Imprinted X-chromosome inactivation at the gamete-to-embryo transition

SUMMARY

In mammals, dosage compensation involves two parallel processes: (i) X-inactivation, which equalizes X-chromosome dosage between males and females, and (ii) X-hyperactivation, which upregulates the active X for X-autosome balance. The field currently favors models whereby dosage compensation initiates “de novo” during mouse development. Here, we develop “So-Smart-seq” to revisit the question and interrogate a comprehensive transcriptome including noncoding genes and repeats in mice. Intriguingly, de novo silencing pertains only to a subset of Xp genes. Evolutionarily older genes and repetitive elements demonstrate constitutive Xp silencing, adopt distinct signatures, and do not require Xist to initiate silencing. We trace Xp silencing backwards in developmental time to meiotic sex chromosome inactivation in the male germ line and observe that Xm-hyperactivation is timed to Xp silencing on a gene-by-gene basis. Thus, during the gamete-to-embryo transition, older Xp genes are transmitted in a “pre-inactivated” state. These findings have implications for the evolution of imprinting.

Graphical Abstract

Whether imprinted X-inactivation occurs exclusively de novo in the embryo or is partially inherited from the male germline remains unclear. Here, Wei et al. demonstrate that silencing of evolutionarily older X-linked genes initiates in the paternal germline and continues into the early mouse embryo, consistent with a Pre-inactivation Model for X-imprinting.

INTRODUCTION

The challenge of regulating gene dosage is seen on both developmental and evolutionary time scales. Nowhere is this more apparent than in sex chromosome biology. Current theories suggest that the X and Y chromosomes evolved from a pair of autosomes and the acquisition of advantageous maleness genes on the Y led to suppression of recombination and progressive loss of genetic material on the Y over time^1–3 (Figure 1A). The incremental loss resulted in sudden dosage imbalances not only between males and females, but also between sex chromosomes and autosomes. Ohno therefore postulated that dosage compensation must involve two simultaneous processes in mammals (Figure 1B)^4,5: (i) X-chromosome inactivation (XCI), which silences one of two X-chromosomes in females to balance between XX and XY individuals⁶, and (ii) hyperactivation of the remaining active X (Xa) to balance the sex chromosome against the rest of the genome^7,8.

Figure 1. So-Smart-seq analysis of Xm-hyperactivation.

(A) Cartoon of the evolution of mammalian sex chromosomes with deduced evolutionary strata⁴⁹. Cartoon adapted from⁵³.

(B) The two tracks of dosage compensation in preimplantation embryos: X-A balancing versus XX-XY balancing. Timing and extent of each form are presently unclear. Am: maternal autosomes; Ap: paternal autosomes; Xm: maternal X; Xp: paternal X.

(C) Schematic diagram of So-Smart-seq of single embryos. Purple: hexamer barcode; Brown: oligo-dT; Blue: Reverse adapter; Green: Forward adapter; Grey: RNAs; black: DNAs.

(D) Two reciprocal interspecific crosses between Mus musculus (C57BL/6J) and Mus castaneus (CAST/EiJ) used in this study.

(E) Ratios of X-linked gene expression in relative to autosomal genes expression (X/A ratio, non-polyA RNAs included) in both male and female embryos from CM (left) and MC (right) crosses, respectively. Grey dot line marks the ratio of 1, indicative of full dosage parity. The number below each boxplot represents the number of embryos used in this analysis. Similar representation is applied to all relevant figures. *p=0.0013, **p=0.0055, ***p=0.0015, ****p=0.019 by Shapiro-Wilk test and unpaired two sample t-test. Centerlines in boxplots represent medians and box limits represent lower and upper quartiles. Whiskers represent 1.5x the interquartile range, and individual samples are represented as dots. Same box plot and dot plot representation were applied to all figures.

(F) Potential scenarios relating to X:A dosage compensation (X:A ratio) in female preimplantation embryos.

(G) Removing non-polyA RNAs from our RNAseq data recapitulates higher X:A ratios seen in published studies, for both male and female embryos. *p<0.001, **p<0.0005, by Shapiro-Wilk test and unpaired two sample t-test.

The mouse has served as a tractable model to study the two arms of dosage compensation. A major advantage of the murine model is the occurrence of imprinted XCI, the form found in evolutionarily earlier mammals, whereby the paternal X (Xp) is invariably inactivated. Although imprinted Xp-inactivation and maternal X (Xm)-hyperactivation are now well-accepted, whether they are already present at zygotic gene activation (ZGA) remains unclear. The prevailing view is that Xp genes reactivate at ZGA in the 2-cell embryo and undergo “de novo” silencing no sooner than the 4- to 8-cell stages^9,10. An alternative hypothesis^11–16, however, is that imprinted XCI may descend from a paternal germ line process known as meiotic sex chromosome inactivation (MSCI)^17–20. The notion that the X-chromosome reactivates globally after MSCI has been overturned, as indeed most X-linked genes remain suppressed within post-meiotic sex chromatin (PMSC) after MSCI^15,21 — further arguing for the hypothesis that Xp may be transmitted to the next generation in a partially, if not wholly, suppressed form to facilitate Xp silencing. “Pre-inactivation” would simultaneously solve the problem of dosage compensation as the Y-chromosome underwent evolutionary loss: Because the Xp is only transmitted to daughters, a pre-silenced Xp would achieve XX and XY balance without the need to evolve a zygote-based XCI. However, there is currently no evidence that Xp transitions into the embryo in a silent state.

Notably, prior studies had mostly focused on polyadenylated coding genes and excluded non-polyadenylated, noncoding, and repeat-associated transcripts that make up the vast majority of the mammalian transcriptome^22–24. Here we revisit the question by developing “So-Smart-seq” to capture a more comprehensive transcriptome. Our analyses reveal unexpectedly that de novo silencing pertains only to a subset of Xp genes and that older Xp genes and repeats are inherited from the male germline in a pre-suppressed state. These findings have implications for the mechanism and the evolution of dosage compensation in mammals.

RESULTS

Strand-optimized Smart-seq

To interrogate the transcriptome comprehensively, we developed Strand-optimized Smart-seq (So-Smart-seq) by combining Smart-seq2²⁵ with modifications to capture polyadenylated (polyA+) and non-polyadenylated (polyA-) RNAs, while excluding highly abundant rRNAs (Figure 1C). By fragmenting total RNA from single preimplantation mouse embryos, adding synthetic polyA tails, and tagging two ends of fragmented cDNAs with distinct barcoded adapters, So-Smart-seq captures all transcripts regardless of polyA status, preserves strand information, and quantifies transcript copy numbers in single preimplantation embryos. We subjected single 1C (1-cell), early 2C (2-cell), late 2C, 4C (4-cell), 8C (8-cell), and 16C (16-cell) embryos, in addition to early blastocysts (earlyB) to So-Smart-seq analysis, with ≥5 female biological replicates per stage (Table S1, S2). Compared to three previously published scRNA-seq methods^26–28, So-Smart-seq was equally effective at capturing polyA+ RNAs, but was superior for polyA- RNAs (as defined in ref²⁹) and low-abundance transcripts (Figure S1A–D). Noncoding RNAs (e.g., Malat1), polyA- mRNAs (e.g., histone clusters), and repetitive RNAs (e.g., MERVL) were more sensitively detected by So-Smart-seq (Figure S1C). So-Smart-seq also yielded better 5’ end and gene body coverage for transcripts >4 kb (Figure S1E,F). We compared our data to three published methods^9,30,31 by performing differential expression (DE) analyses on embryos of the same stage and cross. ~79% of the transcriptome was similar across the datasets. Among DE genes, however, our dataset yielded significantly greater representation of almost all non-PolyA RNAs (Figure S1G), as well as RNAs >4kb in length (Figure S1H). These findings argued for superior performance of So-Smart-seq in capturing non-polyadenylated and longer transcripts.

We then asked if our method is sufficiently quantitative enough to distinguish 2-fold difference in gene expression. To assess, we compared gene expression profiles from Xist knockouts (KO)³² versus WT embryos (Figure S1I), on the principle that Xist KO females would have almost twice the X-lined dosage as WT in the earlyB. stage^9,16. We clearly observed these 1.5-to 2-fold differences, whereas the male embryos (WT) did not exhibit differences (Figure S1J,K). Thus, our method is sensitive enough to detect less than 2-fold changes. Given the sensitivity and inclusion of non-polyA RNA, we sought to exclude RNA degradation products as substantial contaminants. We compared the fold-change (FC) of downregulated genes between 1C and late 2C embryos in our dataset versus published datasets⁹. At the 2C stage, the initiation of ZGA leads to degradation of maternally deposited RNAs³³. If RNA degradation products were substantially represented in our dataset, the DE analysis of 1C versus 2C would reveal larger fold-changes (negative values) in our data than in published data. The ΔlogFC values, however, were very close to 0 (1.19–1.40, 95% confidence interval, by One Sample t-test) — arguing against over-representation of RNA degradation products in our dataset (Figure S1L). The sensitivity and specificity were further demonstrated by Principal component analysis (PCA) showing clear separation of embryos by developmental stages (Figure S2A,B). Expression of stage-specific markers^34–36 was as expected, with high reproducibility among biological replicates (Figure S2C,D). Altogether, these data validated So-Smart-seq as a method of capturing a comprehensive transcriptome in single preimplantation embryos.

Xm-hyperactivation is present at ZGA

To ensure an X:A ratio of ≈ 1.0, the Xa is hypertranscribed in both males and females^5,37 (Figure 1B). The timing and extent of Xa-hyperactivation remain much debated^{7–9,31,38–40}. In the mouse, previous works suggested that Xa hyperactivation begins at the 4C stage or after^7–9,41. Some studies suggested incomplete Xa-hyperactivation and sub-parity between Xs and autosomes^7,8, while others found full X:A parity⁹. The temporal relationship between Xa-hyperactivation and XCI is also debated. Here we explored these questions by performing allele-specific So-Smart-seq in F1 embryos from interspecific crosses between Mus musculus (C57BL/6J) and Mus castaneus (CAST/EiJ). We performed the cross in both directions — (i) M. musculus x M. castaneus (MC) and (ii) M. castaneus x M. musculus (CM) (Figure 1D). ZGA occurs alongside maternal RNA degradation in two waves — one minor (early 2C) and one major (late 2C)⁴² (Figure S3A,B). In early 2C embryos, maternal loaded transcripts remained dominant, but early zygotic expression could already be detected (Figure S3A).

We first examined total expression from two X-chromosomes relative to autosomes and found similar profiles between males and females at all stages, except the late 2C and 4C stages when average X-linked expression was greater in female embryos (Figure S3C), consistent with the lack of chromosome-wide Xp-inactivation^10,16. To examine the dynamics of Xa-hyperactivation (henceforth, “Xm-hyperactivation”, as Xm=Xa), we plotted X:A ratios across developmental stages (Figure 1E). In male embryos, X:A = 1.0 would suggest full Xm-hyperactivation, X:A = 0.5 would suggest absence of Xm-hyperactivation, intermediate values would suggest partial Xm-hyperactivation. Our analysis revealed that the X:A ratio remained similar all across preimplantation male development, with X:A=0.72 in late 2C embryos to X:A≈0.65 thereafter (Figure 1E). These values approximate those estimated for male ES cells⁷ and argue that Xm-hyperactivation is already present by the late 2C stage (Figure S3D), though it does not achieve the full X:A parity. In late 2C female embryos, we observed X:A ≈ 0.85 (Figure 1E). Through pre-implantation development, X:A ratios also did not change significantly, except at the 8C stage when it transiently and modestly increased to ≈1. Thus, apart from a brief and weak dosage increase in 8C embryos, female embryos also displayed an X:A just under <1.0. Because of co-occurring imprinted XCI, an X:A = 0.7–1.0 is consistent with multiple scenarios still to be tested below (Figure 1F): (1) Two Xa’s without hyperactivation; (2) one hyperactivated Xm and one inactivated Xp; or (3) a partially hyperactivated Xm and partially inactivated Xp.

Our X:A estimates appear consistently lower than the X:A>1.0 reported previously^9,31. Notably, an X:A>1.0 would be possible only if Xm and Xp were both hyperactivated (Figure 1F, scenario 4), a possibility not supported by our present analysis (Figure 1E). A key difference is inclusion of polyA- transcripts in our study (Table S3, S4). Interestingly, when we computationally filtered away polyA- transcripts in our dataset, X:A ratios shifted upwards in both sexes and resembled previously reported values (Figure 1G; S3E). Thus, inclusion of a more comprehensive transcriptome using So-Smart-seq leads to a revised X:A ratio (≤1.0).

Given that maternally loaded transcripts are present in 2C embryos (Figure S3A,B)³⁰, one concern was that our calculation could be confounded by admixing maternal and zygotic transcripts. We therefore turned to an orthogonal method, 4sU-seq^43,44 (Figure 2A), which labels nascent transcripts at ZGA using 4-Thiouridine (4sU), allowing us to exclude maternally inherited transcripts during analysis. 2C embryos were treated with 4sU for 18 hours and RNAs were subjected to alkylation prior to RT and RNA-seq, resulting in T to C conversions in cDNAs from nascent zygotic transcripts but not from previously synthesized maternal transcripts. As expected, T->C changes were enriched over (Figure 2B). Furthermore, T->C conversions predominantly occurred in genes known to be expressed at ZGA (Figure 2B–D). Among transcripts with T->C changes, calculation of X:A ratios yielded 0.7 in male embryos and 0.6–0.9 in female embryos (Figure 2E). These estimates agreed with calculations from So-Smart-seq (Figure 1). The 4sU analysis therefore validates the So-Smart-seq pipeline and confirms an X:A ratio of ≤1.0.

Figure 2. Labeling zygotic RNAs in late 2C embryos using 4sU.

(A) Schematic diagram describing the zygotic RNA labeling experiment in late 2C embryos. Briefly, embryos grew in medium containing 4sU. Newly transcribed zygotic RNAs after ZGA incorporated 4sU and were labeled with T->C changes when alkylated by Iodoacetamide (IAA) and converted to cDNAs during RT.

(B) Observed conversion rate of all mutation types in no-4sU control (n=7, top) and 4sU labeled (n=12, bottom) embryos on the sense strand within genes. Total RNAs were divided into three groups based on their allelic origins: Paternal, maternal and neutral which were undistinguishable between two alleles.

(C) Examples in 4sU-seq showing RNA labelling efficiency for zygotic genes with high transcriptional potential at the onset of ZGA, as indicated by the heatmap. Genes were analyzed both in bulk (Total) and individually. Selected genes are indicated by the black arrows. A maximum of one experimental replicate may be removed due to the strong deviation from others.

(D) Examples of RNA labelling efficiency for possible maternally deposited genes with expected low transcriptional potentials after ZGA. Analyses were the same as (C).

(E) 4sU-seq: X:A dosage compensation (X:A ratio) in male and female embryos using 4sU-labeled zygotic polyA transcripts at the late 2C stage.

A class of constitutively silent Xp genes in the oldest evolutionary stratum

Dosage compensation is in equilibrium between two forces: (i) X-to-autosome balancing through X-hyperactivation, and (ii) XX:XY equalization through XCI. An X:A ratio ≤1 at ZGA still leaves open scenarios 1, 2, 3 (Figure 1F). Given that Xm in male embryos is already hyperactivated at the 2C stage (Figure 1E,2E,S3D), one might favor scenario 2 or 3 — both of which are characterized by Xp inactivation (XCI). To distinguish among them, we examined the extent and timing of XCI. Prevailing views posit that imprinted XCI is not initiated “de novo” until the 8C stage^10,30. However, the view that some Xp silencing may be originated in germline MSCI has remained attractive^11,13–15. As the timing differs between these mechanisms, the timing of Xm-hyperactivation could accordingly differ. Notably, previous studies disregarded polyA- transcripts and largely focused on average behavior of X-linked coding genes. Behaviors of individual genes could deviate significantly, however. In gene expression density plots, the global chrX and autosomal profiles were similar at the 2C and 4C stages and differed only after the 8C stage in MC-crossed embryos. CM-crossed embryos showed similar profiles, except that chrX reactivated faster at 2C and 4C stages (Figure S3A,B). Because Xist expression initiates at the late 2C stage (Figure S4A), earlier Xp silencing remained a possibility for specific genes, though not previously observed^9,10,30.

Here we applied So-Smart-seq and tracked individual genes (Figure 3, S4B). Focusing on subsets of genes having allelic information, we observed that, while chr13 genes became progressively more biallelic, some XCI genes remained maternally biased (Figure S4C). To investigate the heterogeneity, we performed K-means clustering on the 102 X-linked expressed genes with allelic information from 4C to early blastocyst (Figure 3A, left). In the MC cross, six distinct categories were revealed (Figure 3A, B). Genes in Categories 1, 2, and 3 mostly showed biallelic expression at the time of 4C and were progressively inactivated on Xp at the 8C, 16C, and early blastocyst stages, respectively (i.e., “early”, “mid”, and “late” de novo silencing). By contrast, Category 4 genes were silent at the 2C stage and appeared “constitutively” silent across preimplantation development (Figure 3B, C). Importantly, corresponding Xm alleles in Category 4 were robustly expressed in both sexes (Figure S4D). Constitutively silent genes accounted for the largest category and had the highest expression of any X-linked category, particularly at the 2C-4C stages. Genes in Category 5 (e.g., Kdm6a, Atp6ap2) consistently showed biallelic expression and were considered “escapees” of XCI. Finally, Category 6 genes showed castaneus-biased expression and were considered “strain-biased”. Average Xp expression differences between the categories revealed a clear resistance to Xp-reactivation among Category 4 genes (Figure 3D).

Figure 3. Identification of constitutively silenced Xp genes.

(A) Heatmap showing the 6 gene categories identified on the basis of their Xp silencing dynamics in both CM and MC crosses. Grey color, not detected. Sat1 and Pdzd11 in category 4 were removed from the downstream analyses due to strong inconsistency between CM and MC crosses.

(B) Boxplots showing the Xp silencing dynamics of 6 categories. N represents the number of genes in each category. The same N numbers are applied to all the boxplots below representing gene categories.

(C) Paternal expression fraction in each category at late 2C stage. P value was calculated by Mann-Whitney U test.

(D) Hierarchical clustering of 6 categories based on similarity in averaged silencing dynamics on Xp.

(E) Nascent RNA FISH of representative constitutive genes (white arrowheads) in late 2C embryos. Two alleles of early silenced gene Gnl3l (white arrows) are also labeled as an internal control. Scale bar: 5μm.

(F) Paternal fraction of reads from zygotic RNA transcripts in late 2C embryos by 4sU-seq. n=20 in category 4, n=16 in category 1 and n=9 in category 2+3. P values were calculated by Mann-Whitney U test.

(G) Linear distance of genes in each category to the Xist locus. *p=0.025, **p=0.001, ***p=0.010 and ****p<0.001 by Mann-Whitney U test.

(H) Interaction frequencies between Xist and genes of each category. P values calculated by Mann-Whitney U test.

(I) Relationship between gene categories and evolutionary age, as determined by location within evolutionary strata⁴⁹. Left: Enrichment of each evolutionary stratum in indicated category. Right: Enrichment of each category in indicated evolutionary stratum. Significance of enrichment determined by fisher’s exact test.

(J) Xm hyperactivation parallel tracks Xp silencing. Maternal RPKM value for each gene is normalized to its male homolog at the same stage. P values were calculated by Wilcoxon Signed-Rank test. P > 0.05 between stages in Category 4 or 5.

We confirmed the resistance to Xp expression using two orthogonal means. First, we examined nascent transcription at the single-cell level using RNA fluorescent in situ hybridization (FISH) and observed that Category 4 genes were either monoallelic or highly skewed to one allele at the late 2C stage (Figure 3E). Second, we analyzed 4sU-seq data and found that the Category 4 genes were clearly suppressed at ZGA, while Category 1–3 genes were biallelic expressors in the same embryos (Figure 3F). Thus, three independent methods — So-Smart-seq, RNA FISH, and 4sU-seq all demonstrate the existence of a large class of constitutively silent Xp genes (Category 4) at ZGA and they remained silent through pre-implantation development. This conclusion was further supported by the reciprocal CM cross, where the same Category 4 genes were also silent at the 2C stage (Figure 3A, right panel). Interestingly, the genes in the CM cross showed greater tendency to escape at the 4C stage before becoming re-inactivated, suggesting that the musculus Xp may be less stringently imprinted, consistent with a general tendency to become more biallelic across the whole Xp in the CM cross (Figure S3A,B) and with previously observed XCI dynamics in MC versus CM crosses⁹.

We then looked for underlying differences that may explain the differential treatment of constitutively silent genes (Category 4). GO analysis found no significant enrichment in any biological process, except nucleotide binding and deubiquitinase activities (FDR=0.054 and 0.074, respectively), in line with massive new transcription and en masse maternal protein degradation/recycling at ZGA. As gene silencing may be affected by distance to Xist^9,14, we noted that constitutively silent and early de novo genes were significantly closer in 2D distance to Xist (Figure 3G). Allele-specific analysis of Hi-C data⁴⁶ showed that they also exhibited more frequent interactions with Xist in 8C embryos (Figure 3H, S4E). Physical proximity to Xist is therefore correlated with the timing and strength of Xp silencing.

Most intriguingly, we observed a correlation with evolutionary age. Mammalian sex chromosome originated from a pair of autosomes that underwent a series of rearrangements and deletions over 300 million years, as reflected in multiple evolutionary strata on the X chromosome^47–49 (Figure 1A). Genes in older strata have had more time to acquire XCI^50,51. We used synteny between mouse and human chrX’s^24,49,51,52 to guide mapping of 6 gene categories. A total of 98 X-linked genes was stratified and all mapped to the 3 most ancient strata that pre-dated the eutherian divergence⁵³. Genes from Stratum 1 (the oldest stratum) showed significant enrichment in constitutively silent genes (Figure 3I, left panel). Stratum 1 genes also showed a higher percentage residency in early de novo genes (most similar to constitutively silent genes), though it did not reach statistical significance. However, early de novo genes showed closer proximity to Stratum 1 in comparison to mid and late de novo genes (Figure S4F). We then performed the reciprocal analysis and asked what gene category is enriched in each evolutionary stratum (Figure 3I, right panel). Significantly, Category 4 genes were also enriched in Stratum 1, while Categories 2+3 genes were relatively depleted. On the other hand, Categories 2+3 genes were significantly enriched in Stratum 2, while Category 4 was depleted. Thus, constitutively silent Xp genes are among the oldest X-linked genes and show closest proximity to Xist.

Xm-hyperactivation coincides with Xp-silencing

Given the heterogeneity of imprinted XCI, we asked if the timing of Xm-hyperactivation varied with the gene category. For each gene category, we calculated maternal RPKM in female embryos and normalized the data to Xm values in stage-matched male embryos. Significantly, the onset of Xm-hyperactivation coincided with initiation of Xp silencing across all categories (Figure 3J). For early de novo genes, Xp reactivation at the 4C stage (Figure 3A, B) coincided with loss of Xm-hyperactivation. When Xp was re-inactivated in 8C embryos, Xm-hyperactivation reappeared. For mid and late de novo genes, Xm-hyperactivation was lost at the 8C stage when embryos displayed a mild paternal expression bias, and resumed at 16C and earlyB stages when Xp initiated inactivation. For constitutively silent genes, however, Xm-hyperactivation was consistently observed across all stages, consistent with constitutive Xp silencing throughout (Figure 3J). For the perceived differences across the stages in Category 4, a one-sample t-test found statistically indistinguishable differences from 1.0 (Fig. S4G), a value representing equal Xm expression between male and female embryos (dash line, Figure 3J). Pair-wise comparisons between two consecutive stages also revealed no significant differences in Xm expression (Wilcoxon Signed-Rank test, p=0.173 between 4C and 8C; p=0.217 between 8C and 16C; and p=0.783 between 16C and early Blastocyst). Thus, constitutively silent genes are Xm-hyperactivated across all stages, consistent with Xp-inactivation across all stages.

Similarly, escapees showed no change in Xm expression across the stages, consistent with their escapee status. For other gene categories, we performed a one-sample t-test before and after Xp silencing. Consistently, we found female Xm expression was initially below 1.0 (p<0.05) before Xp silencing, but became statistically indistinguishable from 1.0 (p>0.05) upon Xp silencing (Figure S4G). We conclude that Xm-hyperactivation is temporally linked to Xp-inactivation and is regulated on a gene-by-gene basis, rather than being globally controlled.

Repression of constitutively silent Xp genes can be traced back to meiotic silencing

To ask how constitutively silent Xp genes are epigenetically distinguished, we first examined an established mark for genomic imprinting — DNA methylation^54–57. Bisulfite sequencing of single embryos revealed progressive demethylation of autosomes (e.g., chr1) during pre-implantation development, with asymmetry between the maternal and paternal genes evident by the 8C stage (Figure 4A,S5A), as expected. Unexpectedly, chrX retained symmetrical DNA methylation at promoters and gene bodies across all stages (Figure 4A, S5A). Category specific analysis also did not reveal any DNA methylation differences (Figure S5B).

Figure 4. Constitutively silenced Xp genes are epigenetically distinct.

(A) Metagene analyses of DNA methylation profiles from single-embryo bisulfite-sequencing across developmental stages. Gene bodies (± 2kb) shown for chrX and representative autosome, chr1. Pat: Paternal chromosome; Mat: maternal chromosome. TSS: Transcription start site; TES: Transcription end site.

(B) Heatmap showing allelic H3K27me3 enrichment on the gene bodies (± 2kb) for 6 gene categories (Cat.) at late 2C and 8C stages⁶⁰. Comp: composite.

(C) Averaged H3K27me3 enrichment profiles in late 2C embryos from (b). *p=9.31×10⁻⁹ in Cat.4, by Wilcoxon Signed-Rank test.

(D) Metagene analysis of chromatin accessibility based on ATAC-seq profiles at promoter regions (TSS ± 2kb) for chr1 and chrX at indicated stages.

(E) Metagene analysis of chromatin accessibility on the gene bodies (± 2kb) for different gene categories at indicated stages based on ATAC-seq profiles.

(F) Representative track views for differential chromatin accessibility and H3K27m3 profiles on Xm versus Xp for Category 4 genes.

We then turned to H3K27me3, an epigenetic mark catalyzed by Polycomb repressive complex 2 (PRC2) that has been associated with maternal-specific imprinting^58,59. Intriguingly, analysis of H3K27me3 ChIP-seq data from preimplantation embryos⁶⁰ showed striking allelic differences on the X-chromosome. Constitutively silent Xp genes (Category 4) were significantly enriched for H3K27me3, whereas other Xp genes showed no inter-allelic differences (Figure 4B,C). The H3K27me3 mark was particularly robust in constitutively silent genes in late 2C embryos. However, unlike previously reported X-linked imprinted genes^61–64, the difference was not evident in mature spermatozoa (Figure S5C). H3K27me3 instead became progressively enriched on constitutively silent genes in the late zygote (Figure S5C,D), suggesting that PRC2 interprets a primary imprint placed in the paternal germline.

Using ATAC-seq⁶⁵, we investigated whether chromatin accessibility may play a role. Chr1 demonstrated no allelic differences across preimplantation development, consistent with biallelic expression, whereas ChrX consistently showed ~2-fold greater signal on Xm relative to Xp (Figure 4D, S5E). This X-linked disparity is consistent with the dataset⁶⁵ being derived from pooled embryos including both sexes (assuming equal sex ratio, Xp signals would be half of Xm signals, if Xp and Xm had equal accessibility). However, when stratified by category, constitutively silent genes showed low accessibility on Xp relative to Xm shortly after ZGA (Figure 4E, F), whereas de novo genes and escapees have relatively equal Xm and Xp accessibility (Figure S5F). Thus, at the gamete-to-embryo transition, constitutively silent (Category 4) genes are differentially marked by enriched H3K27me3 marks and low chromatin accessibility.

We then asked if differences could be further traced backwards in developmental time to the male germ line. Spermatogenesis is complex 2-week-long process initiating with meiosis and is completed by a long post-meiotic maturation phase. Single-cell RNA-seq has profiled this process⁶⁶ and reaffirmed MSCI at pachytene^17–20 (Figure 5A,B, red arrow) and post-meiotic silencing^15,21 (post-meiotic sex chromatin, PMSC, green arrow). Substantially reduced expression of X-linked genes relative to autosomes was evident in both MSCI and PMSC (Figure 5A,B, S6A). X-linked genes generally remained suppressed from round spermatid stages 1–2 (RS1,2) to RS8 and mature spermatozoa (Figure 5A,B). Intriguingly, Category 4 genes were the most suppressed during the transition from meiosis II (MII) to RS8 stages (Figure 5B,C). Furthermore, a few Category 4 genes that showed high reactivation in RS4 became re-inactivated in the terminal maturation stages (RS6–8, Figure S6B). Nascent RNA FISH of Category 4 genes, Htatsf1 and Zrsr2, confirmed the post-meiotic silencing (Figure 5D,E). Furthermore, while chromatin accessibility in mature spermatozoa⁶⁷ was overall lower for X-linked genes than autosomal genes (Figure 5F), accessibility of the gene body was significantly lowest for Category 4 genes (Figure 5G–I, S6C). Thus, Category 4 genes are marked by distinct epigenetic signatures that can be traced back to meiotic silencing.

Figure 5. Constitutively silenced genes on Xp are silenced in spermatogenesis.

(A) Heatmap of X-linked and representative autosomal gene expression in single cells of the testis⁶⁶. Chr1 and chr13, representative autosomes. Spermatogenesis stages shown in chronological order. A1: type A1 spermatogonia; In: intermediate spermatogonia; BS: S phase type B spermatogonia; BG2: G2/M phase type B spermatogonia; G1: G1 phase preleptotene; ePL: early S phase preleptotene; mPL: middle S phase preleptotene; lPL: late S phase preleptotene; L: leptotene; Z: zygotene; eP: early pachytene; mP: middle pachytene; lP: late pachytene; D: diplotene; MI: metaphase I; MII: metaphase II; RS2: steps 1–2 spermatids; RS4: steps 3–4 spermatids; RS6: steps 5–6 spermatids; RS8: steps 7–8 spermatids. MSCI, Meiotic sex chromosome inactivation; PMSC, post-meiotic sex chromatin.

(B) Averaged gene expression values for various X-linked categories and representative autosomal genes for stages shown in (a). *p= 0.0036 at eP, **p= 0.0046 at mP, and ***p= 6.48×10⁻⁶ at lP, between autosomal (chr1 and chr13 combined) and chrX genes, by Mann-Whitney U test.

(C) Left, The mean expression fold-change (FC) of X-linked gene categories at post-meiotic stages (from MII to RS8). P values were calculated by Mann-Whitney U test. Right, the Category 4 genes are exclusively and significantly enriched in the differentially downregulated genes from MII to RS8 among the X-linked gene categories. *p=0.042, by fisher’s exact test. Differentially expressed genes were defined as genes with FDR (False Discovery Rate) <0.05, in Wilcoxon ranked sum test (see Methods).

(D) Nascent RNA FISH of constitutive genes (Htatsf1 (top) and Zrsr2 (bottom)) across spermatogenesis. Cell types were determined based on morphologies of lateral synaptonemal complex by immunofluorescence using SYCP3 antibody. White arrowheads mark the nascent RNA FISH signal. Scale bar: 8μm.

(E) Quantification of nascent RNA FISH signal shown in (d) in all replicates. L: leptotene; Z: zygotene; P: pachytene; D: diplotene; S: spermatids.

(F) Top, metagene analysis of chromatin accessibility by ATAC-seq in mature spermatozoa⁶⁷ for autosomal (chr1) and X-linked genes. Bottom, heatmap of ATAC-seq read coverages. Gene bodies ± 2kb shown. p=3.64×10⁻¹³ between chr1 and chrX, by Wilcoxon Signed-Rank test.

(G) Metagene analysis of ATAC-seq chromatin accessibility in spermatozoa⁶⁷ for various X-linked gene categories. Gene bodies ± 2kb shown.

(H) p-values comparing chromatin accessibility of gene bodies shown in (g) between constitutive genes and other listed gene categories, by paired samples Wilcoxon test.

(I) Heatmap of ATAC-seq read coverages across gene bodies (± 2kb) in spermatozoa for different X-linked gene categories.

Dosage compensation in the absence of Xist

We revisited the role of Xist during different stages of imprinted XCI^9,16,41. To determine if deleting Xist impacts Xm-hyperactivation and Xp-inactivation, we generated paternal Xist knockout female embryos (Xist^Xm/-) by crossing Xist knockout males (Xist^Y/-)³² with wildtype females (CM cross) and then performed So-Smart-Seq. Xist was clearly required for Xp gene silencing in the earlyB stage, as its deletion resulted in nearly equal Xm and Xp expression (Figure 6A,B, S7A,B). At the 2C stage, Xist loss only yielded a small, though significant difference between mutant and WT female embryos when considering all Xp genes in bulk (Figure 6B,S7A). However, stratifying the Xp revealed larger differences. De novo genes reactivated and could not re-inactivate when Xist was deleted (Figure 6A, S7), indicating that Xist is required to initiate and maintain silencing of these genes. In contrast, constitutively silent genes remained mostly suppressed at the 2C stage (Figure 6A,C,D,S7C). Consistent with this, 4SU-seq of 2C Xist mutants also showed suppression of Category 4 genes (Figure 6E). At the 4C stage, constitutively silent Xp genes progressively reactivated in the mutants, in line with that observed in wildtype embryos with the same genetic background (CM cross, Figure S7C). Thus, constitutively silent Xp genes resist reactivation at the 2C stage even without Xist. However, Xist is required to maintain or re-initiate silencing of these genes starting at 8C stage, with timing coinciding with the loss of H3K27me3 marks in 8C embryos (Figure 4B). Overall, average Xp expression did not dramatically increase at the 8C stage in Xist knockouts when compared to the earlyB stage (Figure 6B,F). This relative Xp suppression is consistent with an Xist-independent mechanism descending from the paternal germline.

Figure 6. Effects of Xist ablation on Xp gene and repeat silencing.

(A) Heatmap showing effects on Xp silencing for each gene category in Xist^Xm/- versus WT embryos. WT heatmap duplicated from Figure 3A for comparison. Upper right panel: Values averaged across genes of each category in Xist^Xm/- embryos. Grey color, not detected.

(B) Density plots of paternal expression percentage for X-linked genes in female Xist^Xm/- vs WT (MC cross) embryos. P values by Mann-Whitney U test.

(C) Fraction of Xp expression for constitutive (Cat. 4) versus early de novo (Cat. 1) genes in female Xist^Xm/- embryos. P values by Mann-Whitney U test.

(D) Fraction of Xp expression for constitutive/Category 4 genes in female Xist^Xm/- and WT (MC cross) embryos. P values calculated by Wilcoxon Signed-Rank test.

(E) 4SU-seq analysis: Fraction of Xp reads from zygotic transcripts in late 2C Xist mutant embryos. n=18 in category 4 and n=17 in category 1. P values were calculated by Mann-Whitney U test.

(F) Xp expression fraction for genes (top) and LINEs (bottom) in female Xist^Xm/- and WT embryos at 8C and early blastocyst stage. P values by Mann-Whitney U test. Observation number: for genes, n=128 and 120 in Xist^Xm/- and WT embryos at 8C, respectively, and n=171 and 129 in Xist^Xm/- and WT embryos at early Blastocyst, respectively. For LINEs, n=27 and 22 in Xist^Xm/- and WT embryos at 8C, respectively, and n=30 and 21 in Xist^Xm/- and WT embryos at early Blastocyst, respectively.

(G) Analysis of Xm hyperactivation in the absence of Xp silencing in Xist^Xm/- embryos. Xm RPKM for each gene in female embryos is normalized to that in male embryos at the same stage. p > 0.05 between stages in all categories, by Wilcoxon Signed-Rank test.

(H) X/A ratio in female Xist^Xm/- embryos during preimplantation development (left) and in undifferentiated female mESCs by bulk RNA-seq (right). Male sibling embryos from the same cross served as controls.

(I) Density plots of Xp expression percentage for old versus young LINEs in WT female embryos. P values by Mann-Whitney U test.

(J) Density plots of Xp expression percentage for old LINEs in WT versus Xist^Xm/- embryos. P values by Mann-Whitney U test.

We also asked how Xist loss affects Xm-hyperactivation. In WT embryos, Xm-hyperactivation occurred together with Xp-inactivation (Figure 3J). The overall Xm expression in Xist-mutant embryos was consistently below 1.0 at 8C and earlyB stage, especially for early de novo genes (p=0.005 at 8C stage, and p=0.037 at early blastocyst, by one sample t-test) (Figure 6G). The overall X:A ratio was also approximately 1.0 from 8C to earlyB stage (Figure 6H, left panel). Thus, the loss of imprinted Xp silencing resulted in a loss of Xm-hyperactivation. This X:A ratio resembled that observed in pre-XCI female mouse embryonic stem cells (mESC), which possessed two active X chromosomes (Figure 6H, right panel). These data support the idea that Xm-hyperactivation occurs secondarily to Xp-inactivation, essentially occurring only when Xp-inactivation has occurred. We conclude that Xm-hyperactivation is not pre-determined and is instead tightly linked to XCI.

Constitutive silencing of older Xp LINEs

Although the X chromosome is enriched for repetitive elements²³ and studies have proposed various roles for them during XCI^16,68–70, dosage compensation studies have focused almost exclusively on coding genes. Repeats are inherently challenging to study because of their multicopy nature, less complete annotation of SNPs and indels, and difficulties with aligning and mapping. Here we developed a bioinformatic pipeline to map musculus versus castaneus repeats. We used a de novo assembly of CAST/EiJ genome⁷¹, but kept only uniquely aligned reads from repetitive elements, and determined allelic origins by comparing alignment qualities to musculus versus castaneus reference genomes. To minimize bias, we included only reads with hits on both alleles. Overall, this pipeline retained >50% and >80% of LINE and SINE elements, respectively, with good coverage across the entire X chromosome (Figure S7G,H). To validate the pipeline, we reasoned that few reads (if any) should be assigned to castaneus (Xp) in male MC embryos. This was the case (Figure S7I). Furthermore, in 1C zygotes, reads should be exclusively musculus, due to maternal loading of transcripts. This was also the case (Figure S7J). Third, because SINEs are rapidly demethylated in the paternal genome at the late pronucleus stage⁷², we anticipated some paternal SINE expression from 1C embryos. This was indeed the case (Figure S7J). We also tested the pipeline on a bulk fibroblast RNA-seq dataset derived from the pure Mus castaneus mice and observed exclusive mapping of repeat reads to the castaneus allele (Figure S7K). Thus, reads from repetitive elements can be assigned alleles with high confidence using our pipeline.

We asked if repeats are subject to imprinted XCI. For chrX, the 4C stage showed biallelic expression of almost all repeat families, but later stages showed a tendency towards reduced Xp expression (Figure S7L) — in contrast to autosomal repeats which remained biallelic. Focusing on LINEs because of various proposed roles during XCI^16,68–70, we found that LINEs activated earlier than genes, with paternal expression already evident in early 2C embryos (Figure S7L,M). Across preimplantation development, in contrast to X-linked genes, a large fraction of X-linked LINEs already showed strong Xp silencing at the 8C stage, contrasting sharply with genes (Figure 6F, S7M). Only in the earlyB stage did bulk LINE expression become Xm-biased. This indicates high LINE heterogeneity and escape of some LINEs from imprinted XCI.

Analysis of Xist-mutant embryos revealed that LINE silencing was less dependent on Xist, as a large variance in Xp LINE expression was observed at 8C and beyond (Figure 6F). Notably, LINEs can also be sub-stratified by age into (i) ancestral LINEs (‘old’) present in all eutherians (e.g., L1M2/3) and (ii) younger LINEs (‘young’) that evolved later only in the house mouse (e.g., Lx4/5/6). Indeed, old LINEs behaved differently from young LINEs. First, a majority of old LINEs showed Xp suppression (<25% Xp expression) at ZGA (Figure 6I, yellow), whereas young LINEs showed biallelic expression (green). Interestingly, old LINEs continued to show Xp silencing throughout preimplantation development, while young LINEs did not initiate silencing until the 8C-earlyB stages (Figure 6I, yellow vs. green). The lack of Xp expression was not due to promoter truncation, as the same LINEs were expressed on Xm. Second, silencing of old and young LINEs displayed differential sensitivity to Xist loss: The Xist knockout had no major effect on suppression of old LINEs until the earlyB stage (Figure 6J). Thus, just like old Xp genes, old Xp LINEs tend to be constitutively silenced, are less dependent on Xist for silencing, but require Xist to maintain silencing in blastocysts.

DISCUSSION

X-linked dosage compensation in mammals encompasses two processes: (i) X-to-autosome balancing and (2) XX:XY equalization (Figure 7A,B). In the early embryo, the first is achieved by Xm-hyperactivation, while the second is achieved by imprinted Xp inactivation. Two hypotheses had previously been proposed for the mechanism of Xp silencing — the Pre-inactivation Model¹⁴ and the De novo Model¹⁰. Until now, the field had favored the De Novo Model. Here, with So-Smart-seq, analysis of a comprehensive transcriptome inclusive of polyA-transcripts, noncoding RNAs, and repetitive elements reveals that the X chromosome is more heterogeneous than previously suspected. For imprinted XCI in the early embryo, the existence of constitutive Xp silencing of Category 4 genes and the continuity with MSCI and PMSC in the male germline provide evidence for the Pre-inactivation Model, most notably involving Xp genes in the oldest evolutionary stratum. Older Xp LINEs also follow the pattern of constitutive silencing in the early embryos. Crucially, constitutive silencing is unique to Xp, as Xm counterparts have robust expression at the 2C-4C stages. Rather than being unified by a functional pathway, constitutively silent genes are among the most highly expressed in preimplantation embryos (Figure S4D). Their over-expression may therefore be less tolerable during the sensitive 2C stage⁷³.

Figure 7. Two types of dosage compensation in preimplantation embryos.

(A) For X-linked genes, Xm hyperactivation is present throughout preimplantation development in both male and female embryos. In female embryos, Xm hyperactivation is linked temporally to Xp inactivation, with both occurring on a gene-by-gene basis.

(B) Two parallel tracks of dosage compensation in preimplantation development. Old genes and LINE repeats tend to be constitutively silent on Xp. Xm hyperactivation occurs contemporaneously with Xp silencing on an individual gene basis.

The relative Xist independence^16,41 and the observation that constitutively silent genes and LINEs map to the oldest evolutionary stratum together lend credence to the idea that a male germline-driven X-pre-inactivation may be the ancestral form of imprinted XCI. Indeed, imprinted XCI is widely observed in older mammals, such as marsupials, despite lacking an X-inactivation center and an Xist gene^74–77. In the male germ line of all mammals, the X and Y chromosomes are inactivated together during pachytene (MSCI)^17–20 and remain so after meiosis^15,21. Post-meiotic X-silencing makes possible the transmission of Xp genes to daughters in a pre-inactivated or constitutively silent state^11–14,16. Our analysis demonstrates that constitutively silent Xp genes are differentially treated during the gamete-to-embryo transition, being marked by the lowest chromatin accessibility in late spermiogenesis (Figure 5G–I) and by greatest PRC2/H3K27me3 enrichment in the zygote — consistent with an inactive state during the generational transition. Pre-inactivation would both resolve X-autosome dosage imbalance during Y-chromosome degeneration across evolution, as well as ensure XX-XY balance during early embryo development.

It is tempting to speculate that the unpaired state of the X and Y chromosomes during pachytene initially drives Xp imprinting⁷⁹, perhaps placing the original mark, which is then interpreted by the zygote. H3K27me3 was previously associated with maternal imprinting^58,78, but the current work suggests that H3K27me3 may also be involved in paternal imprinting. It is known that 1–10% of nucleosomes and associated marks, including H3K27me3, can be retained in mature spermatozoa^67,80–83. The retained marks could either serve as the original germline imprint or instead drive zygotic interpretation of the original imprint.

Finally, our study demonstrates that Xm-hyperactivation occurs secondarily to Xp silencing and is timed to silencing of individual genes (Figure 7A,B). Xm-hyperactivation is plastic, rather than pre-determined, and occurs when there is silencing of the Xp counterpart. The timing of Xm-hyperactivation for each gene varies in accordance with timing of Xp silencing. In Xist knockout female embryos, the absence of XCI obviates the need for Xm-hyperactivation altogether, whereas in male embryos with only one X-chromosome, Xm-hyperactivation occurs constitutively at or shortly after the time of ZGA. The linkage between Xm-hyperactivation and Xp silencing may reflect gene-specific feedback mechanisms, rather than a whole-chromosome mechanism. Feedback mechanisms could include post-transcriptional regulation involving miRNA and/or transcription factors that would together control steady state RNA levels. In summary, contrary to the current understanding, our study demonstrates that the mouse X-chromosome is highly stratified and exhibits characteristics of both constitutive Xp silencing and de novo Xp inactivation, with Xm-hyperactivation responding secondarily to Xp silencing dynamics for each gene.

LIMITATIONS OF THE STUDY

So-Smart-seq was designed for a pan-transcriptome view of coding, noncoding and repetitive elements in the early embryo, with better coverage from 5’ to 3’ ends for more accurate allele-specific analysis. While the data argue that Xm-hyperactivation is tightly linked to and responds to XCI in preimplantation embryos, we cannot rule out that Xm (or Xa) hyperactivation may become uncoupled from Xp (or Xi) silencing at other developmental stages. Furthermore, although we have identified a unique subset of constitutively silent genes during the gamete-to-embryo transition and implicated H3K27me3 as an epigenetic mark, the underlying mechanisms by which these genes are silenced in the paternal germline and maintained as silent genes during the transition to the embryo remain to be investigated in full.

STAR★METHODS

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Jeannie T. Lee (lee@molbio.mgh.harvard.edu).

Materials availability

Mouse line generated in this study will be available from the lead contact upon request.

Data and code availability

• All new single-embryo RNAseq and Bisulfite-DNAseq data generated in this study have been deposited at Gene Expression Omnibus (GEO) database under the accession number GSE168455. Microscopy data reported in this paper will be shared by the lead contact upon request.

• All original R code has been deposited at Dataverse with the DOI:10.7910/DVN/W7RJFI.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Experimental model and study participant details

Mice

All mouse experiments presented in this study were conducted in accordance with the animal research guidelines of NIH and approved by the Institutional Animal Care and Use Committee of Massachusetts General Hospital. All wild type preimplantation embryos were derived from reciprocal natural crosses between C57BL/6J and CAST/EiJ. The paternal Xist knockout embryos were obtained by mating wildtype CAST/EiJ females with Xist^−/Y males³² (129S1/SvlmJ). All used mice were at the age of between 8 and 12 weeks.

Cell lines

The mES cell line is a M. musculus/M. castaneus F2 hybrid cell line carrying a mutated Tsix allele was previously described as “Tsix^TST/+”⁸⁴. mES cells were grown in feeder-free 2i medium (DMEM/F12 media, Neurobasal media, 2% Hyclone FBS (Sigma), N2 and B27 supplements, Glutamax, 1x Pen/Strep, 0.1 mM βME, and 1000 U/mL ESGRO recombinant mouse Leukemia Inhibitory Factor (LIF) protein (Sigma, ESG1107)) supplemented with 1uM PD0325901 and 3uM of CHIR99021 (Selleck Chemicals) at 37°C with 5% CO₂

METHOD DETAILS

Preimplantation embryo preparation

Embryos at the stage of PN5 zygote, early2cell, late2cell, 4cell, 8cell, 16cell and early blastocyst (32–64 cell) were harvested at approximately E0.5, E1.1, E1.5, E1.75, E2.25, E2.75, E3.5, respectively. Each embryo was closely examined before experiments to ensure its normal morphology and correct number of blastomeres for the stage of interest. Embryo collections were performed by flushing oviduct and uteri with M2 medium (EMD Millipore) with a syringe and embryos were washed twice in M2 medium. To remove zona pellucida, each embryo was briefly incubated in Acid Tyrode’s solution (Sigma-Aldrich), followed by three washes in PBS containing 1mg/ml acetylated BSA (Sigma). During the wash, gently pipetting was carried out using a mouth pipette to disassociate polar bodies from embryos. Following the wash, each single embryo was manually picked up and transferred into a 0.2ml PCR tube by mouth pipette with minimum amount (< 0.3ul) of carryover solution.

cDNA library preparation from a single embryo

Single embryo was lysed in 1ul lysis solution containing 1% NP-40 and 4U RNase inhibitor (Roche). To fragment total RNAs from the embryo lysis, a combination of 1ul of 5× 1st strand buffer and 0.1 ul of RNA spike-in or water was added into the lysis, followed by the incubation at 82°C for 12 minutes. After quick chill on ice, the RNAs were next end repaired by adding 1ul PNK solution (70mM tris-HCl pH7.6, 10mM MgCl₂, 15mM DTT, 4 units/ul RNase Inhibitor) containing 2 units of T4 PNK (NEB) and by incubating at 37°C for 10 minutes. The RNA samples were further incubated at 37°C for 5 minutes after the addition of 0.5ul mix containing 1 unit of PolyA polymerase (NEB) and 3 picomole ATP. Before reverse transcription, 1ul of pre-RT mix containing 6.25mM dNTPs, 15mM EDTA, and 5uM Reverse-dT Primers (TTTCCCTACACGACGCTCTTCCGATCTNNNNNNTTTTTTTTTTTTTTTVVN) was added to samples. The “RNA identifier” — a unique 6-nt barcode was embedded into each Reverse-dT primer to tag each RNA fragments in the downstream reverse transcription (RT) reaction. For RT, samples were first heated up at 72°C for 2 minutes for denaturation, and then immediately chilled on ice, followed by the addition of 5.5ul fresh made RT mix containing 1X 1st strand Buffer, 10mM DTT, 2U/ul RNase inhibitor, 2M Betaine, 1uM GGrG oligo (iCiGiCGGAGTTCAGACGTGTGCTCTTCCGATCTrGrG+G)⁸⁵ and 20U/ul Superscript II reverse transcriptase²⁵. Samples were then incubated with the following temperature setting: 40°C for 5 mins, 42°C for 30 mins, Two cycles of 50°C for 2 mins followed by 42°C for 2mins, and 15 mins at 70°C. First strand cDNA libraries were next amplified by 15 cycles of first round PCR (F: GTGACTGGAGTTCAGACGTGTGCTCTTC; R: ACACTCTTTCCCTACACGACGCTCTTC) in a total volume of 30ul PCR reactions (Kapa high fidelity PCR mix) and purified with Ampure XP beads at 1:1 ratio.

Ribosome cDNA depletion and 2nd round PCR

RNA probes (Table S5) targeting different loci of ribosomal cDNA were synthesized using MEGAscript T7 transcription kit (ThermoFisher Scientific) with biotin labeled CTP (30% of total CTP). RNA probes were then gel purified and pooled together with equal molarity. To hybridize ribosomal cDNAs with probes, purified cDNAs from PCR were mixed with 130ng of RNA probes in total 30ul of hybridization buffer (10mM Tris-HCl, pH8.0, 0.5M NaCl, 5mM EDTA and 0.2% SDS), and incubated at the following temperatures: 94°C for 2 mins, followed by ramp down to 50°C with 1°C down per 50 seconds, and finally at 50°C for 5 mins. Meanwhile, 50ul of streptavidin C1 beads (ThermoFisher Scientific) were first washed twice with pure water, and then pre-blocked in 50ul of 1x hybridization buffer containing 1ug of 50nt ssDNA and 1.2ug Cre dsDNA fragments (400bp). After 30mins incubation at 50°C with rotation, beads were washed twice with 1x hybridization buffer, and resuspended again in 30ul of 1x hybridization buffer. Beads were then warmed to 50°C while waiting for the cDNA-probe hybridization to complete. To delete ribosomal cDNAs, samples and beads were mixed at 50 °C and incubated for another 5 mins at 50 °C, and immediately put on a magnetic rack. Supernatant was collected and purified with Ampure XP beads. The purified cDNAs were further amplified by 9 cycles of PCR with Multiplexed primers (NEB). The final PCR product was purified and subject to sequencing.

Bisulfite DNAseq library preparation

Bisulfite DNAseq libraries were prepared following a previously published protocol⁸⁶, with some minimal modifications. Briefly, single embryo was lysed in RLT plus buffer (Qiagen), followed by the addition of unmethylated lambda DNA control. Genomic DNAs were treated with CT conversion reagent (EZ methylation kit, Zyon). After the purification using PureLink PCR Purifiation kit (ThermoFisher Scientific), DNAs were amplified 5 times using Preamp oligos and Klenow exo- polymerase (NEB). Amplified DNAs were depleted of single strand DNAs by adding exonuclease I, followed by the purification using Ampure XP beads. The second round of DNA synthesis reaction was carried out using Adapter 2 Oligos. Finally, all DNAs were purified and further amplified using 14 cycles of PCR. The yielded DNAs were purified and ready for bisulfite sequencing. The sex of the libraries was determined based on the expression level of Y-linked genes via qPCR, and only the female libraries were sequenced.

Zygotic RNA labeling with 4-thiouridine in mouse embryos

One-cell zygotes at around E0.75 were collected as described above, and cultured in Advanced KSOM embryo medium (Millipore) containing 90uM 4-thiouridine (Sigma), with 10–15 embryos in every 30ul of medium. Medium was changed very 6 hours. Embryos were harvested at approximately E1.5, and each single embryo was transferred to a 0.2ml PCR tube. After the addition of 3ul of lysis buffer (2.5ul of 200mM Sodium Phosphate buffer pH=8, 0.2ul of RNase inhibitor, 0.1ul of 4sU spike-in, and 0.2ul 10% NP-40), the lysed embryo freezed at −80°C for temporary storage or immediately proceeded to downstream library preparation.

Alkylation-dependent RNAseq library preparation

Libraries were prepared as previous described⁴⁴, with a few modifications. 10ul of Dynabeads (MyOne Dynabeads Strepavidin C1) were transferred to a clean PCR tube, and washed twice with 10ul of Buffer A (0.1M NaOH, 0.05M NaCl), twice with 10ul of Buffer B (0.1M NaCl), followed by the incubation in 20ul oligo-dT buffer containing 5mM Tris-HCl pH7.5, 0.5mM EDTA, 1M NaCl and 50uM oligo-dT (/5Biotin-TEG/AAGCAGTGGTATCAACGCAGAGTACTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTVN, IDT) at room temperature (RT) for 15 min with agitation. The incubated beads were then washed twice with 20ul 1x B&W buffer (5mM Tris-HCl pH7.5, 0.5mM EDTA, 1M NaCl), and resuspended in 30ul freshly prepared resuspension buffer (5.5ul RNase inhibitor (40U/ul) in 26.5ul RNase-free water) on ice. Lysis from each single embryo was kept on ice after thawing from −80°C, and 2ul of processed beads were added into each tube, followed by a brief vortex and incubation for 20 minutes at RT with agitation to allow complete RNA binding to beads. 5ul of fresh alkylation solution (20mM IAA in DMSO) was added into each sample with gentle pipetting. Samples were then spined briefly and incubated at 50°C for 15 min with agitation. After alkylation, samples were briefly spined and immediately put on the magnetic rack, followed by the addition of 10ul of STOP solution (2x Superscript II buffer, 0.3% Tween-20 and 60mM DTT) with gentle pipetting. The samples were spined again and kept on the magnetic rack for 5 min at RT. On the magnetic rack, after removing the supernatant from the beads that binds PolyA RNAs, 10ul of newly prepared RT mix (1X Superscript II RT buffer, 5mM DTT, 0.8U/ul RNase inhibitor, 1M Betaine, 10mM MgCl2, 1mM dNTPs and 1uM GGrG oligo (iCiGiCGGAGTTCAGACGTGTGCTCTTCCGATCTrGrG+G)) was immediately added toward the side of the immobilized beads into the tube to perform the RT using the following temperature: 42°C for 90 min, 10 cycles of 50°C for 2 min followed by 42°C for 2min. The samples were vortexed every 20 min during the RT. The first strand cDNA products from each embryo were next amplified by 22 cycles of PCR (F: GTGACTGGAGTTCAGACGTGTGCTCTTC; R: AAGCAGTGGTATCAACGCAGAGT) in a 30ul reaction (Kapa high fidelity PCR mix) and purified with Ampure XP beads at 1:1 ratio. The PCR products were further processed using Nextera XT DNA library preparation kit (Illumina) with multi-indexes, as previously described⁴⁴. The resulting libraries were pooled and subject to sequencing.

mESC RNA-seq library preparation

Total cell RNA was extracted using TRIzol Reagent (Thermo Fisher Scientific), from which rRNAs were depleted using RiboMinus Eukaryote Kit v2 (Thermo Fisher Scientific) following manufacturer’s protocol. RNA-seq libraries were prepared in two biological replicates using NEBNext^® Ultra II Directional RNA Second Strand Synthesis Module and NEBNext Ultra II DNA Library Prep Kit for Illumina (New England BioLabs) as per manufacturer’s instructions.

Library sequencing

All libraries were pair-end sequenced on the platform of Illumina Hiseq 4000, Hiseq 2500 and Novaseq, following the manufacture instructions.

So-smart-seq data analysis

Raw data was first examined using Trim Galore (The Babraham Bioinformatics Group) to remove adapters from reads using the following parameters: (--stringency 8 --phred33 -e 0.2 -paired --length 43 --r1 44 --r2 44), followed by the PCR duplication removal based on sequences of attached RNA identifier and subsequent base trimming from the 5’end of reads (22nt from read1 and 5nt from read2). The VCF file reporting SNP sites of CAST/EiJ (Cast) based on mm10 was downloaded from Sanger Institute and used to reconstruct the Cast genome from the mm10 genome assembly (ref). Pre-processed reads were aligned to two parental mouse genomes (ref and Cast) separately using STAR (v2.7.10a) in the 2-pass mode, with the allowance of 1 mismatch in every 20 bases and a maximum of 6 mismatches per read pair⁸⁷. Non-canonical introns were not considered. Reads with overlapping mates were only considered with at least 5 nucleotide overlap. All discordant read pairs where two mates were aligned to different chromosomes were discarded. The coordinates of rRNA genes in mm10 mouse genome were downloaded from RepeatMasker track (v4.0.7) in UCSC genome browser and all the reads mapping to these regions were masked and ignored in the final output files (including bam, bed and bw files). To quantify overall expression of each given gene, all unique aligned read pairs overlapping with gene exons were counted using featureCounts (v1.5.0-p1)⁸⁸. Expression levels of Xist and total Y-linked genes were used to determine the sex of embryos from late-2cell to early blastocyst stage. For early-2cell embryos, the total expression level of all X-linked genes together were used for the sex determination. In the differential expression analysis between our data and published datasets, overall gene read counts were used and analysis was performed using edgeR⁸⁹ with default parameters. Differentially expressed genes were defined if a p-value below 0.05 after Benjamini-Hochberg adjustment (FDR). For allelic gene analyses, each unique aligned read that covered strain specific SNPs or indels was analyzed and the alignment quality score of each given read between two genomes was compared to determine its allelic origin. Pairs that differed significantly in alignment score due to mismatches/gaps were classified as allele-specific and the better alignment retained. Pairs with identical alignment scores or scores that differed only slightly due to fragment length penalties were classified as neutral. Each experiment yielded three tracks: paternal, maternal, and composite (neutral, paternal and maternal combined). For better visualization of gene expression between alleles, read mapping coordinates in different parental genomes were also converted to matched coordinates in mm10 mouse genome for view in IGV⁹⁰. Similar to overall expression quantification, only the allelic read pairs overlapping with gene exons were considered for allelic read counts of this gene. Allelic expression level of a given gene was thus calculated by splitting its overall expression in accordance with the allelic expression ratio, which was defined by the ratio of total allelic read count from one allele relative to that of the other allele. For allelic reads in repeats, a de novo assembled CAST/EiJ mouse genome (GenBank assembly: GCA_921999005.2, Wellcome Sanger Institute) and mm10 mouse genome was used for the So-smart-seq read alignment with the same parameters as used in gene alignment. The repetitive portion of genome (LINE, SINEs, LTR and DNA repeats) was determined by RepeatMasker⁹¹. To determine repeat expression, reads that did not overlap with any portion of the repeat elements in either of the two gnomes were filtered out. Only the reads that were uniquely aligned in both two genomes and that overlapped with repeats were considered for downstream analyses. Reads that were uniquely aligned in one genome but had multiple or no hits in the other gnome were also discarded. The alignment qualities of each single read between to parental genomes were compared to determine its allelic origin. The significance of mono allelic or biallelic expression of a given repeat element was determined using the Binomial distribution with Bernoulli p=0.5.

Alkylation Dependent RNAseq analysis

Adapters were removed from raw data using Trim Galore with the following parameters: (--stringency 5 -e 0.2 --length 35 --r1 36 --r2 36). After PCR duplicates removal, pre-processed reads were aligned to two parental mouse genomes separately using STAR (v2.7.10a) in the 2-pass mode, with the same parameters as they were used in the So-smart-seq alignment, except that the default maximum number of mismatches per read pair was used. Allelic reads separation was also carried out based on the approach described in So-smart-seq analysis. For mutation rate analysis, the comparison of each single base between each type of allelic reads (cast, mus or neutral reads) and genome template in the alignment was collected using sam2tsv⁹², from which all types of mismatches along with their associated Reads were retrieved. Only the mutations on the sense stand of the gene exons were considered. Mismatches overlapping with CAST/EiJ SNPs were ignored. The conversion rate of each mutation type was then calculated by dividing the total coverage of a given mutation type by the total coverage (including both matched and mismatched) of the corresponding nucleotide in the whole transcriptome. In WT crosses, genes with at least 6 allelic reads carrying T->C conversion in each sample were considered for gene category analysis, and for Xist-KO crosses, genes with at least 2 allelic reads carrying T->C conversion in each sample were used due to the slightly lower sample qualities.

mESC RNA-seq analysis

Similar to Alkylation Dependent RNAseq analysis, PCR duplicates were removed by Trim Galore and reads separately aligned to reconstructed mus/129 and Cast genomes using STAR with the same parameters, except that a maximum of 8 mismatches were allowed per read pair. Final allele-specific mapping to reference mm10 genome was generated based on SNPs. Uniquely aligned concordantly mapped reads overlapping with gene exons were counted using featureCounts.

H3K27me3 ChIP-seq analysis

Sequencing replicates from each stage were combined. The raw reads were trimmed using Trim Galore (--stringency 12 -e 0.2), and PCR duplicates were removed. Pre-processed reads were aligned to two mouse parental genomes using Novoalign (-i300 100 -t 180 -F STDFQ -rRandom -h180 180 -v180). For allelic read analysis, each unique aligned read that covered strain specific SNPs or indels was analyzed and the alignment quality scores of each given read between two parental genomes were compared to determine its allelic origin. Pairs that differed significantly in alignment scores due to mismatches/gaps were classified as allele-specific and the better alignment was retained. Pairs with identical alignment scores or scores that differed only slightly due to fragment length penalties were classified as neutral. Read mapping coordinates in different mouse genomes were also converted to the coordinates of mm10 reference genome, similar to RNAseq analysis above. Metagene analyses were performed using the software suite DeepTools 3.1.2⁹³ with a bin size of 50 or 100bp.

BS-seq Data analysis

The initial alignments of Bisulfite DNAseq data were carried out as previously described⁸⁶. Briefly, additional adapters at the 3’ end and first 6 nucleotide bases from the 5’ end were trimmed from the raw reads, followed by the PCR duplicates removal. The processed reads were then aligned to both mm10 and pesudo-CAST genome using Bismark⁹⁴ with the following parameters (--non_directional). Due to the nature of bisulfite sequencing, Cast SNPs including C to T, T to C, G to A and A to G were excluded from downstream allelic analyses to avoid false positive allelic decisions.

scRNA-seq analysis on spermatogenic cells

UMI counts of all endogenous genes in all high quality spermatogenic cells were collected and used as raw gene counts. The normalization was then performed by deconvolution using R package SingleCellExperiment⁹⁵ (1.20.1). Briefly, the raw counts were first adjusted by function calculateSumFactors in R package scran-1.26.2⁹⁶ (with the parameter min.mean to 0.1), and then normalized with log transformation by function logNormCounts. The heatmap was generated in R by pheatmap (v1.0.12). Differential gene expression analysis between two cell types (MII and RS8, or RS2 and RS8) was performed using function findMarkers in R package scran. A Wilcoxon ranked sum test was used to calculate the P values. Differentially expressed genes were defined if a p-value below 0.05 after Benjamini-Hochberg adjustment (FDR), in which genes with AUC (area under the curve) <0.5 were considered as downregulated genes, and AUC >0.5 represented upregulated genes.

Allelic skewing analyses

Paternal expression fraction was used to represent the allelic skewing state for both genes and repeat elements on chrX or autosomes. To increase the accuracy and credibility, gene having less than 7 total allelic reads were ignored in each replicate, and only genes that were retained in least two replicates at each stage were considered for skewing analysis. At each stage, the paternal expression fraction of a given gene was calculated by averaging the paternal allelic fractions of this gene from all qualified replicates. In K-mean clustering, to cluster as many X-linked genes as possible, genes that were retained in least one replicate at each stage were considered for skewing analysis or calculating paternal expression fraction. Package NbClust⁹⁷ was used to pre-determine the number of categories in clustering. To perform allelic skewing analysis for repeats on each chromosome, allelic reads from the same repeat genes (regardless of copy numbers) on one chromosome were summarized as the total allelic reads for this repeat gene on this chromosome. Paternal expression fraction of a given repeat on a given chromosome was calculated as the ratio of total paternal allelic reads in relative to total allelic reads of this given repeat on this chromosome, similar to gene allelic analysis. Repeats with less than 5 total allelic reads on one chromosome were filtered out from this chromosome, and repeats that were detected on less than five autosomes and not detected on X chromosome in all replicates combined were also discarded. Repeat allelic ratio was then calculated by averaging the allelic ratio of each given repeat from all qualified replicates. For skewing density analysis of old and young LINEs, allelic reads in all replicates at each stage were combined to increase the allelic read number. All the boxplots and density plots were generated in R (4.0.2) using the following packages: gridExtra (v2.3), gplots (v3.0.4) and ggplot2⁹⁸ (v3.3.2)., and all heatmaps were generated in R using pheatmap (v1.0.12).

X:A ratio

X:A ratio was performed as previously described⁹. The RPKM values were used to represent the expression level of each gene, and the X:A ratio in females (XX:AA) and males (X:AA) were then defined as the mean expression level of X-linked genes divided by the mean expression of autosomal genes. Genes with RPKM < 1 were excluded from the analysis. To ensure the number of expressed gene used for X:A ratio were the same between X and autosomal genes in each sample, the number of expressed X-linked genes was first determined, and the same number of autosomal genes were randomly selected from the total autosomal gene pools. In each selection, X:A ratio was calculated by dividing the mean expression value of X-linked genes by that of selected autosomal genes. The random selection was repeated 1000 times and the final X:A ratio for the embryo was estimated as the median of the 1000 values.

Spermatocyte (germline) nuclear surface spreading and immunostaining

The surface spreads of spermatocytes (germline) were performed as described previously⁹⁹. For immunofluorescence, slides were blocked with 3% non-fat milk in phosphate-buffered saline (PBS) for 30 min and then incubated with primary antibody against SYCP3 (Abcam, ab97672) overnight at 37°C in a humidified chamber. After three washes in PBS containing 0.1% Triton X-100 (PBST), secondary antibody (Alexa Fluor 488 Goat anti-Mouse IgG, Invitrogen) was applied for 1 h at 37°C. Subsequently, the slides were washed three times in PBST and washed once in PBS. Slides were then dehydrated by sequential incubation with 70%, 80%, 90%, and 100% ethanol for 3 min each and air-dried.

Short oligo Probe preparation and nascent RNA FISH

For FISH in male germ cells, Double strand DNA covering a 10kb region in the nascent RNAs of a target gene was first synthesized by PCR, from which the probes were prepared and labeled with Cy3 fluorophore by nick translation using Nick Translation Kit (Roche). RNA-FISH was performed using 200ng nick translated probes in 50 μL of hybridization buffer containing 2X SSC, 50% formamide, 10% dextran sulfate and 100ng/μL mouse Cot-1 DNA (Thermo Fisher Scientific) at 37°C overnight. Prior to hybridization, probes were denatured in hybridization buffer at 95°C for 10 min and pre-annealed at 37°C for 20 min. After being washed once in 25% formamide/2x SSC at 37°C for 20 min and three times in 2x SSC at 37°C for 5 min each, cells were mounted for wide-field fluorescent imaging. Nuclei were counter-stained with Hoechst 33342 (Life Technologies).

For FISH in 2C stage embryos, 50–70 short (40–42nt) DNA oligo probes (Table S6) primarily targeting nascent RNA introns of a given gene were synthesized from IDT, followed by addition of Cy3 or Alexa488 fluorophore labeled dUTP (Enzo life Sciences and Jena BioScience) at the 3’end of probes by terminal transferase (NEB). The FISH was performed as described previously with a few modifications¹⁰⁰. Briefly, F1 embryos were harvested at late 2C stage from the cross between male CAST/EiJ and female C57BL/6J, followed by the removal of zona pellucida in acid Tyrode’s solution and two washes in PBS-BSA (1mg/ml). Embryos were then transferred onto defrosted glass slides with minimal carry-over liquid. Once completely dried out, embryos were then fixed and permeabilized by first incubating in 50ml of fresh 1% PFA in PBS with 0.05% Tergitol (Sigma) for 5 mins on ice, followed by transferring into 50ml of 1% PFA in PBS for another 5 min on ice. After incubation, slides were transferred to 70% ethanol on ice. Prior to hybridization, embryos were dehydrated by sequential incubation with 70%, 80% and 100% ethanol for 2 min each at RT and air-dried. RNA-FISH was then performed using 70–120ng short oligo probes in 20 μL of hybridization buffer containing 2X SSC, 50% formamide, 10% dextran sulfate, 0.1% BSA, 20mM Ribonucleoside-vanadyl complex and 200ng/μL mouse Cot-1 DNA (Thermo Fisher Scientific) at 42°C overnight. After two washes in 50% formamide/2x SSC followed by another two washes in 2x SSC at 42°C for 15 min each, embryos were mounted for fluorescent imaging. Nuclei were stained with DAPI.

Quantification and Statistical Analysis

Statistical analyses were performed by R. Statistical details and results of experiments can be found in the figure legends and figures. p < 0.05 in most statistical tests was considered significance. In the analyses of scRNAseq data from spermatogenic cells, differentially expressed genes were defined if a p-value below 0.05 after Benjamini-Hochberg adjustment (FDR).

Supplementary Material

3

Table S1. Sample replicates at each stage in RNA So-smart-seq and 4sU-seq. Related to Figure 1.

4

Table S2. Mapping details of RNA-seq libraries. Percentage (%) were calculated relative to the original input read number. Related to Figure 1.

5

Table S3. PolyA+ and polyA- gene number included in X:A ratio calculation for MC crossed embryos. Related to Figure 1.

6

Table S4. PolyA- genes included in the X:A ratio calculation. Related to Figure 1.

7

Table S5. Sequences of RNA probes targeting ribosomal cDNA in So-smart-seq. Related to the STAR Methods.

8

Table S6. Short DNA oligo probes used for nascent RNA FISH in embryos. Related to the STAR Methods.

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Primary Ab-SYCP3	Abcam	ab97672
Alexa Fluor 488 Goat anti-Mouse IgG	Thermo Fisher Scientific	A-21121
Bacterial and virus strains
Biological samples
WT C57BL/6J (P) × CAST/EiJ (M) F1 mouse embryo	This paper	N/A
WT CAST/EiJ (P) × C57BL/6J (M) F1 mouse embryo	This paper	N/A
XistXm/- female mouse embryo	This paper	N/A
Chemicals, peptides, and recombinant proteins
Acid Tyrode’s solution	Sigma-Aldrich	Cat#T1788-100ML
acetylated BSA	Sigma-Aldrich	Cat#B8894-5ML
RNase inhibitor	Roche	Cat#3335399001
DTT	Thermo Fisher Scientific	Cat#P2325
ATP	New England BioLabs	Cat#P0756S
Betaine	SIGMA-ALDRICH	Cat#B0300-5VL
Biotin-14-CTP	Thermo Fisher Scientific	Cat#19519-016
RLT plus buffer	Qiagen	Cat#1053393
Unmethylated lambda DNA	Promega	Cat#D1521
4-Thiouridine	Sigma-Aldrich	Cat#T4509-25MG
Iodoacetamide (IAA)	Sigma-Aldrich	Cat#I1149-5G
Cy3-dUTP	Enzo life Sciences	Cat#enz-42501
Aminoallyl-dUTP-XX-AF647	Jena BioScience	Cat#NU-803-XX-AF647-S
Tergitol	Sigma-Aldrich	Cat#NP40S-100ML
Ribonucleoside-vanadyl complex	New England BioLabs	Cat#S1402S
Critical commercial assays
Agencourt AMPure XP Beads	Beckman Coulter	Cat# A63881
Streptavidin C1 beads	Thermo Fisher Scientific	Cat#65001
Nextera XT DNA library preparation kit	Illumina	Cat#FC-131-1024
Nextera XT Index Kit	Illumina	Cat#FC-131-1001
Kapa high fidelity PCR mix	Kapa Biosystems	Cat# KK2601
EZ methylation kit	Zyon	Cat# D5020
PureLink PCR Purifiation kit	Thermo Fisher Scientific	Cat#K310001
Qubit dsDNA Quantification Assay Kit	Thermo Fisher Scientific	Cat#Q32851
Deposited data
Single embryo So-Smart-seq, 4SU-RNAseq, Bisulfite- DNAseq and bulk RNA-seq data	This paper	GSE168455
Early embryo Hi-C data	Du et al.46	GSE82185
Early embryo H3K27me3 ChIP-seq data	Zheng et al.60	GSE76687
Early embryo ATAC-seq data	Wu et al.65	GSE66390
Male germline single-cell RNAseq	Chen et al.66	GSE107644
Sperm ATAC-seq	Jung et al.67	GSE79230
Experimental models: Cell lines
Mouse: ES cell line: TsixTST/+	Ogawa et al.84	N/A
Experimental models: Organisms/strains
Mouse: C57BL/6J: WT	Jackson Laboratory	Cat#000664
Mouse: CAST/EiJ: WT	Jackson Laboratory	Cat#000928
Mouse: 129S1/SvlmJ: Xist−/Y	Marahrens et al.32	N/A
Oligonucleotides
Reverse-dT Primers for So-smart-seq: TTTCCCTACACGACGCTCTTCCGATCTNNNNNNTTTTTTTTTTTTTTTVVN	This paper	N/A
GGrG oligo iCiGiCGGAGTTCAGACGTGTGCTCTTCCGATCTrGrG+G	This paper	N/A
Library PCR forward primer: GTGACTGGAGTTCAGACGTGTGCTCTTC	This paper	N/A
Library PCR reverse primer: ACACTCTTTCCCTACACGACGCTCTTC	This paper	N/A
oligo-dT for Alkylation-dependent RNAseq: /5Biotin-TEG/AAGCAGTGGTATCAACGCAGAGTACTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTVN	Hendriks et al.44	N/A
ISPCR Reverse primer: AAGCAGTGGTATCAACGCAGAGT	Hendriks et al.44	N/A
NEBNext Multiplex Oligos for Illumina	New England BioLabs	E7335S
Oligos for ribosomal cDNA depletion (see Table S5)	This paper	N/A
Short oligo probes for FISH experiments (see Table S6)	This paper	N/A
Nextera XT Index Kit	Illumina	Cat#FC-131-1001
Recombinant DNA
Software and algorithms
edgeR	Robinson et al.89	N/A
Trim Galore	Babraham Bioinformatics Group	N/A
STAR	Dobin et al.87	N/A
RepeatMasker	Smit et al.91	N/A
featureCounts	Liao et al.88	N/A
sam2tsv	Lindenbaum92	N/A
Novoalign	Novocraft	N/A
DeepTools	Ramirez et al.93	N/A
Bismark	Krueger et al.94	N/A
Rstudio	Posit	N/A
SingleCellExperiment	Amezquita et al.95	N/A
scran	Lun et al.96	N/A
pheatmap	https://cran.r-project.org/web/packages/pheatmap/pheatmap.pdf	N/A
NbClust	Charrad et al.97	N/A
gridExtra	https://cran.r-project.org/web/packages/gridxtra/index.html	N/A
ggplot2	Wickham98	N/A
IGV	Thorvaldsdottir et al.90	N/A
R scripts for the plots generated in this paper	This paper	doi:10.7910/DVN/W7RJFI
Other

Highlights.

• Older X-linked genes and LINEs show constitutive paternal silencing in mouse embryos.

• Constitutively silent genes trace repression to the paternal germ line.

• They are epigenetically distinguished by H3K27me3 and low chromatin accessibility.

• Thus, paternal pre-inactivation partially drives imprinted X-inactivation.

DATA AVAILABILITY

The single embryo So-Smart-seq, 4SU-RNAseq, BS-seq and bulk RNAseq data can be accessed in GEO (GSE168455).