A barrier to human embryogenesis involves dysregulated 3D genome reprogramming and ZGA collapse

Abstract

Early embryonic arrest (EEA) is a critical impediment in assisted reproductive technology, leading to lack of viable embryos. Despite its prevalence, the molecular basis of EEA remains elusive, and the contribution of 3D genome reorganization to this arrest remains largely unexplored. Here, we present an integrated analysis of chromatin architecture and transcriptome dynamics in arrested human embryos during zygotic genome activation (ZGA) stage. Our results reveal extensive chromosomal compartment switching, significant alterations in topologically associating domains (TADs) features, and disrupted chromatin looping in arrested embryos. It suggests that 3D genome reorganization and ZGA failure occur in arrested embryos, marked by reduced abundance of ZGA-critical genes, precocious activation of late-stage developmental pathways, suppression of RNA processing and ribosome biogenesis, and compromised energy metabolism. Additionally, we hypothesize that several transcription factors, including KLF17 and several ZNF family members, serve as regulators in the observed structural or functional changes during developmental arrest. Collectively, our findings suggest that defective 3D genome reprogramming and transcriptomic dysregulation during ZGA are associated with human embryonic arrest at the eight-cell stage. This study advances our understanding of human preimplantation development and its development failures by providing a valuable resource and a conceptual framework for future investigations.

Integrative multi-omics analysis reveals association between defective 3D genome reprogramming, transcriptomic dysregulation, and human early embryonic arrest at the ZGA stage.

Introduction

Infertility represents a major challenge to human reproductive health, impacting more than 15% of couples of reproductive ages^1,2. Despite advancements in assisted reproductive technologies, the clinical success rates of these interventions remain suboptimal³. A particularly critical issue in clinical settings is early embryonic arrest during in vitro fertilization (IVF), underscoring the need to elucidate the molecular mechanisms responsible for developmental failure in human embryos.

Human preimplantation embryo development is a highly orchestrated process that entails maternal-to-zygotic transition (MZT) and is susceptible to disruption by maternal, paternal, and embryonic factors^4,5. Zygotic genome activation (ZGA), the first major transcriptional event in embryogenesis, begins with minor ZGA⁶ and culminates in a unique hypertranscription state termed major ZGA⁷. ZGA involves dramatic reprogramming of epigenetic landscapes through the resetting of epigenetic marks, chromatin accessibility, and chromatin organization⁸. The 8-cell stage is a critical temporal window for ZGA, and inhibition of this process can result in the arrest of early embryonic development and failure in embryo implantation^9,10. Arrested 8-cell embryos provide an excellent model for studying human ZGA, as their development is precisely “frozen” at this critical transition, preserving the sequence of key molecular events¹¹.

A pivotal aspect of embryogenesis is the extensive reorganization of 3D chromatin architecture¹². After fertilization, the specialized gametic chromatin is remodeled into a highly relaxed state, largely devoid of conventional architecture, presumably to reset the basal chromatin state for the parental-to-embryonic transition. This organization is progressively rebuilt in early development, a conserved process in many species, including human, mouse, pig, fruit fly, zebrafish, and so on. Despite some degree of ZGA having been observed in arrested human embryos¹³, the epigenetic reprogramming features and their potential contribution to idiopathic embryonic arrest remain unclear.

The advent of Hi-C and related chromatin conformation capture technologies has revolutionized our understanding of 3D genome organization during embryogenesis. These approaches have identified topologically associating domains (TADs) as fundamental units that compartmentalize the genome, facilitating local gene-regulatory interactions while insulating neighboring domains via boundary elements¹⁴. TADs are believed to form through a loop-extrusion process and further organize into higher-order A/B compartments, which correlate with functional genomic features. During ZGA, chromatin accessibility undergoes substantial reorganization, characterized by the loss of a large fraction of pre-existing distal open regions coupled with the establishment of novel accessible sites at putative regulatory elements¹⁵. A recent single-cell multi-omics study compared normally developing and arrested human embryos from the zygote to the 8-cell stage¹⁶, but the landscape of 3D chromatin architecture remains uncharacterized. What’s more, the relationship between transcriptional and 3D genome reorganization remains controversial, with current evidence revealing both cooperative and independent dynamics. For instance, a hypertranscription state in early embryos and the 3D genome architecture engage in bidirectional reinforcement, demonstrating a coregulatory relationship¹⁷, but some studies have revealed that transcriptional suppression during early embryogenesis exerts minimal impact on the establishment and maintenance of TADs^18,19. These seemingly contradictory observations underscore the unresolved complexity of chromatin transcriptome interplay in early mammalian embryos²⁰. A clear understanding of both the structural and transcriptional dimensions is essential to decipher the mechanisms underlying embryonic arrest during this critical developmental window.

In this study, we applied multi-omics sequencing to systematically compare the transcriptomic and 3D genome architectures between arrested and normal human embryos at the 8-cell stage. Our analysis particularly reveals coordinated dysregulation of chromatin remodeling and transcription during embryonic arrest, underscoring the essential role of 3D genome organization in early development. By integrating chromatin interaction data with additional transcriptomes, we observed candidate transcription factors potentially linked to developmental failure. The chromatin conformation maps generated here provide a valuable resource for embryology research. This resource may inform future hypotheses aimed at improving the assessment of developmental potential in IVF embryos, although prospective validation will be required before clinical translation.

Results

Comprehensive overview of the interplay of the 3D genome and transcriptome in human normal and arrested 8-cell embryos

Here, we focused on human embryos at the ZGA stage, comparing normally developing embryos with those that had arrested at the 8-cell stage for unexplained reasons. And then we performed multi-omics sequencing, including Hi-C and RNA-seq, to comprehensively profile the chromatin structure and transcriptome for further bioinformatics analysis (Fig. 1a). Following stringent morphological quality control, we obtained a total of 34 normal and 37 arrested 8-cell embryos (Fig. 1b, Supplementary Fig. 1a), with arrested embryos defined as those that reached the 8-cell stage by day 3 post-fertilization but exhibited no further cell division for 2 days while retaining normal morphology. These embryos were then randomly assigned to independent groups for separate Hi-C and RNA-seq sequencing, enabling comprehensive profiling of chromatin architecture and transcriptome in subsequent bioinformatics analyses. Detailed sample and data information is provided in Supplementary Tables 1 and 2. We obtained a resolution of 40 kb for 8-cell embryos after in-depth sequencing, which enabled us to explore the 3D genome architecture at multi-scale, including compartments, TADs, and chromatin loops. High concordance was observed between replicates in normal and arrested 8-cell embryos in both Hi-C data and RNA-seq data (Supplementary Fig. 1b, c). CNV profiling showed a low fraction of genome altered (FGA = 0.05) in arrested embryos (Supplementary Fig. 1d), suggesting that the small-scale aneuploidy may not contribute to developmental arrest in this cohort. To model chromatin conformation, we employed the PASTIS software²¹ for 3D genome reconstruction. Our analysis reveals that in 8-cell human embryos, each chromosome is spatially confined within discrete chromosomal territories (Fig. 1c), consistent with the organizational pattern observed in other mammalian genomes²². We further analyzed the average intra-chromosomal contact probability on genomic distance and generated a corresponding visualization. As expected, the contact probability decreases monotonically as the genomic distance increases from 0.1 Mb to 100 Mb for these two groups (Fig. 1d). Then, we analyzed the percentages of long-distance interactions between inter-chromosomal and intra-chromosomal contacts, as shown in Fig.1e, and found that in arrested embryos, the genomic interaction landscape was markedly altered, manifested by a genome-wide decrease in intra-chromosomal contacts and a concomitant increase in inter-chromosomal interactions. To further quantify this shift, we analyzed the distribution of intra-chromosomal interactions, finding a significant reduction in the proportion of cis-short (<2 Mb) contacts in arrested embryos (Fig. 1f). This indicated that normal embryos maintain compact chromatin architecture with predominant short-range interactions, whereas arrested embryos exhibited a shift toward long-range interactions. We speculate that this shift may represent a manifestation of 3D genome decompaction²³, a possibility that warrants further investigation.

Fig. 1. Comprehensive overview of the interplay of 3D genome and transcriptome in normally developing and arrested 8-cell embryos.

a Schematic illustration of the study design. Human oocytes were isolated from females and fertilized in vitro to get embryos. Embryos and arrested 8-cell embryos were cultured in vitro for various times before collection. Normal 8-cell embryos of the control groups were collected at day 3 after fertilization, and arrested 8-cell embryos (collected on day 5) were defined as those that reached the 8-cell stage by day 3 but showed no further development while maintaining normal morphology, with no signs of degeneration. And then Hi-C and RNA-seq data were generated from independent groups of embryos and subsequently integrated for bioinformatics analysis. b Representative images of morphologically normal and 8-cell stage-arrested embryos. Scale bars = 50 µm. c Reconstructed 3D genomes based on the beads-on-a-string model. Each bead is 500 bp, and each colored string represents one chromosome. d Contact probability relative to genomic distance. e. Percentage of inter-chromosomal and intra-chromosomal contacts between normal and arrested embryos. f Relative proportion of interaction counts plotted against genomic distance for normal (blue) and arrested (orange) embryos. g Volcano plot showing differentially expressed genes between normal and arrested 8-cell embryos. Significantly higher expressed genes in the arrested group (n = 2164) are shown in red, significantly higher expressed genes in the normal group (n = 2726) in blue, and non-significant genes in gray. Differential expression analysis was performed using limma, with p values adjusted by the Benjamini–Hochberg correction. h Box plots showing the average expression levels of ZGA and zygotic-decay genes in normal and arrested 8-cell embryos. Statistical significance was assessed using the two-sided Wilcoxon rank-sum test. i The plot of z-normalized similarity score of Hi-C data generated from normal and arrested 8-cell embryos. Each dot represents a 4000-kb window. Red dots are significantly dissimilar regions (z-normalized similarity score of ≤−1.2), while dark gray dots and light gray dots are not-significantly dissimilar regions (z-normalized similarity score of >−1.2) and no-score value regions, respectively. j Box plots showing ZGA gene density (number of ZGA genes per 4000-kb window) in structurally stable and variable regions identified by SSIM analysis in (i). significantly different regions exhibited significantly higher ZGA gene density compared to stable regions. Statistical significance was assessed using the two-sided Wilcoxon rank-sum test.

To further characterize the transcriptional landscape associated with developmental arrest, we first performed differential expression analysis between normal and arrested embryos. A total of 2164 genes showed significantly higher abundance in arrested embryos compared to normal controls, whereas 2726 genes showed significantly lower abundance (Fig. 1g). Differentially expressed genes (DEGs) between normal and arrested embryos with lower expression in arrested embryos were enriched for both ZGA genes and maternal genes (Supplementary Fig. 1e, f). Previous studies have reported that failed ZGA²⁴ and delayed clearance of maternal RNAs occur with high frequency in arrested human embryos⁷. Maternal transcripts can be cleared through maternal decay (M-decay) and ZGA-dependent zygotic decay (Z-decay) (Supplementary Fig. 1g). We therefore examined whether zygotic decay and ZGA are affected in arrested 8-cell embryos. ZGA genes showed significantly decreased expression in arrested embryos, whereas Z-decay genes exhibited no significant change (Fig. 1h). These findings suggest that ZGA activation is disorganized in arrested embryos, whereas the expression of Z-decay genes remains relatively unchanged, raising the possibility that the two processes may be differentially affected.

To precisely delineate regions of structural divergence in normal and arrested embryos, we computed similarity scores from Hi-C data and identified 341 genomic regions across all 23 chromosomes that exhibited high-confidence structural variations (z-normalized score of ≤−1.2 and signal-to-noise ratio of ≥0.6). We termed these red-dotted regions as “dissimilar” regions, specifically referring to areas exhibiting highly dissimilar chromatin spatial interaction patterns between the two embryonic states (Fig. 1i). ZGA gene density was significantly higher in dissimilar regions compared with stable regions (Fig. 1j). We next sought to investigate whether alterations in chromatin conformation play a role in embryo arrest. If so, genes located within regions exhibiting significant structural changes should demonstrate more pronounced expression alterations compared to those in structurally stable regions. Supporting this, genes located within these dissimilar regions showed significantly more pronounced expression changes compared to those in structurally conserved regions (Supplementary Fig. 1h). Furthermore, Gene Ontology (GO) enrichment analysis of genes within these regions revealed a critical failure in the post-transcriptional regulatory network in embryo arrest, as evidenced by the significant enrichment of pathways involved in pre-mRNA splicing, RNA subcellular localization, and ribosome biogenesis (Supplementary Fig. 1i).

Collectively, our findings indicate that developmental arrest at the 8-cell stage co-occurs with global 3D genome reorganization and transcriptomic dysregulation, particularly in key RNA-related processes.

Compartment rewiring and consequent transcriptome reprogramming in human embryonic arrest

The crucial hierarchical feature of chromatin organization is A/B compartmentalization, which undergoes dynamic switches during embryogenesis as reflected by compartmentalization. In this study, we identified widespread compartment switching across all chromosomes in developmentally arrested embryos compared with normal controls (Fig. 2a). The genomic distribution and proportional changes in compartment identity between normal and arrested embryos are shown in Fig. 2b, c, and were further validated by downsampling analysis (Supplementary Fig. 2). Compartments labeled “A to A” and “B to B” represent compartments that remained stable under both conditions, whereas “A to B” signified a shift from active in normal embryos to inactive in arrested ones, and “B to A” indicated a transition from inactive to active states in arrested embryos. And comparative analysis of A/B compartment patterns and compartment strength demonstrated that compartmentalization was already relatively indistinct in normal embryos but became further impaired in developmentally arrested human embryos (Fig. 2d, e).

Fig. 2. Rewiring of 3D genome compartmentalization and transcriptional programs during embryonic arrest.

a Comparative analysis of switched compartments across 22 chromosomes between normal and arrested embryos. b Sankey plot showing the dynamic switch of the A/B compartment status. c Percentage of switched A/B compartments status in the normal and arrested groups. d Saddle plots showing that the (A + A) and (B + B) compartment strengths were dynamically changed between the normal and arrested 8-cell embryos. e Compartment strength between the compartments of the normal and arrested embryos. f Individual switching clusters with PC1 and gene expression changes. Genes were grouped by the compartment switching pattern of their resident regions (B-to-A or A-to-B). Shown are genes with log₂ (fold change) (arrested/normal) >0 in the B-to-A group and genes with log₂ (fold change) (arrested/normal) < 0 in the A-to-B group. g Venn diagrams showing the number of genes overlapping among DEGs between normal and arrested embryos, ZGA genes, and genes within A-to-B switching regions and B-to-A switching regions. The numbers in each section represent gene counts. h Gene Ontology enrichment analysis of genes with log₂ (fold change) (arrested/normal) < 0 in A-to-B (389 genes) and genes with log₂ (fold change) (arrested/normal) >0 in B-to-A (485 genes) from (f).

To explore the functional implications of these structural changes, we investigated the correlation between compartment switching and transcriptional alterations. As shown in Fig. 2f, we identified two distinct clusters exhibiting concordant patterns of gene expression changes and compartment switching: A-B conversions associated with reduced abundance genes (389 genes) and B-A conversions with increased abundance genes (485 genes) in the arrested group (Supplementary Data 4). The overlap of these two gene sets with DEGs between normal and arrested embryos and ZGA genes is shown in Fig. 2g. Furthermore, pathway enrichment analysis of the A-B and B-A clusters revealed that processes such as RNA localization, ribosome biogenesis, and ribonucleoprotein complex assembly were significantly reduced in arrested embryos. The upregulated pathways encompassed differentiation pathways, such as limb and appendage morphogenesis (Fig. 2h). In contrast to the essential housekeeping processes, these programs are typically activated in later embryogenesis to direct the formation of specific organs. This pattern indicates that one of the problems in arrested embryos may be a loss of developmental timing, rather than a simple growth cessation.

Taken together, these findings demonstrate that arrested embryos display dramatic alterations in chromatin compartmentalization and extensive transcriptomic dysregulation.

Structural disorganization of TADs and higher-order TAD clusters in arrested embryos

To investigate the dynamics of TADs during embryonic arrest, we mapped and compared high-order chromatin architectures across key stages of early development, including the 4-cell (minor ZGA), 8-cell (major ZGA), and blastocyst (post-ZGA) stages, comparing normal and arrested embryos. The characteristic triangular TAD topology, absent in 4-cell embryos, emerged faintly in both normal and arrested 8-cell embryos and became sharply observed in blastocyst stage. Consistent with this pattern, TAD signal intensity progressively increased throughout early human embryogenesis, with a pronounced surge from the 8-cell to blastocyst stage (Fig. 3a).

Fig. 3. Comparative analysis reveals the differences of TAD and TAD clusters architecture in arrested embryos.

a Snapshot of the interaction heatmaps for high-order chromatin structures, directional index (DI), and insulation score (IS) in human embryos, including 4-cell, 8-cell, arrested 8-cell, and inner cell mass (ICM) at a 40-kb resolution. Hi-C data of 4-cell and ICM were from public data reanalyzed⁶⁵. And all samples (public and in-house) originated from the same clinical cohort, underwent the same standardized experimental protocols, and were sequenced in the same batch. b Number of TADs and c TAD boundaries in normal and arrested embryos. d TAD scores and e insulation scores in normal and arrested embryos. f Box plots comparing the ZGA gene expression levels (log₂(FPKM + 1)) at variable TAD boundaries between normal and arrested embryos. Statistical significance was assessed using the two-sided Wilcoxon rank-sum test. g Box plots showing the TE counts per variable TAD boundary in normal and arrested embryos. TE annotations were retrieved from the UCSC Genome Browser Database. Statistical significance was assessed using the two‑sided Wilcoxon rank‑sum test. h The number of TADs per chromosome is strongly correlated with the size of the chromosome. (normal r = 0.414, p value = 0.494, arrested r = 0.383, p value = 0.0713). i TAD interaction networks in normal and arrested embryos (Orange circle: high-degree nodes). j Sanky plot of TAD Clusters of interacting TADs and non-interacting TADs transition from normal to arrested embryos.

Next, we systematically evaluated and compared the fundamental characteristics of TADs in between normal and arrested embryos. By stratifying TAD regions into variable and invariant categories based on intergroup comparisons, we found that the majority of TADs (∼98.6%, n = 2191) were shared between normal and arrested embryos (invariant TADs, Fig. 3b). In contrast, TAD boundaries exhibited considerable variability: 479 of 1,343 (∼36%) boundaries were specific to normal embryos, while 582 (∼43%) were specific to arrested embryos (Fig. 3c). These results indicate that TAD domains are substantially more conserved than TAD boundaries.

Notably, the TAD domain score, reflecting intra-TAD interaction intensity, was significantly elevated in arrested embryos within invariant regions (Fig. 3d). This enhancement was further corroborated by regional analysis in arrested embryos (Supplementary Fig. 3a). TAD boundaries play a critical role in 3D genome organization by constraining enhancer-promoter interactions within domains, thereby preventing gene dysregulation or diseases^25,26. We therefore assessed the genomic features associated with variable TAD boundaries. Quantitative assessment of boundary strength revealed statistically significant weakening in developmentally arrested embryos within variable regions, though the effect size was modest (Fig. 3e). Additionally, ZGA gene density showed no significant difference between the arrested and normal groups (Supplementary Fig. 3b), whereas ZGA gene expression levels were significantly lower in arrested embryos (Fig. 3f). Transposable elements (TEs) and TE-derived sequences affect 3D genome architecture at multiple levels. And TEs represent a substantial proportion of boundary elements that delimit TADs and sub-TADs, contributing to boundary diversity²⁷. We found that variable TAD boundaries in arrested embryos exhibited significantly higher TEs density (Fig. 3g). Collectively, in arrested embryos, variable TAD boundaries are characterized by both aberrant ZGA gene expression and significant enrichment of TEs. What’s more, TAD sizes were significantly larger in arrested embryos compared to normal controls (Supplementary Fig. 3c). And consistent with theoretical expectations, our analysis revealed a positive correlation between chromosomal size and TAD number in both groups (Fig. 3h). TAD cliques are known to associate with key organizational processes, such as stabilizing heterochromatin at the nuclear periphery through lamina-associated domains during cell differentiation²⁸. We therefore examined TAD-TAD interaction patterns and found that the identity of highly interconnected hub TADs was substantially altered in arrested embryos, indicating a reorganization of TAD clustering (Fig. 3i). A Sankey diagram further revealed profound alterations in the size distribution of TAD clusters in arrested embryos (Fig. 3j).

Collectively, these results establish that the structural rewiring of TADs and their higher-order clusters is a key feature of arrested embryos.

Integrative analysis of TAD-associated transcriptomes and ZGA dysregulation identifies cellular energy metabolism underlying embryonic arrest

To elucidate the functional relationship between TAD architecture and transcriptional regulation, we analyzed DEGs and ZGA genes in normal versus arrested 8-cell embryos. We first identified group-specific genes and gene groups with different spatial distribution relative to TADs (Fig. 4a). Expression analysis revealed that ZGA-related genes were significantly lower in arrested embryos, both within TADs and TAD boundaries (p < 0.01, Wilcoxon rank-sum test; Fig. 4b), indicating a pronounced disruption in ZGA. Through sequential intersection of ZGA genes, DEGs between normal and arrested embryos, and condition-specific TAD and TAD boundary genes, we generated four non-overlapping gene sets (Groups 1–4) for subsequent functional analysis (Supplementary Fig. 4a, Supplementary Data 5). Further GO enrichment analysis of these genes revealed both shared and exclusive patterns between normal and arrested embryos (Fig. 4c). Ribosome-related pathways and RNA processing were central to ZGA in both groups, which is consistent with prior studies²⁹. However, “ribonucleoprotein complex biogenesis” was enriched among ZGA genes within TADs in normal embryos but shifted to boundary-associated ZGA genes in arrested embryos. Similarly, “RNA metabolic process” and “RNA biosynthetic process” were exclusively enriched in boundary-located ZGA genes in the arrested group, suggesting that mislocalization of ZGA genes may contribute to developmental arrest. We also observed distinct metabolic signatures: TAD boundary genes in normal embryos were enriched in “glutamine catabolic process”, whereas arrested embryos showed enrichment for “glutamate biosynthetic process”. Given that glutamine serves as a TCA cycle substrate³⁰ and enhances development to the 8-cell stage³¹, the lower expression of GLUD1 and GLS in arrested embryos (Supplementary Fig. 4b) may be indicative of metabolic imbalances that contribute to developmental arrest^32,33. In arrested embryos, both genes were located within chromatin loops (Supplementary Fig. 4b) and GLS mapped to regions undergoing A-to-B compartment switching, whereas GLUD1 remained in the A compartment. These results reveal that profound metabolic dysregulation represents a hallmark feature of developmental arrest.

Fig. 4. Integrative analysis of TAD-associated transcriptomes identifies potential regulators underlying ZGA dysregulation in embryonic arrest.

a UpSetR plot illustrates the number of genes identified across distinct groups. Horizontal bar chart shows the gene numbers of the groups including genes involved in ZGA, DEGs between normal and arrested embryos, as well as genes located within TADs and at TAD boundaries in both normal and arrested embryos. And the column height of vertical bar chart depends on how many groups are selected as shown by color circles. b Violin plot of gene expression in TADs and TAD boundaries of normal and arrested embryos. c Dot plot illustration of Gene Ontology (GO) pathway analysis for the four protein groups: specifically, and significantly differentially expressed proteins localized to inner-TAD and TAD boundary regions in both the normal and arrested groups. GO analysis was performed on Gene ontology online website (https://www.geneontology.org/) and enrichment scores were calculated automatically. The size of a dot represents the number of genes associated with the GO term, and the color of dots represents the adjusted p value. p < 0.05 was considered statistically significant. d Scatter plot of the distribution of gene numbers within TADs of varying sizes is shown for the normal and arrested groups. Points are colored by their ZGA gene content: yellow (1–9 ZGA genes), pink (≥10 ZGA genes), and blue (no ZGA genes). e Heatmap of differentially expressed ZGA genes (FDR < 0.05) located in the TAD highlighted by the green box in (d). Expression levels are shown as log₂ (fold change (arrested vs. normal), with red indicating higher expression in arrested embryos and blue indicating lower expression.

To further explore the relationship between genes, especially ZGA genes, and TAD structures, we examined their distribution and found a low correlation between TAD size and gene number in both normal and arrested embryos (r = 0.025 in normal, r = 0.027 in arrested, Fig. 4d), while also observing that TAD boundary size was not correlated with the number of associated genes (Supplementary Fig. 4c). ZGA genes were broadly distributed, present in 82.71% of TADs in normal embryos (2.61% containing >10 genes) compared to 77.56% in arrested embryos (2.54% containing >10 genes). Then we systematically examined TADs enriched in ZGA genes (Fig. 4d) and selected the TAD with the most ZGA genes (green box) for further analysis. Overlap analysis of ZGA genes within these two TADs revealed that 7 ZGA genes were present only in the normal group (Supplementary Fig. 4d, Supplementary Data 6). Notably, SCO1 encodes a key regulator of mitochondrial copper homeostasis and energy production^34,35, and copper homeostasis is critical for embryogenesis³⁶. SCO1 was significantly lower in arrested embryos, making it an interesting candidate for further functional investigation. Consistent with the widespread ZGA dysregulation we observed earlier (Supplementary Fig. 1e, f), the majority of ZGA genes within the selected TAD also showed lower expression in arrested embryos (Fig. 4e).

Linking chromatin loop disruption to ZGA dysregulation in embryonic arrest with highlights on key transcriptional factors

Chromatin loops, defined by the spatial proximity of distal loci, are universal structural-functional units of the 3D genome³⁷. To investigate potential structural anomalies in chromatin looping, we performed a quantitative comparison of TADs and chromatin loops across all chromosomes in normal versus arrested embryos. Our analysis revealed that TAD numbers remained largely stable, while chromatin loops varied markedly in several chromosomes, such as chromosomes 1 and 2 (Supplementary Fig. 5a, b). Notably, on chromosome 14, a reduction in TAD number in arrested embryos coincided with an increase in chromatin loops. To explore this further, we analyzed loop distribution across TADs and observed a pronounced increase in loop density within specific TAD regions in the arrested group, as highlighted by green boxes in Fig. 5a. Visualization of the corresponding region on chromosome 14 confirmed a marked intensification of loop interactions in arrested embryos (Supplementary Fig. 5c).

Fig. 5. Linking chromatin loop disruption to ZGA dysregulation in embryonic arrest with highlights on key transcriptional factors.

a Visualization of the distribution of chromatin loops in TADs across various chromosomes. The x-axis represents the genomic position in megabases (Mb) across different chromosomes, and the y-axis represents the number of loops per TAD. The plot shows data for two groups, normal and arrested, with the different groups indicated by the colors of the data points. b Scatter plot showing the relationship between loop contact frequency and gene expression. The y-axis represents the log₂(fold change) of chromatin loop contact frequency, and the x-axis represents the log₂(fold change) of gene expression for the corresponding genes. A weak but statistically significant positive correlation was observed (Pearson’s R = 0.1, p = 0.0078). c Bubble plot of differentially expressed transcription factors with enriched motifs. Anchor sequences of chromatin loops to both normal and arrested embryos were extracted. Motif enrichment was evaluated using HOMER (v4.11). Transcription factor motifs with p-adjust values < 0.05 in either condition were included in subsequent analyses. Bubble size corresponds to the enrichment significance (−log₁₀ p value), and color represents the TF’s differential expression level (red, higher expression in arrested group; blue, lower expression in arrested versus normal embryos). And the source data is in Supplementary Data 7. d Venn diagram illustrating the intersection of three gene sets: DEGs between normal and arrested embryos, human ZGA genes, and JASPAR-derived human transcription factors. The 56 overlapping genes and genes extracted from the literature. were defined as candidate ZGA-associated TFs, and their expression patterns are displayed in (e). e A comparative visualization of the 56 genes overlapping set from the central core in (d), shared among TFs, DEGs between normal and arrested embryos, and ZGA groups, and genes extracted from the literature. Circle size corresponds to the magnitude of the fold change, while color represents expression level (FPKM). f Protein–Protein interaction (PPI) network analysis for the candidate TFs using the STRING database. Proteins are clustered and color-coded based on interaction evidence. Focused PPI subnetworks for three key TFs, ZNF281, ZNF597, and ZNF622, of interest. Connecting lines represent predicted or known functional associations from the STRING database.

Prior research suggests that even subtle variations in chromatin loop contact frequency can influence gene expression and histone modifications such as H3K27ac³⁸. We therefore examined whether loop contact frequency correlated with transcriptional changes. A modest but statistically significant correlation (r = 0.1, p = 0.0078) was observed between gene expression fold change and loop contact frequency (Fig. 5b). Since cohesin and condensin complexes are essential for chromosome folding³⁹, we assessed their expression. Most cohesin subunit genes, including SMC2, SMC4, RAD21, STAG2, CTCF, and so on, were significantly reduced in arrested embryos (Supplementary Fig. 5d). This is consistent with reports that CTCF depletion leads to altered TAD sizes⁴⁰, mirroring our observation of enlarged TADs in arrested embryos (Supplementary Fig. 3c). Together, these results indicate that embryo arrest involves extensive reorganization of the chromatin loop interactome and may be associated with transcriptional dysregulation.

To link structural alterations to regulatory function and discover candidate transcriptional factors (TFs), we integrated TF-motif enrichment of chromatin loops with transcriptomic analysis (Fig. 5c). We observed several transcription factor motifs significantly enriched in either normal or arrested embryos, with bubble size reflecting enrichment significance and color indicating the differential expression levels of the corresponding transcription factors. Given that ZGA involves coordinated action of multiple transcription factors, we compared normal and arrested 8-cell embryos to pinpoint potential ZGA regulators. We took the intersection of (i) DEGs between normal and arrested embryos, (ii) human ZGA gene set, and (iii) human transcription factors from the JASPAR ChIP-seq database⁴¹ (Fig. 5d). Expression profiles of candidate regulators are shown in Fig. 5e. Among these candidates, KLF17 emerged as a potential regulator through two independent analytical criteria: its expression was significantly lower in the arrested embryos, and its binding motifs were significantly enriched (Fig. 5c). KLF17 is known to be upregulated during the 8-cell stage in humans⁴², and in mice, it facilitates ZGA by recruiting RNA polymerase II⁴³. Although a similar mechanism has not been confirmed in humans, its protein interaction network suggests involvement with central regulatory partners (Fig. 5f). Based on these findings, we speculate that reduced KLF17 abundance may be associated with altered 3D genome organization, potentially involved in developmental arrest.

Other critical transcription factors, including TPRX family members (TPRX1, TPRX2)¹⁰, ZSCAN5B⁴⁴, KLF4, MYC, and YY1¹⁶, were also expressed at significantly lower expression levels in arrested embryos (Fig. 5e). We also noted substantial representation of zinc-finger protein (ZNF) family genes, a class of versatile DNA/RNA-binding regulators controlling transcription, DNA repair, and signal transduction⁴⁵. Two ZNFs, ZNF143, which collaborate with CTCF/cohesin in loop formation⁴⁶, and ZNF675, essential for proper ZGA activation, were present at significantly lower levels in arrested embryos⁴⁷. To uncover potential ZNF regulators, we constructed a protein–protein interaction (PPI) network from the 56 candidate TFs and genes extracted from the literature using the STRING database⁴⁸. This analysis highlighted ZNF281, ZNF597, and ZNF622 as potentially prominent genes among TF ZNFs. Further examination of their interaction partners provided initial clues to their potential roles in early development (Fig. 5f). Among them, ZNF281 is a known pluripotency factor in embryonic stem cells⁴⁹, though its role in human ZGA remains unexplored.

In summary, arrested embryos display chromatin loop reorganization. We hypothesize that reduced KLF17 abundance and its specific motif enrichment indicate its involvement in developmental arrest, and that reduced expression of several ZNF family members may contribute to ZGA dysregulation. However, the proposed functions of these TFs require further experimental validation.

Discussion

With infertility affecting 12.6–17.5% of couples globally and projected to become the third most significant health challenge of the twenty-first century⁵⁰, IVF has emerged as a critical treatment despite its limited success per cycle⁵¹. Mammalian preimplantation embryogenesis involves a sequence of critical events, fertilization, ZGA, and lineage segregation, that are tightly regulated both transcriptionally and epigenetically. Although single-cell and low-input omics have recently illuminated the molecular landscape of early development, the functional role of 3D genome architecture in embryonic arrest at the ZGA stage remains poorly understood.

Our study addresses this gap through an integrated analysis of the 3D chromatin architecture and transcriptomes of developmentally arrested 8-cell human embryos. Our findings indicate that arrested embryos exhibit multi-scale rewiring of the chromatin landscape and ZGA transcriptional dysregulation. At the compartment level, we observed shifts in distal-to-proximal interaction patterns and dynamic A/B compartment transitions. At the TAD level, arrested embryos exhibited structural alterations in TAD features and changes in the transcriptional activity of major ZGA genes. Analysis of integrated 3D genome and transcriptomic data showed premature activation of late-stage developmental pathways and dysregulation of energy metabolism-related genes in arrested embryos. We propose that reduced KLF17 abundance and arrest-specific motif enrichment contribute to arrested development, and that reduced expression of several ZNF factors, as suggested by PPI analysis, may be linked to ZGA dysregulation. Unlike earlier studies reliant on heterogeneous public datasets, our work offers a de novo, matched multi-omics resource derived from a unique collection of human embryos, enabling insights with high internal consistency.

While our Hi-C data reveal widespread alterations in chromatin organization in arrested embryos, the underlying mechanisms remain to be elucidated. Interpretation of these findings should also consider that selecting normal 8-cell controls based solely on morphology cannot entirely exclude embryos that might arrest upon extended culture, as many causes of developmental arrest are not morphologically detectable, a challenge also faced in IVF embryo selection. Conventional light microscopy has an intrinsic resolution limit⁵², rendering it incapable of reporting on dynamic molecular processes, such as defects in 3D chromatin architecture, which is central to this study, are entirely invisible to morphological inspection. Multiple cellular mechanisms have been implicated in this developmental failure, including epigenetic dysregulation^53,54, chromosomal aneuploidy⁵⁵, and single gene disorders⁵⁶. Aberrant chromosomal compaction⁵⁷, cell-cycle or mitotic defects⁵⁸, nuclear collapse or early apoptotic changes⁵⁹ may lead to genomic instability and subsequently contribute to embryonic developmental arrest. Distinguishing between these possibilities would require complementary approaches. While our findings reveal a total of approximately 50% of the genome undergoes compartment switching (including A-to-B and B-to-A transitions), to our knowledge, such dramatic compartment changes have primarily been associated with differentiation^60,61, reprogramming⁶². However, other stimuli, such as DNA damage⁶⁰ or loss of CTCF/cohesin⁶³, induce only minor or no effects⁶⁴. Therefore, we speculate that the extensive chromatin reorganization observed in arrested embryos is unlikely to result from a single, simple cause. Instead, it may reflect the convergence of multiple complex and interrelated factors, warranting further investigation into the multifactorial nature of human embryonic arrest.

Our findings lay a foundation for understanding embryonic arrest through 3D genome dysregulation, yet several limitations highlight important directions for future research. A limitation of this study is the inability to directly correlate chromatin structure and gene expression from the same single embryo. Our results revealed population-level defects in 3D genome reprogramming and the ZGA process. Due to inherent molecular heterogeneity among individual embryos, combining Hi-C data with transcriptomic data from different embryo groups may fail to accurately represent the true molecular landscape of a single embryo under a specific condition. But stringent experimental design and careful sample selection ensured high-quality and reproducible data, as evidenced by the strong consistency across biological replicates in both normal and arrested groups. We agree that the future application of the single-cell multi-omics sequencing approach will be invaluable to dissect intra-embryo heterogeneity and further causally link genome folding to transcriptional failure during ZGA. What’s more, given the complexity of genome architecture during embryogenesis, integrating additional epigenomic profiles (e.g., H3K27ac or ChIP-seq data) would facilitate a more comprehensive understanding of the mechanisms underlying embryonic arrest.

In summary, this study provides a comparative map of 3D genome reorganization during developmental arrest at the crucial ZGA stage. The chromatin interaction datasets generated here represent a valuable resource for deciphering the regulatory blueprint of human embryogenesis. And it may help guide strategies for evaluating embryo developmental potential in IVF, pending prospective validation to establish clinical application.

Methods

Ethics statement

This study was approved by the institutional review boards of The Affiliated Ren Ji Hospital of Shanghai Jiao Tong University (2,019,122,701). The study was conducted in accordance with the measures of the People’s Republic of China on the administration of human-assisted reproductive technology, the ethical principles of human-assisted reproductive technology, and the Helsinki declaration.

All early human embryos used in this study were obtained with written informed consent from the donors. The embryos were sourced from Shanghai Jiao Tong University Affiliated Ren Ji Hospital. All ethical regulations relevant to human research participants were followed.

Human 8-cell embryos collection

All early human embryos utilized in this study exhibited normal developmental progression and morphological integrity. To prevent potential contamination from cumulus cells or sperm, all two-pronuclei (2PN) 8-cell embryo samples were obtained through Intracytoplasmic Sperm Injection (ICSI) rather than conventional IVF. Fertilization was confirmed 16–19 h post-ICSI, with embryo cleavage status documented at 24-h intervals. Both normally developing and arrested 8-cell embryos selected for analysis displayed high-quality morphology. To ensure sample consistency, all donated embryos were vitrified in liquid nitrogen for storage and subsequent batch processing. Prior to Hi-C library preparation, the zona pellucida was uniformly removed using acid Tyrode’s solution.

In this study, 34 normal 8-cell embryos were acquired from 20 donors, and 37 arrested 8-cell embryos were acquired from 37 donors (aged 30–45 years old, Han ethnicity, female) with tubal-factor infertility following written informed consent. Normal 8-cell-stage embryos (8A1) were defined as day 3 embryos exhibiting eight blastomeres of uniform size and absence of cytoplasmic fragments. Arrested human 8-cell embryos were defined as those that developed to the 8-cell stage by day 3 post-fertilization but remained arrested at this stage for the following 2 days, maintaining the 8-cell morphology on day 4 and day 5, while normally developing embryos progressed to the blastocyst stage by day 5.

In situ Hi-C library preparation and sequencing

Generation of Hi-C libraries with a low cell number of embryos was performed according to modified protocols we published before⁶⁵. Embryos were removed from the zona pellucida with acidic Tyrode’s solution (Sigma) before fixation in 1% formaldehyde (10 min, room temperature), and then were quenched with glycine. Subsequently, cells were pelleted (3000 × g, 5 min, 4 °C) and lysed in ice-cold Hi-C lysis buffer (≥30 min). Chromatin was digested overnight with DpnII (NEB) at 37 °C after SDS/Triton X-100 treatment. Restriction ends were filled with biotin-14-dATP (Life Technologies) and DNA Polymerase I (NEB; 24 °C, 4 h), followed by proximity ligation with T4 DNA ligase (room temperature, 4 h; then 4 °C, 16 h, both with low-speed rotation). After proteinase K digestion (55 °C, 30 min), DNA was precipitated using DNA-Be-Down (Sangon Biotech, B544732-0001), 2.5 × volumes of pure ethanol and 0.23 × volumes of 3 M sodium acetate, pH 5.2.

After mixing by inversion and incubation at −80 °C for 15 min, the sample was centrifuged at 13,000 rpm for 15 min at 4 °C. The pellet obtained was washed twice with cold 75% ethanol and dissolved in 43 μl of 1× Tris buffer (10 mM Tris-HCl, pH 8). Subsequent steps included end repair, A-tailing, adapter ligation (Vazyme ND607-01), USER enzyme (NEB) treatment to open adapters, and streptavidin bead capture (MyOne T1 Dynabeads, Life Technologies). Final libraries were amplified (15 cycles) with indexed primers (NEB E7335S) and VAHTS HiFi mix, then purified with AMPure XP beads for sequencing.

RNA-seq library preparation and sequencing

RNA sequencing (RNA-seq) was performed using the Smart-seq protocol⁶⁶. In brief, lysates from human embryo samples were subjected to reverse transcription with oligo-dT primers targeting polyadenylated (poly-A) tails of mRNAs, enabling selective cDNA synthesis. The resulting cDNA was preamplified with KAPA HiFi HotStart polymerase and purified. Purified amplified cDNA was then used for library preparation, which included a single-strand circularization step to produce single-stranded circular DNA libraries containing >200 DNA nanoball copies. Final sequencing was carried out on the BGISEQ500 platform.

RNA-seq data analysis

Clean reads were produced by BBmap (version 38.16) and then mapped to the human reference genome (hg19) with TopHat2 (version 2.1.1) and Cufflinks (version 2.2.1) to measure the transcript abundance in fragments per kilobase per million (FPKM) as previously described^67,68. DEGs between normal and arrested embryos were identified using the limma package based on FDR-adjusted multiple t-test (FDR < 0.05 and |log₂(fold change)| ≥ 2). GO analysis of DEGs was performed using the R package ClusterProfiler with a threshold FDR < 0.05 and was visualized using the R package ggplot2⁶⁹.

Assessment of biological replicate reproducibility

To evaluate the reproducibility between biological replicates for both the normal and arrested groups, we assessed consistency at both chromatin structure and transcriptome levels. For Hi-C data, reproducibility was assessed using HiCExplorer (version 3.7.6). Specifically, the hicCorrelate tool was employed to calculate Pearson correlation coefficients and generate scatter plots at a 2 Mb resolution. For RNA-seq data, biological reproducibility was assessed by calculating Pearson correlation coefficients between replicates for both groups. FPKM were log₂-transformed (log₂(FPKM + 1)). Consistency was visualized using scatter plots with points color-coded by Gaussian kernel density estimation.

Identification of major ZGA genes

For human major ZGA genes, we identified significant upregulated genes with FDR < 0.01 and |log₂(fold change)| > 2 in human normal 8-cell embryos compared with human normal 4-cell embryos⁷⁰.

Identification of maternal genes

To define maternal mRNA degradation dynamics during human MZT, we analyzed published single-cell RNA-seq data of GV oocytes following the criteria established by Sha et al.⁷. Briefly, genes with FPKM > 2 in GV oocytes and reliable sequence annotations were selected as maternal transcripts. The maternal genes list is provided in Supplementary Data 2.

Hi-C data processing

Hi-C data were processed as previous described^65,71,72. Briefly, raw reads were trimmed with Cutadapt (version 1.0) and BBmap (version 38.16). Then, clean reads were aligned, processed, and iteratively corrected using HiC-Pro (version 2.11.1)⁷³. We obtained raw and ICE (iterative correction and eigenvector decomposition)⁷⁴ normalized matrices for further analysis. And the HiTC R package was used to calculate the A/B compartment status by using the normalized matrices at 500 kb resolution with the pca.hic function with default parameters⁷⁵. The compartment strength was calculated by CoolTools (version 0.5.2) with a saddle plot. TADs and TAD boundaries were identified at 500 kb on 40kb-binned contact maps as previously described methods⁷⁶. Finally, we used PSYCHIC software to identify the enhancer and promoter chromatin loops with default parameters under p < 0.0001⁷⁷.

Quantification of chromatin interaction dissimilarity

To systematically compare chromatin interaction patterns between normal and arrested human 8-cell embryos, we developed a computational approach that integrates the structure similarity index measure (SSIM) with sliding-window genomic scanning. This method effectively transforms abstract Hi-C matrix comparisons into quantifiable “dissimilarity scores.” Hi-C contact matrices for both conditions were first normalized using a z-score transformation to ensure comparability. For each 100 kb genomic window, we computed the SSIM between the interaction matrices of normal and arrested embryos, generating a genome-wide dissimilarity score profile. Windows with z-score ≤ −1.2 (corresponding to the bottom 5% of the SSIM distribution) were classified as highly dissimilar regions, indicating significant chromatin structural reorganization potentially associated with developmental arrest. Windows exhibiting a low signal-to-noise ratio (SNR < 0.6) were designated as noisy regions and excluded from downstream comparative analyses to minimize technical artifacts arising from low sequencing depth or repetitive sequence interference.

Variable TADs and TAD boundaries analysis

For the comparison of TAD, we first used the Directional Index (DI) value to calculate the TADs. Then, we used bedtools to calculate the overlapped TADs and TAD boundaries. Two TADs or TAD boundaries are defined as invariable if their ranges differ by less than 200 kb (upstream and downstream), they are considered variable.

Identification of TAD-TAD interactions and TAD clusters

We calculated the TAD-TAD interactions and TAD clusters as previously described with modification⁷⁸. We applied the NCHG model for all pairs of TADs on the same chromosome to identify the significance of TAD-TAD interactions with the Benjamini–Hochberg adjusted p value (BH < 0.01). Then the NetworkX Python library was used to identify the maximal size of TAD clusters using the Bron-Kerbosch algorithm and plot the TAD cluster dynamics. The hub TADs were identified based on more than five connected nodes.

3D modeling of chromatins

The three-dimensional (3D) chromatin architecture for both normal and arrested embryos was inferred using the multidimensional scaling (MDS) algorithm implemented in the PASTIS software²¹. This inference was based on Hi-C contact matrices binned at a 500-kb resolution. The resulting 3D models were subsequently visualized using the Mayavi2 package⁷⁹.

Motif analysis

Motif enrichment analysis was followed as previously described⁶⁷. Briefly, we used HOMER software (version 4.11) to identify enriched transcription factor motifs. The input regions were defined as the anchors of chromatin loops. For each anchor, a ±200 bp window centered on the loop boundary was extracted for analysis.

As a control background, we used size-matched random genomic regions, matched for GC content and chromosome distribution. Motif discovery was performed using HOMER with parameters (-len 8,10,12), searching against a comprehensive database of known transcription factor motifs. Enriched motifs were identified using the binomial test with Benjamini–Hochberg correction for multiple testing, applying a significance threshold of adjusted p value < 0.05.

Copy number variation (CNV) analysis and FGA calculation

To infer CNV from Hi-C data, we used the CNV function of the HiNT (Hi-C for Detection of Copy Number Variations and Translocations) program⁸⁰. Here, we directly used the previously generated .cool files of 8-cell embryos as input and predicted CNV events at 100-kb resolution. To quantify genomic instability, we calculated the FGA for each embryo⁸¹. FGA is defined as the proportion of the autosomal genome affected by copy number alterations, based on segmented log₂ copy number ratios from HiNT output. Genomic regions with |log₂ ratio| > 0.2 from the baseline were considered altered. FGA was then computed as the total length of such regions divided by the total length of all autosomes.

Statistics and reproducibility

No statistical methods were used to predetermine sample sizes. Sample sizes were determined based on previous studies of early embryonic development in humans. Biological replicates are defined as independent pools of multiple embryos randomly combined. For each condition (normal and arrested), two biological replicates were collected for RNA-seq and two for Hi-C analysis, as is standard in the field. Due to the nature of the morphological assessment, data collection and analyses were not performed blind to embryo group allocation. Embryos that failed quality control were excluded from both the count statistics and subsequent analyses. Each replicate served as the experimental unit for all statistical analyses. Data are presented as means ± standard deviations (SD). Statistical analysis was conducted by R and Microsoft Excel. Statistical significance was determined using one-way ANOVA, Student’s t-test, or Wilcoxon rank-sum test, as indicated in the corresponding figure legends. All FDR corrections throughout the study were performed using the Benjamini–Hochberg (BH) procedure. A p value < 0.05 indicated a statistically significant difference.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Y.C. and G.T. designed, performed data analysis, and wrote the manuscript. C.H. acquired clinical samples and performed sample processing. S.W. performed 3D genome construction analysis. Q.L., S.Y., and M.Z. were concerned with the manuscript. Y.S. provided clinical samples. J.W. and G.T. provided constructive guidance and detailed inspection. All authors edited and approved the manuscript.

Peer review

Peer review information

Communications Biology thanks Mrinmoy Pal and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: Wee-Wei Tee and Mengtan Xing. A peer review file is available.

Data availability

Raw and processed sequencing data generated in this study have been deposited in the China National Center for Bioinformation, National Genomics Data Center under accession numbers PRJCA049418 HRA014395. Previously published data that were re-analyzed in this study are available as follows: The human 4-cell and ICM embryos’ Hi-C raw sequence data are available from the China National Center for Bioinformation, National Genomics Data Center under accession number PRJCA016057 HRA004381; the ZGA and Z-decay gene lists were generated using scRNA-seq data from human early embryos (GEO: GSE36552). RNA-seq data of human oocyte used for generating maternal genes were from dataset GSE101571. All data supporting the findings of this study are available within the paper and its Supplementary Information.

Code availability

All scripts, including R scripts and Python scripts used for data processing, statistical analysis, and figure generation in this study, are publicly available at GitHub (https://github.com/skytguuu/2026cyd_human_embryo_arrested_data) and have been archived in Zenodo with https://doi.org/10.5281/zenodo.20257948⁸².

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s42003-026-10614-0.