Single-cell co-mapping reveals relationship between chromatin state and gene expression in early zebrafish development

  1. Vivek Bhardwaj
  2. Alberto Griffa
  3. Helena Viñas Gaza
  4. Peter Zeller
  5. Alexander van Oudenaarden

Reviewed by

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

Log in to save this article

Abstract

Establishing a cell-type-specific chromatin landscape is crucial for the maintenance of cell identity during embryonic development. However, our knowledge of how this landscape is set during vertebrate embryogenesis has been limited, due to the lack of methods to jointly detect chromatin modifications and gene expression in the same cell. Here we present a multimodal measurement of full-length transcriptome and histone modifications in individual cells during early embryonic development in zebrafish. We show that before the formation of germ layers, the chromatin and transcription states of cells are uncoupled, and become progressively connected during gastrulation and somitogenesis. Silencing of developmental genes is achieved by local spreading of repressive chromatin together with cell-type specific demethylation. Combining transcription factor (TF) expression and chromatin states within an interpretable machine learning model, we classify TFs as lineage-specific activators and repressors and identify a subset of TFs that are epigenetically regulated. Altogether, our data resolves the dynamic relationship between chromatin and transcription during early vertebrate development and clarifies how these two layers interact to establish cell identity.

Article activity feed

  1. Review Commons

    Note: This response was posted by the corresponding author to Review Commons. The content has not been altered except for formatting.

    Learn more at Review Commons

    Reply to the reviewers

    *Reviewer #1 (Evidence, reproducibility and clarity (Required)): *

    *The authors have a longstanding focus and reputation on single cell sequencing technology development and application. In this current study, the authors developed a novel single-cell multi-omic assay termed "T-ChIC" so that to jointly profile the histone modifications along with the full-length transcriptome from the same single cells, analyzed the dynamic relationship between chromatin state and gene expression during zebrafish development and cell fate determination. In general, the assay works well, the data look convincing and conclusions are beneficial to the community. *

    Thank you for your positive feedback.

    *There are several single-cell methodologies all claim to co-profile chromatin modifications and gene expression from the same individual cell, such as CoTECH, Paired-tag and others. Although T-ChIC employs pA-Mnase and IVT to obtain these modalities from single cells which are different, could the author provide some direct comparisons among all these technologies to see whether T-ChIC outperforms? *

    In a separate technical manuscript describing the application of T-ChIC in mouse cells (Zeller, Blotenburg et al 2024, bioRxiv, 2024.05. 09.593364), we have provided a direct comparison of data quality between T-ChIC and other single-cell methods for chromatin-RNA co-profiling (Please refer to Fig. 1C,D and Fig. S1D, E, of the preprint). We show that compared to other methods, T-ChIC is able to better preserve the expected biological relationship between the histone modifications and gene expression in single cells.

    *In current study, T-ChIC profiled H3K27me3 and H3K4me1 modifications, these data look great. How about other histone modifications (eg H3K9me3 and H3K36me3) and transcription factors? *

    While we haven't profiled these other modifications using T-ChIC in Zebrafish, we have previously published high quality data on these histone modifications using the sortChIC method, on which T-ChIC is based (Zeller, Yeung et al 2023). In our comparison, we find that histone modification profiles between T-ChIC and sortChIC are very similar (Fig. S1C in Zeller, Blotenburg et al 2024). Therefore the method is expected to work as well for the other histone marks.

    *T-ChIC can detect full length transcription from the same single cells, but in FigS3, the authors still used other published single cell transcriptomics to annotate the cell types, this seems unnecessary? *

    We used the published scRNA-seq dataset with a larger number of cells to homogenize our cell type labels with these datasets, but we also cross-referenced our cluster-specific marker genes with ZFIN and homogenized the cell type labels with ZFIN ontology. This way our annotation is in line with previous datasets but not biased by it. Due the relatively smaller size of our data, we didn't expect to identify unique, rare cell types, but our full-length total RNA assay helps us identify non-coding RNAs such as miRNA previously undetected in scRNA assays, which we have now highlighted in new figure S1c .

    *Throughout the manuscript, the authors found some interesting dynamics between chromatin state and gene expression during embryogenesis, independent approaches should be used to validate these findings, such as IHC staining or RNA ISH? *

    We appreciate that the ISH staining could be useful to validate the expression pattern of genes identified in this study. But to validate the relationships between the histone marks and gene expression, we need to combine these stainings with functional genomics experiments, such as PRC2-related knockouts. Due to their complexity, such experiments are beyond the scope of this manuscript (see also reply to reviewer #3, comment #4 for details).

    *In Fig2 and FigS4, the authors showed H3K27me3 cis spreading during development, this looks really interesting. Is this zebrafish specific? H3K27me3 ChIP-seq or CutTag data from mouse and/or human embryos should be reanalyzed and used to compare. The authors could speculate some possible mechanisms to explain this spreading pattern? *

    Thanks for the suggestion. In this revision, we have reanalysed a dataset of mouse ChIP-seq of H3K27me3 during mouse embryonic development by Xiang et al (Nature Genetics 2019) and find similar evidence of spreading of H3K27me3 signal from their pre-marked promoter regions at E5.5 epiblast upon differentiation (new Figure S4i). This observation, combined with the fact that the mechanism of pre-marking of promoters by PRC1-PRC2 interaction seems to be conserved between the two species (see (Hickey et al., 2022), (Mei et al., 2021) & (Chen et al., 2021)), suggests that the dynamics of H3K27me3 pattern establishment is conserved across vertebrates. But we think a high-resolution profiling via a method like T-ChIC would be more useful to demonstrate the dynamics of signal spreading during mouse embryonic development in the future. We have discussed this further in our revised manuscript.

    *Reviewer #1 (Significance (Required)): *

    *The authors have a longstanding focus and reputation on single cell sequencing technology development and application. In this current study, the authors developed a novel single-cell multi-omic assay termed "T-ChIC" so that to jointly profile the histone modifications along with the full-length transcriptome from the same single cells, analyzed the dynamic relationship between chromatin state and gene expression during zebrafish development and cell fate determination. In general, the assay works well, the data look convincing and conclusions are beneficial to the community. *

    Thank you very much for your supportive remarks.

    *Reviewer #2 (Evidence, reproducibility and clarity (Required)): *

    *Joint analysis of multiple modalities in single cells will provide a comprehensive view of cell fate states. In this manuscript, Bhardwaj et al developed a single-cell multi-omics assay, T-ChIC, to simultaneously capture histone modifications and full-length transcriptome and applied the method on early embryos of zebrafish. The authors observed a decoupled relationship between the chromatin modifications and gene expression at early developmental stages. The correlation becomes stronger as development proceeds, as genes are silenced by the cis-spreading of the repressive marker H3k27me3. Overall, the work is well performed, and the results are meaningful and interesting to readers in the epigenomic and embryonic development fields. There are some concerns before the manuscript is considered for publication. *

    We thank the reviewer for appreciating the quality of our study.

    *Major concerns: *

      • A major point of this study is to understand embryo development, especially gastrulation, with the power of scMulti-Omics assay. However, the current analysis didn't focus on deciphering the biology of gastrulation, i.e., lineage-specific pioneer factors that help to reform the chromatin landscape. The majority of the data analysis is based on the temporal dimension, but not the cell-type-specific dimension, which reduces the value of the single-cell assay. *

    We focused on the lineage-specific transcription factor activity during gastrulation in Figure 4 and S8 of the manuscript and discovered several interesting regulators active at this stage. During our analysis of the temporal dimension for the rest of the manuscript, we also classified the cells by their germ layer and "latent" developmental time by taking the full advantage of the single-cell nature of our data. Additionally, we have now added the cell-type-specific H3K27-demethylation results for 24hpf in response to your comment below. We hope that these results, together with our openly available dataset would demonstrate the advantage of the single-cell aspect of our dataset.

    1. *The cis-spreading of H3K27me3 with developmental time is interesting. Considering H3k27me3 could mark bivalent regions, especially in pluripotent cells, there must be some regions that have lost H3k27me3 signals during development. Therefore, it's confusing that the authors didn't find these regions (30% spreading, 70% stable). The authors should explain and discuss this issue. *

    Indeed we see that ~30% of the bins enriched in the pluripotent stage spread, while 70% do not seem to spread. In line with earlier observations(Hickey et al., 2022; Vastenhouw et al., 2010), we find that H3K27me3 is almost absent in the zygote and is still being accumulated until 24hpf and beyond. Therefore the majority of the sites in the genome still seem to be in the process of gaining H3K27me3 until 24hpf, explaining why we see mostly "spreading" and "stable" states. Considering most of these sites are at promoters and show signs of bivalency, we think that these sites are marked for activation or silencing at later stages. We have discussed this in the manuscript ("discussion"). However, in response to this and earlier comment, we went back and searched for genes that show H3K27-demethylation in the most mature cell types (at 24 hpf) in our data, and found a subset of genes that show K27 demethylation after acquiring them earlier. Interestingly, most of the top genes in this list are well-known as developmentally important for their corresponding cell types. We have added this new result and discussed it further in the manuscript (Fig. 2d,e, , Supplementary table 3).

    *Minors: *

      • The authors cited two scMulti-omics studies in the introduction, but there have been lots of single-cell multi-omics studies published recently. The authors should cite and consider them. *

    We have cited more single-cell chromatin and multiome studies focussed on early embryogenesis in the introduction now.

    *2. T-ChIC seems to have been presented in a previous paper (ref 15). Therefore, Fig. 1a is unnecessary to show. *

    Figure 1a. shows a summary of our Zebrafish TChIC workflow, which contains the unique sample multiplexing and sorting strategy to reduce batch effects, which was not applied in the original TChIC workflow. We have now clarified this in "Results".

    1. *It's better to show the percentage of cell numbers (30% vs 70%) for each heatmap in Figure 2C. *

    We have added the numbers to the corresponding legends.

    1. *Please double-check the citation of Fig. S4C, which may not relate to the conclusion of signal differences between lineages. *

    The citation seems to be correct (Fig. S4C supplements Fig. 2C, but shows mesodermal lineage cells) but the description of the legend was a bit misleading. We have clarified this now.

    *5. Figure 4C has not been cited or mentioned in the main text. Please check. *

    Thanks for pointing it out. We have cited it in Results now.

    *Reviewer #2 (Significance (Required)): *

    *Strengths: This work utilized a new single-cell multi-omics method and generated abundant epigenomics and transcriptomics datasets for cells covering multiple key developmental stages of zebrafish. *

    *Limitations: The data analysis was superficial and mainly focused on the correspondence between the two modalities. The discussion of developmental biology was limited. *

    *Advance: The zebrafish single-cell datasets are valuable. The T-ChIC method is new and interesting. *

    *The audience will be specialized and from basic research fields, such as developmental biology, epigenomics, bioinformatics, etc. *

    *I'm more specialized in the direction of single-cell epigenomics, gene regulation, 3D genomics, etc. *

    Thank you for your remarks.

    *Reviewer #3 (Evidence, reproducibility and clarity (Required)): *

    *This manuscript introduces T‑ChIC, a single‑cell multi‑omics workflow that jointly profiles full‑length transcripts and histone modifications (H3K27me3 and H3K4me1) and applies it to early zebrafish embryos (4-24 hpf). The study convincingly demonstrates that chromatin-transcription coupling strengthens during gastrulation and somitogenesis, that promoter‑anchored H3K27me3 spreads in cis to enforce developmental gene silencing, and that integrating TF chromatin status with expression can predict lineage‑specific activators and repressors. *

    *Major concerns *

    1. *Independent biological replicates are absent, so the authors should process at least one additional clutch of embryos for key stages (e.g., 6 hpf and 12 hpf) with T‑ChIC and demonstrate that the resulting data match the current dataset. *

    Thanks for pointing this out. We had, in fact, performed T-ChIC experiments in four rounds of biological replicates (independent clutch of embryos) and merged the data to create our resource. Although not all timepoints were profiled in each replicate, two timepoints (10 and 24hpf) are present in all four, and the celltype composition of these replicates from these 2 timepoints are very similar. We have added new plots in figure S2f and added (new) supplementary table (#1) to highlight the presence of biological replicates.

    *2. **The TF‑activity regression model uses an arbitrary R² {greater than or equal to} 0.6 threshold; cross‑validated R² distributions, permutation‑based FDR control, and effect‑size confidence intervals are needed to justify this cut‑off. *

    Thank you for this suggestion. We did use 10-fold cross validation during training and obtained the R2 values of TF motifs from the independent test set as an unbiased estimate. However, the cutoff of R2 > 0.6 to select the TFs for classification was indeed arbitrary. In the revised version, we now report the FDR-adjusted p-values for these R2 estimates based on permutation tests, and select TFs with a cutoff of padj supplementary table #4 to include the p-values for all tested TFs. However, we see that our arbitrary cutoff of 0.6 was in fact, too stringent, and we can classify many more TFs based on the FDR cutoffs. We also updated our reported numbers in Fig. 4c to reflect this. Moreover, supplementary table #4 contains the complete list of TFs used in the analysis to allow others to choose their own cutoff.

    *3. **Predicted TF functions lack empirical support, making it essential to test representative activators (e.g., Tbx16) and repressors (e.g., Zbtb16a) via CRISPRi or morpholino knock‑down and to measure target‑gene expression and H3K4me1 changes. *

    We agree that independent validation of the functions of our predicted TFs on target gene activity would be important. During this revision, we analysed recently published scRNA-seq data of Saunders et al. (2023) (Saunders et al., 2023), which includes CRISPR-mediated F0 knockouts of a couple of our predicted TFs, but the scRNAseq was performed at later stages (24hpf onward) compared to our H3K4me1 analysis (which was 4-12 hpf). Therefore, we saw off-target genes being affected in lineages where these TFs are clearly not expressed (attached Fig 1). We therefore didn't include these results in the manuscript. In future, we aim to systematically test the TFs predicted in our study with CRISPRi or similar experiments.

    *4. **The study does not prove that H3K27me3 spreading causes silencing; embryos treated with an Ezh2 inhibitor or prc2 mutants should be re‑profiled by T‑ChIC to show loss of spreading along with gene re‑expression. *

    We appreciate the suggestion that indeed PRC2-disruption followed by T-ChIC or other forms of validation would be needed to confirm whether the H3K27me3 spreading is indeed causally linked to the silencing of the identified target genes. But performing this validation is complicated because of multiple reasons: 1) due to the EZH2 contribution from maternal RNA and the contradicting effects of various EZH2 zygotic mutations (depending on where the mutation occurs), the only properly validated PRC2-related mutant seems to be the maternal-zygotic mutant MZezh2, which requires germ cell transplantation (see Rougeot et al. 2019 (Rougeot et al., 2019)) , and San et al. 2019 (San et al., 2019) for details). The use of inhibitors have been described in other studies (den Broeder et al., 2020; Huang et al., 2021), but they do not show a validation of the H3K27me3 loss or a similar phenotype as the MZezh2 mutants, and can present unwanted side effects and toxicity at a high dose, affecting gene expression results. Moreover, in an attempt to validate, we performed our own trials with the EZH2 inhibitor (GSK123) and saw that this time window might be too short to see the effect within 24hpf (attached Fig. 2). Therefore, this validation is a more complex endeavor beyond the scope of this study. Nevertheless, our further analysis of H3K27me3 de-methylation on developmentally important genes (new Fig. 2e-f, Sup. table 3) adds more confidence that the polycomb repression plays an important role, and provides enough ground for future follow up studies.

    *Minor concerns *

    1. *Repressive chromatin coverage is limited, so profiling an additional silencing mark such as H3K9me3 or DNA methylation would clarify cooperation with H3K27me3 during development. *

    We agree that H3K27me3 alone would not be sufficient to fully understand the repressive chromatin state. Extension to other chromatin marks and DNA methylation would be the focus of our follow up works.

    *2. Computational transparency is incomplete; a supplementary table listing all trimming, mapping, and peak‑calling parameters (cutadapt, STAR/hisat2, MACS2, histoneHMM, etc.) should be provided. *

    As mentioned in the manuscript, we provide an open-source pre-processing pipeline "scChICflow" to perform all these steps (github.com/bhardwaj-lab/scChICflow). We have now also provided the configuration files on our zenodo repository (see below), which can simply be plugged into this pipeline together with the fastq files from GEO to obtain the processed dataset that we describe in the manuscript. Additionally, we have also clarified the peak calling and post-processing steps in the manuscript now.

    *3. Data‑ and code‑availability statements lack detail; the exact GEO accession release date, loom‑file contents, and a DOI‑tagged Zenodo archive of analysis scripts should be added. *

    We have now publicly released the .h5ad files with raw counts, normalized counts, and complete gene and cell-level metadata, along with signal tracks (bigwigs) and peaks on GEO. Additionally, we now also released the source datasets and notebooks (.Rmarkdown format) on Zenodo that can be used to replicate the figures in the manuscript, and updated our statements on "Data and code availability".

    *4. Minor editorial issues remain, such as replacing "critical" with "crucial" in the Abstract, adding software version numbers to figure legends, and correcting the SAMtools reference. *

    Thank you for spotting them. We have fixed these issues.

    *Reviewer #3 (Significance (Required)): *

    The method is technically innovative and the biological insights are valuable; however, several issues-mainly concerning experimental design, statistical rigor, and functional validation-must be addressed to solidify the conclusions.

    Thank you for your comments. We hope to have addressed your concerns in this revised version of our manuscript.

  2. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #3

    Evidence, reproducibility and clarity

    This manuscript introduces T‑ChIC, a single‑cell multi‑omics workflow that jointly profiles full‑length transcripts and histone modifications (H3K27me3 and H3K4me1) and applies it to early zebrafish embryos (4-24 hpf). The study convincingly demonstrates that chromatin-transcription coupling strengthens during gastrulation and somitogenesis, that promoter‑anchored H3K27me3 spreads in cis to enforce developmental gene silencing, and that integrating TF chromatin status with expression can predict lineage‑specific activators and repressors.

    Major concerns

    1. Independent biological replicates are absent, so the authors should process at least one additional clutch of embryos for key stages (e.g., 6 hpf and 12 hpf) with T‑ChIC and demonstrate that the resulting data match the current dataset.
    2. The TF‑activity regression model uses an arbitrary R² {greater than or equal to} 0.6 threshold; cross‑validated R² distributions, permutation‑based FDR control, and effect‑size confidence intervals are needed to justify this cut‑off.
    3. Predicted TF functions lack empirical support, making it essential to test representative activators (e.g., Tbx16) and repressors (e.g., Zbtb16a) via CRISPRi or morpholino knock‑down and to measure target‑gene expression and H3K4me1 changes.
    4. The study does not prove that H3K27me3 spreading causes silencing; embryos treated with an Ezh2 inhibitor or prc2 mutants should be re‑profiled by T‑ChIC to show loss of spreading along with gene re‑expression.

    Minor concerns

    1. Repressive chromatin coverage is limited, so profiling an additional silencing mark such as H3K9me3 or DNA methylation would clarify cooperation with H3K27me3 during development.
    2. Computational transparency is incomplete; a supplementary table listing all trimming, mapping, and peak‑calling parameters (cutadapt, STAR/hisat2, MACS2, histoneHMM, etc.) should be provided.
    3. Data‑ and code‑availability statements lack detail; the exact GEO accession release date, loom‑file contents, and a DOI‑tagged Zenodo archive of analysis scripts should be added.
    4. Minor editorial issues remain, such as replacing "critical" with "crucial" in the Abstract, adding software version numbers to figure legends, and correcting the SAMtools reference.

    Significance

    The method is technically innovative and the biological insights are valuable; however, several issues-mainly concerning experimental design, statistical rigor, and functional validation-must be addressed to solidify the conclusions.

  3. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #2

    Evidence, reproducibility and clarity

    Joint analysis of multiple modalities in single cells will provide a comprehensive view of cell fate states. In this manuscript, Bhardwaj et al developed a single-cell multi-omics assay, T-ChIC, to simultaneously capture histone modifications and full-length transcriptome and applied the method on early embryos of zebrafish. The authors observed a decoupled relationship between the chromatin modifications and gene expression at early developmental stages. The correlation becomes stronger as development proceeds, as genes are silenced by the cis-spreading of the repressive marker H3k27me3. Overall, the work is well performed, and the results are meaningful and interesting to readers in the epigenomic and embryonic development fields. There are some concerns before the manuscript is considered for publication.

    Major concerns:

    1. A major point of this study is to understand embryo development, especially gastrulation, with the power of scMulti-Omics assay. However, the current analysis didn't focus on deciphering the biology of gastrulation, i.e., lineage-specific pioneer factors that help to reform the chromatin landscape. The majority of the data analysis is based on the temporal dimension, but not the cell-type-specific dimension, which reduces the value of the single-cell assay.
    2. The cis-spreading of H3K27me3 with developmental time is interesting. Considering H3k27me3 could mark bivalent regions, especially in pluripotent cells, there must be some regions that have lost H3k27me3 signals during development. Therefore, it's confusing that the authors didn't find these regions (30% spreading, 70% stable). The authors should explain and discuss this issue.

    Minors:

    1. The authors cited two scMulti-omics studies in the introduction, but there have been lots of single-cell multi-omics studies published recently. The authors should cite and consider them.
    2. T-ChIC seems to have been presented in a previous paper (ref 15). Therefore, Fig. 1a is unnecessary to show.
    3. It's better to show the percentage of cell numbers (30% vs 70%) for each heatmap in Figure 2C.
    4. Please double-check the citation of Fig. S4C, which may not relate to the conclusion of signal differences between lineages.
    5. Figure 4C has not been cited or mentioned in the main text. Please check.

    Significance

    Strengths: This work utilized a new single-cell multi-omics method and generated abundant epigenomics and transcriptomics datasets for cells covering multiple key developmental stages of zebrafish. Limitations: The data analysis was superficial and mainly focused on the correspondence between the two modalities. The discussion of developmental biology was limited.

    Advance: The zebrafish single-cell datasets are valuable. The T-ChIC method is new and interesting.

    The audience will be specialized and from basic research fields, such as developmental biology, epigenomics, bioinformatics, etc.

    I'm more specialized in the direction of single-cell epigenomics, gene regulation, 3D genomics, etc.

  4. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #1

    Evidence, reproducibility and clarity

    The authors have a longstanding focus and reputation on single cell sequencing technology development and application. In this current study, the authors developed a novel single-cell multi-omic assay termed "T-ChIC" so that to jointly profile the histone modifications along with the full-length transcriptome from the same single cells, analyzed the dynamic relationship between chromatin state and gene expression during zebrafish development and cell fate determination. In general, the assay works well, the data look convincing and conclusions are beneficial to the community.

    There are several single-cell methodologies all claim to co-profile chromatin modifications and gene expression from the same individual cell, such as CoTECH, Paired-tag and others. Although T-ChIC employs pA-Mnase and IVT to obtain these modalities from single cells which are different, could the author provide some direct comparisons among all these technologies to see whether T-ChIC outperforms?

    In current study, T-ChIC profiled H3K27me3 and H3K4me1 modifications, these data look great. How about other histone modifications (eg H3K9me3 and H3K36me3) and transcription factors?

    T-ChIC can detect full length transcription from the same single cells, but in FigS3, the authors still used other published single cell transcriptomics to annotate the cell types, this seems unnecessary?

    Throughout the manuscript, the authors found some interesting dynamics between chromatin state and gene expression during embryogenesis, independent approaches should be used to validate these findings, such as IHC staining or RNA ISH?

    In Fig2 and FigS4, the authors showed H3K27me3 cis spreading during development, this looks really interesting. Is this zebrafish specific? H3K27me3 ChIP-seq or CutTag data from mouse and/or human embryos should be reanalyzed and used to compare. The authors could speculate some possible mechanisms to explain this spreading pattern?

    Significance

    The authors have a longstanding focus and reputation on single cell sequencing technology development and application. In this current study, the authors developed a novel single-cell multi-omic assay termed "T-ChIC" so that to jointly profile the histone modifications along with the full-length transcriptome from the same single cells, analyzed the dynamic relationship between chromatin state and gene expression during zebrafish development and cell fate determination. In general, the assay works well, the data look convincing and conclusions are beneficial to the community.

  5. Review Commons

    Note: This response was posted by the corresponding author to Review Commons. The content has not been altered except for formatting.

    Learn more at Review Commons

    Reply to the reviewers

    *Reviewer #1 (Evidence, reproducibility and clarity (Required)): *

    *The authors have a longstanding focus and reputation on single cell sequencing technology development and application. In this current study, the authors developed a novel single-cell multi-omic assay termed "T-ChIC" so that to jointly profile the histone modifications along with the full-length transcriptome from the same single cells, analyzed the dynamic relationship between chromatin state and gene expression during zebrafish development and cell fate determination. In general, the assay works well, the data look convincing and conclusions are beneficial to the community. *

    Thank you for your positive feedback.

    *There are several single-cell methodologies all claim to co-profile chromatin modifications and gene expression from the same individual cell, such as CoTECH, Paired-tag and others. Although T-ChIC employs pA-Mnase and IVT to obtain these modalities from single cells which are different, could the author provide some direct comparisons among all these technologies to see whether T-ChIC outperforms? *

    In a separate technical manuscript describing the application of T-ChIC in mouse cells (Zeller, Blotenburg et al 2024, bioRxiv, 2024.05. 09.593364), we have provided a direct comparison of data quality between T-ChIC and other single-cell methods for chromatin-RNA co-profiling (Please refer to Fig. 1C,D and Fig. S1D, E, of the preprint). We show that compared to other methods, T-ChIC is able to better preserve the expected biological relationship between the histone modifications and gene expression in single cells.

    *In current study, T-ChIC profiled H3K27me3 and H3K4me1 modifications, these data look great. How about other histone modifications (eg H3K9me3 and H3K36me3) and transcription factors? *

    While we haven't profiled these other modifications using T-ChIC in Zebrafish, we have previously published high quality data on these histone modifications using the sortChIC method, on which T-ChIC is based (Zeller, Yeung et al 2023). In our comparison, we find that histone modification profiles between T-ChIC and sortChIC are very similar (Fig. S1C in Zeller, Blotenburg et al 2024). Therefore the method is expected to work as well for the other histone marks.

    *T-ChIC can detect full length transcription from the same single cells, but in FigS3, the authors still used other published single cell transcriptomics to annotate the cell types, this seems unnecessary? *

    We used the published scRNA-seq dataset with a larger number of cells to homogenize our cell type labels with these datasets, but we also cross-referenced our cluster-specific marker genes with ZFIN and homogenized the cell type labels with ZFIN ontology. This way our annotation is in line with previous datasets but not biased by it. Due the relatively smaller size of our data, we didn't expect to identify unique, rare cell types, but our full-length total RNA assay helps us identify non-coding RNAs such as miRNA previously undetected in scRNA assays, which we have now highlighted in new figure S1c .

    *Throughout the manuscript, the authors found some interesting dynamics between chromatin state and gene expression during embryogenesis, independent approaches should be used to validate these findings, such as IHC staining or RNA ISH? *

    We appreciate that the ISH staining could be useful to validate the expression pattern of genes identified in this study. But to validate the relationships between the histone marks and gene expression, we need to combine these stainings with functional genomics experiments, such as PRC2-related knockouts. Due to their complexity, such experiments are beyond the scope of this manuscript (see also reply to reviewer #3, comment #4 for details).

    *In Fig2 and FigS4, the authors showed H3K27me3 cis spreading during development, this looks really interesting. Is this zebrafish specific? H3K27me3 ChIP-seq or CutTag data from mouse and/or human embryos should be reanalyzed and used to compare. The authors could speculate some possible mechanisms to explain this spreading pattern? *

    Thanks for the suggestion. In this revision, we have reanalysed a dataset of mouse ChIP-seq of H3K27me3 during mouse embryonic development by Xiang et al (Nature Genetics 2019) and find similar evidence of spreading of H3K27me3 signal from their pre-marked promoter regions at E5.5 epiblast upon differentiation (new Figure S4i). This observation, combined with the fact that the mechanism of pre-marking of promoters by PRC1-PRC2 interaction seems to be conserved between the two species (see (Hickey et al., 2022), (Mei et al., 2021) & (Chen et al., 2021)), suggests that the dynamics of H3K27me3 pattern establishment is conserved across vertebrates. But we think a high-resolution profiling via a method like T-ChIC would be more useful to demonstrate the dynamics of signal spreading during mouse embryonic development in the future. We have discussed this further in our revised manuscript.

    *Reviewer #1 (Significance (Required)): *

    *The authors have a longstanding focus and reputation on single cell sequencing technology development and application. In this current study, the authors developed a novel single-cell multi-omic assay termed "T-ChIC" so that to jointly profile the histone modifications along with the full-length transcriptome from the same single cells, analyzed the dynamic relationship between chromatin state and gene expression during zebrafish development and cell fate determination. In general, the assay works well, the data look convincing and conclusions are beneficial to the community. *

    Thank you very much for your supportive remarks.

    *Reviewer #2 (Evidence, reproducibility and clarity (Required)): *

    *Joint analysis of multiple modalities in single cells will provide a comprehensive view of cell fate states. In this manuscript, Bhardwaj et al developed a single-cell multi-omics assay, T-ChIC, to simultaneously capture histone modifications and full-length transcriptome and applied the method on early embryos of zebrafish. The authors observed a decoupled relationship between the chromatin modifications and gene expression at early developmental stages. The correlation becomes stronger as development proceeds, as genes are silenced by the cis-spreading of the repressive marker H3k27me3. Overall, the work is well performed, and the results are meaningful and interesting to readers in the epigenomic and embryonic development fields. There are some concerns before the manuscript is considered for publication. *

    We thank the reviewer for appreciating the quality of our study.

    *Major concerns: *

      • A major point of this study is to understand embryo development, especially gastrulation, with the power of scMulti-Omics assay. However, the current analysis didn't focus on deciphering the biology of gastrulation, i.e., lineage-specific pioneer factors that help to reform the chromatin landscape. The majority of the data analysis is based on the temporal dimension, but not the cell-type-specific dimension, which reduces the value of the single-cell assay. *

    We focused on the lineage-specific transcription factor activity during gastrulation in Figure 4 and S8 of the manuscript and discovered several interesting regulators active at this stage. During our analysis of the temporal dimension for the rest of the manuscript, we also classified the cells by their germ layer and "latent" developmental time by taking the full advantage of the single-cell nature of our data. Additionally, we have now added the cell-type-specific H3K27-demethylation results for 24hpf in response to your comment below. We hope that these results, together with our openly available dataset would demonstrate the advantage of the single-cell aspect of our dataset.

    1. *The cis-spreading of H3K27me3 with developmental time is interesting. Considering H3k27me3 could mark bivalent regions, especially in pluripotent cells, there must be some regions that have lost H3k27me3 signals during development. Therefore, it's confusing that the authors didn't find these regions (30% spreading, 70% stable). The authors should explain and discuss this issue. *

    Indeed we see that ~30% of the bins enriched in the pluripotent stage spread, while 70% do not seem to spread. In line with earlier observations(Hickey et al., 2022; Vastenhouw et al., 2010), we find that H3K27me3 is almost absent in the zygote and is still being accumulated until 24hpf and beyond. Therefore the majority of the sites in the genome still seem to be in the process of gaining H3K27me3 until 24hpf, explaining why we see mostly "spreading" and "stable" states. Considering most of these sites are at promoters and show signs of bivalency, we think that these sites are marked for activation or silencing at later stages. We have discussed this in the manuscript ("discussion"). However, in response to this and earlier comment, we went back and searched for genes that show H3K27-demethylation in the most mature cell types (at 24 hpf) in our data, and found a subset of genes that show K27 demethylation after acquiring them earlier. Interestingly, most of the top genes in this list are well-known as developmentally important for their corresponding cell types. We have added this new result and discussed it further in the manuscript (Fig. 2d,e, , Supplementary table 3).

    *Minors: *

      • The authors cited two scMulti-omics studies in the introduction, but there have been lots of single-cell multi-omics studies published recently. The authors should cite and consider them. *

    We have cited more single-cell chromatin and multiome studies focussed on early embryogenesis in the introduction now.

    *2. T-ChIC seems to have been presented in a previous paper (ref 15). Therefore, Fig. 1a is unnecessary to show. *

    Figure 1a. shows a summary of our Zebrafish TChIC workflow, which contains the unique sample multiplexing and sorting strategy to reduce batch effects, which was not applied in the original TChIC workflow. We have now clarified this in "Results".

    1. *It's better to show the percentage of cell numbers (30% vs 70%) for each heatmap in Figure 2C. *

    We have added the numbers to the corresponding legends.

    1. *Please double-check the citation of Fig. S4C, which may not relate to the conclusion of signal differences between lineages. *

    The citation seems to be correct (Fig. S4C supplements Fig. 2C, but shows mesodermal lineage cells) but the description of the legend was a bit misleading. We have clarified this now.

    *5. Figure 4C has not been cited or mentioned in the main text. Please check. *

    Thanks for pointing it out. We have cited it in Results now.

    *Reviewer #2 (Significance (Required)): *

    *Strengths: This work utilized a new single-cell multi-omics method and generated abundant epigenomics and transcriptomics datasets for cells covering multiple key developmental stages of zebrafish. *

    *Limitations: The data analysis was superficial and mainly focused on the correspondence between the two modalities. The discussion of developmental biology was limited. *

    *Advance: The zebrafish single-cell datasets are valuable. The T-ChIC method is new and interesting. *

    *The audience will be specialized and from basic research fields, such as developmental biology, epigenomics, bioinformatics, etc. *

    *I'm more specialized in the direction of single-cell epigenomics, gene regulation, 3D genomics, etc. *

    Thank you for your remarks.

    *Reviewer #3 (Evidence, reproducibility and clarity (Required)): *

    *This manuscript introduces T‑ChIC, a single‑cell multi‑omics workflow that jointly profiles full‑length transcripts and histone modifications (H3K27me3 and H3K4me1) and applies it to early zebrafish embryos (4-24 hpf). The study convincingly demonstrates that chromatin-transcription coupling strengthens during gastrulation and somitogenesis, that promoter‑anchored H3K27me3 spreads in cis to enforce developmental gene silencing, and that integrating TF chromatin status with expression can predict lineage‑specific activators and repressors. *

    *Major concerns *

    1. *Independent biological replicates are absent, so the authors should process at least one additional clutch of embryos for key stages (e.g., 6 hpf and 12 hpf) with T‑ChIC and demonstrate that the resulting data match the current dataset. *

    Thanks for pointing this out. We had, in fact, performed T-ChIC experiments in four rounds of biological replicates (independent clutch of embryos) and merged the data to create our resource. Although not all timepoints were profiled in each replicate, two timepoints (10 and 24hpf) are present in all four, and the celltype composition of these replicates from these 2 timepoints are very similar. We have added new plots in figure S2f and added (new) supplementary table (#1) to highlight the presence of biological replicates.

    *2. **The TF‑activity regression model uses an arbitrary R² {greater than or equal to} 0.6 threshold; cross‑validated R² distributions, permutation‑based FDR control, and effect‑size confidence intervals are needed to justify this cut‑off. *

    Thank you for this suggestion. We did use 10-fold cross validation during training and obtained the R2 values of TF motifs from the independent test set as an unbiased estimate. However, the cutoff of R2 > 0.6 to select the TFs for classification was indeed arbitrary. In the revised version, we now report the FDR-adjusted p-values for these R2 estimates based on permutation tests, and select TFs with a cutoff of padj supplementary table #4 to include the p-values for all tested TFs. However, we see that our arbitrary cutoff of 0.6 was in fact, too stringent, and we can classify many more TFs based on the FDR cutoffs. We also updated our reported numbers in Fig. 4c to reflect this. Moreover, supplementary table #4 contains the complete list of TFs used in the analysis to allow others to choose their own cutoff.

    *3. **Predicted TF functions lack empirical support, making it essential to test representative activators (e.g., Tbx16) and repressors (e.g., Zbtb16a) via CRISPRi or morpholino knock‑down and to measure target‑gene expression and H3K4me1 changes. *

    We agree that independent validation of the functions of our predicted TFs on target gene activity would be important. During this revision, we analysed recently published scRNA-seq data of Saunders et al. (2023) (Saunders et al., 2023), which includes CRISPR-mediated F0 knockouts of a couple of our predicted TFs, but the scRNAseq was performed at later stages (24hpf onward) compared to our H3K4me1 analysis (which was 4-12 hpf). Therefore, we saw off-target genes being affected in lineages where these TFs are clearly not expressed (attached Fig 1). We therefore didn't include these results in the manuscript. In future, we aim to systematically test the TFs predicted in our study with CRISPRi or similar experiments.

    *4. **The study does not prove that H3K27me3 spreading causes silencing; embryos treated with an Ezh2 inhibitor or prc2 mutants should be re‑profiled by T‑ChIC to show loss of spreading along with gene re‑expression. *

    We appreciate the suggestion that indeed PRC2-disruption followed by T-ChIC or other forms of validation would be needed to confirm whether the H3K27me3 spreading is indeed causally linked to the silencing of the identified target genes. But performing this validation is complicated because of multiple reasons: 1) due to the EZH2 contribution from maternal RNA and the contradicting effects of various EZH2 zygotic mutations (depending on where the mutation occurs), the only properly validated PRC2-related mutant seems to be the maternal-zygotic mutant MZezh2, which requires germ cell transplantation (see Rougeot et al. 2019 (Rougeot et al., 2019)) , and San et al. 2019 (San et al., 2019) for details). The use of inhibitors have been described in other studies (den Broeder et al., 2020; Huang et al., 2021), but they do not show a validation of the H3K27me3 loss or a similar phenotype as the MZezh2 mutants, and can present unwanted side effects and toxicity at a high dose, affecting gene expression results. Moreover, in an attempt to validate, we performed our own trials with the EZH2 inhibitor (GSK123) and saw that this time window might be too short to see the effect within 24hpf (attached Fig. 2). Therefore, this validation is a more complex endeavor beyond the scope of this study. Nevertheless, our further analysis of H3K27me3 de-methylation on developmentally important genes (new Fig. 2e-f, Sup. table 3) adds more confidence that the polycomb repression plays an important role, and provides enough ground for future follow up studies.

    *Minor concerns *

    1. *Repressive chromatin coverage is limited, so profiling an additional silencing mark such as H3K9me3 or DNA methylation would clarify cooperation with H3K27me3 during development. *

    We agree that H3K27me3 alone would not be sufficient to fully understand the repressive chromatin state. Extension to other chromatin marks and DNA methylation would be the focus of our follow up works.

    *2. Computational transparency is incomplete; a supplementary table listing all trimming, mapping, and peak‑calling parameters (cutadapt, STAR/hisat2, MACS2, histoneHMM, etc.) should be provided. *

    As mentioned in the manuscript, we provide an open-source pre-processing pipeline "scChICflow" to perform all these steps (github.com/bhardwaj-lab/scChICflow). We have now also provided the configuration files on our zenodo repository (see below), which can simply be plugged into this pipeline together with the fastq files from GEO to obtain the processed dataset that we describe in the manuscript. Additionally, we have also clarified the peak calling and post-processing steps in the manuscript now.

    *3. Data‑ and code‑availability statements lack detail; the exact GEO accession release date, loom‑file contents, and a DOI‑tagged Zenodo archive of analysis scripts should be added. *

    We have now publicly released the .h5ad files with raw counts, normalized counts, and complete gene and cell-level metadata, along with signal tracks (bigwigs) and peaks on GEO. Additionally, we now also released the source datasets and notebooks (.Rmarkdown format) on Zenodo that can be used to replicate the figures in the manuscript, and updated our statements on "Data and code availability".

    *4. Minor editorial issues remain, such as replacing "critical" with "crucial" in the Abstract, adding software version numbers to figure legends, and correcting the SAMtools reference. *

    Thank you for spotting them. We have fixed these issues.

    *Reviewer #3 (Significance (Required)): *

    The method is technically innovative and the biological insights are valuable; however, several issues-mainly concerning experimental design, statistical rigor, and functional validation-must be addressed to solidify the conclusions.

    Thank you for your comments. We hope to have addressed your concerns in this revised version of our manuscript.

  6. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #3

    Evidence, reproducibility and clarity

    This manuscript introduces T‑ChIC, a single‑cell multi‑omics workflow that jointly profiles full‑length transcripts and histone modifications (H3K27me3 and H3K4me1) and applies it to early zebrafish embryos (4-24 hpf). The study convincingly demonstrates that chromatin-transcription coupling strengthens during gastrulation and somitogenesis, that promoter‑anchored H3K27me3 spreads in cis to enforce developmental gene silencing, and that integrating TF chromatin status with expression can predict lineage‑specific activators and repressors.

    Major concerns

    1. Independent biological replicates are absent, so the authors should process at least one additional clutch of embryos for key stages (e.g., 6 hpf and 12 hpf) with T‑ChIC and demonstrate that the resulting data match the current dataset.
    2. The TF‑activity regression model uses an arbitrary R² {greater than or equal to} 0.6 threshold; cross‑validated R² distributions, permutation‑based FDR control, and effect‑size confidence intervals are needed to justify this cut‑off.
    3. Predicted TF functions lack empirical support, making it essential to test representative activators (e.g., Tbx16) and repressors (e.g., Zbtb16a) via CRISPRi or morpholino knock‑down and to measure target‑gene expression and H3K4me1 changes.
    4. The study does not prove that H3K27me3 spreading causes silencing; embryos treated with an Ezh2 inhibitor or prc2 mutants should be re‑profiled by T‑ChIC to show loss of spreading along with gene re‑expression.

    Minor concerns

    1. Repressive chromatin coverage is limited, so profiling an additional silencing mark such as H3K9me3 or DNA methylation would clarify cooperation with H3K27me3 during development.
    2. Computational transparency is incomplete; a supplementary table listing all trimming, mapping, and peak‑calling parameters (cutadapt, STAR/hisat2, MACS2, histoneHMM, etc.) should be provided.
    3. Data‑ and code‑availability statements lack detail; the exact GEO accession release date, loom‑file contents, and a DOI‑tagged Zenodo archive of analysis scripts should be added.
    4. Minor editorial issues remain, such as replacing "critical" with "crucial" in the Abstract, adding software version numbers to figure legends, and correcting the SAMtools reference.

    Significance

    The method is technically innovative and the biological insights are valuable; however, several issues-mainly concerning experimental design, statistical rigor, and functional validation-must be addressed to solidify the conclusions.

  7. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #2

    Evidence, reproducibility and clarity

    Joint analysis of multiple modalities in single cells will provide a comprehensive view of cell fate states. In this manuscript, Bhardwaj et al developed a single-cell multi-omics assay, T-ChIC, to simultaneously capture histone modifications and full-length transcriptome and applied the method on early embryos of zebrafish. The authors observed a decoupled relationship between the chromatin modifications and gene expression at early developmental stages. The correlation becomes stronger as development proceeds, as genes are silenced by the cis-spreading of the repressive marker H3k27me3. Overall, the work is well performed, and the results are meaningful and interesting to readers in the epigenomic and embryonic development fields. There are some concerns before the manuscript is considered for publication.

    Major concerns:

    1. A major point of this study is to understand embryo development, especially gastrulation, with the power of scMulti-Omics assay. However, the current analysis didn't focus on deciphering the biology of gastrulation, i.e., lineage-specific pioneer factors that help to reform the chromatin landscape. The majority of the data analysis is based on the temporal dimension, but not the cell-type-specific dimension, which reduces the value of the single-cell assay.
    2. The cis-spreading of H3K27me3 with developmental time is interesting. Considering H3k27me3 could mark bivalent regions, especially in pluripotent cells, there must be some regions that have lost H3k27me3 signals during development. Therefore, it's confusing that the authors didn't find these regions (30% spreading, 70% stable). The authors should explain and discuss this issue.

    Minors:

    1. The authors cited two scMulti-omics studies in the introduction, but there have been lots of single-cell multi-omics studies published recently. The authors should cite and consider them.
    2. T-ChIC seems to have been presented in a previous paper (ref 15). Therefore, Fig. 1a is unnecessary to show.
    3. It's better to show the percentage of cell numbers (30% vs 70%) for each heatmap in Figure 2C.
    4. Please double-check the citation of Fig. S4C, which may not relate to the conclusion of signal differences between lineages.
    5. Figure 4C has not been cited or mentioned in the main text. Please check.

    Significance

    Strengths: This work utilized a new single-cell multi-omics method and generated abundant epigenomics and transcriptomics datasets for cells covering multiple key developmental stages of zebrafish. Limitations: The data analysis was superficial and mainly focused on the correspondence between the two modalities. The discussion of developmental biology was limited.

    Advance: The zebrafish single-cell datasets are valuable. The T-ChIC method is new and interesting.

    The audience will be specialized and from basic research fields, such as developmental biology, epigenomics, bioinformatics, etc.

    I'm more specialized in the direction of single-cell epigenomics, gene regulation, 3D genomics, etc.

  8. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #1

    Evidence, reproducibility and clarity

    The authors have a longstanding focus and reputation on single cell sequencing technology development and application. In this current study, the authors developed a novel single-cell multi-omic assay termed "T-ChIC" so that to jointly profile the histone modifications along with the full-length transcriptome from the same single cells, analyzed the dynamic relationship between chromatin state and gene expression during zebrafish development and cell fate determination. In general, the assay works well, the data look convincing and conclusions are beneficial to the community.

    There are several single-cell methodologies all claim to co-profile chromatin modifications and gene expression from the same individual cell, such as CoTECH, Paired-tag and others. Although T-ChIC employs pA-Mnase and IVT to obtain these modalities from single cells which are different, could the author provide some direct comparisons among all these technologies to see whether T-ChIC outperforms?

    In current study, T-ChIC profiled H3K27me3 and H3K4me1 modifications, these data look great. How about other histone modifications (eg H3K9me3 and H3K36me3) and transcription factors?

    T-ChIC can detect full length transcription from the same single cells, but in FigS3, the authors still used other published single cell transcriptomics to annotate the cell types, this seems unnecessary?

    Throughout the manuscript, the authors found some interesting dynamics between chromatin state and gene expression during embryogenesis, independent approaches should be used to validate these findings, such as IHC staining or RNA ISH?

    In Fig2 and FigS4, the authors showed H3K27me3 cis spreading during development, this looks really interesting. Is this zebrafish specific? H3K27me3 ChIP-seq or CutTag data from mouse and/or human embryos should be reanalyzed and used to compare. The authors could speculate some possible mechanisms to explain this spreading pattern?

    Significance

    The authors have a longstanding focus and reputation on single cell sequencing technology development and application. In this current study, the authors developed a novel single-cell multi-omic assay termed "T-ChIC" so that to jointly profile the histone modifications along with the full-length transcriptome from the same single cells, analyzed the dynamic relationship between chromatin state and gene expression during zebrafish development and cell fate determination. In general, the assay works well, the data look convincing and conclusions are beneficial to the community.

  9. Version published to 10.1101/2024.09.23.614335 on bioRxiv