Integrative omics analysis reveals gene regulatory mechanisms distinguishing organoid-derived hepatocytes from primary human hepatocytes

  1. Haoyu Wu
  2. Annie S. P. Yang
  3. Suzan Stelloo
  4. Floris J. M. Roos
  5. René H. M. te Morsche
  6. Anne H. Verkerk
  7. Maria V. Luna-Velez
  8. Laura Wingens
  9. Johannes H. W. de Wilt
  10. Robert W. Sauerwein
  11. Klaas W. Mulder
  12. Simon J. van Heeringen
  13. Monique M. A. Verstegen
  14. Luc J. W. van der Laan
  15. Hendrik Marks
  16. Richárd Bártfai

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

Background and Aims

Hepatic organoid cultures are considered a powerful model system to study liver development and diseases in vitro . However, hepatocyte-like cells differentiated from such organoids remain immature compared to primary human hepatocytes. Therefore, a comprehensive understanding of differences in gene regulatory mechanisms between primary human hepatocytes and hepatic organoids is essential to obtain functional hepatocyte-like cells in vitro for fundamental and therapeutic applications.

Methods

We obtained primary human hepatocytes at high purity from all zones of the liver lobule using an optimized two-step perfusion protocol. We captured the single-cell transcriptome and chromatin accessibility landscape using scRNA-seq and ATAC-seq, respectively. We identified key transcription factors and compared the gene regulatory mechanisms in primary human hepatocytes and (un)differentiated intrahepatic cholangiocyte organoids. Using siRNA-mediated perturbations, we showed the functional relevance of an organoid-enriched transcription factor during in vitro differentiation of hepatocyte-like cells.

Results

Our integrative omics analysis revealed that Activator Protein 1 (AP-1) family members cooperate with hepatocyte-specific transcription factors, including HNF4A, in maintaining cellular functionality of mature human hepatocytes. Comparative analysis identified distinct transcription factor sets specifically active in human hepatocytes and organoids. Amongst these ELF3 is unique to intrahepatic cholangiocyte organoids and its expression level negatively correlate with expression of hepatic marker genes. Functional analysis of ELF3 furthermore revealed that ELF3 depletion optimizes the formation of hepatocyte-like cells from intrahepatic cholangiocyte organoids.

Conclusions

Collectively, our integrative analysis provides insights into the transcriptional regulatory networks of human hepatocytes and hepatic organoids, thereby informing future strategies for better establishment of urgently-needed hepatic model systems in vitro .

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

    We appreciate the positive and constructive comments of the reviewers on our paper. Below please find our point-by-point response to their comments.

    Reviewer #1:

    Main comments:

    1. The expression levels of many genes, including some major TFs (like CEBPa or HNF4) in isolated primary hepatocytes greatly differ from that in normal liver. This is due to the disruption of cell-cell contacts. For this reason, single nuclei sequencing is more reliable and it is the preferred method. It is not indicated how many biological replicates were used and what level of variability was observed between different preparations.

    We thank the reviewer for pointing out the immediate response of hepatocytes to dissociation, including in expression of CEBPa or HNF4 (this reviewer) and stress-related genes (reviewer 3), which we were aware of.

    Unfortunately, however no perfect method exists to explore only hepatocytes in the context of the liver and single nuclei RNA-seq, which was not available at the start of our study, also has its limitations (e.g. substantial ambient RNA contamination, a lower median number of genes detected and potential for biases and higher doublet rates due to increased amplification steps (PMID: 34515767)).

    Importantly, in our current study, we were interested in exploring gene regulatory networks in hepatocytes by the combination of RNA-seq and ATAC-seq. In our hands, data that we obtained from single cell ATAC-seq was far too shallow and noisy to predict gene regulatory networks. Hence, we needed to rely on pure populations of hepatocytes to perform our studies with bulk ATAC-seq, for which we optimized perfusion and subsequent density gradient centrifugation. While we succeeded in obtaining a very pure hepatocyte population, we agree with the reviewer that due to dissociation-associated changes the results that we obtain might not fully reflect the events happening in hepatocytes in the liver.

    To address this issue brought up by reviewer 1 and 3, i) we will better indicate our rationale within the manuscript, and the limitations as indicated by both reviewer 1 and 3; ii) to provide an overview of potential changes that were induced by the perfusion procedure that we applied, we will compare the hepatocyte RNA-seq transcriptomes that we obtained with in vivo liver RNA-seq, with specific attention to transcription factors and stress-related genes (see reviewer 3, point 1); iii) we will better separate in the figures data obtained from hepatocytes versus data obtained from liver (see also point 2 from this reviewer).

    Additionally, we will indicate how many replicated were used, and the level of variability between different preparations (donors).

    1. The regulome studies involved analysis of ENCODE data sets (ChIP-seq), while the RNA-seq data were obtained in the current work. Due to the different source of the data (e.g primary hepatocytes used for ENCODE consortia members and this study) differences are expected. In the present study the cells were FACS-sorted immediately after isolation, while the ones used to produce ENCODE data sets were not subjected to sorting and were also probably cultured. This limits the accuracy of comparisons. Furthermore, the authors should indicate exactly which ENCODE data-sets were used.

    It is also unusual to observe broad distribution of the ATF3, JUND and EGR1 ChIP-seq reads over the PCK1 gene or the Alb gene (Fig S3). Peaks called by MACS should be indicated. Have the authors verified this distribution, e.g by ChIP-PCR or other means? It is quite unlikely that binding motifs are present all over the gene bodies. Is it possible that these factors interact with elongating RNA Pol-II complexes? What is the situation in other actively transcribing gene bodies?

    In the first paragraph of this comment, the reviewer rightfully points out that we use data from different sources in the first part of our study: scRNA-seq and ATAC-seq from perfusion-obtained hepatocytes (this study) and ENCODE ChIP-seq data which, in contrast to what the reviewer seems to assume, is obtained from liver (as profiled by ENCODE).

    We did choose to use ChIP-seq data from liver tissue to corroborate our findings in isolated hepatocytes in the tissue of origin (largely composed of hepatocytes). Indeed, the near perfect co-localization of HNF4A and ATF3/EGR1 in liver tissue and the enrichment of corresponding DNA motifs in our ATAC-seq data strongly suggests interaction between bZIP family members and hepatocyte-specific transcription factors (including HNF4A) and hence support our conclusion.

    To further address this issue, we will better separate the data obtained from hepatocytes versus data obtained from liver in the figures and include additional data for liver if available (see also point 1 from this reviewer). Additionally, we will indicate exactly which ENCODE datasets were used (see table below). Where relevant, we will explicitly mention the limitations/confounding factors of our analysis.

    EGR1-liver ChIP-seq

    ENCODE Project Consortium

    ENCFF389LQC, ENCFF132PDR

    JUND-liver ChIP-seq

    ENCODE Project Consortium

    ENCFF215GBK, ENCFF978CPC

    ATF3-liver ChIP-seq

    ENCODE Project Consortium

    ENCFF522PUA, ENCFF094LXX

    HNF4A-liver ChIP-seq

    ENCODE Project Consortium

    ENCFF302XOK, ENCFF500ZBE

    FOXA1-liver ChIP-seq

    ENCODE Project Consortium

    ENCFF765EAP, ENCFF945VNK

    CTCF-liver ChIP-seq

    ENCODE Project Consortium

    ENCFF002EXB

    RAD21-liver ChIP-seq

    ENCODE Project Consortium

    ENCFF643ZXX, ENCFF171UDL

    EGR1- K562 ChIP-seq

    ENCODE Project Consortium

    ENCFF000PZK, ENCFF000PZP

    JUND- K562 ChIP-seq

    ENCODE Project Consortium

    ENCFF000YSC, ENCFF000YSE

    ATF3- K562 ChIP-seq

    ENCODE Project Consortium

    ENCFF000PWC, ENCFF000PWA

    With respect to the second paragraph: We obtained these liver tissue ChIP-seq profiles from ENCODE, in which these have gone through thorough validation procedures. Furthermore, we do observe very similar patterns with a complementary, but independent approach, ATAC-seq in hepatocytes. Hence, we do not think that further validation by ChIP-qPCR will have much added value.

    We will follow the advice of the reviewer by i) indicating MACS peaks in our examples, ii) check whether ChIP-seq peaks in coding regions are typical for these datasets. If not, we will show better examples. If they are, we will are investigate potential motifs present in gene bodies, iii) investigate literature for a possible link between these factors and elongating RNA Pol-II complexes; and iv) investigate actively transcribing gene bodies

    1. The synergism between AP1 and HNF4 is based on RNA and ChIP data in Primary hepatocytes. The main evidence for the synergism are co-binding of the two factors and the regulome profiles in the individual cells. In ICOs where both factors are expressed at high levels ChIP-seq data are not available and the potential binding distribution is estimated by the presence of binding motifs in ATAC-seq positive areas. Considering the concern described in point 2, it is important to obtain ChIP-seq data in ICOs too.

    We would like to point out that, we make the central observations on overlapping regulatory modules in perfusion-derived hepatocytes, the ChIP-seq data to show co-binding of AP-1 and other factors with HNF4A (Fig 2c-f; Fig S3c-e) is all based on liver tissues. By showing this in the tissue or origin, we feel we provide sufficient evidence for the (potential) interplay between these factors in the liver, making ChIP-seq in ICOs redundant and beyond the scope of this study.

    In addition, more direct experimental evidence for the synergism is needed. For example, demonstrating the synergism between HNF4 and some AP1 factors in specific genes by co-transfection experiments.

    With regards to the potential synergy between HNF4 and AP1 in adult hepatocytes: previous studies have shown an essential role for c-Jun (part of AP1) in normal hematogenesis, with hepatocytes being rounded and detached in c-Jun KO mice (PMID: 8371760). This clearly shows the critical role of c-Jun in liver development and support to a potential interaction with HNF4 factors.

    Yet, we agree with the reviewers that co-transfection (or knock down) experiments would be an elegant means to further support our conclusion. Unfortunately, however, PHHs are refractory to transfection making this experiment nearly impossible. Hence, instead we will tone down our statements about cooperation between these factors, instead referring to overlapping regulatory modules and co-binding as we observe.

    1. Transcriptome comparisons between primary hepatocytes and intrahepatic cholangiocyte organoids (ICO) or ICOs cultured in hepatocyte differentiation medium (DM-ICO) were performed before (Ref. 6). These cells were derived from the same donor. In the current study ICOs were obtained from a biobank, thus they were from different donors. Differences between the expression patterns of primary cells and EM-IOC and DM-IOC organoid cultures are expected even if they derived from the same donor. In Ref.6 it is clearly demonstrated that DM-IOCs closely mimic many, but not all aspects of the liver phenotype. The present paper therefore provides only incremental new knowledge about the usefulness of organoid cultures in general. On the other hand, the scRNA-seq data with cells from the organoids point to the lack of zonation, which is an important new information, not analysed in Ref.6

    We agree with the reviewer that the EM-ICOs and DM-ICOs have been well characterized in the ground-breaking works Reference 6. Indeed, in Figure 5d of Reference 6, it is shown that DM-ICOs display more comparable expression profile to hepatocytes than EM-ICOs. However, there are also clear differences between hepatocytes and DM-ICOs, indicating incomplete differentiation of the later. In our study, we now make the important observation that the differentiation potential of ICOs at least in part depends on the expression of ELF3 (Figure 3B).

    To address this issue, we will put emphasis on the findings in Ref 6, and we will put our observations in better perspective in relation to Ref 6.

    1. In the methods section the description of ICO culture conditions are very epigrammatic. It refers to previously published protocols but also mentions the addition of BMP7 in the first round of culturing without explaining why was this important. It would be useful if the authors describe exactly the culture conditions they used. Were the ICOs from the biobank established under culture conditions described in Ref 6 or by previous protocols?

    We apologize for this being unclear. We will include this information in the revised manuscript.

    1. The results about ELF3 function are interesting and convincing. This is a novel finding and may worth to perform a global transcriptome analysis and some immunostainings with specific markers in siELF3 cells to further strengthen its regulatory role in cholangiocyte-hepatocyte conversion.

    We agree with the reviewer. To follow this up, we will perform RNA-seq during differentiation of ICOs towards hepatocytes, with and without siRNA-mediated ELF3 knockdown. This will further reveal the precise regulatory role of ELF3 in during hepatocyte differentiation.

    Reviewer #2:

    Comments:

    1. Hepatocyte nuclear factors do not form a transcription factor (TF) family, they are from different TF families: the nuclear receptor, homeobox, and forkhead TF (super)families.

    We thank the reviewer for pointing the mistakes in points 1 to 6 with regards to the naming of protein and protein families in our manuscript, we apologize for these inaccuracies. We will correct these naming and references, and check for any further inconsistencies.

    1. AP-1 is not a TF family either. It is basically a heterodimer of FOS and JUN (sub)family members, which are part of the bZIP (super)family such as C/EBPs and ATF3, which latter is related to JDP2.

    We will adapt this.

    1. EGR1 is not a bZIP protein, it is a zinc finger protein from the EGR family. Was the motif of EGRs enriched? Only the motif of C/EBPs is shown on Fig. 2D.

    We will adapt this. We will also analyze whether the motif of EGRs is enriched

    1. RAD21 is not a TF, it is part of the Cohesin ring, which is associated to the insulator-binding CTCF.

    We will adapt this.

    1. EP300 (Fig. 2A) and PPARGC1A (Fig. 3B) are not TFs, they are co-regulators, basically co-activators, which can interact with several TFs. EP300 is otherwise not so specific, its presence in the chromatin is one of the major active enhancer marks.

    We will adapt this.

    1. DNA sequence motifs are typically not specific for a single TF, rather for a TF (sub)family, so based on a motif, it is usually not possible to identify a certain TF (Fig. 3F). Are there other nuclear receptors, SOX or ETS proteins that can bind to the identified motifs? (For example, FLI1 and several other ETS proteins can bind to the motif of ELF3/EHF, or there are several DR1-binding nuclear receptor dimers like HNF4/HNF4 or PPAR/RXR.)

    We agree with the reviewer. We will analyze this and adapt the manuscript according to our findings.

    &) Although the manuscript is easy to follow and understand, it needs to be checked for grammar.

    We have asked a native speaker to proofread and adapt the manuscript.

    Reviewer #3:

    1. It is well known that perfusion of primary hepatic tissues (mice and human) results in immediate genetic responses, which will be captured right away in the performed RNASeq analysis. Stress pathways are upregulated and will normalize when the cells are put in culture for a couple of days. (Not too long, as they then undergo EMT and de-differentiate into non-parenchyma cells.) These responses can influence the expression profiles observed.

    We thank the reviewer for this comment. Please see how we will address this concern in our reply to reviewer 1, issue 1, who raised a very similar point.

    1. Why were the organoid cultures not differentiating properly into hepatocytes using different media cocktails (EM versus DM)? They seem to maintain cholangiocyte features, which questions the culture conditions used.

    We thank the reviewer for the chance to clarify this important point. We like to stress that we do use the standard differentiation protocol as published (which we will also better detail in our material methods) and it does lead to differentiation towards hepatocyte like cells (both morphologically and gene expression-wise). However, what is not highlighted in previous publications, but broadly observed in the field, is that this differentiation is far from being complete and that the extent to which proper differentiation occurs varies between organoids from different donors. In our study, we now make the important observation that the differentiation potential of ICOs at least in part depends on the expression of ELF3 (Figure 3B).

    1. The authors found the up-regulation of the AP-1 family proteins such as ATF3 and EGR1 which are known to induce apoptosis/cell death. Hepatic organoids are often found to have the un-intended necrotic core development which is caused by the oxygen diffusion matter and this issue is highly likely relevant to the size of the organoids. So, it would be advisable to specify the size of hepatic organoids (i.e., diameter) and check the necrosis-related genes.

    To follow-up on this comment of the reviewer: We will measure the size of our organoids. These organoids indeed are typically hollow inside and hence we will check the expression of necrosis related genes and adjust our conclusions accordingly.

    1. The KD approach with ELF3 in the ICOs is a good way forward, however only a minor number of hepatocellular genes are recovered, questioning the central role of ELF3 in driving the hepatocellular program. Functional assays, such as albumin release, bile acid production and CYP450 response should be coupled with the gene expression analysis.

    In line with the response to reviewer 1 (point 6) we will perform RNA-seq to better characterize ELF3 KD-associated genes expression changes including genes typical and functionally relevant for hepatocyte function (e.g. albumin release and bile acid secretion)

    1. The manuscript should be supplemented by adding the statement regarding the specific reason why a different set of donors was selected for two transcriptomics. The authors used three different donors for scRNA-seq and other two donors for the ATAC-seq. It seems better if all five donors were used for both transcriptomics analyses to reduce the inconsistent proportion of primary human hepatocytes (PHHs) from each donor. In addition, the donors which are selected should have identical genetic backgrounds for in-depth analysis of PHHs. The various backgrounds such as age, sex and ethnicity cause the transcriptional and translational heterogeneity. The authors need to explain the criteria on the selection of the donors.

    We do agree with the reviewer that ideally all experiments are performed on the same set of donors. However, PHHs are obtained from surgical margins and hence provide a very limited source, leading to different experiments being performed on different donors. Importantly, the replicates for each experiment type have been obtained from multiple donors enabling us to capture common rather than donor specific expression/chromatin accessibility signatures.

    Within the revised manuscript, we will include a paragraph on the criteria on the selection of the donors, and why a different set of donors was selected for two transcriptomics. Also, we will provide information with respect to the background of the donors.

  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

    In this study the authors aimed at characterizing the differences between primary human hepatocytes and hepatic organoids derived from human intrahepatic cholangiocytes (ICO), which can differentiate into hepatocytes, using scRNA-seq and ATAC-seq approaches. Their goal was to identify gene regulatory signatures that differ between the two models and to single out transcription factors that could drive hepatocellular functionality. They found the AP-1 family members to be associated with increased hepatic function together with known hepatocyte identity markers, such as HNF4A, FOXA1 and FOXA2. In ICOs they observed an increase of ELF3, which represent cholangiocyte-like features. KD of this factor induced the expression of known hepatocellular marker genes, such as ALB, CYP3A4, TTR, GC, and GLUL, indicating ELF3 may function as a barrier in hepatocyte differentiation.

    Although this is an interesting approach to decipher, which transcription factors are involved in the development of proper human related hepatic organoids, it requires a more thorough analysis and ideally an improvement in the culture conditions to support their claims.

    1. It is well known that perfusion of primary hepatic tissues (mice and human) results in immediate genetic responses, which will be captured right away in the performed RNASeq analysis. Stress pathways are upregulated and will normalize when the cells are put in culture for a couple of days. (Not too long, as they then undergo EMT and de-differentiate into non-parenchyma cells.) These responses can influence the expression profiles observed.
    2. Why were the organoid cultures not differentiating properly into hepatocytes using different media cocktails (EM versus DM)? They seem to maintain cholangiocyte features, which questions the culture conditions used.
    3. The authors found the up-regulation of the AP-1 family proteins such as ATF3 and EGR1 which are known to induce apoptosis/cell death. Hepatic organoids are often found to have the un-intended necrotic core development which is caused by the oxygen diffusion matter and this issue is highly likely relevant to the size of the organoids. So, it would be advisable to specify the size of hepatic organoids (i.e., diameter) and check the necrosis-related genes.
    4. The KD approach with ELF3 in the ICOs is a good way forward, however only a minor number of hepatocellular genes are recovered, questioning the central role of ELF3 in driving the hepatocellular program. Functional assays, such as albumin release, bile acid production and CYP450 response should be coupled with the gene expression analysis.
    5. The manuscript should be supplemented by adding the statement regarding the specific reason why a different set of donors was selected for two transcriptomics. The authors used three different donors for scRNA-seq and other two donors for the ATAC-seq. It seems better if all five donors were used for both transcriptomics analyses to reduce the inconsistent proportion of primary human hepatocytes (PHHs) from each donor. In addition, the donors which are selected should have identical genetic backgrounds for in-depth analysis of PHHs. The various backgrounds such as age, sex and ethnicity cause the transcriptional and translational heterogeneity. The authors need to explain the criteria on the selection of the donors.

    Significance

    General assessment: This study used two powerful transcriptomics methods. The liver zonation was considered in the analysis which is reasonable. Limitations are related to cell culture conditions and lack of validations.

    Advance: This study extends the knowledge in human in vitro model system (mostly technical, but also clinical field).

    Audience: The audience from the basic and clinical research will be interested in this study.

    The field of expertise: Liver metabolism, pathophysiology of liver diseases, pre-clinical investigation

  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

    Comments:

    1. Hepatocyte nuclear factors do not form a transcription factor (TF) family, they are from different TF families: the nuclear receptor, homeobox, and forkhead TF (super)families.
    2. AP-1 is not a TF family either. It is basically a heterodimer of FOS and JUN (sub)family members, which are part of the bZIP (super)family such as C/EBPs and ATF3, which latter is related to JDP2.
    3. EGR1 is not a bZIP protein, it is a zinc finger protein from the EGR family. Was the motif of EGRs enriched? Only the motif of C/EBPs is shown on Fig. 2D.
    4. RAD21 is not a TF, it is part of the Cohesin ring, which is associated to the insulator-binding CTCF.
    5. EP300 (Fig. 2A) and PPARGC1A (Fig. 3B) are not TFs, they are co-regulators, basically co-activators, which can interact with several TFs. EP300 is otherwise not so specific, its presence in the chromatin is one of the major active enhancer marks.
    6. DNA sequence motifs are typically not specific for a single TF, rather for a TF (sub)family, so based on a motif, it is usually not possible to identify a certain TF (Fig. 3F). Are there other nuclear receptors, SOX or ETS proteins that can bind to the identified motifs? (For example, FLI1 and several other ETS proteins can bind to the motif of ELF3/EHF, or there are several DR1-binding nuclear receptor dimers like HNF4/HNF4 or PPAR/RXR.)
    7. Although the manuscript is easy to follow and understand, it needs to be checked for grammar.

    Significance

    Haoyu Wu and his colleagues investigated the gene regulatory mechanisms contributing to human hepatocyte differentiation and maintenance integrating scRNA-seq, ATAC-seq, and ChIP-seq data and applying knock-down experiments. They differentiated the hepatocytes of the individual liver zones, identified the "lineage-determining" transcription factors of hepatocytes and intrahepatic cholangiocyte organoids, showed the co-localization of hepatocyte-specific and other, e.g., AP-1 transcription factors, and showed that the knock-down of ELF3 enhances hepatocyte characteristics. Although several findings and conclusions of the manuscript are available from the literature in some form, and some results could be interpreted better, this manuscript provides a novel insight in liver biology with results useful to the field. After thorough revision, this reviewer recommends the manuscript for publication.

  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

    In this manuscript Wu et al., present results from a comparative analysis of transcriptomes of primary hepatocytes and organoid cultures derived from intrahepatic cholangiocytes. The authors also performed scRNA-seq and ATAC-seq experiments. Using SCENIC R computational tool and ENCODE ChIP-seq data they performed regulon analysis. The main findings of the paper are the following: 1) The cell-to-cell heterogeneity of primary hepatocytes are mainly due to their zonal expression patterns within porto-central axis. 2) AP1 family of factors (JUN, FOS, ATF3 and others) co-occupy gene regulatory regions bound by HNF4 and the associated transcriptome profiles suggest that AP1 factors cooperate with liver-specific factors to regulate hepatic genes. 3) Different sets of transcription factors are active in primary hepatocytes and intrahepatic cholangiocyte organoids that were differentiated in vitro to hepatocyte-like cells. 4) Identification of ELF3 as a factor required for cholangiocyte to hepatocyte conversion. The findings are interesting, although in many cases are expected. There are some issues that need to be addressed.

    Main comments:

    1. The expression levels of many genes, including some major TFs (like CEBPa or HNF4) in isolated primary hepatocytes greatly differ from that in normal liver. This is due to the disruption of cell-cell contacts. For this reason, single nuclei sequencing is more reliable and it is the preferred method. It is not indicated how many biological replicates were used and what level of variability was observed between different preparations.
    2. The regulome studies involved analysis of ENCODE data sets (ChIP-seq), while the RNA-seq data were obtained in the current work. Due to the different source of the data (e.g primary hepatocytes used for ENCODE consortia members and this study) differences are expected. In the present study the cells were FACS-sorted immediately after isolation, while the ones used to produce ENCODE data sets were not subjected to sorting and were also probably cultured. This limits the accuracy of comparisons. Furthermore, the authors should indicate exactly which ENCODE data-sets were used. It is also unusual to observe broad distribution of the ATF3, JUND and EGR1 ChIP-seq reads over the PCK1 gene or the Alb gene (Fig S3). Peaks called by MACS should be indicated. Have the authors verified this distribution, e.g by ChIP-PCR or other means? It is quite unlikely that binding motifs are present all over the gene bodies. Is it possible that these factors interact with elongating RNA Pol-II complexes? What is the situation in other actively transcribing gene bodies?
    3. The synergism between AP1 and HNF4 is based on RNA and ChIP data in Primary hepatocytes. The main evidence for the synergism are co-binding of the two factors and the regulome profiles in the individual cells. In ICOs where both factors are expressed at high levels ChIP-seq data are not available and the potential binding distribution is estimated by the presence of binding motifs in ATAC-seq positive areas. Considering the concern described in point 2, it is important to obtain ChIPs-seq data in ICOs too. In addition, more direct experimental evidence for the synergism is needed. For example, demonstrating the synergism between HNF4 and some AP1 factors in specific genes by co-transfection experiments.
    4. Transcriptome comparisons between primary hepatocytes and intrahepatic cholangiocyte organoids (ICO) or ICOs cultured in hepatocyte differentiation medium (DM-ICO) were performed before (Ref. 6). These cells were derived from the same donor. In the current study ICOs were obtained from a biobank, thus they were from different donors. Differences between the expression patterns of primary cells and EM-IOC and DM-IOC organoid cultures are expected even if they derived from the same donor. In Ref.6 it is clearly demonstrated that DM-IOCs closely mimic many, but not all aspects of the liver phenotype. The present paper therefore provides only incremental new knowledge about the usefulness of organoid cultures in general. On the other hand, the scRNA-seq data with cells from the organoids point to the lack of zonation, which is an important new information, not analysed in Ref.6
    5. In the methods section the description of ICO culture conditions are very epigrammatic. It refers to previously published protocols but also mentions the addition of BMP7 in the first round of culturing without explaining why was this important. It would be useful if the authors describe exactly the culture conditions they used. Were the ICOs from the biobank established under culture conditions described in Ref 6 or by previous protocols?
    6. The results about ELF3 function are interesting and convincing. This is a novel finding and may worth to perform a global transcriptome analysis and some immunostainings with specific markers in siELF3 cells to further strengthen its regulatory role in cholangiocyte-hepatocyte conversion.

    Referees cross-commenting

    I fully agree with the comments of Reviewer 2. Addressing them would clearly improve the paper. I also fully agree with Reviewer 3. In my opinion, special emphasis should be put on addressing point 1, 3, 4 and 5. Properly addressing these points would also answer at least partially my concerns (Reviewer 1) described in point 1, 2, 3 and 6. I would recommend the authors focus on the above issues.

    Significance

    Strengths: combines scRNA-seq with regulome analysis to identify synergism between different classes of transcription factors.

    Weaknesses: The data sets come from different sources. Key conclusions drawn from computational analysis are not validated experimentally. Comparing expression patterns of primary cells with those of organoid cultures is risky due to a number of technical limitations.

  5. Version published to 10.1101/2023.12.05.570132v1 on bioRxiv