To generate a mouse model carrying a conditional Rnf12 allele we flanked the coding region of exon 5 with LoxP sites (Fig. S2). Exon 5 encodes 517 out of RLIM’s total of 600 amino acids ⁹ including the RING finger. We targeted the Rnf12 KO (Δ) to oocytes using transgenic mice that express Cre recombinase (Cre) under the control of the mouse mammary tumor virus long terminal repeat (MMTV-LTR). Several MMTV-Cre mouse lines exist some of which target the female germline ¹¹. Cre expression in oocytes was verified by crossing MMTV-Cre (line F) mice to Rosa26-loxP-stop-loxP-lacZ (R26R) animals ¹² (Fig. 1C). Ovaries from Rnf12^fl/fl × MMTV-Cre (line F) females (fl/fl-Cre) showed lack of RLIM in oocytes but not in the surrounding granulosa cells or in stromal cells (Fig. 1D), confirming a Rnf12 KO in oocytes. The correct targeting was corroborated by the fact that we obtained KO males with germline deletion of the Rnf12 gene (Δ/Y), lacking RLIM in all somatic tissues examined (Fig. 1E; not shown). Because Δ/Y males are viable and fertile, these results demonstrate that Rnf12/RLIM is not required for basic cellular or developmental functions, for the maturation of oocytes or for meiotic sex chromosome inactivation.

Pups born by fl/fl females displayed normal gender ratios and a normal transmission of the paternal X chromosomes (p) was observed in matings using wt, fl or Δp males (Fig. 2A, mating schemes 1–3). However, no female offspring carrying a maternally transmitted KO (Δm) allele was born by fl/fl-Cre or wt/fl-Cre females crossed with wt, fl or Δp males (Fig. 2A, schemes 4, 5; not shown) whereas the Δm allele transmitted efficiently to male pups. The probability of obtaining male versus female pups from fl/fl-Cre or wt/fl-Cre females was highly significant (P<2 × 10⁻⁸). In contrast, wt/fl-Cre females transmitted the wt allele normally and the probability of producing male wt/Y versus male fl/Y + Δm/Y offsprings was similar (P>0.13) (scheme 5). We also performed matings with the MMTV-Cre line D that does not target to the female germline ¹¹ and observed normal Mendelian distributions for male and females pups (n=177; not shown) sired by fl/fl-Cre (line D) females. These findings combined with a decreased mean litter size for matings 4 and 5 indicated that the deletion of Rnf12 in the maternal germ line leads to embryonic lethality. As mice were bred in a congenic C57BL/6 background to eliminate strain-specific influences, our results reveal a sex-specific parent-of-origin effect.

Next, we examined embryos that received either a Δm or Δp allele. While male Δm/Y embryos appeared normal, the maternal transmission of the KO allele to female conceptuses in the same litters resulted in severe growth defects that were apparent as early as E7.5 (Fig. 2B–E). However, generating this heterozygous genotype with paternal transmission of the KO allele resulted in female fl/Δp embryos that appeared normal (Fig. 2F). Embryonic components of Δm conceptuses differed in size and showed various degrees of disorganization (Figs 2G; H). The embryo shown in Fig. 2G was the largest observed female Δm embryo. No obvious differences in severity of growth defects between heterozygous Δm/fl and Δ/Δ female embryos were detected at E7.5, E8.5 or E9.5 (Fig. 2; Figs. S3A–C). We were unable to recover Δm female conceptuses at stages later than E11.5 presumably due to reabsorption. Quantifications of phenotypes showed that all recovered Δm female embryos displayed growth defects (Fig. S3D). In 25% of deciduas we were unable to recover a conceptus. As this corresponds to the number of missing female embryos expected for a Mendelian ratio it is highly likely that these correspond to Δm females.

Examining Δm blastocyst outgrowths at pre-implantation stages, trophoblast migration, cell number and expression of the early trophoblast marker Troma-1 ¹³ appeared comparable to controls (Fig. S4A, B). For analyses at early post-implantation stages we sectioned through entire E5.5 and E6.5 deciduas sired by a fl/Δ-Cre × Δ/Y cross. From this mating around 50% of the embryos should correspond to Δm females. Indeed, Hematoxylin/Eosin (HE) stainings revealed around 50% of mildly or severely disorganized embryos (Fig. S4C, D). These results suggest that growth defects in Δm embryos first occur around implantation or very shortly thereafter. Next, we examined the effects of the Rnf12 KO on the development of extraembryonic tissues and analyzed placentae of Δm embryos in the cases when a conceptus was found. HE stainings revealed that most if not all tissues derived from the extraembryonic trophoblast were missing in Δm/fl placentae at E10.5 (n=4) and E9.5 (n=5), whereas the maternal deciduas appeared normal (Figs. S5A, B). This was accompanied by a lack of trophoblast markers PAI-1 and Cdx2 ^14,15 as early as E8.5 during placental development (n=4) (Fig. S5C; not shown). Because these phenotypes are reminiscent of those described for female Xist KO embryos caused by a paternally inherited Xist-KO allele ¹⁶, we hypothesized that Rnf12 may regulate XCI.

We first examined random XCI in wt/Δp adult females and found that virtually all somatic cells in ovaries of wt/Δp adult females (n=3) stained RLIM-positive (Fig. S6A), and general RLIM levels in somatic tissues were similar in wt/Δp and wt adult females (Fig. S6B). Because the ratio of fl/Δp female to male Δ/Y pups (Fig. 2A, scheme 3) was normal these results suggest that in wt/Δp adult females random XCI is skewed towards the mutated allele. Staining mouse embryonic fibroblasts (MEFs) of fl/Δp embryos at E12.5 with RLIM and H3K27me3 ¹⁷ antibodies revealed that 94% (n=300 of 3 embryos) of MEFs stained positive for RLIM and H3K27me3 (Fig. 3A). In addition, in RNA fluorescent in-situ hybridization (RNA FISH) experiments using a double strand Xist probe that recognizes Xist and Tsix showed that 97% (n=300 of 3 embryos) of the stained MEFs showed specific Xist paints on one X chromosome.

To test the status of the Rnf12 gene on the Xi, we performed RNA FISH with probes against Xist and Rnf12. For Rnf12 we used a 12 kb genomic probe located 5′ of the deletion site that recognizes wt and Δ transcripts of the Rnf12 gene equally well (Fig. S7B). Results revealed that 92% of MEFs that expressed Xist also expressed Rnf12 in a monoallelic fashion similar to fl/fl control MEFs (Fig. 3B). Next, we generated ES cells lacking Rnf12 and isolated the heterozygous fl/Δp line 7 and Δ/Δ ES lines 4, 6, 11 and 23 (Fig. S7A; not shown). Consistent with skewed random XCI we found that 98% of H3K27me3-positive EB-differentiated fl/Δp #7 ES cells expressed RLIM protein, whereas Δ/Δ ES cells did not (Fig. S7C, not shown). All Δ/Δ ES lines stained positive for H3K27me3 and developed Xist clouds upon differentiation (Fig. 3C, D; data not shown). A time course comparing wt female ES cells with lines fl/Δp #7, Δ/Δ #4 and Δ/Δ #23 revealed a slower initiation of XCI at days 2 and 4 during embryoid body (EB)-differentiation (Fig. 3C). However, at days 6 and 8 these differences were no longer significant. The rates of Xist cloud formation of Δ/Δ and fl/Δp ES cells were similar (P>0.05). To assess X silencing in mutant ES cells we co-hybridized cells with Xist and Rnf12 probes in RNA FISH at 2, 4, 6 and 8 days of EB-differentiation. Focusing on cells with Xist clouds (=100%) we compared distribution of Rnf12 signals in nuclei displaying either monoallelism, biallelism or no signal. Silencing of the Rnf12 gene was slower in all Rnf12 mutant ES cells at 2, 4 and 6 days of differentiation when compared to wt (P<0.05). Again, mutated ES cells did not differ significantly among themselves (Fig. 3D). Silencing of Pgk1, another X-linked gene, was also observed in Rnf12 mutant ES cells after 6 days of EB-differentiation (Fig. S7D). These results combined with our finding that in teratoma assays, injecting ES cells in kidney capsules of immunodeprived NOD-SKID mice, Δ/Δ cells participate in the formation of ectodermal, endodermal and mesodermal germ layers and derived cell types (Fig. S7E), indicate that ES cell lacking RLIM/Rnf12 initiate XCI. Our data also suggest that random XCI is skewed toward the mutated Rnf12 allele in fl/Δp females.

Because these results did not explain the observed embryonic lethalities (Fig. 2), we next investigated imprinted XCI in E3.5 and E4.5 blastocyst outgrowths via RNA FISH. When required, embryo gender was determined via isolation of the inner cell mass (ICM) after image recording and genotyping for the presence of the Zfy gene and, as control, β-actin using PCR. A high percentage of central cells in the ICM of control and Δm E4.5 blastocyst outgrowths developed Xist clouds or single pinpoints, indicating transcription foci (Fig. 4A; Movie S1 and S2; S8A). This suggests that in contrast to ES cells in culture, random XCI occurs with similar kinetics in blastocysts. However, although quantification of ICM stainings was not possible due to high cell density, the decreased number of E4.5 ICM cells staining positive for H3K27me3 ¹⁷ in Δm when compared to wt blastocysts (Fig. S8B) suggested that imprinted XCI is inhibited in primitive endoderm cells. Importantly, focusing on trophoblasts that also undergo imprinted XCI, only around 10% of Δ/Δ and Δm cells displayed Xist clouds compared to more than 90% of fl/fl and fl/Δp cells (S9A–D). Higher magnification revealed the presence of pinpoints in around 20% of Δm trophoblasts at E4.5 and 30% at E3.5. Generally 1 pinpoint per trophoblast was detected although some cells developed 2 pinpoints likely due to the development of polyploidism ¹⁸. The parental origin of these Xist signals was not examined and because we used a double-stranded Xist probe we cannot distinguish between Xist and Tsix pinpoints. Co-stainings with antibodies directed against RLIM and H3K27me3 revealed that only around 10% of Δ/Δ trophoblasts displayed H3K27me3 signals compared to 94% in fl/fl trophoblasts (Fig. S10A–C). To quantitatively compare Xist expression in blastocysts and to monitor expression of Tsix²,^19,20, we performed RT-qPCR on E3.5 and E4.5 blastocysts comparing Δ/Δ with fl/fl embryos (Fig. S11A). In agreement with results obtained in RNA FISH, Xist expression was reduced in Δ/Δ females. Tsix levels were only mildly affected in E3.5 embryos and decreased at E4.5, suggesting that RLIM does not induce Xist by repression of Tsix RNA transcription. Increased Pgk1 levels suggested defects at the level of Xp silencing.

To investigate Xp silencing we co-hybridized E3.5 blastocyst outgrowths with the Xist probe and with probes recognizing X-linked genes Pgk1 or Rnf12. While both the Pgk1 or Rnf12 genes were efficiently silenced on the Xp in control trophoblasts, we observed at least two spots of Pgk1 or Rnf12 in a high percentage of Δm trophoblasts with at least one signal in proximity of a Xist pinpoint (Fig. 4B; S11B,C). Furthermore, the formation of nuclear compartments around the Xp that exclude Cot-1 RNA or general transcription factors such as TATA-Box binding protein (TBP) ^21,22 was inhibited in Δm trophoblasts (Fig. S12; not shown). Defects in Rnf12 silencing were observed as early as the 8 cell stage (Fig. 4C). Because this is the earliest stage when X-linked genes are silenced ^3,4, these results indicate that defects do not occur at the maintenance level. This is also in agreement with the finding that Δm cells did not form initial Xist clouds in 8 or 4 cell-staged embryos (Figs. 4C–D). However, because we detected one Xist transcription foci in a significant number of cells and Xist accumulation is regulated at the transcriptional level ²³, our data are consistent with a crucial role for RLIM/Rnf12 for the transcriptional upregulation of Xist. The fact that a small but significant number of Δm trophoblasts developed Xist clouds (Fig. S9) indicates that RLIM is not absolutely required for Xist upregulation but rather for it to occur reliably at the appropriate time. Indeed, RLIM is able to modulate the transcriptional activity of various classes of transcription factors ^5,8.

While our manuscript was in preparation a paper was published which showed that overexpression of RLIM/Rnf12 initiates random XCI in ES cells ²⁴. While the reported data are in general agreement with our results, our finding that Rnf12−/− ES cells initiate random XCI was surprising, suggesting the existence of several competence factors that may compensate for the lack of RLIM/Rnf12. Our in vivo data show that RLIM/Rnf12 is required for imprinted XCI suggesting that it may be the only competence factor present in high concentrations during imprinted XCI. Indeed, the observed female parent-of origin effect in XCI in Δm females (Fig. 2) suggests that the maternal RLIM/Rnf12 protein/mRNA drives imprinted XCI and several observations support this view: 1. RLIM protein/mRNA accumulates to high levels in oocytes (Fig. 1); 2. RLIM is undetectable in two cell stage embryos carrying a Δm allele (Fig. S13); 3. Overexpression of RLIM protein can initiate random XCI ²⁴; 4. The paternal contribution of RLIM/Rnf12 appears irrelevant for imprinted XCI (Fig. 2); and 5. It explains why imprinted XCI occurs in fl/Δp females (Figs. 2; S9). Indeed, the maternal deposit of mRNA/protein to control early embryogenesis represents a mechanism that is widespread among many species ²⁵ including mice ²⁶. Thus, we propose that maternal Rnf12/RLIM acts as a crucial regulator for the initiation of imprinted XCI in mice (Fig. S14).

1

2

3