Non-canonical imprinting in the spotlight Free

Unlike most vertebrates, mammalian uniparental embryos derived from two maternal (parthenotes; see Glossary, Box 1) or two paternal (androgenotes; see Glossary, Box 1) genomes fail to develop beyond the mid-gestation stage, even in carefully controlled laboratory settings (Surani et al., 1984; McGrath and Solter, 1984; Barton et al., 1985). Close examination of the failed uniparental embryos revealed gross abnormalities in placenta size: the placenta of parthenotes is drastically smaller and, conversely, the placenta of androgenotes is enlarged. Why does mammalian development, in particular placenta size homeostasis, require exactly one maternal and one paternal genome?

Subsequent studies suggested that the failure of uniparental zygotes to develop was due to the requirement of maintenance of parent-specific DNA methylation (see Glossary, Box 1) at so-called ‘imprinting control regions’ (ICRs; see Glossary, Box 1), of which there are roughly 21 in mouse and human (Tucci et al., 2019). ICRs are the basis of canonical imprinting (see Glossary, Box 1) and control the expression of ∼150 genes in cis in the mouse genome. Misexpression of canonical imprinted genes in embryonic (precursors to the adult soma) and extra-embryonic cells (precursors to the placenta) has been associated with uniparental embryo lethality (Walsh et al., 1994; Weinberg-Shukron et al., 2022). Gene dose misregulation is frequently associated with pathologies, including cancers, fragile-X syndrome, Huntington's disease and imprinting disorders, such as Angelman and Prader-Willi syndromes (Tucci et al., 2019). Therefore, deregulation of imprinted genes in uniparental and cloned embryos due to a lack of imprinting may drive the disruption of placenta homeostasis and engender developmental arrest. For more on the regulation of gene expression in the placenta, we refer the reader to recent reviews (Hanna, 2020; Pastor and Kwon, 2022).

Recent work has identified a non-canonical form of imprinting (see Glossary, Box 1) , mediated by histone methylation. In this Spotlight, we discuss how these non-canonical imprints were discovered, their importance in development and their conservation across mammalian species. Although non-canonical imprinting is found in the early morula stage embryo, this Spotlight focuses on extra-embryonic imprinting in the mouse, as it is the most well-characterized barrier to the development of uniparental and cloned embryos.

Oocyte and sperm chromatin differ drastically in DNA methylation levels, and there are thousands of so-called differentially methylated regions between gametes. However, parental DNA methylation level differences are largely harmonized by the blastocyst stage (Fig. 1A) (Wang et al., 2014; Guo et al., 2014; Peat et al., 2014; Xia and Xie, 2020). Exceptionally, ICRs maintain asymmetric parental DNA methylation levels throughout embryogenesis and into adulthood (Fig. 1B). Mechanistically, parental methylation levels are maintained through the action of sequence- and DNA methylation-specific DNA-binding proteins that recruit chromatin modifiers to the methylated allele (Li et al., 2008; Strogantsev et al., 2015; Takahashi et al., 2019). Canonical imprints exert their control over nearby genes through a variety of cis- acting mechanisms, including repression via DNA methylation and histone 3 lysine 9 trimethylation (H3K9me3; see Glossary, Box 1), through long non-coding RNA transcription, 3D genome organization and recruitment of other chromatin regulating complexes (reviewed by Tucci et al., 2019).

Gametic chromatin asymmetry extends beyond DNA methylation. The sperm genome is packaged with protamines (see Glossary, Box 1) that are replaced by histones soon after fertilization (Nonchev and Tsanev, 1990). Meanwhile, the oocyte exhibits a specific histone modification landscape, with broad domains of the repressive Polycomb Repressive Complex 2 (PRC2)-deposited histone 3 lysine 27 trimethylation (H3K27me3; see Glossary, Box 1) (Zheng et al., 2016). Interestingly, a subset of these parental chromatin differences persists throughout preimplantation development (Mei et al., 2021; Chen et al., 2021; Lismer et al., 2020) (Fig. 1C). Thus, like DNA methylation, histone modifications deposited in gametes and maintained during development may regulate gene expression in a parent-of-origin manner. In recent years, it has become clear that H3K27me3 confers a non-canonical form of imprinted gene expression in developing mice (Inoue et al., 2017, 2018; Chen et al., 2019; Santini et al., 2021) (Fig. 1D).

Beyond canonical imprints, an additional layer of imprinting was hinted at, with the discovery that the paternal-specific expression of Gab1, Sfmbt2 and Slc38a4 in the placenta is independent of oocyte DNA methylation (Okae et al., 2012, 2014). Notably, Gab1, Sfmbt2 and Slc38a4 have important roles in placental development, e.g. in nutrient transport, and are highly expressed in the placenta (Itoh et al., 2000; Miri et al., 2013; Bogutz et al., 2019; Xie et al., 2022). This suggests that previous placental phenotypes in uniparental embryos may not be entirely due to canonical imprinting.

The identification of the molecular basis of non-canonical imprinting, as well as the discovery of further candidates that are subject to this form of regulation, was enabled by technical advancements in low-input, high-throughput sequencing assays that allowed for genome-wide characterization of rare cell populations such as mouse oocytes and early embryos. Indeed, genome-wide profiling of zygotes uncovered an asymmetry in chromatin accessibility between the parental genomes, which is reflected in discordant accessibility patterns between parthenogenic and androgenetic morula stage embryos (Inoue et al., 2017). Genetic ablation of Eed, which encodes a core subunit of PRC2, and overexpression of Kdm6b, which encodes a H3K27me3 demethylase, demonstrated that this asymmetric parental chromatin accessibility depends on H3K27me3 levels deposited in oocytes (Inoue et al., 2017, 2018) (Fig. 2A,B). Importantly, lack of the H3K27me3 imprint in oocytes is associated with deregulation of a few dozen genes in the early embryo and post-implantation placental precursor cells, including Gab1, Jade1, Smoc1, Sfmbt2 and Slc38a4 (Inoue et al., 2018). Subsequent studies confirmed non-canonical imprinting by maternal H3K27me3, but also showed that, interestingly, H3K27me3 is generally lost on both alleles by the post-implantation stage in rodents (Zheng et al., 2016) and even earlier in other mammals (Lu et al., 2021). Indeed, maternal-specific H3K27me3 is not maintained at non-canonical imprints in the developing placenta, where imprinted expression occurs (Fig. 1D) (Chen et al., 2019; Hanna et al., 2019). As such, although H3K27me3 may act as the primary genomic imprint, it cannot be responsible for imprinted gene expression in the placenta, per se. Rather, maternal-specific DNA methylation is observed at non-canonical imprints in the placenta (Chen et al., 2019; Hanna et al., 2019). Thus, maternal H3K27me3 is seemingly supplanted by DNA methylation via an undefined recruitment mechanism.

Further studies implicated the histone 3 lysine 9 di-methylation (H3K9me2; see Glossary, Box 1) methyltransferases G9a and GLP in the establishment and maintenance of non-canonical imprinting (Zeng et al., 2021b). Removing G9a and/or GLP in the growing oocyte results in upregulation of non-canonical imprinted genes Gab1 (Demond et al., 2023) and Sfmbt2 (Meng et al., 2022). Why only two non-canonically imprinted genes are upregulated in oocytes lacking G9a is unknown, but it may be related to the similar observation that oocytes lacking DNA methylation also show few changes in canonical imprinted gene expression levels (Xu et al., 2019). Instead, the failure to establish putative H3K9me2 imprints in oocytes results in misexpression later in development. For example, G9a maternal knockout (KO) embryos show biallelic expression of Gab1, Smoc1 and Jade1 in placenta precursor cells (Zeng et al., 2021b; Andergassen et al., 2021). Similarly, G9a zygotic mutations also result in biallelic expression of Gab1, Smoc1, Sall1 and Sfmbt2 in the early placenta (Zeng et al., 2021b; Andergassen et al., 2021), and of Slc38a4 in the embryo (Auclair et al., 2016), hinting at a role for H3K9me2 in the establishment and maintenance of non-canonical imprinting. In other words, maternal knockouts of two distinct chromatin modifying pathways – PRC2 (H3K27me3) and G9a/GLP (H3K9me2) – both result in loss of non-canonical imprinting (Wagschal et al., 2008; Inoue et al., 2018; Zeng et al., 2021b; Matoba et al., 2022). Do both complexes exert the primary imprint in a non-redundant manner? A side-by-side comparison of zygotic Eed and G9a/Glp KOs demonstrated that G9a/GLP play a dominant role in non-canonical imprint maintenance (Andergassen et al., 2021). Given the established association between H3K9me3, DNA methylation and canonical imprinting (Fig. 1B) (Wang et al., 2018; Yang et al., 2022), it is plausible that H3K9me2 is required for the maintenance, and perhaps establishment, of non-canonical imprinting. To formally address the role of H3K27me3 and H3K9me2 in non-canonical imprinting establishment and maintenance, a higher order mutant for Eed and G9a/Glp would potentially reveal the epistatic or synergistic relationship. Additionally, genome-wide maps of H3K9me2 levels in the developing embryo and placenta are required to determine whether the histone modification is maintained beyond implantation and in the placenta.

Finally, non-canonically imprinted genes Jade1, Sfmbt2 and Smoc1 are transcriptionally upregulated in Smchd1 maternal KO placentas (Wanigasuriya et al., 2020). Smchd1 encodes a protein that mediates repressive chromatin over long ranges; initially discovered as a regulator of X-chromosome inactivation (Blewitt et al., 2008), subsequent studies have implicated SMCHD1 in repressing Hox genes (Benetti et al., 2022) and the silent allele of non-canonical imprinted genes, including Xist (see Glossary, Box 1) (Wanigasuriya et al., 2020). How SMCHD1 controls expression of non-canonical imprints without altering levels of DNA methylation (Wanigasuriya et al., 2020) or H3K27me3 (Chen et al., 2015) remains an active area of research.

Careful examination of non-canonical gene expression revealed that the residual solo long terminal repeats (LTRs) of the ERVK family of retrotransposons (see Glossary, Box 1) often serve as alternate promoters when genes are paternally expressed in the placenta (Hanna et al., 2019). Indeed, the genetic basis of non-canonical imprints was confirmed by excising the non-canonically imprinted ERVK alternate promoter of the Gab1 gene, which resulted in a moderate loss of imprinted expression of Gab1 in extra-embryonic cells (Hanna et al., 2019).

It was subsequently shown that non-canonically imprinted ERVK elements are regulated by G9a/H3K9me2 (Zeng et al., 2021b; Andergassen et al., 2021), which appears to be a general mode of ERVK control during mouse embryogenesis (Zeng et al., 2021a). Taken together, it is tempting to speculate that G9a/H3K9me2 may regulate ERVKs and non-canonical imprints, akin to of SETDB1/H3K9me3 repressing other families of LTR retrotransposons (Matsui et al., 2010) and canonical imprints (Wang et al., 2018; Yang et al., 2022). Although imprinted ERVK alternate promoters do not have any obvious shared motifs (Hanna et al., 2019), future studies will hopefully address the mechanism that targets G9a/H3K9me2 to these regions.

Notably, species-specific ERV insertions drive species-specific DNA methylation establishment over canonical imprints in the female germline (Brind'Amour et al., 2018; Bogutz et al., 2019). Perhaps functional non-canonical imprints can be identified through species-specific ERVK element insertions near genes that show species-specific imprinted expression. Indeed, a rat-specific ERVK insertion in the promoter of Slc38a1 coincides with rat-specific non-canonical imprinting (Richard Albert et al., 2023). Whether by establishing canonical imprints in oocytes or by promoting non-canonical paternal expression in extraembryonic tissues, both cases represent fascinating examples of transposable elements as vectors of potentially adaptive gene regulation.

While elucidating the mechanism of non-canonical imprinting has deepened our understanding of the extent of intergenerational epigenetic inheritance and of gene regulation in general, the findings raise a key question: the precise expression levels of which non-canonically imprinted genes are crucial for normal placenta homeostasis and, consequently, for normal mammalian in utero development?

Several recent studies have explored the developmental role of non-canonical imprints using complementary and overlapping genetic approaches. Using a candidate-based approach of previously defined non-canonical imprinted genes (Inoue et al., 2017, 2018), four of these studies tested whether restoring dose expression of deregulated non-canonically imprinted genes in somatic cell nuclear transfer (SCNT, see Glossary, Box 1) embryos – where this mode of imprinting is normally lost (Fig. 2C) – is sufficient to facilitate normal development (Wang et al., 2020; Xie et al., 2022; Matoba et al., 2018; Inoue et al., 2020).

Modulating expression of the master regulator of X-chromosome inactivation, the long noncoding RNA gene Xist, resulted in moderate increase in developmental success, yet placental defects remained in SCNT embryos (Matoba et al., 2018). Slc38a4 encodes an amino acid carrier expressed from the paternal genome in the placenta (Bogutz et al., 2019). Paternal KO of Slc38a4 results in a diminished placenta size and intrauterine growth restriction in biparental embryos (Matoba et al., 2019). Conversely, loss of imprinted expression of Slc38a4, as occurs in SCNT embryos, results in placenta overgrowth (Xie et al., 2022). Interestingly, restoring Slc38a4 expression levels in SCNT embryos partially rescues the placenta overgrowth phenotype but does not result in an increase in live births (Xie et al., 2022). The fact that a partially rescued placenta phenotype did not coincide with an increase in live births was unexpected given the clear link between placenta homeostasis and embryonic outcomes as discussed above. Finally, an independent group showed that correcting expression levels of either Sfmbt2, Jade1, Gab1 or Smoc1 in SCNT embryos partially rescues the placenta overgrowth phenotype and results in a notable increase in live birth rates (17/162 compared with 0/565 in control SCNT), with Sfmbt2 having the largest effect (Wang et al., 2020).

Using a complementary approach to SCNT, Matoba et al. used naturally fertilized Eed KO oocytes to generate embryos devoid of non-canonical imprints with much greater efficiency. Xist imprinting was again shown to be crucial for proper development, especially in male embryos with only one X chromosome (Matoba et al., 2022). However, even though correcting for Xist dose enabled more embryos to develop to term, developmental delays and placenta overgrowth were observed (as noted above), likely due to loss of imprinted expression of other non-canonically imprinted genes. Indeed, paternal deletion of Slc38a4 partially rescued the placenta overgrowth phenotype in maternal Eed+Xist KO embryos. Furthermore, careful genetic manipulation to correct expression levels of the imprinted intronic miRNA cluster of Sfmbt2 showed the greatest rescue of placenta overgrowth. Taken together, these six recent studies suggest that the correct dose of non-canonically imprinted genes Xist, Slc38a4, Sfmbt2 and Sfmbt2 miRNAs helps orchestrate placenta size homeostasis. However, it is possible that this is only a partial list of functional non-canonical imprints; more research is required to elucidate the number and function of this class of genes during development.

A striking aspect of canonical imprints is their general conservation in eutherian mammals (Tucci et al., 2019; Kobayashi, 2021), although it should be stressed that canonical imprinting has generally been investigated based on homology with known imprinted genes identified in the mouse: the chief mammalian model organism. Thus, a comprehensive analysis of imprinted gene expression – canonical and non-canonical – in embryonic and extra-embryonic tissues in a range of mammals is required to determine the relative conservation of imprinting.

What is clear is that therian (eutherians and marsupials) genomes exhibit canonical imprinting, whereas there is no evidence of imprinting in monotremes, which are egg-laying mammals (Renfree et al., 2009). Even less is known about the conservation of non-canonically imprinted gene expression (Kobayashi, 2021). For example, analysis of H3K27me3 level dynamics in early embryos suggests that rodents may be unique in their retention of the mark, at least compared with cow, pig and human (Lu et al., 2021). Indeed, H3K27me3 profiling in uniparental macaque embryos found only sparse associations between paternally expressed genes and maternally biased H3K27me3 (Chu et al., 2021). Primates, including crab eating macaques and humans, clearly show extensive placenta-specific imprinting, albeit at non-overlapping sets of genes (Chu et al., 2021; Xu et al., 2021; Monteagudo-Sánchez et al., 2019), and many of these imprints are associated with a germline DMR (Xu et al., 2021), suggesting that non-canonical imprinting could be less important in the primate lineage. However, it should be noted that in macaque placenta, over 200 paternally expressed genes were not obviously associated with maternal DNA methylation (Chu et al., 2021). This at least opens up the possibility that non-canonical imprinting is occurring, if largely by a different mechanism than that reported in mice and rats.

Intriguingly, a recent screen in the rat model identified non-canonical imprinting of Sfmbt2 and Slc38a4, which are demonstrably important for mouse placenta development (as discussed above) (Richard Albert et al., 2023). Furthermore, eight rat-specific imprints were identified, all of which were transcribed from the paternal allele exclusively in extra-embryonic cells, suggesting a non-canonical mode of imprinted regulation. Compared with canonical imprints identified in rat, all of which were previously characterized in mouse or human, these results suggest non-canonical imprinting is rapidly evolving.

Thus, although the data were limited at the time of writing, it appears that non-canonical imprinting controls expression of a subset of genes in rodents, although the mechanisms dictating the phenomenon as well as the sets of genes implicated may vary between species. Transposable elements are known to be drivers of evolution, including of the placenta (Chuong, 2013; Chuong et al., 2013; Frost et al., 2023), and may have helped accelerate the appearance of imprinted genes. Determining whether non-canonical imprinting occurs in all mammals, and whether ERVKs or different ERV family members may contribute, will provide insights into the convergent evolution of such a mode of gene regulation.

Non-canonical imprinting represents an emerging and fascinating subject for epigenetics and developmental research (Raas et al., 2021; Inoue, 2022). As discussed here, many questions still remain regarding the underlying mechanisms, the scope and the evolutionary conservation of this mode of gene regulation. Although mechanistic studies outside the mouse model remain a general challenge, agnostic exploratory studies, such as those recently described in rats (Richard Albert et al., 2023), will hopefully begin to shed light on the other aspects.

A striking recent study provided strong evidence that genomic imprints remain a barrier for uniparental offspring in mammals. Modification of the methylation status of seven canonical imprints [paternally methylated H19 and Gtl2 (Meg3)/Dlk1 and maternally methylated Igf2r, Snrpn, Kcnq1ot1, Nespas (Gnasas1) and Peg10 ICRs] using epigenome editing in artificially activated diploid oocytes resulted in the generation of an adult mouse (Wei et al., 2022). However, this was an extremely rare outcome, as it represented the only survivor out of 192 implanted embryos. This low efficiency may suggest inefficient DNA methylation editing or that an insufficient number of canonical ICRs are modified. It is also possible that the biallelic expression of non-canonical imprints may have contributed to the very low rate of developmental success (Fig. 2D). Epigenome editing can be used to modify chromatin marks besides DNA methylation (Policarpi et al., 2022 preprint). An interesting avenue to pursue will be the modification of non-canonical imprints, perhaps in combination with canonical imprints, in order to assess the importance of parental asymmetries for proper developmental progression. Although the technology is still in its infancy, epigenome editing could be conducted in vitro using recently developed ex utero development systems (Aguilera-Castrejon et al., 2021; Amadei et al., 2022), which may mitigate both the ethical considerations and labour intensiveness of generating in vivo transgenic material. Hopefully, continued research will continue to unravel the epigenetic regulatory mechanisms, as well as the evolutionary underpinnings, of the fascinating biology of imprinted gene control.

We thank members of the Greenberg lab for useful discussions, and the reviewers for substantially improving the article.