Modification of internal RNA domains from the long non-coding RNA XIST refines roles in silencing and heterochromatin recruitment

Abstract

XIST is a paradigmatic long non-coding RNA required for the initiation of the epigenetic silencing of X chromosomes, leading to dosage compensation in mammalian females. XIST includes tandem repeat domains (A-F) that serve as binding platforms for protein complexes involved in silencing, localization, and heterochromatinization of the inactive X chromosome. XIST/Xist transgenes can silence autosomes as well as the X in both early development and somatic cells. In our quest to create a minimal but highly functional XIST transgene, we have tested both removal of non-repeat regions of XIST and duplication of repeat regions previously established as critical for XIST function. Targeting inducible XIST transgenes to an autosomal FRT site in somatic cells allows us to compare the activity of induced constructs on silencing and chromatin recruitment. We have previously identified that the CCC-dense mouse polycomb-interaction domain (PID) could restore functionality to a minimal human XIST construct lacking the B and D repeats. We now demonstrate that the density of the CCC motifs is not the sole contributor to PID functionality in the human transgene. Surprisingly, a duplication of repeat A reduces the expression from the exogenous promoter, as does removal of the region around the F repeats including six YY1 binding sites. In this context of an inducible XIST in somatic cells the internal small exons were not required; however, duplication of the E repeat region augmented functionality. Overall, the strongest correlations we observed across our diverse set of constructs were between chromatin deposition and silencing.

Introduction

XIST was one of the first long non-coding RNAs (lncRNAs) discovered, and associates in cis with the chromosome from which it is expressed [1–3]. XIST, and the mouse ortholog Xist, are essential for X-chromosome inactivation [4, 5]. XIST/Xist is proposed to establish the facultative heterochromatin and gene silencing of the inactive X (Xi) by recruitment of chromatin remodelers primarily through the unique tandem repeat modules (labelled A-F) within the RNA (reviewed in [6]). Transgenes have demonstrated that XIST/Xist can act to silence autosomal genes, not only during an early developmental window [7] but also in somatic cells [3]. This has led to the proposed use of XIST transgenes to silence supernumerary autosomes for chromosomal therapy [8]. There are alternative isoforms of both human and mouse XIST/Xist, with the shorter human full-length cDNA transgene (FullXIST) being 14 kb, a challenging size to deploy experimentally or therapeutically. We have previously generated a miniXIST construct that combined 1.85 kb of the 5′ end of XIST containing the A and F repeat regions with the mouse region XR-PID (PID), the minimal region of Xist required for polycomb recruitment [9] and another 2.7 kb including the internal exons and the E repeat to generate a partially functional transgene [10]. Subsequent duplication of the PID domain (PID-PID) improved the ability of this miniXIST to silence and recruit heterochromatic marks, although still not to the level of FullXIST transgenes [11]. The silencing capacity of these constructs was proportional to the number of CCC motifs and dependent on HNRNPK recruitment, which binds to the CCC motif. HNRNPK heterozygous deletions had more impact on the recruitment of H2AK119 ubiquitinylation (H2AK119ub) than on H3K27trimethylation (H3K27me3) [11].

We had previously tested various combinations of human and mouse B repeat regions, with PID including all of the mouse B repeats and a part of the mouse-specific C repeats, and BhB being a fusion of the human-specific Bh and human B repeats [10, 11]. As we observed an augmentation of silencing and chromatin recruitment in proportion to the number of CCC in the B region, we now hypothesized that we would improve silencing by incorporating five copies of the compact BhB repeat. Such a construct would mimic the total number of CCC all across XIST, but focused in one small region, and would also be entirely derived from human sequences, as the PID region is mouse Xist-derived.

Most transgenic experiments with XIST/Xist have assessed the role of the repeat regions through deletions (reviewed in [6]), and thus, there is limited information on whether additional copies of critical elements would augment silencing. The A repeat is responsible for silencing in both human and mouse transgenes ([12, 13], reviewed in [6]); removal deleteriously impacts expression from the endogenous promoter and also an exogenous promoter in some contexts (summarized in [14]). We therefore hypothesized that an additional A repeat region might improve silencing; although across mammals the number of A repeats is quite constrained [15]. Additionally, the E repeat is considered responsible for matrix association of the RNA through binding with CIZ1 and a self-aggregating complex of PTBP1, MATR3, TDP-43 and CELF1; leading to localization of XIST, perhaps through phase separation [16, 17]. Removal of the E repeat from XIST/Xist leads to loss of localization of XIST/Xist with transgenes or the endogenous locus (e.g. [18, 19]). We therefore asked whether an additional E repeat region might improve silencing and/or heterochromatin recruitment.

In addition to adding regions to our constructs, we also sought to minimize the construct size necessary while maintaining functionality. We therefore tested the impact of removal of the poorly conserved F repeat region, which overlaps a series of YY1 DNA binding sites, and has been reported to function as an alternative promoter or enhancer for XIST; these functions should not be necessary in our inducible exogenous promoter system. However, the region is also reported to be the site of tethering of XIST through YY1 interactions [20]. We also queried the impact of removing the conserved internal exons two to five. XIST was derived from a protein-coding gene with exaptation of repeat elements in the large first and sixth exons [21, 22]. There are reports of impact of these internal exons, particularly exons four or five, on XIST/Xist transcription and function [23, 24]; however, our recent deletion of these regions in somatic cells revealed limited impact on XIST localization or maintenance of silencing [19].

In addition to developing a minimal functional XIST transgene, comparing the functionality of these constructs reveals the essential or redundant nature of domains within XIST. Our findings refuted the simple CCC model of PID functionality and demonstrated that duplication of the E repeat region improved XIST transgene functionality. The internal exons were not required for functionality of XIST. Removal of the F region, or surprisingly, duplication of the A repeat region impacted transcription, even from an exogenous inducible promoter. These results further our understanding of the critical domains of XIST, and our ability to synthesize a smaller long non-coding RNA as a cis-acting silencer.

Results

Expansion of human B repeat region is insufficient to replicate mouse PID functionality

We previously established a doxycycline (Dox)-inducible system in the male fibrosarcoma cell line HT1080 by inserting an FRT site on the short arm of chromosome 8 (8p) to allow for targeted site-specific integration of XIST transgenes [3]. With this system, we have shown that inclusion of two mouse PID domains gave the best functionality of our reduced XIST transgenes, and that the number of CCC motifs appeared to correlate with how well the transgenes silenced genes and recruited heterochromatin marks [11]. In order to examine the impact of increased number of CCC motifs on XIST functionality, we integrated the BhB(5x) trangene, which contains five copies of the human BhB repeat in the backbone AFE construct at 8p (Fig. 1A; [10]). This construct has the same number of HNRNPK-binding CCC motifs as the FullXIST, which is almost double that of PID-PID. Two independent clones for each construct were treated in triplicate with Dox to induce transgene expression. Expression levels of induced FullXIST, PID-PID and BhB(5x) were measured by RT-qPCR and showed no statistical difference in XIST expression, albeit with greater variability between clones of PID-PID (Fig. 1B). We then assessed the capacity of BhB(5x) to silence four endogenous genes in cis by pyrosequencing to determine allelic expression before and after XIST induction. Percent silencing was calculated as the change in allelic ratio after Dox relative to no induction (ND). CTSB, DLC1, SLC25A37 and STC1 genes are located at 5.1, 6.4, 16.9 and 17.1 Mb proximal to the transgene on 8p (Fig. 1A and C, Supplemental Table S1). Despite the number of CCC motifs in BhB(5x), FullXIST was significantly better at silencing all four genes on 8p. PID-PID was significantly better than BhB(5x) at silencing DLC1 and SLC25A37, although not CTSB and STC1. Previous work in our lab [11] found that silencing by PID-PID compared to FullXIST was lower but not significant; here, all four genes were significantly better silenced by FullXIST.

Figure 1.

BhB(5x) demonstrates poorer silencing capacity than PID-PID. (A) Schematic of transgenes of interest. Each construct is denoted by a unique colour. The number of CCC motifs is shown on the right. Bottom shows the location of the four genes examined for silencing and their distance from the FRT site of transgene integration. (B) RT-qPCR showing relative expression of transgenes relative to UBC after Dox induction. Each datapoint represents a cDNA sample, and the shape of the datapoints indicates the two independent clones for each transgene. (C) Silencing of CTSB, DLC1, SLC25A37, and STC1 by FullXIST, PID-PID, and BhB(5x) as measured using the change of allelic ratio before and after Dox induction. Each datapoint represents a unique induced RNA sample, and clones for each construct aredistinguished by shape. (D-F) enrichment scores of H2AK119ub (D-E) and H3K27me3 (F) under the XIST foci, obtained from IF-RNA FISH analysis. Two independent clones were analyzed for each transgene and pooled together to generate each boxplot. Sample size for (D), (E), and (F) is 95, 73, and 88 nuclei per clone, respectively. The tick on the y-axis indicates the estimated median t-score of FullXIST (blue) or PID-PID (purple) as a reference. H3K27me3 and H2AK119ub recruitment scores in PID-PID were shown to be ~ 80% of FullXIST in our previous study [11]. Statistical significance was determined for (B-C) by comparing all means of each construct using Tukey–Kramer HSD. Non-parametric comparisons to the PID-PID or FullXIST transgene were done using the steel method for (D-F). Significance: P < 0.0001 ***; P = 0.0001–0.01 **; P = 0.01–0.05 *; not significant, ns.

We also investigated the ability of BhB(5x) to recruit two heterochromatin marks, H2AK119ub and H3K27me3, to 8p by IF-RNA FISH. BhB(5x) was statistically less able to recruit either H2AK119ub or H3K27me3 than FullXIST, although it was not significantly different from PID-PID for H2AK119ub recruitment (Fig. 1D-F). The per-clone enrichment values and representative FISH images are shown in Supplemental Fig. S1 and S2. PID-PID was previously seen to recruit H2AK119ub and H3K27me3 at approximately 80% of FullXIST levels, but those differences were not significant [11]. Overall, BhB(5x) did not perform as well as PID-PID despite having almost twice the number of CCC motifs. Therefore, the contribution of the PID domain to silencing and heterochromatin recruitment is not simply attributable to its number of CCC motifs.

Duplication of the E repeats improves silencing

Given that adding more CCCs did not improve silencing, and that PID-PID tends to be less effective than FullXIST in recruiting heterochromatin and silencing [11], we tested manipulation of other regions to improve functionality. We chose to duplicate the A repeat as it is known to be required for silencing (+A transgene). Separately, we duplicated the E repeat (+E transgene) for its ability to bind CIZ1 [25] as well as PTBP1, MATR3, TDP-43 and CELF1 [17] (Fig. 2A). Both of these transgenes were created with PID-PID as the backbone. While the +E transgene showed robust expression with Dox treatment, even greater than the PID-PID or FullXIST controls, very low levels of +A transgene expression were observed for the two clones examined (Fig. 2B). Silencing ability was measured with pyrosequencing by comparing the allelic ratio of the four genes on 8p before and after Dox induction (Fig. 2C). The +E transgene had equivalent silencing ability as FullXIST for all genes but DLC1, where it was more similar to PID-PID. On the other hand, limited silencing was observed for the +A transgene: only STC1 had more than 30% silencing, and all but SLC25A37 were significantly worse than PID-PID.

Figure 2.

Improving efficacy of XIST transgenes. (A) Schematic of transgenes of interest. Each construct is denoted by a unique colour. (B) RT-qPCR showing relative expression of transgenes relative to UBC after Dox induction. Each datapoint represents a cDNA sample, and the shape of the datapoints indicates the two independent clones for each transgene. The data shown for FullXIST and PID-PID are the same as for Fig. 1B. (C) Silencing of CTSB, DLC1, SLC25A37, and STC1 by FullXIST, PID-PID, +a, and + E as measured using the change of allelic ratio before and after Dox induction. Each datapoint represents a unique induced RNA sample, and clones for each construct are distinguished by shape. The data shown for FullXIST and PID-PID are the same as for Fig. 1C. (D-E) enrichment scores of H2AK119ub (D) and H3K27me3 (E) under the XIST foci, obtained from IF-RNA FISH analysis. Two independent clones were analyzed for each transgene and pooled together to generate each boxplot. Sample size for (D) and (E) is 95 and 88 nuclei per clone, respectively. The tick on the y-axis indicates the estimated median t-score of PID-PID (purple) as a reference. H3K27me3 and H2AK119ub recruitment scores in PID-PID were shown to be 80% of FullXIST in our previous study [11]. Statistical significance was determined for (B-C) by comparing all means of each construct using Tukey–Kramer HSD. Non-parametric comparisons to the FullXIST transgene were done using the steel method for (D-E). Significance: P < 0.0001 ***; P = 0.0001–0.01 **; P = 0.01–0.05 *; not significant, ns.

Our IF-RNA FISH results also showed that the +E construct was not significantly different from FullXIST for recruiting H2AK119ub or H3K27me3 (Figs. 2D and E; Fig. S1 and S2). As +A exhibited significantly lower expression and poorer silencing, it was excluded from further analysis on recruitment of heterochromatin marks. Overall, these results suggest that elements within the E region can improve the functionality of the PID-PID transgene when duplicated, resulting in a construct larger than PID-PID, but still much smaller than FullXIST.

The F repeat region, but not the internal exons, are important for XIST functionality

To attempt to reduce the size of transgene necessary to establish silencing, we individually removed two internal regions containing either the F repeat (-F transgene) or spanning exons two to five (−Ex2–5 transgene). In the -F transgene, six YY1 binding sites were removed, as well as the short F repeat, which is more prominent in mouse. This deletion shortens the PID-PID transgene by ~ 600 bp, keeping the ~ 450 bp of the tandem A repeats intact. The -Ex2–5 transgene removed ~ 815 bp including the internal exons (Fig. 3A). RT-qPCR (Fig. 3B) revealed that -Ex2–5 was expressed significantly higher than PID-PID, while -F significantly dropped XIST levels. Pyrosequencing silencing data (Fig. 3C) indicated that the -Ex2–5 transgene was statistically equivalent to FullXIST for all but DLC1, for which it had a silencing ability more similar to PID-PID. The lower-expressed -F transgene had statistically lower silencing than PID-PID for all genes except CTSB, where the trend was still present.

Figure 3.

Testing functionality of -Ex2–5 and -F transgenes. (A) Schematic of the transgenes of interest and their respective sizes. Each construct is denoted by a unique colour. (B) RT-qPCR showing the relative expression of transgenes relative to UBC after Dox induction. Each datapoint represents a cDNA sample, and the shape of the datapoints indicates the two independent clones for each transgene. (C) Silencing of CTSB, DLC1, SLC25A37, and STC1 by FullXIST, PID-PID, −Ex2–5, and -F as measured using the change of allelic ratio before and after Dox induction. Each datapoint represents a unique induced RNA sample, and clones for each construct are distinguished by shape. The data shown for FullXIST and PID-PID are the same as for Fig. 1C and 2C. (D-E) enrichment scores of H2AK119ub (D) and H3K27me3 (E) under the XIST foci, obtained from IF-RNA FISH analysis. Two independent clones were analyzed for each transgene and pooled together to generate each boxplot. Sample size for (D) and (E) is 126 and 98 nuclei per clone, respectively. The tick on the y-axis indicates the estimated median t-score of FullXIST (blue) or PID-PID (purple) as a reference. Statistical significance was determined for (B-C) by comparing all means of each construct using Tukey–Kramer HSD. Non-parametric comparisons to the PID-PID transgene were done using the steel method for (D-E). Significance: P < 0.0001 ***; P = 0.0001–0.01 **; P = 0.01–0.05 *; not significant, ns.

By IF-RNA FISH, both the -F and -Ex2–5 transgenes showed localized XIST signals, and the induced XIST foci became enriched with H2AK119ub and H3K27me3 (Fig. 3D and E; Fig. S3 and S4). The -F transgene showed significantly less enrichment for both polycomb-deposited marks relative to PID-PID. The -Ex2–5 transgene showed comparable H2AK119ub and H3K27me3 enrichment compared to PID-PID. Overall, we observed a substantial impact of loss of the F repeat region on XIST expression, silencing and polycomb recruitment. This impact cannot be attributed to the decreased size of the construct as the removal of the internal exons resulted in a smaller construct that functions similarly to, or better than, PID-PID.

Silencing correlates with chromatin recruitment better than XIST expression level

The reduced ability of the +A and -F transgenes to silence and/or recruit heterochromatin marks led us to consider that, although all transgenes use the same inducible promoter, the variability in XIST expression levels across our clones might be impacting transgene functionality. Silencing of the four endogenous genes are significantly correlated with each other, and also with XIST (Fig. S5); however, the correlation with XIST is lost when the low-expressing -F and + A clones are excluded. We therefore performed a non-parametric Kendall’s correlation analysis to examine the relationships between the averaged percent silencing of the four endogenous genes, XIST expression level, and H3K27me3 and H2AK119ub recruitment on all of our clones. In this analysis we again see a moderate correlation between XIST expression and gene silencing (ⲧ = 0.495, P = 0.0138); however, silencing was most strongly correlated with both H3K27me3 (ⲧ = 0.788, P = 0.0004) and H2AK119ub (ⲧ = 0.667, P = 0.0026). H2AK119ub recruitment was also positively correlated with H3K27me3 (ⲧ = 0.455 P = 0.0397) and XIST expression (ⲧ = 0.485, P = 0.0282). The only comparison that was not significant was between XIST expression and H3K27me3 recruitment (ⲧ = 0.303, P = 0.1702). Thus across diverse constructs altering different domains of XIST, silencing shows a stronger correlation with chromatin recruitment than with levels of XIST expression.

Discussion

The mouse PID region, which is present in two copies in our base construct, consists of ~ 300 bp of mouse B repeat and 300 bp of mouse C repeat. In mouse, the C repeat is over 1.4 kb, while in humans the C repeat is less than a single copy and the 45 bp present in humans has limited conservation with the region in PID. In our prior work [10], a base construct including the repeats A, F and E neither silenced nor recruited marks (beyond CIZ1 protein binding). The addition of PID enabled silencing and recruitment of H3K27me3, H2AK119ub and H4K20me1. Doubling of the PID improved this silencing, and including the human BhB repeat also enhanced function (see Fig. 4B, [11]). The correlation between each mark and number of CCC motifs in the transgenic constructs was over 90%, leading us to hypothesize that more CCC motifs would further improve effectiveness. This was clearly not the case, with BhB(5x) being similar to, or weaker than, PID-PID despite having almost double the CCC motifs. Thus, it appears that the C component of the PID is contributing to functionality, and not replaceable by the human BhB. The PID region was defined in mouse as the region that recruits HNRNPK, and thereby the PCGF3/5 non-canonical PRC1, to the Xi [9]. The proximal 300 bp of the PID is comprised of the B repeat region while the distal 300 bp is comprised of the first three of the ~ 14 copies of the mouse C repeat. In mouse, deletion of the proximal region abolished PRC1/2 recruitment, while deletion of the distal region only decreased polycomb-deposited marks [9], supporting a partial role for this region in the recruitment of heterochromatin (see also [26]), which would be missing in our expanded BhB construct. Our previous work demonstrated more impact of HNRNPK loss on H2AK119ub than H3K27me3 recruitment, so it is feasible that the B-repeat binds HNRNPK while the C repeat region of the PID has an additional role, perhaps replaced by the expanded D repeat in humans as humans lack C repeats [27, 28]. Given the importance of the full PID structure, a future approach to improving transgene functionality could be inclusion of a third copy of the region (see Fig. 4C).

Figure 4.

Correlations between chromatin mark recruitment and XIST-mediated silencing. (A) Correlations between XIST transgene expression, average percent silencing of the four endogenous genes and t-scores of heterochromatic marks H3K27me3 and H2AK119ub. Correlation significance was assessed using two-tailed non-parametric Kendall correlation tests. The correlation coefficients and their P-values are shown in the heatmap. All transgenes are included in the correlation analysis, and each datapoint represents the average for a clone. All t-score and XIST expression datapoints were normalized to the average of the two FullXIST clones, or the estimated FullXIST average based on PID-PID results in the IF-RNA FISH experiments where FullXIST was not included. H3K27me3 and H2AK119ub recruitment scores in PID-PID were shown to be 80% of FullXIST in our previous study [11]. (B) Summation of the capacity to silence endogenous genes and recruit H3K27me3 and H2AK119ub of transgenes examined in the current and past publications [10, 11]. AFE + PID was called miniXIST. From current work only a single +A clone was examined for H2AK119ub. (C) Schematic of new constructs predicted to demonstrate improved capacity to silence endogenous genes and recruit heterochromatin marks.

The transgenic approach we utilized included an inducible promoter to limit variability in expression levels between different transgenic constructs. However, we observe some variability in XIST expression levels both between and within constructs. This variability is not explained by proportion of cells expressing XIST, which was similar between all clones. Notably, we were only able to induce XIST expression to a minor extent in clones with a duplication of the A repeats (+A) or a deletion of the F repeats (-F). Deletion of the A repeats from the endogenous locus consistently results in loss of XIST expression, even when the promoter region is intact (summarized in [19]). In addition, the A repeats have been argued to impact transcription of some transgenic XIST constructs, potentially by promoting transcriptional read-through (summarized in [14]). At the RNA level, the repeats are believed to form inter- and intra-repeat structures that might be impacted by a duplication of the region [29, 30]. Indeed, the number of A repeats is constrained across mammals [15], unlike other repeats within XIST. Thus, the A repeats are essential for transcription of endogenous XIST but have additional impacts on transcription that are still not understood. The F repeat has been suggested to act as a DNA-based enhancer, and the region deleted in -F transgenes includes six canonical YY1 motifs; however, the inducible promoter should not require additional enhancer elements. It is possible that the 600 bp removed in the -F construct could contribute to the functionality of a minimal XIST construct; for example, the YY1 motifs have been suggested to act as a ‘tether’ on the DNA [20], although we did not observe a loss of localization (Fig. S3B and S4B). Alternatively, if the F repeat normally acts to prevent self-silencing it is possible that deletion allows partial silencing of the transgene. Finally it is possible that our changes to the construct resulted in unforeseen interactions between elements within the RNA and or the DNA.

Two constructs, +E and -Ex2–5, each gained expression relative to the FullXIST and PID-PID controls, making it challenging to determine if the gains in silencing ability are simply attributable to XIST expression levels. We consider this less likely, given the weakening of the overall correlation between silencing and XIST expression when the low-expressors (-F and + A) are removed (Supplementary Fig. S5). Furthermore, FullXIST clones have consistently less XIST expression than PID-PID, yet are better silencers, thus the transgene content is able to overcome minor expression differences. The E repeat region is complex (reviewed in [6]), and is the largest repeat region remaining in our minimized construct; however, we felt it important to duplicate the region to determine if there was potential to improve the functionality of a minimal construct. The resulting 7.2 kb construct is the largest tested (after FullXIST), but did demonstrate equivalent activity to FullXIST. In the future it may be possible to combine the +E and -Ex2–5 to generate a smaller highly functional XIST transgene (see Fig. 4C). The E repeats are known to recruit both CIZ1 and a complex of PTBP1, MATR3, CELF1 and TDP-43 forming Xist-seeded protein assemblies with multivalent interactions enabling phase separation [16, 17]. This molecular crowding can involve many of the proteins binding to XIST/Xist with intrinsically disordered domains ([31], reviewed in [32]), and may contribute to the broad correlation we see between chromatin recruitment and silencing.

In Fig. 4B we summarize the various transgenes examined in this report, as well as previous studies using this system [10, 11]. The four genes we assessed for silencing were located from 5.1 to 17.1 Mb from the integration site, yet there was no consistent trend with proximity to XIST. While surprising, previous studies on the spread of silencing on autosomes translocated with the X chromosome have shown discontinuous spread of inactivation [33, 34]. Altogether, our results show that modifications to individual parts of XIST can significantly impact its function, while other regions appear non-essential for functionality of an XIST transgene. Removing the small exons two to five did not have a negative impact, despite their conservation across species, with some of the exons showing vestiges of the ancestral origin of XIST from the Lnx3 protein-coding gene [21, 22]. We saw that adding extra copies of repeat E and the PID region noticeably improved silencing and heterochromatic changes; however, the BhB(5x) demonstrated that this improvement is not as simple as increasing the number of CCC motifs present in the BhB repeats, suggesting that the content or the specific organization of these elements are needed for a full XIST functionality. This work reveals which elements are potentially necessary for XIST activity and reaffirms the possibility of creating a short, fully functional XIST transgene.

Materials and methods

Transgene generation and cell culture

The PID-PID transgene [11] was used as the backbone for +A, +E, -F, and -Exons constructs, while the AFE transgene [10] served as the backbone for BhB(5×). All transgenic constructs were synthesized by GeneScript. We include the sequence of PID-PID in Supplementary Fig. S6, with the regions added/deleted highlighted and the alternative BhB(5x). We used the HT1080 cell line for transgene integration into a previously established FRT site on chromosome 8 [35], using the FLP-In recombination System (Invitrogen #K6010–01). We selected two independent clones for each transgene. The cells were grown in DMEM media plus 10% fetal bovine serum, 5 mL of L-glutamine, non-essential amino acids, and penicillin–streptomycin. The expression of the transgene was induced for 5 days with Dox (1 μg/ml media concentration). No ethics was required for cell culture work.

RNA extraction and RT-qPCR

RNA extraction from all samples was performed using TRIzol (Invitrogen #15596–018) reagent according to the manufacturer’s protocol. Next, the RNA samples were treated with DNase I (Roche #4716728001) 1 U/μl per sample for 20 minutes at 37°C, followed by 10 minutes at 75°C to deactivate the enzyme. The RNA was used to generate the cDNA by reverse transcription using M-MLV Reverse Transcriptase (Promega #M1705), with the following conditions for each 20 μL reaction: 4 μL of 5x buffer, 0.4 μL of 25 μM dNTPs, 1 μL of 100 μM random hexamers, 0.5 μL of 40 U/μl RiboLock RNase inhibitor, 1 μL of M-MLV Reverse Transcriptase, and 1 μg of the RNA sample. Incubation was at 42°C for 2 h, followed by 95°C for 5 min. The cDNA was stored at −20°C. qPCR was performed using qPCRBIO SyGreen Blue Mix Lo-ROX (PCR Biosystems) with 1.5 μL of cDNA. The qPCR was done with a QuantStudio 3 (Thermo Fisher) thermocycler, using MicroAmp Fast Optical 96-Well Reaction Plates (Applied Biosystems), and the PCR program was: 95°C for 3 minutes, 40 cycles of (95°C for 5 seconds, 65°C for 30 seconds), followed by a melt curve stage. Gene expression was quantified using the ∆Ct method, normalizing to the endogenous expressed gene UBC. Each qPCR reaction was done in technical triplicates for each cDNA sample. Statistical analysis of expression was performed in JMP (JMP® [19] JMP Statistical Discovery LLC, Cary, NC, USA).

Pyrosequencing

STC1, DLC1, SLC25A37, and CTSB genes were amplified by PCR using cDNA samples from triplicate samples of two clones per construct, with the following conditions per reaction: 2.5 μL of 10x reaction Buffer, 0.2 μL of dNTPs 25 μM, 0.25 μL of primers forward and reverse 25 uM, 0.125 μL of Taq polymerase (NEB #0273S), and 1 μL of cDNA. The final reaction volume was 25 μL. The PCR products were then used for Pyrosequencing using the PyroMark® Q48 AutoPrep system (QIAGEN). Before each run, the machine cartridges were cleaned with high-purity water. Following the protocol instructions, we added 10 μL of PCR product and 3 μL of magnetic beads to each well of the Q48 disc. In each run, the amounts of nucleotides, buffers, substrate and enzyme were indicated by the PyroMark® Q48 software. The software automatically calculated the allelic ratios for the analyzed SNPs. Pyrosequencing was carried out for each construct under Dox and ND conditions. The percentage of gene silencing was calculated as the relative reduction in allelic ratio upon Dox induction, expressed as (ND—Dox)/ND, where ND is the average of triplicate untreated samples. Changes in allelic frequency were presumed to reflect inactivation. Statistical analyses were performed in JMP (JMP® [19] JMP Statistical Discovery LLC, Cary, NC, USA).

IF-RNA fish

The cells were grown on coverslips and induced with Dox (1 μg/ml of media) for 5 days. Next, the coverslips were fixed with the following protocol: the media was removed, and the coverslips were washed with 1x PBS, then rinsed in CSK buffer, then incubated for 8 minutes in CSK buffer containing 0.47% Triton X-100 (Sigma-Aldrich #93443). Afterward, coverslips were fixed in 4% PFA for 8 minutes and rinsed in 70% ethanol prepared with DEPC-treated H₂O. Samples were stored in snap-cap vials with 70% ethanol at 4°C. For the RNA FISH combined with IF, the coverslips were first rinsed with 1x PBS. Then, the coverslips were incubated facing down for 20 minutes with a mix of 100 μL of PBT (1% (wt/vol) BSA, 0.1% (vol/vol) Tween-20 (Sigma-Aldrich #P2286) in PBS) plus 1 μL RiboLock (ThermoFisher #EO0384) in parafilm, forming a small pocket for each coverslip. Next, the coverslips were moved to another Parafilm and incubated with a solution containing 100 μL of PBT, 1 μL of RiboLock and 1–1.5 μL of primary antibody (Supplemental Table S2) overnight at 4°C. The next day, the coverslips were washed with 0.1% Tween-20 in 1x PBS for 5 minutes at room temperature three times. Cells were then incubated with 1 μL of secondary antibody in 100 μL of PBT and 1 μL of RiboLock for 60 minutes at room temperature in the dark. From this step onward, coverslips were kept in the dark. Cells were washed three times with 1x PBS containing 0.1% Tween-20, then fixed with 4% paraformaldehyde for 5 minutes, followed by a final 1x PBS wash for another 5 minutes at room temperature. Then, the RNA FISH started by dehydrating the cells with a graded ethanol series (70%, 80%, and 100%) for 2 minutes at each concentration. The coverslips were then allowed to air-dry before the addition of the probe mixture. DNA probes were generated using the Nick Translation DNA Labelling System 2.0 (Enzo, #ENZ-GEN111–0050). For each coverslip, 6 μL of the labelled probe was combined with 12 μL of Human Cot-1 DNA (ThermoFisher, #15279011) and 2 μL of UltraPure™ Salmon Sperm DNA Solution (ThermoFisher, #15632011). The resulting mix was concentrated in a SpeedVac for at least 1.5 hours. The dried probe was subsequently dissolved in 10 μL of deionized formamide (Sigma-Aldrich, #F7503) and denatured at 80°C for 10 minutes. Following denaturation, 10 μL of hybridization buffer (20% BSA and 50% dextran sulphate in 4x SSC) was added and gently mixed. The final 20 μL volume was placed onto Parafilm, and coverslips were put cell side down over the drop, avoiding air bubbles. The coverslips were sealed within Parafilm envelopes and incubated overnight at 37°C. On the third day, the coverslips were gently rinsed in a mixture of deionized formamide and 4x SSC (Invitrogen, #15557044) for 20 minutes at 37°C. They were then transferred to 2x SSC for another 20 minutes at the same temperature, followed by a final rinse in 1x SSC at room temperature for 20 minutes. The coverslips were briefly rinsed in 1x PBS, and mounted with the cells facing down on glass slide using VECTASHIELD® Vibrance™ Antifade Mounting Medium with DAPI (VectorLabs #H-1800-2). The slides were stored at 4°C.

Pipeline for assessing recruitment of chromatin marks by XIST

To evaluate the recruitment of chromatin marks to XIST, images were acquired on a Leica SP5 confocal microscope using a 63x/1.4 oil immersion objective. We used four channels: DAPI to visualize nuclei, green for XIST, red for the mark of interest, and a merged channel. Between ten and twelve pictures per clone of each transgene were captured. All images were collected as single planes. The raw images were processed using a custom workflow in KNIME Analytics Platform (v5.4.4) (https://www.knime.com/). Each channel underwent background subtraction, and global intensity thresholds were set to segment XIST signals based on the levels at which they were clearly detectable. Nuclear segmentation was performed with Cellpose (https://www.cellpose.org/) node available from the KNIME community hub, using the DAPI signal to define individual nuclei, which allowed a first classification of cells as XIST-positive or XIST-negative. Then, the mean intensity of the target mark was measured in each XIST-positive nucleus, both under and outside the XIST focus. These mark intensity values were used to calculate an enrichment score under XIST focus (t-score) with the following formula:

where XIST⁺Int and XIST⁻Int represent the mean mark intensity co-localizing and not co-localizing with XIST focus in the nucleus; SD is the standard deviation of the mark intensity and Size is the number of pixels used to measure the mean intensities. A positive t-score would indicate an enrichment of the mark under XIST focus, whereas a negative t-score would suggest a depletion of the mark under XIST. This approach allowed us to evaluate whether different XIST transgenes were capable of recruiting specific chromatin marks, providing a quantitative measure. Statistical analysis was performed in JMP (JMP® [19] JMP Statistical Discovery LLC, Cary, NC, USA).

The data underlying this article is available in the online supplementary material or will be shared on reasonable request to the corresponding author.

Author contributions

Maria Jose Navarro-Cobos (Data curation, Formal analysis, Investigation, Writing—original draft), Bronwyn Julia Posynick (Data curation, Writing—review & editing), Christine Yang (Data curation, Writing—review & editing), Carolyn Janet Brown (Conceptualization, Data curation, Formal analysis, Supervision, Writing—original draft, Writing—review & editing).

Conflict of interest statement

All authors state that they have no conflicts of interest.

Funding

Support for the research was from the Canadian Institutes of Health Research (PJT-156048).