Jo et al. use a combination of micropatterned differentiation, single cell RNA sequencing and pharmacological treatments to study primordial germ cell (PGC) differentiation starting from human pluripotent stem cells. Geometrical confinement in conjunction with a pre-differentiation step allowed the authors to reach remarkable differentiation efficiencies. While Minn et al. already reported the presence of PGC-like cells in micropatterned differentiating human cultures by scRNA-Seq (as acknowledged by the authors), the careful characterization of the PGC-like population using immunostainings and scRNA-Seq is a strength of the manuscript. The attempt at mechanistically dissecting the signaling pathways required for PGC fate specification is somehow weaker. The authors do not present sufficient evidence supporting the ability to specify PGC fate in the absence of Wnt signaling and the importance of the relative signaling levels of BMP to Nodal pathways; the wording of the text should be amended to better reflect the presented evidence or the authors should perform additional experiments to support these claims. Show We thank the reviewer for this comment. As described in more detail in the responses below, we have significantly strengthened the evidence for the rescue of Wnt inhibition by exogenous Activin treatment and have nuanced our interpretation. We believe that our data suggest low levels of Wnt may be required directly for PGC competence, while much higher levels are required indirectly to induce Nodal, with Nodal signaling being the limiting factor for PGC specification under the reference condition with BMP4 treatment only. We describe this in detail in the manuscript but summarize it in Author response image 1 in a simplified diagram: We have also carried out additional experiments that match model predictions demonstrating the importance of relative BMP and Nodal signaling levels and amended the text to reflect the evidence as suggested. More details are provided below.
We believe the mechanism by which cells confined to small colonies differentiate to PGCLCs more efficiently is explained by a larger fraction of the cells being exposed to the necessary levels of BMP and Nodal signaling. In large colonies BMP signaling was shown to be restricted to a distance of 50-100 μm from the colony edge through receptor localization and secretion of inhibitors (Etoc et al., Dev Cell 2016). From this one would expect that BMP signaling extends a similar distance from the edge in small colonies, so that a larger fraction of cells are receiving the BMP signal needed to differentiate to PGCLCs. Because it was not previously shown that the length scale of BMP signaling and downstream signals are preserved as colony size is reduced, we have now included an analysis of BMP signaling (pSmad1 levels) and Nodal signaling (nuclear Smad2/3 levels) as a function of colony size (Figure 5i-k). This confirms our hypothesis and provides a potential mechanism.
We thank the reviewer for this suggestion. Experimentally we have now tested the effect of 5x5 = 25 different combinations of BMP and Activin doses on PGCLC differentiation. We then challenged the mathematical model to predict the ‘phase diagram’ corresponding to this data with good agreement (Figure 6f). It is important to note here that the model was fit using only data with 50ng/ml of BMP, making this a true prediction. We also point out that the phase diagram predicted in this way is different from the one shown in Figure 6d, not only because of the lower resolution, but because Figure 6f shows the steady state after uniform stimulation in space and time (i.e. the response on the very edge), whereas the predicted phase diagram shows average expression at 42h in a 100um range from the colony edge using the previously measured spatiotemporal gradients of BMP and Activin response. Finally, the data in Figure 6f shows mean expression levels as opposed to the percentage double positive cells for the same data in Figure 4q because our model does not simulate individual cells and noise, only allowing us to compare mean expression. We explain all this in the text now. As a minor change to facilitate comparison of data and model we have now plotted the concentrations of BMP and Activin in Figure 6 rather than the scaled model parameters from 0 to 1, we also further optimized the model parameters without qualitative changes.
We have significantly extended our analysis of the effect of WNT inhibition and subsequent rescue of PGCs by Activin treatment. This includes staining for TFAP2C,NANOG,PRDM1 and staining for LEF1 as a measure of WNT signaling. Figure 4 and Figure 4—figure supplement 1 now also include treatment with IWR-1, a different small molecule inhibitor of WNT signaling, as well inhibition by IWR-1 and IWP2 at different times and different doses.
As suggested, we have now included a range of BMP concentrations. The reduction in PGCs at lower BMP doses is in line with our model and does not contradict a dependence on the relative signaling levels of BMP and Nodal by which we mean that optimal dose of Activin for PGCLC specification depends on the level of BMP and vice versa. We have amended the text to state this more clearly.
We thank the reviewer for pointing this out, the colors match now.
We thank the reviewer for pointing out this mistake and we have corrected it.
The revised manuscript no longer contains this panel but shows scatterplots for two conditions instead (Figure 5fg).
We believe that the discrepancies are due to timing. PGCLCs in our manuscript at 42h are likely less mature than those at day 2 (48h) in Kojima et al. We examined PGCLCs in our system a day later and found that they express NANOG and POU5F1 at much higher levels than pluripotent cells, similar to those in Kojima at 48h. We now show this with immunofluorescence staining for NANOG and POU5F1 at 48h, 72h, 96h in Figure 2hg and Figure 2—figure supplement 2, and have added a discussion regarding NANOG and POU5F1 levels to reflect these observations. We also collected scRNA-seq data at 48, 72, and 96 hours and similarly found that NANOG and POU5F1 substantially increase over time in PGCLCs relative to pluripotent cells, including from 42h to 48h (we show the data and provide further discussion in the next section of this response). The fact that our PGCLCs initially express NANOG and POU5F1 at levels similar to pluripotent epiblast cells is consistent with immunofluorescence data on nascent PGCs in cynomolgus monkeys, where the earliest PGCs appear to express NANOG at levels similar to the epiblast (Sasaki et al. Dev Cell 2016 Figure S5B).
In summary: We carried out scRNA-seq analysis comparing our micropatterned hPGCLCs to ‘conventional’ hPGCLCs (Chen et al., Cell Rep 2019), and to PGCs from the CS7 human gastrula (Tyser et al., Nature 2021). We conclude that our PGCLCs appear at least as similar to in vivo PGCs as other hPSC-derived PGCLCs. We discuss the details of our analysis below. Extending our analysis of NANOG and POU5F1, we first compared scRNA-seq data that we collected for micropatterned hPGCLCs at 42, 48,72, and 96h across the full panel of marker genes used to establish PGC identity in Figure 4e of Kobayashi et al., Nature 2017. We also added PDPN and KIT to this panel as they are also commonly used cell surface markers for PGCs (e.g. Sasaki et al. Dev Cell 2016). In addition to NANOG and POU5F1 we also see a clear increase in other early PGC markers over time. We share this data in Author response image 2 but have not included it in the main manuscript because the full dataset is incredibly rich and describing it in detail is beyond the scope of this manuscript, so we are in the process of writing a separate manuscript about it. Micropatterned PGCLCs.We see a similar gene signature as in Kobayashi d4 hPGC+Cy with a few discrepancies. First PGCs in Kobayashi also appear to express some mesoderm and endoderm markers, which may represent imperfect FACS sorting since their data is bulk RNA-seq, or represent a difference in the induction strategy. Furthermore, we appear to lack DND1 and DPPA3 and SYCP3 in comparison to their d4 hPGC+Cy. This again may reflect a difference in experimental setup, for example we see small amounts of DDPA3 transcripts in the mesoderm but not the early PGCs and the bulk RNAseq may combine those. Therefore, we also compared our scRNA-seq data with that for hPGCLCs from Chen et al. Cell Rep 2019 and the Tyser et al. Nature 2021 CS7 human gastrula (in vivo data) and processed each sample in the same way (Author response image 3): We see that the transcriptional profiles of the PGCLCs from Chen et al. are very similar to ours and to the profile from the CS7 human gastrula. It is difficult to say with certainty whether differences are due to technical limitations or biology. For example, we appear to be missing low levels of DND1 relative to Chen et al., but in Author response image 2 we are showing z-scores, i.e. relative expression compared to the other cell populations in the dataset. If we look at absolute expression in Author response image 4 we see that DND1 is barely expressed in Chen et al. and for us may have simply been below the detection threshold, since sequencing depth in scRNAseq is typically on the order of one read per gene per cell. As another example, one might conclude from Author response image 3 that our PGCLCs upregulate KLF4 and TFAP2C more robustly than Chen et al., but again that is not supported when looking at absolute expression. Overall, we prefer to not overinterpret the data and view the expression profiles as very similar. In that regard it is also striking that the human embryo data lacks expression significant expression of PRDM1 in the PGCs. This may again simply be due to noise, especially since their PGC population is only 8 cells. It may also be a clustering error: PRDM1+TFAP2C+ cells are present in that dataset but not annotated as PGC but rather as non-neural/amniotic ectoderm. Again, similar considerations apply to other differences.
We thank the reviewer for pointing this out. We found that there was an error in the figure since the correlation was calculated on the wrong set of highly variable genes. It is now calculated on the intersection of the top 10% most highly variable genes from each data sets. With the new calculation PGCLCs in 42h micropatterned colonies correlate most strongly to PGCs, but the PGCLCs still show relatively low correlation with in vivo PGC and relatively high correlation with non-neural ectoderm (amniotic ectoderm) and endoderm. In summary: we believe that this is also due to a combination of (1) our PGCLCs at 42h being nascent / relatively immature and (2) because of technical limitations in how correlations between datasets are being calculated. To support that low correlation at 42h has to do with the developmental stage of our PGCLCs relative to the CS7 PGCs, we calculated correlation in an identical way for our 48, 72, 96h timepoints (Author response image 5). We see dramatic increases in correlation between PGCLCs and PGCs and relative decreases in correlation with other cell types. The correlation of PGCLCs at 42h with non-neural/amniotic ectoderm and endoderm may hint at shared early transcriptional profiles, consistent with our model placing PGC in between these cell types (Figure 6d). While relative correlations in these Author response images are suggestive, absolute correlation should not be quantitatively compared with numbers elsewhere, since correlations are calculated on normalized expression (z-score) and are highly sensitive to the overall cell content of the dataset (e.g. the PGCLC transcriptome after normalization will be different depending on whether endodermal cells are present or not in the same dataset) as well as various preprocessing and integration steps. Our previous calculations of correlation in the manuscript did not integrate datasets because we have found results to strongly depend on the integration method. However, to illustrate the effect we did integrate our 42h micropattern data with the human gastrula data using the most popular method: Seurat and have included this in Figure 1—figure supplement 2f. As might be expected correlations after integration are much higher, with PGCLC-PGC correlation going from 0.24 to 0.73. To give basis of comparison of our PGCLCs to other PGCLCs, we also calculated the correlation matrices of the data from Chen et al. Cell Rep 2019 using the exact same methods as ours, without integration (Author response image 6). These should be compared to our correlation matrices in Author response image 5 and in Figure 2f, left. We did not fully annotate this dataset and only labeled the PGC cluster, which is unambiguous. We see that by 96h, PGCLC-PGC correlations are similar to, but in fact somewhat lower than for our system. We conclude that our PGCLCs appear at least as similar to in vivo PGCs as other hPSC-derived PGCLCs.
Indeed, the expression levels are not uniform in each cell population. Gene expression levels are heterogeneous because of biological noise and on top of that there is measurement noise. Moreover, because of the spatiotemporal dynamics of signaling, cells may be at different stages of differentiation in different positions adding more heterogeneity. Nevertheless, coloring scatterplots for density shows clearly defined populations when population sizes are similar, as happens on the small micropatterns, and we now show this in Figure 5—figure supplement 1c (if another population is much larger as is the case for large micropatterns, it is hard to see the density in the smaller PGC cluster). We believe our observations are generally expected and have no knowledge of any study showing single cell quantitative analysis of protein expression in this manner where the studied cell populations are found to have more uniform expression.
We thank the reviewer for this point of caution. We are aware of the risks of denoising/imputation and believe we did make careful use of it. None of our conclusions rely on MAGIC. As stated explicitly in the methods section, MAGIC was used for two purposes only: 1) Visualization in Figure 1. It is easier to see the gene expression patterns on PHATE maps after some smoothing with MAGIC. We have now included the same data without MAGIC in Figure 1—figure supplement 2d and have also include raw data in Figure 1—figure supplement 2e. 2) Relationships between pairs of genes in Figure 1—figure supplement 1k,n. While none of our conclusions rest on these figures, we believe that they actually demonstrate the value of imputation, because while raw single cell RNA-seq data is so noisy one cannot discern any relationship between any pair of genes, denoising yields relationships between genes that closely resemble those found from immunofluorescence data providing cross-validation for both approaches. We also discuss the differences found in the main text and state imputation artefacts are one possible explanation for these differences. Again, we have now included raw data in Figure 1—figure supplement 1lo to emphasize this. We only used magic for these two purposes and have revised the text to make this clear. For all other analysis, such as cluster comparison with heatmaps (e.g. Figure 1k), and differential expression analysis, we used the raw data, without MAGIC.
We have now included CT values relative to GAPDH in Figure 2—figure supplement 1e.
As also described above in response to a comment by reviewer 2, we have significantly extended and nuanced our analysis of the effect of WNT inhibition and subsequent rescue of PGCs by Activin treatment. Figure 4 and Figure 4—figure supplement 1 now include treatment with IWR-1, a different small molecule inhibitor of WNT signaling, as well inhibition by IWR-1 and IWP2 at different times and different doses.
We have now explicitly stated in the methods section that mTeSR1 contains FGF2 and TGFβ and added additional description of the differentiation protocol.
We have removed several instances of these and similar words.
We have changed “established” to “suggested”. The work we referred to is now published in Nature Communications.
We agree and thank the reviewer for this comment. We have added a paragraph relating our results to in vivo findings. References Chen, Di, Na Sun, Lei Hou, Rachel Kim, Jared Faith, Marianna Aslanyan, Yu Tao, et al. 2019. “Human Primordial Germ Cells Are Specified From Lineage-Primed Progenitors..” Cell Reports 29 (13): 4568–4582.e5. doi:10.1016/j.celrep.2019.11.083. Etoc, Fred, Jakob Metzger, Albert Ruzo, Christoph Kirst, Anna Yoney, M Zeeshan Ozair, Ali H Brivanlou, and Eric D Siggia. 2016. “A Balance Between Secreted Inhibitors and Edge Sensing Controls Gastruloid Self-Organization..” Developmental Cell 39 (3): 302–15. doi:10.1016/j.devcel.2016.09.016. Kobayashi, Toshihiro, Haixin Zhang, Walfred W C Tang, Naoko Irie, Sarah Withey, Doris Klisch, Anastasiya Sybirna, et al. 2017. “Principles of Early Human Development and Germ Cell Program From Conserved Model Systems..” Nature 546 (7658): 416–20. doi:10.1038/nature22812. Kojima, Yoji, Kotaro Sasaki, Shihori Yokobayashi, Yoshitake Sakai, Tomonori Nakamura, Yukihiro Yabuta, Fumio Nakaki, et al. 2017. “Evolutionarily Distinctive Transcriptional and Signaling Programs Drive Human Germ Cell Lineage Specification From Pluripotent Stem Cells..” Cell Stem Cell 21 (4): 517–532.e5. doi:10.1016/j.stem.2017.09.005. Sasaki, Kotaro, Tomonori Nakamura, Ikuhiro Okamoto, Yukihiro Yabuta, Chizuru Iwatani, Hideaki Tsuchiya, Yasunari Seita, et al. 2016. “The Germ Cell Fate of Cynomolgus Monkeys Is Specified in the Nascent Amnion..” Developmental Cell 39 (2): 169–85. doi:10.1016/j.devcel.2016.09.007. Tyser, R.C.V., Mahammadov, E., Nakanoh, S. et al. Single-cell transcriptomic characterization of a gastrulating human embryo. Nature 600, 285–289 (2021). https://doi.org/10.1038/s41586-021-04158-y What controls the differentiation of cells?Gene Expression Regulates Cell Differentiation.
What genes control cell differentiation during development?In the adult body, Hox genes are among others responsible for driving the differentiation of tissue stem cells towards their respective lineages in order to repair and maintain the correct function of tissues and organs.
Does DNA control cell differentiation?Answer and Explanation: The DNA basically contains genetic information that controls cell functions. In cell differentiation, the DNA determines what kind of cell the young cell will be specialized into.
What controls the differentiation of cells in a multicellular organism?Roles of DNA and RNA in Cell Differentiation
Dexoyribonucleic Acid, or DNA, controls the way cells function. It also determines what type of specialized cells will be made. Stem cells are cells that have the ability to become any type of specialized cell in the body.
|