Single-cell transcriptomics of a dynamic cell behavior in murine airways

Curation statements for this article:
  • Curated by eLife

    eLife logo

    Evaluation Summary:

    This is an interesting manuscript presenting an ex vivo explant model that allows combining live cell imaging with single-cell transcriptomic analyses. Using mouse models with specific fluorescent reporters that can be used to characterize cellular behaviour in the transplanted tissue and mark individual cells, the authors show that this approach can be used to identify transcriptional differences between cells that differ in cellular movement features during epithelial repair after injury. This is a first step to further expanding the description of cellular heterogeneity, including cellular behavioural as well as transcriptomic features. This manuscript is of broad interest to cell biologists as it describes a new method that links cellular behaviour in intact tissues to single cell sequencing. The method, which relies on the use of a transgenic strain, was demonstrated for cell migration in mouse airway regeneration. It begins to bridge the gap between cellular and molecular phenotyping of single cells but the authors should be clearer the limitations of the technique.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 and Reviewer #3 agreed to share their name with the authors.)

This article has been Reviewed by the following groups

Read the full article See related articles

Abstract

Despite advances in high-dimensional cellular analysis, the molecular profiling of dynamic behaviors of cells in their native environment remains a major challenge. We present a method that allows us to couple the physiological behaviors of cells in an intact murine tissue to deep molecular profiling of individual cells. This method enabled us to establish a novel molecular signature for a striking migratory cellular behavior following injury in murine airways.

Article activity feed

  1. Author Response

    Reviewer #1 (Public Review):

    They established a "behavioral transcriptomics" platform as they cultured mouse primary cell explant on an apparatus, imaged the cells over time, and analyzed cells with differential physiological status by scRNA-seq. They showed evidence that the system recapitulated physiological features of airway cells, including chemical-induced damage response. They further utilized the system to isolate cells of different cellular features and analyzed gene expression through scRNA-seq. The study demonstrates an interesting establishment and application of an in vitro system mimicking in vivo.

    However, several major concerns need to be resolved.

    First, whereas the overall study seems to focus on the establishment of airway epithelial cell explant apparatus and its application, take home messages that are delivered by the authors seem to emphasize the transcriptome analysis part. The authors introduced "spatial transcriptomics" and"behavioral transcriptomics" in the abstract but it is hard to appreciate that the study resolves spatial transcriptomics. This causes unnecessary confusion. Second, probably related to the first question, it is hard to find the novelty of the study. Third, probably the last and most important part of the manuscript is to analyze the cells by Smart-seq. But the analysis was performed on the SO2 injured animal only and lacked experiment on wildtype mice. If the authors tried to prove the feasibility of the technique rather than resolving physiological mechanism here, then I would recommend explaining why wild type experiment was not performed.

    The method described in the manuscript consists of two components: a novel tissue imaging platform, and characterization of a cellular behavior. Both steps can be generalized to different tissue contexts and different cellular behaviors, respectively. We have revised the title and abstract to specify the scope of this study and have also revised the text accordingly.

    Live imaging allows us to observe cell behaviors in intact tissues but does not provide information on cell type. By profiling cells that are observed by live imaging to share a behavior at single-cell resolution rather than bulk, we can separate out sources of transcriptional variation, like cell type identity, in order to identify the transcriptional signatures that reflect cell behaviors.

    Single-cell sequencing (via Smart-seq) has been previously performed in wild-type mouse trachea (Montoro et al., 2018), and identified underlying cellular heterogeneity. However, the steady state tracheal epithelium is largely quiescent, characterized by slow turnover and a lack of visible cell motility. We performed daily imaging of trachea explants from uninjured mice over 4 days and did not observe any significant displacement of epithelial cells. Furthermore, we also imaged an uninjured explanted tracheal epithelium every 40 minutes for over 19 hours with no significant directional movement (new Movie 3). We added the following text to the manuscript: “Imaging of trachea explant controls from uninjured mice over 19 hours revealed no cellular displacement in the airway epithelium (Movie 3).”

    In contrast, regeneration activates cell motility followed by cell proliferation. Therefore, we chose tissue regeneration as the more suitable biological context for this study to examine cellular dynamics. We leveraged the gene signatures derived from the previous wild-type study (Montoro et al., 2018) to identify different cell types and make like-for-like comparisons. We used an independent regeneration dataset in the same tissue but with a different injury model (Plasschaert et al., 2018) to test whether the molecular signatures derived in our study that differentiate moving and non-moving cells are generalizable to other contexts.

    Reviewer #2 (Public Review):

    Kwok et al. devise a method that uses a transgenic mouse line to make the link between cell behaviour in intact living tissue and subsequent dissociation into distinct groups forsingle cell sequencing. Specifically, they set up a mouse airway culture system in which it is possible to maintain live cells for multiple days and then preserve the same tissue. The analysed tissue section can be fixed and known cell types identified via classical staining protocols. In this system they imaged a number of tissue phenotypes such as ciliary beating, mucociliary clearing and airway regeneration. With respect to airway regeneration they observe that there was cellular heterogeneity between cells with the capacity to move and so-called non-movers, which the authors were able to quantitively track.To make the link with single cell sequencing, they use the Kaede transgenic mouse lines,which contains a green fluorescent reporter gene, that can be converted into a red fluorescent reported gene by illuminating a defined tissue section, in this case regions enriched for movers or non-movers. After dissociation of the tissue, cells were FACSsorted using the reporter protein. Subsequent single cell RNAseq revealed distinct gene signatures that were associated with the mover versus the non-mover phenotype. These phenotypes could also be detected in previously published data sets.

    The conclusions of the paper are supported by the data that is presented, but the comparison to existing mouse injury data could be improved. A weakness of the paper is the implication that the technique can be used for any of the phenotypes that they have examined. However, in order to be assessed by this method,there need to be a reasonably large number of cells that show similar behaviour in a region that can be photoconverted. If it is indeed possible to do the photoconversion at the single cell level, the authors should demonstrate that such resolution is possible, or otherwise clearly state this limitation of the technique they have developed.

    We recognize that the approach in this study does not involve photoconversion at single-cell resolution. While single-cell photoconversion and subsequent intermittent live imaging has been demonstrated in other systems such as zebrafish (Green and Smith, 2018) and mouse skin (Park et al., 2017), the throughput of doing downstream single-cell analysis would be limited, especially in a cell type-specific manner. Having observed a relatively homogeneous behavior of cells within a small region (~200 μm diameter, Movie 1 and Movie 2) of the airway epithelium, we photoconverted a small area with several hundred cells. Subsequent single cell sequencing allowed us to compare differences in gene expression between basal cells of slow/non-moving regions to basal cells of fast/moving regions.

    Reviewer #3 (Public Review):

    In this manuscript, the authors identify a pressing need to couple visualized in situ cell behaviour with deep molecular profiling of visualized cells, aiming to move beyond inferences made from time-lapse tissue sampling approaches or the analysis of transcriptional kinetics to identify the molecular pathways that drive cellular behaviour in situ. The authors identify live cell imaging combined with deep molecular profiling of theimaged cells as one possible solution. To this end, the authors establish a novel platform for live cell imaging of tracheal epithelial cells using explants of mouse trachea that allows long-term visualization of cell behaviour, and try to couple live-cell imaging to the transcriptional cell states.

    Combining single-cell RNA-seq analyses with live cell imaging offers the unique opportunity to link transcriptional and anatomic, morphological or movement phenotypes of individual cells. To be able to do this in intact tissues at baseline and in response to injury would allow a far more detailed and integral analysis of cellular behaviour in their physiological context. As such, the approach of the authors is interesting and clearly focused on achieving this goal. The only data that can support a claim of successfully achieving this ambitious goal are presented in figure 3, where an advanced mouse model(the Kaede-Green mouse) is used that allows labelling individual cells by photo-conversion, followed by isolation of individual cells by flow cytometry and plate-basedsc RNA-seq analysis of sorted cells. By taking this approach, the authors are able to identify transcriptional differences at the group level between tracheal epithelial cell subsets that differ in their movement after injury.

    While this in itself is a remarkable accomplishment, and an interesting observation, the relationship between the 'behaviour' of the cells observed with live cell imaging (the movement after injury) versus the transcriptional phenotype remains rather elusive. One explanation could be that active movement of cells depends on a specific transcriptional program, that is lacking from the non-moving cells. Another explanation could be that the tracheal epithelial cells are inherently heterogeneous, and one subset has the capacity to move whereas others do not, and the transcriptional profile merely identifies these heterogeneous populations. The observation that non-mover cell populations contain both basal and club cells, whereas mover regions only have basal cells seems to support this notion to some extent. However, the authors then claim to use basal-cell derived signatures (excluding the club cells) from mover and non-mover regions and compare this to literature data from another injury model to show that these signatures also identify distinct subsets in a mouse model of polidocanol-induced injury. How the distinction basal vs club cells in the non-mover regions is made remains unclear, and would seem challenging from the number of cells analyzed (as presented in figure 3).

    The identification of two behavioural phenotypes of basal cells (mover vs non-mover) in this manuscript is based on group-level phenotypes: the cells belong to a region of moversor a region of non-movers. This is relevant for figures 2 (including supplemental) and 3. In figure 2 supplemental 2C, it seems evident that within one region (or focussing only on all moving regions?), the behaviour of all cells within that region/selection is quite uniform:the variation is really very limited, and all cells seem to speed up and slow down in a highly coordinated fashion within the selected regions shown. At the same time, in figure2D, the distribution of regions across speed categories at 26-36 hours pi (the peak of the movement in suppl 2C) seems almost bimodal, with regions belonging either to non-mover(range 0.5 - 2.5 uM/hr) or mover (range 3.0-7.0 uM/hr) phenotypes. However, all regions display an increased movement at 16h pi compared to the pre-injury movements (Figure2C), indicating that all cells will be induced to induce movement to some extent.

    My main concern with this analysis is that the behavioural phenotype of the epithelial cells is assumed to be homogeneous within each region, allowing a contrast to be made in figure3 for the transcriptional phenotypes on the basis of moving phenotypes rather than on the basis of the main variation within the dataset.

    For instance, from the t-SNE plot (3B) - for what it's worth of course - and the heatmap (3C) there seems to be at least one non-mover cell that transcriptionally has a higher resemblance to the mover cells than to the other non-mover cells. Of course that can just be the variability present in the dataset, but it could also indicate that non-mover regions are not completely homogeneous, and even more so, that the moving vs non-moving associated transcriptional phenotype is a gradual transition rather than 2 clearly separate sub-phenotypes.

    All-in-all, this manuscript describes an interesting technical advance and shows some of the applications thereof. However, the approach also has its limitations: The requirement to mark cells with specific behavioural features for follow-up transcriptomic analysis (such as by photoconversion) necessitates the division of the epithelial cells into major categories on the basis of certain cellular phenotypes (such as movement) that can be visualized by live cell imaging. This limits the analysis opportunities to group-based contrasts in cellular behaviour as also used here by the authors.

    Also, the use of explanted tissue is of course less ideal than in vivo imaging, but most likely the only technically feasible approach at this moment. At the same time, the capacity to combine image-based features with single-cell transcriptomic data is an important advance, even when initially only possible in explanted tissue from mouse models carrying all kinds of fluorescent reporters. To strengthen the manuscript, it would therefore be important to discuss the limitations of the approach, as well as to provide a more comprehensive overview of the possible applications that the authors foresee.

    We thank the reviewer for the feedback. Our data demonstrates that the movement behavior is an injury-induced phenotype. 24 hours after injury (hpi), the “mover” transcriptional program is transiently enriched, while the “non-mover” transcriptional program is also transiently decreased, consistent with a cell state that is induced by injury (see Figure 4A, 24-hpi).

    SO2 removes nearly all the luminal cells (Rock et al., 2009) so we removed the club cells to compare injury response in basal cells. Distinguishing basal vs club cells is done by hierarchical clustering and comparison to established cell type signatures (Montoro et al., 2018). We apologize that the initial presentation did not make this clear. In the revised manuscript, we have provided an additional figure supplement demonstrating the hierarchical clustering (Figure 3 - figure supplement 1A), and the disjoint expression of canonical markers Krt5 (basal) and Scgb1a1 (club), which enabled us to assign unambiguous cell-type identities to discovered clusters (Figure 3 - figure supplement 1B).

    We agree with the reviewer that all cells, including cells that we classified as “mover” and “nonmover” are induced to move compared to pre-injury as suggested by Figure 2c. However, “mover” and “non-mover” cells differ dramatically in the amplitude and collective directionality of movement. We investigated the movement phenotypes in detail, including high-resolution imaging at shorter time intervals (10 min). We found that the slow “non-movers” had a large circular directionality variance (akin to oscillations), whereas the rapid “movers” moved directionally across the field of view. We quantified this with particle image velocimetry in Figure 2 – figure supplement 3C-D, and we revised the text to provide additional details about this result.

    The reviewer also raises concern about whether the movement is homogeneous enough to account for the variation in the datasets. We used our imaging data to determine the time points in which the mover and non-mover phenotypes varied the most (around 40 hrs post injury) between different regions (Figure 2 - figure supplement 2A, C) but we have also demonstrated that the movement within each region is indeed relatively homogeneous (~200 μm diameter, Movie 1 and Movie 2).

    We acknowledge that the presented data did not eliminate the possibility of another main variation within the dataset. We now perform PCA on the dataset, which confirmed that while the first principal component (PC) is associated with a solitary pulmonary neuroendocrine cell, the second PC is strongly associated with the difference between moving and non-moving cells (p=0.003, Wald test). When analyzing only the basal cells, we find that PC-1 provides a very clean separation and overlaps perfectly with the moving vs non-moving distinction (p<2 x 10-16, Wald test, Figure 3 - figure supplemental 2a). Taken together, with this additional analysis we can confirm that our focus on this behavioral phenotype reflects the main variation within the dataset.

    We appreciate the reviewer’s nuanced question about the single outlier cell. While we do observe a transcriptional phenotype that is clearly distinct, as the reviewer points out, there is a very small degree of overlap between the two cell type clusters visible on the t-SNE plot in Figure 3B. Given that the physical process of movement is a matter of degree, it is possible that this particular cell is simply not moving as much, and thus activating movement-related transcriptional programs to a lower degree. To analyze this question further in response to this question, we analyzed the separability of these groups by training a machine learning (k-nearest neighbor) classifier to distinguish these clusters (new Figure 3 - figure supplement 2b). We found that the groups could be distinguished with a high accuracy of 98.7% (95% CI: 92.7-99.9) using 5 or more of the signature genes that we defined in Figure 3C. This additional analysis we continue to conclude that while the groups have a very small degree of overlap, the moving and non-moving phenotypes are strongly separable.

    We acknowledge the limitations of this approach to groups of cells (see response to Reviewer 1) and both the limitations and advantages of using a tissue rather than cells, and we added these points to the discussion section.

  2. Evaluation Summary:

    This is an interesting manuscript presenting an ex vivo explant model that allows combining live cell imaging with single-cell transcriptomic analyses. Using mouse models with specific fluorescent reporters that can be used to characterize cellular behaviour in the transplanted tissue and mark individual cells, the authors show that this approach can be used to identify transcriptional differences between cells that differ in cellular movement features during epithelial repair after injury. This is a first step to further expanding the description of cellular heterogeneity, including cellular behavioural as well as transcriptomic features. This manuscript is of broad interest to cell biologists as it describes a new method that links cellular behaviour in intact tissues to single cell sequencing. The method, which relies on the use of a transgenic strain, was demonstrated for cell migration in mouse airway regeneration. It begins to bridge the gap between cellular and molecular phenotyping of single cells but the authors should be clearer the limitations of the technique.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 and Reviewer #3 agreed to share their name with the authors.)

  3. Reviewer #1 (Public Review):

    They established a "behavioral transcriptomics" platform as they cultured mouse primary cell explant on an apparatus, imaged the cells over time, and analyzed cells with differential physiological status by scRNA-seq. They showed evidence that the system recapitulated physiological features of airway cells, including chemical-induced damage response. They further utilized the system to isolate cells of different cellular features and analyzed gene expression through scRNA-seq. The study demonstrates an interesting establishment and application of an in vitro system mimicking in vivo.

    However, several major concerns need to be resolved. First, whereas the overall study seems to focus on the establishment of airway epithelial cell explant apparatus and its application, take home messages that are delivered by the authors seem to emphasize the transcriptome analysis part. The authors introduced "spatial transcriptomics" and "behavioral transcriptomics" in the abstract but it is hard to appreciate that the study resolves spatial transcriptomics. This causes unnecessary confusion. Second, probably related to the first question, it is hard to find the novelty of the study. Third, probably the last and most important part of the manuscript is to analyze the cells by Smart-seq. But the analysis was performed on the SO2 injured animal only and lacked experiment on wild type mice. If the authors tried to prove the feasibility of the technique rather than resolving physiological mechanism here, then I would recommend explaining why wild type experiment was not performed.

  4. Reviewer #2 (Public Review):

    Kwok et al. devise a method that uses a transgenic mouse line to make the link between cell behaviour in intact living tissue and subsequent dissociation into distinct groups for single cell sequencing. Specifically, they set up a mouse airway culture system in which it is possible to maintain live cells for multiple days and then preserve the same tissue. The analysed tissue section can be fixed and known cell types identified via classical staining protocols. In this system they imaged a number of tissue phenotypes such as ciliary beating, mucociliary clearing and airway regeneration. With respect to airway regeneration they observe that there was cellular heterogeneity between cells with the capacity to move and so-called non-movers, which the authors were able to quantitively track.

    To make the link with single cell sequencing, they use the Kaede transgenic mouse lines, which contains a green fluorescent reporter gene, that can be converted into a red fluorescent reported gene by illuminating a defined tissue section, in this case regions enriched for movers or non-movers. After dissociation of the tissue, cells were FACS sorted using the reporter protein. Subsequent single cell RNAseq revealed distinct gene signatures that were associated with the mover versus the non-mover phenotype. These phenotypes could also be detected in previously published data sets.

    The conclusions of the paper are supported by the data that is presented, but the comparison to existing mouse injury data could be improved.

    A weakness of the paper is the implication that the technique can be used for any of the phenotypes that they have examined. However, in order to be assessed by this method, there need to be a reasonably large number of cells that show similar behaviour in a region that can be photoconverted. If it is indeed possible to do the photoconversion at the single cell level, the authors should demonstrate that such resolution is possible, or otherwise clearly state this limitation of the technique they have developed.

  5. Reviewer #3 (Public Review):

    In this manuscript, the authors identify a pressing need to couple visualized in situ cell behaviour with deep molecular profiling of visualized cells, aiming to move beyond inferences made from time-lapse tissue sampling approaches or the analysis of transcriptional kinetics to identify the molecular pathways that drive cellular behaviour in situ. The authors identify live cell imaging combined with deep molecular profiling of the imaged cells as one possible solution. To this end, the authors establish a novel platform for live cell imaging of tracheal epithelial cells using explants of mouse trachea that allows long-term visualization of cell behaviour, and try to couple live-cell imaging to the transcriptional cell states.

    Combining single-cell RNA-seq analyses with live cell imaging offers the unique opportunity to link transcriptional and anatomic, morphological or movement phenotypes of individual cells. To be able to do this in intact tissues at baseline and in response to injury would allow a far more detailed and integral analysis of cellular behaviour in their physiological context. As such, the approach of the authors is interesting and clearly focused on achieving this goal. The only data that can support a claim of successfully achieving this ambitious goal are presented in figure 3, where an advanced mouse model (the Kaede-Green mouse) is used that allows labelling individual cells by photo-conversion, followed by isolation of individual cells by flow cytometry and plate-based scRNA-seq analysis of sorted cells. By taking this approach, the authors are able to identify transcriptional differences at the group level between tracheal epithelial cell subsets that differ in their movement after injury.

    While this in itself is a remarkable accomplishment, and an interesting observation, the relationship between the 'behaviour' of the cells observed with live cell imaging (the movement after injury) versus the transcriptional phenotype remains rather elusive. One explanation could be that active movement of cells depends on a specific transcriptional program, that is lacking from the non-moving cells. Another explanation could be that the tracheal epithelial cells are inherently heterogeneous, and one subset has the capacity to move whereas others do not, and the transcriptional profile merely identifies these heterogeneous populations. The observation that non-mover cell populations contain both basal and club cells, whereas mover regions only have basal cells seems to support this notion to some extent. However, the authors then claim to use basal-cell derived signatures (excluding the club cells) from mover and non-mover regions and compare this to literature data from another injury model to show that these signatures also identify distinct subsets in a mouse model of polidocanol-induced injury. How the distinction basal vs club cells in the non-mover regions is made remains unclear, and would seem challenging from the number of cells analyzed (as presented in figure 3).

    The identification of two behavioural phenotypes of basal cells (mover vs non-mover) in this manuscript is based on group-level phenotypes: the cells belong to a region of movers or a region of non-movers. This is relevant for figures 2 (including supplemental) and 3. In figure 2 supplemental 2C, it seems evident that within one region (or focussing only on all moving regions?), the behaviour of all cells within that region/selection is quite uniform: the variation is really very limited, and all cells seem to speed up and slow down in a highly coordinated fashion within the selected regions shown. At the same time, in figure 2D, the distribution of regions across speed categories at 26-36 hours pi (the peak of the movement in suppl 2C) seems almost bimodal, with regions belonging either to non-mover (range 0.5 - 2.5 uM/hr) or mover (range 3.0-7.0 uM/hr) phenotypes. However, all regions display an increased movement at 16h pi compared to the pre-injury movements (Figure 2C), indicating that all cells will be induced to induce movement to some extent. My main concern with this analysis is that the behavioural phenotype of the epithelial cells is assumed to be homogeneous within each region, allowing a contrast to be made in figure 3 for the transcriptional phenotypes on the basis of moving phenotypes rather than on the basis of the main variation within the dataset. For instance, from the t-SNE plot (3B) - for what it's worth of course - and the heatmap (3C) there seems to be at least one non-mover cell that transcriptionally has a higher resemblance to the mover cells than to the other non-mover cells. Of course that can just be the variability present in the dataset, but it could also indicate that non-mover regions are not completely homogeneous, and even more so, that the moving vs non-moving associated transcriptional phenotype is a gradual transition rather than 2 clearly separate sub-phenotypes.

    All-in-all, this manuscript describes an interesting technical advance and shows some of the applications thereof. However, the approach also has its limitations: The requirement to mark cells with specific behavioural features for follow-up transcriptomic analysis (such as by photoconversion) necessitates the division of the epithelial cells into major categories on the basis of certain cellular phenotypes (such as movement) that can be visualized by live cell imaging. This limits the analysis opportunities to group-based contrasts in cellular behaviour as also used here by the authors. Also, the use of explanted tissue is of course less ideal than in vivo imaging, but most likely the only technically feasible approach at this moment. At the same time, the capacity to combine image-based features with single-cell transcriptomic data is an important advance, even when initially only possible in explanted tissue from mouse models carrying all kinds of fluorescent reporters. To strengthen the manuscript, it would therefore be important to discuss the limitations of the approach, as well as to provide a more comprehensive overview of the possible applications that the authors foresee.