X-chromosome upregulation: hypotheses, observations and mechanisms

Longstanding controversies include whether or not XCU is present in mammals, and if so, whether it is global or affects only a subset of genes, and to which extent (whether it achieves complete dosage compensation or only partial). We have compiled a list of the studies that have investigated XCU in mammals and included their conclusions, approaches and data used for analysis (Table 1). One of the main reasons why different studies reached different conclusions is their approach when determining XCU. Many authors have compared the expression levels of X-linked genes with that of autosomal genes across several tissues in different mammals. Using this approach, the vast majority of studies have reported similar global levels of expression of X-linked and autosomal genes (based on expression ratios and/or distributions), thus concluding that upregulation of the single active X in mammals occurs. The two exceptions are a study using low-coverage proteomics data and a study in which non-expressed genes were not discarded from the analysis (Deng et al., 2011; He et al., 2011; Kharchenko et al., 2011). Concluding that XCU takes place from the fact that expression levels of the single active present-day X chromosome are similar to expression levels of autosomes assumes that, before sex-chromosome differentiation, expression levels of genes on the (ancestral) proto-sex chromosomes were similar to those on the ancestral autosomes. Such an assumption is not directly derived from Ohno's hypothesis, which did not postulate similar levels of expression but balanced levels of expression, which had to be preserved upon sex-chromosome differentiation. Thus, other authors have argued that comparing X-linked expression levels to autosomal expression levels is not a real test of Ohno's hypothesis (He et al., 2011; Julien et al., 2012; Lin et al., 2012). Instead, they should be compared to expression levels in the ancestral proto-sex chromosomes, for which these authors proposed to use, as a proxy, expression levels of the genes in monotremes and birds that are (autosomal) orthologs of the therian X-linked genes. Based on these comparative analyses, no upregulation was observed, leading the authors to refute Ohno's hypothesis. This approach makes assumptions too, and whether contemporary mammals can be directly compared to contemporary birds was initially questioned (Disteche, 2016). Meanwhile, consistent results have been achieved using other outgroups and various sets of autosomal genes (Julien et al., 2012; Marin et al., 2017; Wang et al., 2020), and although no XCU was found in placental mammals, full global XCU was demonstrated in marsupials (Julien et al., 2012). Current-to-ancestral comparisons may not always be feasible (if relevant data, including for outgroups, is not available; e.g. during specific developmental stages), so careful X-to-autosome comparisons can still be relevant and informative.

Table 1.

Conclusions of studies addressing XCU in mammals

Open in a separate window

These considerations mirror the challenge of defining sex-chromosome dosage compensation. Some authors adopt a broader definition, such as ‘the regulatory mechanisms that balance gene expression between the autosomes and sex chromosomes in the heterogametic sex’ (Mank et al., 2011), whereas others explicitly include the notion of the evolutionary history of the chromosomes; for example, ‘the maintenance of ancestral expression levels of sex-linked genes relative to autosomal expression in the heterogametic sex’ (Gu and Walters, 2017).

So, is there XCU or not in placental mammals? XCU as predicted in Ohno's hypothesis, which refers to higher expression of the present-day X compared with that of a single proto-sex chromosome, cannot be directly tested. As we reviewed, different approaches to address this question have different assumptions and have reached different conclusions. The question is not only whether the X is upregulated or not, but to what extent. Based on current-to-ancestral comparisons in placentals, a twofold upregulation is not achieved at the mRNA level but upregulation takes place – at least for some genes, as proposed by many (Naik et al., 2022; reviewed by Gu and Walters, 2017; Mank et al., 2011; Pessia et al., 2014). Recent single-cell, single-allele RNA-sequencing has confirmed higher expression from genes on the active X chromosome (Lentini et al., 2022), as discussed further below. At the molecular level, the (active) X chromosome is enriched in features that are all consistent with a ‘hyperactive’ transcriptional state compared with autosomes (Fig. 1). Its gene promoters show higher transcriptional burst frequencies (Larsson et al., 2019; Talon et al., 2021) and are enriched in the initiation form of RNA polymerase II, active histone marks, including histone acetylation (H4K16ac), and the corresponding acetyltransferase (MOF) that mediates XCU in Drosophila (Deng et al., 2011, 2013; Yildirim et al., 2011). Concomitantly, the active X shows higher chromatin accessibility than autosomes, as profiled by single-cell ATAC-seq (Talon et al., 2021). This investigation has identified increased chromatin accessibility on the active X chromosomes in mouse XX fibroblasts and XY mouse embryonic stem cells (mESCs), but not on the active X chromosomes of XX induced pluripotent stem cells (iPSCs) or mESCs. Interestingly, these results match the observations that the X chromosomes in mouse XX fibroblasts and XY mESCs are upregulated, whereas X chromosomes in XX mESCs are not (Larsson et al., 2019; Lentini et al., 2022). Recently, the BRD4 protein (containing bromodomains, which recognise acetylated lysine residues such as those in histones) has been implicated in the transcriptional activation of X-linked genes showing upregulation (Lyu et al., 2022), but this has been contested (Lentini and Reinius, 2023; Lyu et al., 2023).

Open in a separate window

Fig. 1.

Balancing the scales: the choreography of upregulation and inactivation of the X chromosome during mouse embryonic development. (A) A hypothetical model of chromosome conformation dependent on the status of the X chromosome: active (Xa), upregulated (Xu), inactive (Xi). For the Xa state, an open chromatin configuration is illustrated, along with topologically associating domains (TADs) represented by orange and red triangles, and interactions between TADs shown as black connections. In the Xu state, it has been suggested that the number of interactions between TADs increases (Lentini et al., 2022), and there is ongoing debate regarding the potential involvement of BRD4 in transcriptional upregulation. In the Xi state, heterochromatin configuration is illustrated, along with the presence of the two megadomains (MD1; MD2) separated by the tandem repeat DXZ4. (B) Overview of the timing and dynamics during mouse pre- and early post-implantation development in both sexes. In XX embryos, from pre-implantation stages (E0.5-E4.5), the maternal X chromosome (Xm) undergoes upregulation (XCU) while the paternal X chromosome (Xp) is subject to inactivation (iXCI, Xist RNA in blue). At the blastocyst stage, Xm undergoes downregulation (XCdr), whereas Xp undergoes reactivation (XCR). Following this, in peri- to post-implantation stages, random XCI takes place, followed by (random) XCU on the other chromosome. In XY embryos, the sole X present (Xm) undergoes XCU immediately after ZGA, maintaining this upregulated state throughout development.

A seemingly absent full dosage compensation (at the transcriptional level) has led to alternative hypotheses for the origin of XCI and partial XCU, unrelated to dosage compensation (Chandra, 1985, 2022; Engelstädter and Haig, 2008; Gribnau and Grootegoed, 2012; Haig, 2006; Iwasa and Pomiankowski, 2001; Mank et al., 2011; Pessia et al., 2014). Recently, however, the Kaessmann lab has proposed a reconciliatory perspective. Translatome analysis of tissues from four therian species (but not from platypus, a monotreme) revealed ‘translation upregulation’, with higher ratios of current-to-ancestral expression for the translatome than for the transcriptome (Wang et al., 2020). This was also associated with higher translation efficiencies and protein abundance (Wang et al., 2020). Combined with transcriptional upregulation, translational upregulation appears to have largely restored ancestral expression levels and, thus, X-to-autosome balance (Wang et al., 2020). Accordingly, previous studies have reported that X-linked transcripts have significantly higher ribosome density (Faucillion and Larsson, 2015) and longer half-lives (Deng et al., 2013; Faucillion and Larsson, 2015; Rücklé et al., 2023) than autosomal transcripts. Recently, depletion of RNA-associated N6-methyladenosine (m6A) modification led to a reduction in the ratio of X:autosome expression in both mouse and human cells, mainly through an increase in the stability of autosomal transcripts (Rücklé et al., 2023). X-linked transcripts were mostly unaffected, which is explained by their low(er) levels of m6A (Rücklé et al., 2023). This appears to be an intrinsic feature of X-linked transcripts, which show a depletion of the GGACH sequence, the m6A consensus motif (Rücklé et al., 2023). This suggests that the higher stability of X-linked mRNAs is hard-wired in the X-chromosome DNA sequence; how this has evolved remains an intriguing open question. In summary, full dosage compensation in placental mammals appears to be happening through a combination of transcriptional and post-transcriptional upregulation of gene expression (Wang et al., 2020), and thus Ohno's hypothesis stands after all.

Unlike XCU in Drosophila and XCI in mammals, mammalian XCU occurs in both sexes and does not appear to rely on a chromosome-wide mechanism – two aspects that we believe have contributed to hinder our understanding of this enigmatic process. Despite an increasingly better molecular understanding of XCU, many questions remain unanswered: namely, which mechanisms confer specificity to the upregulation (i.e. how do they target X-linked genes), and whether all X-linked genes or a subset need to be upregulated? Of note, alternative mechanisms have been reported that have (potentially) allowed compensation the hemizygosity of X-linked genes during sex chromosome evolution; these include the downregulation of autosomal genes that are partners of X-linked genes (Julien et al., 2012), retention of a functional gene copy on the Y chromosome (Bellott et al., 2014; Cortez et al., 2014), duplication of genes on the X (Julien et al., 2012), and relocalisation of proto-Y genes to autosomes (Carelli et al., 2016; Hughes et al., 2015; Potrzebowski et al., 2008). Interestingly, in rodents that have lost the Y chromosome completely, some ‘Y-linked’ genes (presumably the dosage-sensitive ones) are found on the X or autosomes (Arakawa et al., 2002; Kuroiwa et al., 2010; Mulugeta et al., 2016).

X-chromosome inactivation: convergent evolution in therian mammals

Based on several genetic studies and observations in mammals, Mary Lyon put forward the idea of X-chromosome inactivation in 1961, by proposing that the dark-staining X chromosome in female somatic cells (Barr and Bertram, 1949; Ohno et al., 1959) was inactivated (Lyon, 1961). The 60th anniversary of her seminal proposal was celebrated recently (Moyano Rodriguez and Borensztein, 2023). Insightfully, Lyon also anticipated that this inactive X could be ‘either paternal or maternal in origin in different cells of the same animal’ (what is referred to as random XCI, true for placental mammals but not for marsupials) and that it occurred early in embryonic development (Lyon, 1961). A truly epigenetic process, XCI is heritable through mitosis and can be reversed (X-chromosome reactivation), which happens in specific developmental stages and pathological contexts (Panda et al., 2020; Spaziano and Cantone, 2021; Talon et al., 2019). How XCI is triggered specifically in XX individuals, how it affects only one X chromosome and the molecular mechanisms that are implicated in the transcriptional silencing of the X, which is accompanied by heterochromatinisation and chromosome refolding, have been recently reviewed elsewhere (Kanata et al., 2024; Keniry and Blewitt, 2023; Loda et al., 2022; Mutzel and Schulz, 2020; Schwämmle and Schulz, 2023). Here, we cover the evolutionary diversity observed in XCI across mammalian species and its implications for our understanding of dosage regulation and compensation.

According to Ohno's hypothesis, XCI was the ‘second step’ needed for X-linked dosage compensation upon differentiation of the mammalian sex chromosomes; XCI evolved in XX individuals to counteract the effects of XCU, which balanced X-linked gene expression to autosomes in XY individuals but created a problem for XX. Often in the field, XCI is mentioned as having evolved to ‘equilibrate X-linked gene expression between the sexes’, an oversight because of course selection does not work on the balance between the sexes, but on the individual (Vicoso and Bachtrog, 2009). Besides being imprecise, such formulation reinforces the idea that compensated X-linked gene expression is expected to be the same between the sexes, although this might not be the case. Instead, it just needs to be compatible with life and reproduction in each sex [‘incomplete but sufficient’ (Gu and Walters, 2017)]. This means, for example, that the X:autosome expression ratio (for each gene) does not have to be exactly the same in XX and XY individuals.

Remarkably, XCI is present in both marsupials and placental mammals (Fig. 2), but it appears to have evolved independently in these two lineages (reviewed by Shevchenko et al., 2013). Marsupial and placental XCI do share certain features: relying on the activity of long noncoding RNAs (lncRNAs), having the same functional outcome (silencing of X-linked genes), the inactive X being targeted by H3K27 trimethylation and to the perinucleolar compartment (Mahadevaiah et al., 2009), but their genetic origins are not homologous. At the forefront of orchestrating XCI in placental mammals stands the lncRNA, Xist, discovered more than 30 years ago (https://thenode.biologists.com/xist-discovery/discussion/). Xist is essential for XCI in mice (Marahrens et al., 1997; Penny et al., 1996), but no Xist gene has ever been found in marsupials (Davidow et al., 2007; Deakin et al., 2009; Hore et al., 2007; Okamoto and Heard, 2009; Shevchenko et al., 2007; Waters et al., 2005). The Xist gene is proposed to have emerged de novo in eutherians, exhibiting remnants traceable to mobile elements spanning diverse classes and to Lnx3, a protein-coding gene present in birds and marsupials but that no longer exists in eutherians (Duret et al., 2006; Elisaphenko et al., 2008). In marsupials, XCI is associated with a different lncRNA, Rsx. Although not being a sequence homologue of Xist, their RNAs share many functional attributes, such as the female-specific expression, the association and ‘coating’ of the inactive X, the activity in cis, and a similar protein interactome (Grant et al., 2012; Mahadevaiah et al., 2020; McIntyre et al., 2024). Formal genetic evidence of Rsx being essential for marsupial XCI is still lacking; however, expression of Rsx in mESCs from an autosomal transgene resulted in gene silencing in cis (Grant et al., 2012). It is noteworthy that marsupial XCI is imprinted (it is always the paternal X that is inactivated in XX somatic cells, whereas in placentals XCI is random) and comparatively more incomplete than the Xist-driven process (reviewed by Deakin et al., 2009).

Open in a separate window

Fig. 2.

The evolutionary diversity of dosage compensation mechanisms among mammals. The phylogenetic tree (generated with timetree.org) displays various mammalian species. On the right, we have indicated what is known regarding X-linked dosage compensation processes in the species presented (iXCI, imprinted X-chromosome inactivation; rXCI, random X-chromosome inactivation; XCU, X-chromosome upregulation). Boxes in orange denote presence, in black denote absence, in red denote conflicting research and in white denote absence of studies. White asterisks indicate that a given process is assumed to be present but has not been formally demonstrated. The pictograms used for the tree are solely for visualisation purposes and might not correspond to specific species. References can be found in Table 1.

Marsupial and placental XCI are thus a compelling illustration of convergent evolution, driven by two independently evolved lncRNAs that silence the activity of nearly an entire chromosome (McIntyre et al., 2024). Importantly, as mentioned before, the sex chromosomes in marsupials and placentals are homologous, whereas the dosage compensation mechanisms in XX individuals are not; it is unclear whether these were preceded by an older dosage compensation mechanism in the last common ancestor. Some authors have suggested an initial process potentially involving other noncoding RNAs, possibly acting gene-by-gene (as in monotremes, Box 3), which was replaced by chromosome-wide regulation by Xist and Rsx to facilitate more efficient silencing (Gribnau and Grootegoed, 2012; McIntyre et al., 2024). It is also possible that Rsx represents the ancestral regulator, whether acting gene-by-gene or chromosome-wide. Interestingly, a recent study in chicken has shown that a microRNA contributes to the downregulation of Z-linked genes in ZZ individuals, which are upregulated in ZW individuals (Fallahshahroudi et al., 2024 preprint).

Placental XCI: a diversity of roads leading to random XCI

Investigating XCI in placentals other than the mouse (e.g. human, macaque, rabbit, cow and pig) has revealed evolutionary flexibility in the regulation and dynamics of XCI (Fig. 2). Some of the truths we learnt about XCI with the mouse appear to be more of an exception than the rule in placental mammals, probably representing recently evolved characteristics rather than common ancestral traits – a likely reflection of the extensive evolutionary radiation of rodents (Fabre et al., 2012) and of the mouse genome evolving faster than that of larger mammals due to more generation cycles per unit of time (Svoboda, 2018).

When does XCI take place during development? In all species examined so far, the random XCI pattern observed in somatic cells is first detected in post-implantation embryos. This corresponds, for example, to embryonic day (E)6-E7 in the mouse and E15-E17 in the macaque, which raises interesting questions regarding the time and extent to which gene dosage compensation is needed (Heard and Rougeulle, 2021; Okamoto et al., 2021). However, it is also important to consider the differences between chronological time and developmental time (Dubansky, 2018; Garcia-Ojalvo and Bulut-Karslioglu, 2023). Even if macaque embryos go through more days without dosage compensation, in terms of ‘developmental time’ dosage compensation appears to be required at a similar stage as in the mouse (Fig. 3).

Open in a separate window

Fig. 3.

Colourful development: X-linked states in mice, macaques and humans. The dynamics of both X chromosomes during development in mice, monkeys and humans are summarised from available studies. Colours represent the status of one X chromosome (X1) or the other (X2), illustrating how XCI patterns during early development are quite variable across species. The ‘active’ status indicates either the expression of one or more X-linked genes assessed using fluorescence in situ hybridisation (FISH) or by single-cell RNA sequencing. Schemes inspired by Nakamura et al. (2021), Okamoto et al. (2021), Saiba et al. (2018), Shevchenko et al. (2019) and Zhai et al. (2022).

Not all species, however, reach random XCI in the same way. In the mouse, there is an earlier wave of XCI, which is paternally imprinted, similar to marsupials (another likely example of convergent evolution) and occurs during preimplantation stages (reviewed by Furlan and Galupa, 2022). Murine imprinted XCI is maintained in the extra-embryonic tissues and reverted in the inner-cell mass of the blastocyst, which gives rise to the epiblast cells, in which random XCI occurs some days later. In placental mammals, imprinted XCI is restricted to extra-embryonic tissues and appears to also be present in rat and bovine embryos (Dindot et al., 2004; Magaraki et al., 2019; Wake et al., 1976; Xue et al., 2002; Yu et al., 2020), but absent in human, macaque, rabbit, pig and horse, where extra-embryonic tissues show random XCI (Beckelmann et al., 2012; Goszczynski et al., 2021; Moreira de Mello et al., 2010; Okamoto et al., 2011, 2021; Ramos-Ibeas et al., 2019; Romagnano et al., 1987; Wang et al., 2012; Zou et al., 2019). Which species show imprinted XCI does not appear to be associated with the type of placental structure (Laudon et al., 2024). Instead, it has been linked to earlier zygotic genome activation (ZGA) and faster development (Migeon, 2002), but this does not appear to hold in bovine embryos (Svoboda, 2018). Evolutionary considerations about imprinted XCI have recently been reviewed by Furlan and Galupa (2022).

In the mouse, the expression of Xist is tightly coupled to XCI. Intriguingly, this is not the case for other mammalian species. During preimplantation development, XIST RNA is detected coating the X chromosome for several days without inducing gene silencing in human, monkey, rabbit and bovine embryos (Okamoto et al., 2011, 2021; Yu et al., 2020). Moreover, XIST RNA is detected coating both X chromosomes in XX embryos or even the X chromosome in XY human and macaque embryos for several days (Okamoto et al., 2011, 2021; Vallot et al., 2017). It is still unclear why and how XIST RNA accumulation is uncoupled from XCI, and what is the switch/trigger that allows these pre-XCI but XIST-associated states to eventually be resolved into random XCI in XX embryos and no XCI in XY embryos. Nevertheless, in human embryos, XIST presence might not be without consequences: it coincides with the downregulation of X-linked gene expression, which has been termed X-chromosome ‘dampening’ (XCD) (Petropoulos et al., 2016). Dampening does not appear to be present in other primates, such as the macaque or marmoset (Cidral et al., 2021; Okamoto et al., 2021) and remains contested in human (Moreira de Mello et al., 2017; Mandal et al., 2020). Recently, it has been shown in human ESCs that deletion of XIST leads to derepression of X-linked expression (suggesting that XIST is responsible for XCD) and that SPEN, the transcriptional repressor that is essential for initiating gene silencing during XCI (Dossin et al., 2020), is also involved (Alfeghaly et al., 2024; Dror et al., 2024). Again, what prevents the expression of XIST and recruitment of SPEN to lead to full XCI remains unknown.

New insights into the regulation of Xist across placental species

The regulation of Xist preceding random XCI also shows species-specific variations, at least, as evaluated by studies on cultured mouse and human cells. Xist is embedded in a regulatory landscape with many other noncoding loci (reviewed by Luchsinger-Morcelle et al., 2024), some of which evolved via pseudogenisation from protein-coding genes along with Xist (Chureau et al., 2002; Duret et al., 2006). Whether Tsix, which runs antisense to Xist and is essential for XCI regulation in mice, is as important in other species is an old debate (reviewed by Galupa and Heard, 2018). For now, genetic evidence to support or disprove such a role is still missing. More recently, Claire Rougeulle's lab has spearheaded functional analyses in primate ESCs of some of the noncoding neighbours of XIST, the JPX and FTX loci that, in mouse, are important positive regulators of Xist and random XCI (Furlan et al., 2018; Gjaltema et al., 2022; Sun et al., 2013; Tian et al., 2010). Interestingly, FTX functions are not conserved in human (Rosspopoff et al., 2023), and JPX is also a major regulator of XIST regulation in human but not in macaque or marmoset (Cazottes et al., 2023 preprint; Rosspopoff et al., 2023). Yet, between human and mouse there are differences in how Jpx/JPX regulates Xist/XIST. In mouse, the Jpx RNA mediates Xist regulation, at the post-transcriptional level; in human, the JPX RNA is dispensable for XIST regulation. Rather, it is JPX transcription that is required for proper XIST expression, in cis, probably via influencing RNA polymerase II recruitment to the XIST promoter (Rosspopoff et al., 2023). In macaque and marmoset ESCs, no indications have been found of ongoing transcription at syntenic positions of Tsix, Linx or Xite (also known as Rr18) (Cazottes et al., 2023 preprint), other noncoding loci recognised as significant Xist repressors in mice (Galupa et al., 2020; Hierholzer et al., 2022; Lee, 2000; Lee and Lu, 1999; Ogawa and Lee, 2003; Sado et al., 2001).

Unravelling the diverse mechanisms and regulatory strategies governing XCI across eutherian mammals sheds light on the intricate evolutionary dynamics of this dosage compensation process. As stated by Okamoto and colleagues, the existing diversity ‘probably reflects the fact that developmental processes are constantly changing during evolution and that the regulation of processes such as XCI have to display substantial plasticity to accommodate these changes’ (Okamoto et al., 2011).

We thank Estelle Nicolas for critical feedback on the manuscript and Ikuhiro Okamoto for critical feedback on the figures. We apologise to authors whose work we overlooked or did not discuss or cite due to limits in the number of references. We thank the anonymous reviewers for pointing us to additional literature and for their constructive feedback. Figures were prepared with BioRender.com.