Research Article

Resetting histone modifications during human parental-to-zygotic transition

See allHide authors and affiliations

Science  26 Jul 2019:
Vol. 365, Issue 6451, pp. 353-360
DOI: 10.1126/science.aaw5118

Epigenetics of human embryonic development

Substantial epigenetic reprogramming occurs in mammalian early development. Xia et al. investigated the reprogramming dynamics of three key histone modifications in human early embryonic development. The reprogramming in humans is highly species-specific and different than that in mice. A globally permissive chromatin state connects parental and zygotic epigenomes during maternal-to-zygotic transition. Inner-cell mass-specific, but not trophectoderm-specific, genes are asymmetrically patterned by a histone mark during early lineage segregation.

Science, this issue p. 353


Histone modifications regulate gene expression and development. To address how they are reprogrammed in human early development, we investigated key histone marks in human oocytes and early embryos. Unlike that in mouse oocytes, the permissive mark trimethylated histone H3 lysine 4 (H3K4me3) largely exhibits canonical patterns at promoters in human oocytes. After fertilization, prezygotic genome activation (pre-ZGA) embryos acquire permissive chromatin and widespread H3K4me3 in CpG-rich regulatory regions. By contrast, the repressive mark H3K27me3 undergoes global depletion. CpG-rich regulatory regions then resolve to either active or repressed states upon ZGA, followed by subsequent restoration of H3K27me3 at developmental genes. Finally, by combining chromatin and transcriptome maps, we revealed transcription circuitry and asymmetric H3K27me3 patterning during early lineage specification. Collectively, our data unveil a priming phase connecting human parental-to-zygotic epigenetic transition.

Histone modifications are essential for regulating spatiotemporal gene expression in development (1). Different types of histone modifications have diverse functions and demarcate distinct genomic elements (2). For example, trimethylated histone H3 lysine 4 (H3K4me3) is a conserved hallmark of promoters and transcriptional initiation (3). H3K27me3 is a repressive histone mark preferentially occupying the promoters of developmental genes (4). H3K27ac is widely used as an active mark for promoters and enhancers (5). Mutations of enzymes depositing these marks often result in embryonic lethality in mice and are linked to human diseases (6, 7). Upon fertilization, terminally differentiated gametes undergo drastic epigenetic reprogramming to become totipotent zygotes (8, 9). In mice, several histone marks show noncanonical distributions and functions in oocytes and early embryos (1014). Both H3K4me3 and H3K27me3 display widespread distal domains in mouse oocytes, which can be briefly inherited into early embryos and regulate zygotic gene expression (10, 11, 1416). However, little is known about such a process in human early development. Here, we used CUT&RUN [cleavage under targets and release using nuclease (17)] to study the reprogramming of key histone marks in human early development. Our data reveal distinctive chromatin states during the parental-to-zygotic transition in humans and diverse epigenetic reprogramming modes among different species in early animal embryogenesis.

Canonical H3K4me3 in human oocytes and pervasive H3K4me3 in CpG-rich regulatory regions in human prezygotic genome activation (pre-ZGA) embryos

By optimizing CUT&RUN to allow the generation of high-quality data using as few as 50 cells (fig. S1, A to C; see materials and methods), we performed CUT&RUN for H3K4me3 and H3K27me3 in human germinal vesicle (GV) oocytes; two-cell (for H3K27me3), four-cell (pre-ZGA), and eight-cell (peri-ZGA) embryos (normal and triploid embryos, 3PN); and inner cell masses (ICMs) from blastocysts (post-ZGA) (Fig. 1A) and validated these data (Fig. 1B and figs. S2, A to E, and S3, A to D). We also profiled and validated H3K27ac in eight-cell embryos and ICMs (Fig. 1, A and B, and fig. S4, A to C). Finally, we performed ATAC-seq (assay for transposase-accessible chromatin using sequencing) in human GV oocytes and included published H3K4me3 and H3K27me3 data from sperm (18) for analyses (tables S1 and S2).

Fig. 1 Mapping histone modifications in human gametes and preimplantation embryos.

(A) University of California Santa Cruz (UCSC) genome browser view showing H3K4me3, H3K27me3, and H3K27ac signals in human sperm (18), GV oocytes, preimplantation embryos, and hESCs. H3K27ac data in hESCs are from ENCODE (29). 3PN, embryos derived from zygote with three pronuclei. (B) H3K4me3, H3K27me3, and H3K27ac enrichment near representative genes and their expression levels (20). For HOXD, the average expression level across all HOXD genes is shown.

H3K4me3 occurs as strong peaks at promoters in human GV oocytes (Fig. 1A), which is in contrast to that in mouse oocytes, where it exists in a noncanonical pattern [noncanonical (nc)H3K4me3] as broad domains in partially methylated domains (PMDs) (fig. S5, A and B) (1013). We confirmed the presence of PMDs in human oocytes (fig. S5C) and that these GV oocytes were at late stages of growth with the hallmark of a surrounded nucleolus (19) (fig. S5, D and E). MI oocytes also show a genome-wide H3K4me3 pattern that is very similar to that of GV oocytes (fig. S5, F and G). Such species-specific differences in H3K4me3 are not due to the different methods used (fig. S5H). We then focused on H3K4me3 in human early embryos. Strong promoter H3K4me3 is already found at the pre-ZGA four-cell stage, and this H3K4me3 is also wider than that at other stages (Fig. 1A and fig. S6A). About 53% of these promoters retain H3K4me3 and become preferentially activated in eight-cell-stage embryos (Fig. 2, A and B). The remaining promoters, which associate with genes functioning in cell differentiation and development, lose H3K4me3 and remain inactive upon ZGA. The accessibility in these promoters follows the dynamics of H3K4me3 (Fig. 2, A and B). Indeed, four-cell embryos have the most (n = 2546) stage-specific H3K4me3 promoters among all stages examined (fig. S6B). Although four-cell H3K4me3 is not simply correlated with future transcription (Fig. 2A and fig. S6C), it is well correlated with CpG densities in a continuous and monotonic manner (Fig. 2C). This applies to both genes that are to be activated and those that remain silenced upon ZGA (fig. S7A). In particular, promoters with medium CpG levels are preferentially marked by H3K4me3 in four-cell embryos (Fig. 2A and fig. S7B) but become transcription correlated in post-ZGA stages (fig. S7C). A cluster of promoters is marked by H3K4me3 in four-cell embryos but not in GV oocytes, MI oocytes, or sperm (fig. S7D), supporting de novo H3K4me3 at this stage.

Fig. 2 Dynamics of H3K4me3 in human early development.

(A) Heat maps showing H3K4me3 enrichment at the promoters marked by H3K4me3 either in both four-cell and eight-cell embryos (n = 1272) or only in four-cell embryos (n = 1125) (left). Maternally expressed genes were excluded. ATAC-seq (20), RNA-seq, promoter CpG density (middle), and gene ontology analysis results (right) are also shown. (B) H3K4me3, ATAC-seq, and CpG density levels are shown at representative promoters (left, shaded) or distal accessible regions (right, shaded). Gene expression (left) and gamete DNA methylation (right) are also shown. (C) Scatter plots comparing promoter H3K4me3 with CpG density in human oocytes, embryos, and hESCs. (D) Heat maps showing enrichment of ATAC-seq and H3K4me3 around stage-specific distal ATAC-seq peaks in human early embryos. DNA methylation (29, 42, 43) and CpG density are also mapped.

Four-cell embryos also show widespread distal H3K4me3 (Fig. 1A). Our previous study identified a large number of distal accessible regions in pre-ZGA human embryos (20). Four-cell–specific distal accessible chromatin is also enriched for H3K4me3 (Fig. 2, B and D). These regions are relatively CpG rich (although they have less CpG than do promoters) and hypomethylated in both oocytes and early embryos. This is also the case for distal four-cell H3K4me3 sites (fig. S8A). A moderate H3K4me3 enrichment in oocyte PMD is observed in four-cell embryos but not in oocytes (fig. S5B), likely reflecting de novo deposition of distal H3K4me3 in inherited maternal PMDs (Fig. 2D). Indeed, immunofluorescence (IF) analysis showed increased H3K4me3 after fertilization (fig. S5E) (21). It then quickly decreases in eight-cell embryos, accompanied by the up-regulation of KDM5B and down-regulation of KMT2B, a methyltransferase for ncH3K4me3 in mouse oocytes (13) (fig. S8B). Although reminiscent of ncH3K4me3 in mouse oocytes, human four-cell distal H3K4me3 is much weaker (relative to promoter) (figs. S5, A and B, and S8C) and preferentially resides in gene-dense regions (compartment A). By contrast, ncH3K4me3 in mouse oocytes is present in both gene-dense regions (compartment A) and gene deserts (compartment B) (fig. S8D). In sum, these data reveal widespread distal H3K4me3 in CpG-rich and hypomethylated regions in pre-ZGA human embryos.

Resetting H3K27me3 during human preimplantation development

In human oocytes, H3K27me3 appears to occur in both canonical developmental gene promoters and PMDs (fig. S9, A and B). However, unlike that in mice, oocyte PMD H3K27me3 in humans is not limited to oocytes and early embryos but is also found in somatic tissues. Unlike mouse maternal H3K27me3, which persists until the blastocyst stage (14), H3K27me3 becomes nearly absent in eight-cell human embryos, as shown by IF and CUT&RUN (Fig. 3A and fig. S10A) (21). Coincidently, the PRC2 core components EED and SUZ12 (4) are expressed only in mouse, not human, oocytes and pre-ZGA embryos (Fig. 3B and fig. S10B) (14, 22). We speculate that H3K27me3 in human GV oocytes may be established at an earlier stage when PRC2 is expressed (fig. S10B) and persists in nondividing oocytes. Intriguingly, IF analysis showed that H3K27me3 is present in half of the chromatin in a human zygote (Fig. 3A, arrow). CUT&RUN analysis confirmed that oocyte-specific but not sperm-specific H3K27me3 is retained in two-cell embryos (Fig. 3C). Thus, H3K27me3 undergoes global erasure with the maternal allele briefly delayed. Correlated with the activation of EED and SUZ12 at the eight-cell stage, H3K27me3 is readily found at PRC2 classic targets in ICMs (Fig. 3D). ICM H3K27me3 targets only partially overlap with those in sperm or oocytes (Fig. 3D and fig. S10C), suggesting that the resetting of H3K27me3 involves distinct sets of genes from human gametes to early embryos.

Fig. 3 H3K27me3 resetting in human early development and promoter priming and resolving.

(A) IF of H3K27me3 in human GV oocyte, one- to eight-cell embryos (3PN), and blastocyst. (B) Heat maps showing the expression of H3K27me3-related enzymes. (C) Heat maps showing sperm-specific, oocyte-specific, and shared peaks of H3K27me3 at the promoter and distal regions in human gametes and two-cell embryos. (D) UCSC genome browser views showing H3K27me3 at the gamete–ICM common, gamete-specific, or ICM-specific H3K27me3 targets. (E) Heat maps showing H3K4me3, H3K27me3, and chromatin accessibility at “PcG targets,” “all active” and “all inactive” promoters, and corresponding gene expression. H3K4me3 (this study) and chromatin accessibility (20) of TBEs are also shown.

In mice, persisting maternal H3K27me3 regulates allelic expression in early embryos (15, 16, 23). In particular, inherited oocyte H3K27me3 represses maternal Xist, the master regulator of X chromosome inactivation (XCI), leading to paternal-specific Xist expression and XCI (15, 24). The global loss of H3K27me3 by ZGA in humans predicts the absence of such imprinting events. Indeed, H3K27me3 is absent at the human XIST locus from four-cell embryos to ICMs (fig. S11A), consistent with the lack of imprinted XCI in human embryos (25). Intriguingly, the X chromosome is globally enriched for H3K27me3 in GV oocytes, resembling the inactive X chromosome in female somatic tissues (fig. S11, B and C). As human oocytes show a comparable X chromosome–to–autosome expression ratio (26), these data raised an interesting possibility that H3K27me3 may be involved in X chromosome dosage compensation in human GV oocytes. We then examined putative H3K27me3-controlled imprinted genes identified in mice (16). Among 29 genes that have human orthologs, 11 show promoter H3K27me3 in human oocytes and ICMs, with none showing H3K27me3 at the eight-cell stage and only one (OTX2) expressed in early embryos (fig. S12, A and B). We also examined a list of paternally expressed genes in human morula (27). H3K27me3 in human morula (27) is globally similar to that in ICM but is distinct from that in oocytes, suggesting that H3K27me3 is likely established de novo (fig. S13, A and B). Only 4 out of 27 genes have strong H3K27me3 in GV oocytes or ICM, yet they lack H3K27me3 at the eight-cell stage (fig. S13, C and D). Thus, the paternal expression of these genes is likely regulated by alternative mechanisms. Whether oocyte H3K27me3-mediated gene imprinting exists in humans warrants further investigation.

Promoter and enhancer priming and resolving during human maternal-to-zygotic transition

The widespread de novo H3K4me3 and global loss of H3K27me3 before human ZGA indicated prevalent genome priming. Intriguingly, our previous study in zebrafish also found widespread deposition of histone acetylation at promoters, including those for developmental genes, before ZGA, which subsequently resolve to either active or repressed states after ZGA (28). Such promoter priming is linked to proper ZGA and embryonic development. To ask whether this is also true for human, we identified PcG (Polycomb group protein) target genes [defined in human embryonic stem cells (hESCs)] and compared them with genes that are expressed or inactive at all stages (Fig. 3E). The promoters of the PcG target group are strongly enriched by H3K27me3 in GV oocytes and ICMs. However, they show elevated H3K4me3 and chromatin accessibility in pre-ZGA four-cell embryos, which then decrease upon ZGA (eight-cell stage) (Fig. 3E and fig. S14, A and B). The “all active” but not the “all inactive” gene group showed strong H3K4me3 and ATAC-seq before and after ZGA (Fig. 3E and fig. S14, A and B). We then investigated whether such a priming-resolving epigenetic transition also occurs in the distal regions. Distal H3K4me3 sites are enriched for regulatory elements, as 71.9% (compared with 28.2% of random) of them overlap with candidate cis-regulatory elements (CcREs) identified by ENCODE project (29). Distal four-cell H3K4me3-marked CcREs show high chromatin accessibility (fig. S15, A and B). After ZGA, one class (fig. S15A, “active”) remain accessible and are marked by H3K27ac in post-ZGA embryos and are associated with basic biological processes (GREAT analysis). By contrast, the other class (fig. S15A, “poised”) are associated with H3K27me3 in ICM and somatic cells and reside preferentially near developmental genes (figs. S15A and S16A). They do not show H3K4me3 in hESCs and somatic cells, confirming that they are not unannotated promoters (fig. S16B). These enhancers, however, can acquire H3K27ac and become activated in a tissue-specific manner (fig. S16, C and D), suggesting that they may be “poised enhancers.” Notably, both active and poised CcREs show low accessibility in GV oocytes (fig. S15, A and B), which may have distinct enhancer usages compared with embryos and somatic tissues. Finally, the priming-resolving transition of H3K4me3 and chromatin accessibility is prevented in transcription-blocked embryos (TBEs) treated with alpha-amanitin (Fig. 3E and figs. S14, A and B, S15, A and B, and S17, A to C). These data indicate that chromatin may be globally reset to a permissive state prior to ZGA, before they resolve after ZGA in a transcription-dependent manner.

Transcription circuitry and asymmetric epigenetic patterning during human early lineage segregation

We then sought to determine transcription circuitry in human post-ZGA development. We first identified distal H3K27ac sites that are also accessible [as assayed by ATAC-seq (20)] as putative active enhancers. These active enhancers are hypomethylated in ICMs and hESCs but not in eight-cell embryos, where TET genes, the DNA demethylases (30), are down-regulated (30) (fig. S18, A to C). Next, we identified the binding motifs for transcription factors (TFs) enriched in these enhancers (fig. S19A). GSC and OTX2 are specifically enriched in eight-cell embryos. The GATA and KLF families and TFAP2A/C are enriched in eight-cell embryos and ICMs. Consistently, GATA factors are critical regulators of primitive endoderm (PE), which is part of the ICM (31), and KLF factors support the naïve pluripotency (32). Finally, TFs in ICMs and hESCs include pluripotency factors and the TEAD family, with the latter being involved in mouse ICM and trophectoderm (TE) segregation (33). Thus, these results show stage-specific usages of the TF repertoire in human embryonic development.

Human embryos develop into epiblast (EPI), PE, and TE lineages concurrently (25). We searched for putative factors that regulate these lineages using MARINa (master regulator inference algorithm) (34), which identifies key TF regulator candidates whose putative targets are enriched cell-specific transcription programs. By predicting the TF targets (35) and integration with single-cell RNA-sequencing (RNA-seq) data in human early embryos (25), MARINa identified putative regulators for each lineage (Fig. 4A and fig. S19B-C; see materials and methods). Encouragingly, this “transcription lineage circuitry map” (Fig. 4A) correctly predicted many known lineage-related TFs, including NANOG, OCT4, and SOX2 for the EPI lineage; GATA4 and GATA6 for the PE lineage; and TEAD1, TEAD3, TFAP2C, GATA2, and GATA3 for the TE lineage. For instance, GATA4/6 are canonical markers for PE in both mice and humans (33). By contrast, GATA2/3 double-knockout mutant disrupts mouse TE formation (36). We also identified additional candidate regulators of human lineage specification. TEAD4, a mouse TE regulator (37), is predicted to be an EPI and PE regulator in human (fig. S19D). TEAD2, which has functions in neural crest (38), is also classified as an EPI regulator. OTX2, a human early embryo–specific TF (20), is predicted to regulate PE development. Notably, OTX2 is enriched in the eight-cell embryo but not in the ICM (fig. S19A), suggesting that it may regulate lineage circuitry before lineage segregation. Therefore, our analysis identified both conserved and species-specific TFs that may orchestrate human early development and lineage segregation.

Fig. 4 Putative master regulators for lineage segregation and asymmetric H3K27me3 patterning in human post-ZGA embryos.

(A) (Left) Schematic model for the identification of lineage master regulators. (Right) PCA biplot showing the identified putative master regulators for the EPI, PE, and TE lineages identified by MARINa. Each candidate TF regulator is associated with three enrichment scores during pairwise comparison of EPI, TE, and PE. Pseudopoints (as coordinate references) were added to represent the ideal lineage-specific regulators. (B) (Left) Heat maps showing the expression of EPI-, PE-, and TE-lineage–specific genes in day 5 (D5) to D7 human embryos (25). The absolute and relative (row-normalized) expression levels among the three lineages are shown. (Middle) H3K27me3 from GV oocyte to D6 or D7 ICM and TE (this study) and morula (27). (Right) Fractions of lineage-specific genes marked by H3K27me3 in hESCs, tissue (brain and liver), and ICM or TE.

It has been reported that during the initial human lineage specification, both PE- and EPI-specific genes are preferentially silenced in TE cells (25). However, TE-specific genes are down-regulated but still retain certain levels of expression in PE and EPI cells even at a late blastocyst stage (D7) (25) (Fig. 4B, compare expression versus relative expression, and fig. S20A). Notably, in ICM (D6), H3K27me3 marks substantial portions of ICM-specific genes (34% of EPI-specific and 38% of PE-specific genes) but few (8%) TE-specific genes (Fig. 4B). This is also true in TE cells and also at a later stage (Fig. 4B and fig. S20A), when we examined TE (D6) and later-stage ICM and TE (D7) separated as described previously (20) (fig. S20B; see materials and methods). By contrast, such biased H3K27me3 patterning was not found in hESCs or somatic cells (Fig. 4B). These data revealed asymmetric epigenetic patterning of ICM versus TE-specific genes in human early embryos. As ICMs include mixed EPI and PE cells, future studies are needed to determine whether H3K27me3 is differentially deposited at EPI- and PE-specific genes at the single-cell level. The biased patterning appears as early as the morula stage (Fig. 4B and fig. S20A), before lineage segregation. If true, the differential expression of EPI- and PE-specific genes may involve lineage-specific erasure of H3K27me3. Alternatively, early-stage cells may consist of heterogeneous lineage-primed subpopulations with different H3K27me3 states.


A long-standing question in the field is to what extent the modes of epigenetic reprogramming in early embryos are conserved between model organisms and humans. Earlier studies have shown that reprogramming of DNA methylation is highly similar between mouse and human (39, 40). However, the reprogramming of histone modifications is much more complex. A recurrent theme is that parental chromatin marks appear to ultimately undergo resetting during the parental-to-zygotic transition, although often in a stage- and allele-specific manner. Distinct species-specific features of epigenetic reprogramming also clearly exist (fig. S20C). H3K4me3 exists as noncanonical forms in mouse oocytes (10, 11, 13, 14) but shows a primarily canonical pattern in human oocytes. Mouse distal H3K27me3 domains can persist beyond ZGA and regulate allelic gene expression (15, 16, 23). However, this does not seem to be evident in humans, consistent with the lack of imprinted XCI (25). These differences highlight pronounced divergence of epigenetic reprogramming modes in early animal embryos during evolution.

Intriguingly, CpG-rich regulatory elements, which preferentially carry unstable nucleosomes (41), appear to acquire pervasive H3K4me3 and accessible chromatin prior to ZGA, before undergoing ZGA-dependent resolution. To distinguish it from ncH3K4me3 in mouse oocytes, we termed such pervasive H3K4me3 in promoter and distal regulatory elements before ZGA as “priming H3K4me3.” We propose that such dynamic epigenetic reprogramming is analogous to computer system rebooting and can thus be referred to as “epigenome rebooting,” which includes the erasure of the parental epigenetic memory (memory clearance), the configuration of the default epigenome (loading of basic configuration) by priming regulatory elements, and the establishment of the zygotic epigenome for subsequent developmental programs (loading of advanced functional programs) (Fig. 5A).

Fig. 5 Dynamic reprogramming of histone modifications during the human maternal-to-zygotic transition.

Schematic model showing “epigenome rebooting” during the maternal-to-zygotic transition. After fertilization, the parental H3K27me3 is globally lost from the PcG targets. De novo H3K4me3 and accessible chromatin are established at CpG-rich promoters (including developmental genes) and distal regions, possibly as a “default” chromatin state (basic configuration). The zygotic epigenome (advanced configuration) is established through ZGA-induced reprogramming followed by the establishment of H3K27me3. This leads to resolution of primed promoters to “active” promoters with H3K4me3 or “poised” developmental gene promoters with H3K27me3. A similar transition occurs for putative enhancers. Early lineage specification is accompanied by asymmetric H3K27me3 patterning, with ICM (EPI or PE)–specific, but not TE-specific, genes preferentially marked. As ICM includes mixed EPI or PE cells, the H3K27me3 states in individual EPI or PE cells remain unknown (question marks).

We also found preferential H3K27me3 patterning on ICM-specific but not TE-specific genes in the initial human early lineages. One possible reason for this is that TE-specific genes are more compatible with EPI and PE, whereas the EPI- and PE-specific genes are incompatible with other lineages and thus require H3K27me3 to stay repressed. Our study reveals both conserved and diverse reprogramming modes that connect parental epigenomes and zygotic epigenomes and paves the way for future investigations of epigenetic reprogramming in human early development.

Supplementary Materials

Materials and Methods

Figs. S1 to S20

Tables S1 to S3

References (4464)

References and Notes

Acknowledgments: We thank S. Henikoff for sharing pA-MNase and advising on CUT&RUN; P. Wang, Y. Zhu, and F. Duan for sharing hESCs; lab members for discussion; and the animal and biocomputing facility at Tsinghua University for support. Funding: This work was supported by National Natural Science Foundation of China (81820108016 and 81741042 to Y.S., 31422031 and 31725018 to W.Xie, and 31870817 to J.W.X.), Henan Provincial Obstetrical and Gynecological Diseases (Reproductive Medicine) Clinical Research Center (to Y.S. and J.W.X), National Key R&D Program (2016YFC0900300 to W.Xie and 2017YFA0102802 to N.J.), National Basic Research Program of China (2015CB856201 to W.Xie), Beijing Municipal Science & Technology Commission (Z181100001318006 to W.Xie), and THU-PKU Center for Life Sciences (to W.Xie). W.Xie is a Howard Hughes Medical Institute international research scholar. Author contributions: Y.-P.S. and W.Xie conceived and supervised the project. J.X. performed ATAC-seq and managed the human sample collection. CUT&RUN in human embryo (W.Xia), mouse embryos, and hESCs (G.Yu and K.X.) were performed with help from B.L., L.L., N.Z. and D.S. J.X., G.Yao, and N.Z. performed TBE experiments. G.Yao, X.M., S.S., Y.-P.S., and H.J. prepared the human samples. X.M. and T.L. performed IF. K.X. and Z.L. collected mouse oocyte and embryos. N.Z. and Q.W. performed next-generation sequencing. X.C. and J.N. provided hESCs. Y.W., Z.B., W.S., and L.H. recruited the volunteers and performed oocyte retrieval. G.Yu analyzed data and prepared most of the figures. G.Yu, W.Xia, and W.Xie interpreted the data. W.Xia and W.Xie wrote the manuscript with help from G.Yu, J.X., and the remaining authors. Competing interests: The authors declare no competing financial interests. Data and materials availability: All data have been deposited to the Gene Expression Omnibus (GEO) with accession number GSE124718.

Stay Connected to Science

Navigate This Article