween 20 in 2× SSC at 60°C water bath twice for 15 min each and once at room temperature for 15 min.

After the XIST RNA FISH or ATRX intron RNA FISH, the coverslip was embedded with 4% (w/v) polyacrylamide gel to immobilize the probes in place, preventing the detachment of the oligo probes during RNase A digestion in the chromatin tracing primary probe hybridization procedure. To embed the sample, we freshly prepared a degassed gel solution containing 4% (w/v) acrylamide and bis-acrylamide solution, 300 mM NaCl, 60 mM tris-HCl (pH 8.0), ammonium persulfate (0.3 mg/ml), 0.15% (v/v) N,N,N',N'-Tetramethyl ethylenediamine (TEMED), and 1:100,000 yellow-green fiducial beads. We prepared a 2″ by 3″ glass slide covered with gel slick solution and dried for 10 min in air. We removed the excess liquid on the glass slide before adding gel solution. We dropped 50 μl of gel solution onto the slick glass slide and then gently flipped the coverslip onto the gel solution without causing bubbles. After 1.5 hours, the polyacrylamide gel solidified, and the coverslip was gently removed from the glass slide with a razor. The sample was proceeded to chromatin tracing primary probe hybridization starting from the HCl treatment.

Chromatin tracing

Image system setup

For chromatin tracing, we used homebuilt microscope system as described previously (81), which contained a Nikon Ti2-U body, a Nikon CFI Plan Apo Lambda 60× oil objective lens (numerical aperture of 1.40), and a Hamamatsu Orca Flash 4.0 V3 camera, an active focus lock system. For epifluorescence illumination, we used 647, 560, and 488 nm lasers to excite and image the Alexa Fluor 647 dye, Cy3 and ATTO 565 dye, and the yellow-green fiducial beads, respectively. The laser lines were directed to the sample using a penta-band dichroic mirror on the excitation path, and the fluorescent emission was filtered using a penta-band emission filter on the emission path. During the imaging, the sample was automatically scanned across different fields of view with a motorized xy sample stage and scanned in z direction with a piezo z positioner. The image size was 1536 pixels by 1536 pixels or 1024 pixels by 1024 pixels when using the Dual-View setup as described previously (27). An automated fluidic system (58, 80) was used for buffer exchange during sequential hybridization of secondary probes and imaging.

Sequential hybridization of secondary probes and imaging

After the chromatin tracing primary probe hybridization or the combined XIST RNA FISH with chromatin tracing primary probe hybridization, we proceeded to sequential hybridization of secondary probes and imaging. All following steps were performed at room temperature. The sample was assembled into a Bioptech’s FCS2 flow chamber and connected to a homebuild automated fluidics system (25, 80). To visualize the X chromosome regions, we imaged the cells in 2× SSC in a 647 nm channel to capture the chromatin tracing primary probe signal (hyb 0). For samples treated by combined XIST RNA FISH or ATRX intron RNA FISH with chromatin tracing primary probe hybridization, we also imaged in a 561 nm channel to capture the XIST RNA FISH signal or ATRX intron RNA FISH signal.

Before secondary probe hybridization, all signals were photobleached in 2× SSC. Next, the secondary probes labeled with Alexa Fluor 647 or ATTO 565 dyes in secondary hybridization buffer composed of 20% (v/v) ethylene carbonate (EC) in 2× SSC at a final concentration of 6 nM each were introduced to the sample and incubated for 30 min. The sample was then washed by 10% (v/v) EC in 2× SSC for 3 min and switched to 2× SSC buffer. Then, we took z-stepping images in 2× SSC with 647, 560, and 488 nm laser illuminations for Alexa Fluor 647– and ATTO 565–labeled secondary probes and fiducial beads, respectively. The secondary probe hybridization and imaging procedure was repeated for 20 rounds until all 40 TADs were imaged. Dye-labeled secondary probes for ChrX chromatin tracing were reported before (25) and relisted in table S2.

Silanization of coverslips

Coverslips were silanized to cross-link the polyacrylamide gel. We followed the same silanization protocol as previously described (82). Briefly, we immersed the coverslips in a 1:1 mixture of 37% (v/v) HCl and methanol for 30 min at room temperature in a fume hood. We rinsed the coverslips with Milli-Q water three times and once with 70% (v/v) ethanol and then dried in a 70°C oven for 20 to 30 min. Next, we immersed the coverslips in silanization solution composed of 0.1% (v/v) triethylamine and 0.2% (v/v) allyltrichlorosilane in chloroform for 30 min at room temperature. We washed the coverslips with chloroform once and 100% ethanol once and baked in a 70°C oven for 1 hour to dehydrate the silane layer. The silanized coverslips could be stored in a desiccated chamber at room temperature for weeks.

Chromatin tracing analysis

Determination of 3D chromatin traces and mean spatial distance matrix

All analyses in this work were performed with MATLAB R2017b or later versions. The analysis procedures and pipelines are similar to that described in the previous work (81). Briefly, we first calibrated the color shift between 647 and 561 nm channels with TetraSpeck beads. Then, we determined the sample drift between hybridization rounds by fitting the fiducial beads. Next, we fitted the TAD foci from the sequential hybridization rounds of images and then corrected the sample drift and the color shift from the raw TAD positions. The corrected TAD positions were then linked into chromatin traces on the basis of their spatial clustering patterns. We attempted to refit the missing TADs within the chromatin trace region in the corresponding hybridization rounds and added the refitted loci to the chromatin traces. We used the refitted chromatin traces to calculate the spatial distances among TADs and averaged all individual chromosomes under each cell line/condition to get the mean spatial distance matrix. Analyzed chromatin tracing data and chromatin tracing analysis codes are available at https://campuspress.yale.edu/wanglab/hPSC.

Domain boundary analysis

Domain boundaries were determined on the basis of the mean spatial distance matrix. As illustrated in fig. S17, a sliding window of 5 × 5 TADs (shown as the green square) moves along the diagonal of the matrix. We averaged the distances within this window and defined it as the “interdomain” score, as well as averaged the distances within the two triangles by the two sides of the sliding window (shown as the two yellow triangles; the two equal sides of each isosceles right triangle have the same length as the sliding window) and defined it as the “intradomain” score. The boundary score was then calculated by dividing the difference of the “inter-” and “intra-” domain scores by the sum of the “inter-” and “intra-” domain scores and was assigned to the center TAD. The formula is shown below

$$\text{Boundary score}=\frac{\mathrm{mean}(\mathrm{interdomain})-\mathrm{mean}(\mathrm{intradomain})}{\mathrm{mean}(\mathrm{interdomain})+\mathrm{mean}(\mathrm{intradomain})}$$

This procedure was repeated for TAD 6 to TAD 35 along the diagonal because of the restrictions of the size of the sliding window. We plotted the boundary score profile and called peaks by finding the local maxima above a threshold. The superdomain boundaries were determined as the TAD locations of the called peaks.

Radius of gyration calculation

The radius of gyration of a chromatin trace is defined as the root mean square distance of all observed TADs to the center position of all TADs of the chromatin trace, which reflects the compactness of the chromosome folding, as calculated in the following formula

$$\text{Radius of gyration}=\sqrt{\frac{1}{n}(r_1^2+r_2^2+\cdots +r_n^2)}$$

where n represents the number of detected TADs in the individual chromatin trace, and r₁, r₂, … r_n represent the spatial distance between the centroid and TAD 1, TAD 2, … TAD N of the trace.

Chromatin traces with no less than 90% detection rate (no less than 36 TADs were detected of the 40 targeted TADs) were used for radius of gyration calculation. The chromatin tracing data of Xa/Xi chromosomes in human somatic IMR90 cells were downloaded from the previous work (83).

Determination of the Xa-like and Xi-like chromosomes

Clustering of chromatin conformation

To cluster chromosomes on the basis of chromatin conformation, we used the chromatin traces with greater than 90% TAD detection efficiency. For each trace, the spatial distance matrix was generated on the basis of the 3D positions of TADs. The missing values in the matrix were filled by linear interpolation of neighboring, nonmissing values. The spatial distances between TAD pairs were extracted from the matrix and taken as entries for the Louvain-Jaccard clustering algorithm, and the resulting trace clusters were visualized with t-SNE, using cosine distance.

Compartmentalization analysis of macrodomains

We identified the macrodomain compartments using an algorithm previously described (25) with some modifications. Briefly, for each trace cluster, we first calculated the mean spatial distance and the genomic distance between each TAD pair. Then, we calculated the running average of spatial distance using the 30 closest genomic distance entries for each data point, which yielded the expected spatial distance for each TAD pair at their genomic distance. We then normalized the mean spatial distance by the expected spatial distance for each TAD pair. On the resulting normalized mean spatial distance matrix, we calculated the Pearson correlation coefficient between each row pair or column pair. On the resulting Pearson correlation matrix, we further performed principal components analysis and used the coefficients of the first principal component as compartment scores for TADs. We assigned TADs with positive compartment scores to one macrodomain and those with negative compartment scores to another macrodomain.

Determination of the Xa-like or Xi-like identity

We compared the compartmentalization profiles between trace clusters of hPSCs and Xa/Xi chromosomes of IMR90 somatic cells. The trace clusters that were compartmentalized at the centromere or near the centromere position were determined as Xa-like chromosomes, while the trace clusters that were compartmentalized at the DXZ4 or near the DXZ4 position were determined as Xi-like chromosomes.

Determination of XIST^+/− and ATRX^+/− chromosomes

In the chromatin tracing experiments combined with XIST RNA FISH or ATRX intron RNA FISH, the images taken from the 561 nm channel during hyb 0 captured the XIST RNA FISH signal or the ATRX intron RNA FISH signal. We projected the maximum intensity value of each pixel along the z axis of these images and overlaid them with the previously determined chromatin traces. Through visual inspection, we manually determined the XIST⁺ or ATRX⁺ chromosomes when the chromatin traces are colocalized with the XIST or ATRX intron RNA FISH signal, and those without colocalization are marked as XIST⁻ or ATRX⁻ chromosomes.

Polarization index calculation

After we determined the macrodomain compartments described in the previous section, we calculated the polarization index between the two macrodomain compartments in individual chromosome copies as previously described (25). Briefly, we built 3D convex hulls for the two macrodomain compartments for each chromatin trace with more than 90% of TADs detected. Then, we calculated the 3D volume of these two hulls (referred to as V₁ and V₂) and the overlapped volume (referred to as V_S). The polarization index was calculated as followed

$$\text{Polarization index}=\sqrt{(1-\frac{V_{\mathrm{S}}}{V_1})(1-\frac{V_{\mathrm{S}}}{V_2})}$$

The value of a polarization index ranges from 0 to 1. A higher polarization index of a chromatin trace indicates that the two macrodomains are more spatially segregated and polarized. A lower polarization index indicates that the two macrodomains are more overlapping.

Supplementary Materials

This PDF file includes:

Figs. S1 to S17

Tables S1, S2, and S4

Legend for table S3

Other Supplementary Material for this manuscript includes the following:

Table S3