Extracellular matrix phenotyping by imaging mass cytometry defines distinct cellular matrix environments associated with allergic airway inflammation
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (Review Commons)
Abstract
The extracellular matrix (ECM) forms the scaffold in which cells reside and interact. The composition of this scaffold guides the development of local immune responses and tissue function. With the advent of multiplexed spatial imaging methodologies, investigating the intricacies of cellular spatial organisation are more accessible than ever. However, the relationship between cellular organisation and ECM composition has been broadly overlooked. Using imaging mass cytometry, we investigated the association between cellular niches and their surrounding matrix environment during allergic airway inflammation in two commonly used mouse strains. By first classifying cells according to their canonical intracellular markers and then by developing a novel analysis pipeline to independently characterise a cells ECM environment, we integrated analysis of both intracellular and extracellular data. Applying this methodology to three distinct tissue regions we reveal disparate and restricted responses. Recruited neutrophils were dispersed within the alveolar parenchyma, alongside a loss of alveolar type I cells and an expansion of alveolar type II cells. This activated parenchyma was associated with increased proximity to hyaluronan and chondroitin sulphate. In contrast, infiltrating CD11b + and MHCII + cells accumulated in the adventitial cuff and aligned with an expansion of the subepithelial layer. This expanded subepithelial region was enriched for closely interacting stromal and CD11b + immune cells which overlaid regions enriched for type-I and type-III collagen. The cell-cell and cell-matrix interactions identified here will provide a greater understanding of the mechanisms and regulation of allergic disease progression across different inbred mouse strains and provide specific pathways to target aspects of remodelling during allergic pathology.
Graphical Abstract
Article activity feed
-
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
Below is a point-by-point response to reviewers comments. We appreciate the reviewers' thoughtful consideration of the manuscript and __suggestions
Reviewer #1
Evidence, reproducibility and clarity
In this study, Parkinson et al. investigated lung extracellular matrix using imaging mass cytometry (IMC) in mouse models. Overall, the paper is well-written, and the data are clear, although major points outlined below need to be addressed.
In its current form, the paper appears more like a methods-focused study since, to my understanding, no new biological responses are described. The methods employed are very interesting, particularly the extracellular …
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
Below is a point-by-point response to reviewers comments. We appreciate the reviewers' thoughtful consideration of the manuscript and __suggestions
Reviewer #1
Evidence, reproducibility and clarity
In this study, Parkinson et al. investigated lung extracellular matrix using imaging mass cytometry (IMC) in mouse models. Overall, the paper is well-written, and the data are clear, although major points outlined below need to be addressed.
In its current form, the paper appears more like a methods-focused study since, to my understanding, no new biological responses are described. The methods employed are very interesting, particularly the extracellular matrix analysis.
As the reviewer points out a large part of this manuscript is the development of a novel methodology for analyzing the spatial ECM changes in a model of allergic airway inflammation. However, there are several novel responses described in the manuscript. Firstly, differing spatial organisation of immune cells across different mouse strains has not been shown before, particularly in a model of chronic allergic pathology that shares features of severe steroid-resistant asthma in people. Secondly, we show that specific macrophage-fibroblast interactions are occurring in the subepithelial region during DRA-induced allergic airway inflammation. Finally, we integrate all these established and novel findings with detailed spatial analysis of the cellular ECM environment, something which is sorely needed in the field.
However, the scope of the study is quite limited, as all the experiments were performed with mouse samples, which are relatively easy to work with, and the cell organisation is simple compared to humans.
Whilst we appreciate that the dataset in this study is limited, imaging mass cytometry studies, especially when optimizing reagents, are costly, time consuming, and have limited throughput, not to mention the time required to develop new computational tools for data analysis. Investigating cell-matrix changes in mouse data is vitally important for understanding the mechanistic role of pathways and interactions during disease processes. Whilst we have not provided human datasets in this study, staining, data acquisition and analysis has been performed on FFPE samples, making our pipelines applicable to archival tissue banks. Regardless, we are currently preparing a publication showing the applicability of this technique to human samples. Many ECM components are well conserved between humans and mice and the cellular structure and architecture of the lung shares a lot of similarities. Many papers (PMID: 39437149, 38758780, 38581685, and 38142637) have used this imaging technology in the analysis of human cancer, which shows an even more complicated and dense cellular organisation.
The authors do not discuss how this analysis pipeline could be applied to human samples. Furthermore, the entire paper relies on imaging mass cytometry, and additional techniques could have been used to confirm some of the observations, especially given the availability of mouse samples.
As mentioned above, we have taken steps to show that this technology is applicable to humans, though this is outside the scope of this already lengthy manuscript. Additionally, Steinbock, the main analysis pipeline, is well published in human datasets (PMID: 38758780, 39905080, 39759522, and 39761010) and the homology between ECM components is strong between mouse and human. The technology itself is completely species agnostic, so there is no reason to think that there would be issues when applying to humans, other than some differences in the marker expression of certain populations, which is well characterised in many cases.
The reviewer’s comment regarding the use of additional techniques is valid. Firstly, these murine lung pathology samples are derived from the same mouse experiments used in our previous publication (PMID: 33587776), where we have analysed histology, immune mediators and cells using a variety of techniques including flow cytometry and ELISA. We will ensure this point is made clearer in the manuscript. In addition, for revision we plan to compliment IMC data presented with fluorescent immuno-staining to characterize cell populations in greater resolution and also using 3D precision cut lung slices to better characterize and visualize cell populations of interest in greater depth, directly addressing the reviewer’s concerns.
The introduction mentions the DRA model without providing an explanation of what it involves. Non-specialist readers may not be familiar with this abbreviation, and further clarification should be provided.
As the DRA model has been characterized previously, we provided references in the text in order to save space. However, we agree with the reviewer and will provide this information up front in the introduction to make the manuscript more approachable for a non-specialist.
In the methods section, it is not mentioned whether the lungs were inflated before tissue collection, which is crucial for preserving normal cellular organization. The authors should clarify whether this was performed.
Lungs were inflated prior to tissue collection. We agree that this is important information to include in the methods and we will update the manuscript to reflect this.
Figure 1 provides a brief summary of the methods employed in the study but could be enriched with additional information. In its current state, it does not provide meaningful insights beyond what is described in the methods section. It would be helpful if the authors clarified whether the mice used were adults and whether both male and female animals were included.
We agree with the reviewer. The idea behind this figure was to have an approachable introduction to the manuscript. However, in line with the reviewer’s previous comments about focusing more on the biology we will move this to supplementary to keep the importance focused on the biological results. Mouse age and gender were included in the methods of the paper, aligning to the ARRIVE guidelines for reporting animal research. We will additionally clarify that these are adult mice
Additionally, they could present examples of the cell segmentation approach with zoomed-in images at the cellular level to illustrate the analysis.
This is a great idea and appreciate the reviewer’s suggestion. We will provide maps (with zoomed inserts) of the cell segmentation and cell classification across representative ROIs to show not only the segmentation but to provide an overview of how the cell types localise across the lung. This addition will also highlight the caveat of IMC around image resolution of 1μm2 which limits the sensitivity of cell segmentation. We will discuss such limitations of the technique in general in the manuscript in response to this and later reviewer comments.
The first set of data in Figure 2 suggests that C57Bl/6 mice did not respond to allergen treatment, as shown by the non-significant increase in cell numbers. The authors should provide evidence that their model induced inflammation through alternative methods, such as assessing eosinophil counts or pathology.
We know that these exact animals are allergic as their immunological responses were characterized in a previous publication (PMID: 33587776) demonstrating eosinophil counts and cytokine responses measured by flow cytometry. However, in light of the reviewer’s comment, we will add histological images of the lung to this current manuscript. Such data, together with enhanced expression of RELMα and Ym2 from airway epithelial cells (Sup Fig 6) and the shift from ATI to ATII cells in both C57BL/6 and BALB/c mice after DRA treatment (Fig 5 g) will provide thorough evidence that the DRA model induces allergic airway inflammation and pathology in both mouse strains.
The UMAP representation indicates significant overlap between cell clusters, which raises concerns about the accuracy of cell segmentation. For example, the heatmap in Supplementary Figure 1 shows endothelial cells expressing markers such as VWF, aSMA, Vimentin, and PDGFRα, suggesting that the cell cluster may contain a mixture of endothelial cells, vascular smooth muscle cells, and fibroblasts.
UMAP reductions of IMC do not separate as clearly as those from single cell RNAseq or flow cytometry. This is because the staining intensity from IMC is much lower. Rather than being on a log scale, as for single cell or flow cytometry, the values are much closer to linear. Additionally, due to the limitations in IMC resolution and the fact that we did not have distinct membrane markers in our panel, cell mask generation is often non-optimal. This is particularly evident in regions where cells are in close proximity and where the limitations of, an effectively, two-dimensional 5-micron thick tissue section mean that there can be overlap between one cell and another. Whilst we acknowledge that some populations will be a mix of cell types we are limited by the number of markers we can use in IMC, as well as the limitations mentioned above. We have accounted for this by using methodologies to identify and focus on tissue regions (lisaClust) and correlate changes to differences in these regions rather than single cells per se.
Examples of segmented cells should be shown to validate this approach.
As per the reviewers comment above, we will provide maps (with zoomed inserts) of the cell segmentation and cell classification across representative ROIs to show not only the segmentation but to provide an overview of how the cell types localised across the lung.
It is unclear what Figure 2e represents. If it is simply to show that certain clusters can be grouped together, such as AEC, AT1, and AT2 as epithelial cells, this could be conveyed in a simpler way.
We apologise that the reviewer found Figure 2e confusing. The aim of this figure was to provide a simple diagram to highlight how different classifications of cell types aligned. This was required because there were variations in the specificity of some clusters and to address specific questions it made more sense to analyse cells at a broader level. i.e. merging resting and activated ATI/II cells or grouping specific immune cell clusters into larger groups. We did consider a table, but we did not feel this was a “simpler” way to do it. As it is simply for reference, we will move Figure 2e to supplemental.
The analysis of extracellular matrix components presented in Figure 3 provides a novel method for studying these acellular structures, which is a challenge in the field. The authors should be commended for their efforts in this area.
We thank the reviewer for their comment here. We agree that this is a vital area that needs to be addressed as the immunomatrix becomes ever more important in understanding disease pathogenesis. We developed this novel method to begin to understand key spatial interactions between cells and ECM molecules, something missing from the majority of high-dimensional imaging datasets.
However, the parameters investigated in Figures 4-6 do not report any novel findings. While IMC appears to work effectively to analyse these parameters simultaneously, the induction of immune foci and changes in tissue organisation following allergen challenges are already well-documented in both mouse models and human samples.
We disagree with the reviewer on this point. Figure 4 shows that immune cell infiltration in the adventitial cuff is different between BALB/c and C57BL/6 mice. This is a new discovery and provides nuance to our previously published data (PMID: 33587776), which showed that in the bronchoalveolar lavage from these same mice there were no differences in immune cell populations at these chronic time points. Therefore, analysis of lavage cells or lung histology in isolation does not provide a full picture of allergic immune responses.
Figure 5 shows neutrophils localised with alveolar macrophages in the alveolar parenchyma in this chronic DRA model completely distinct from the spatial advential cuff region occupied by other CD11b+ cells. In addition, we show that we can identify perturbations in the alveolar parenchyma by IMC and these correlate with known differences in allergy and asthma such as alterations in ATI/ATII balance, which has also not been shown in this model.
Figure 6 demonstrates that we can identify a tissue region termed “subepithelial cells” which is the site of where remodelling events are known to occur in asthma and allergic pathology. This ECM-rich region is strongly associated with fibroblasts and immune cells which leads in to figure 7 showing that these cells are interacting.
In addition to all of this the main focus of this manuscript is to link these analysis parameters to changes in the ECM environment and we have included this in each of these figures showing how these correlates with allergic changes and how they may be important in understanding these processes. In response to this reviewer’s point, we will highlight and make these novel findings clearer within the text of the manuscript.
In Figure 5, the authors show a decrease in neutrophil numbers in challenged mice. This is unexpected, as this model is widely known to induce strong neutrophil recruitment. The authors should clarify this finding and investigate whether neutrophil chemoattractants are increased in these samples.
This is a keen observation by the reviewer. We were interested in this finding however as it was not the focus of the paper we did not investigate further. In our previous publication we show that there are increased neutrophil numbers in the BAL of these animals (PMID: 33587776) and as mentioned above, we show in figure 5 that neutrophils are found mainly in the alveolar parenchyma. This perhaps means that they are more sensitive to being washed out in the BAL and perhaps there are differences in their “stickiness” in BALB/c and C57BL/6 animals or during DRA-induced allergy. This is in contrast to eosinophils (likely within our CD11b+ cells) which are found in the adventitial cuff, a region is not likely to be captured by BAL wash, though we know that these cells are actively present in the BAL. Overall, though this is an interesting result it was not the focus of this already lengthy paper and is best investigated in another project.
When analysing epithelial cells, the authors separate AT1 and AT2 cells based on podoplanin expression. However, data in Supplementary Figure 4b suggest that both cell types express similar levels of podoplanin. The authors do not provide statistical validation for the claim that AT1 cells express higher levels. Additionally, as podoplanin is expressed by various cell types, such as lymphatic endothelial cells, additional markers are required to confirm the identity of AT1 cells.
Again, the reviewer is entirely correct here. The cells we have identified are labelled as ATI as a best guess and correlate with ATII cells based on anatomical location – though this is likely shared by some of the populations mentioned by the reviewer. The majority of cells in this population are likely ATIs, as they are localized in the alveolar parenchyma and are cells that are not SPC+, though we cannot say for sure without more markers and we were already at the limit of the number of markers that we can run with one IMC panel. It is likely that there are contaminating lymphatic endothelial cells in this cluster. However, these will be a relatively minor population and do not change the main findings presented in the paper. To address this and other comments by the reviewer we plan to include a limitation section to the discussion that highlights exactly these points for future studies.
The last set of data in Figure 7 is interesting and shows that immune cells interact with a population of S100a4 fibroblasts. This finding could be expanded further, as CD11b and Ly6C are expressed by a variety of immune cells. The authors should include additional staining to identify the specific cell types involved, such as monocytes, eosinophils, or airway macrophages. Furthermore, the authors should speculate on why these fibroblast regions attract immune cells. Are these regions enriched in chemokines or other factors?
We thank them for this suggestion. To answer this point, we will conduct immunofluorescent imaging to provide further characterization of these cells in greater depth, as we agree, this will be important to consider. To best visualize cells and their interactions in this adventitial region, we plan to use 3D precision cut lung slices from PBS versus DRA mice in combination with confocal imaging. This method will allow us to utilize antibodies and markers that do not work in the FFPE sections such as SiglecF (eosinophils), CD11c (DCs, macrophages), CD64 and CD169 (macrophages).
The discussion is engaging but focuses more on methodological aspects than new biological insights. Without mechanistic links, it is challenging to draw meaningful biological conclusions.
We agree that the discussion could be used to reinforce the importance of the biological discoveries we have made (listed previously) in the discussion. However, we also believe that it is important to discuss the methodology as this is a novel way to explore ECM-cell interactions in the tissue as highlighted by the reviewer. There are many limitations to using IMC and similar techniques that should be highlighted for future studies so that we can develop better ways of quantifying the ECM environment during disease.
Significance
The study of Parkinson et al. provides interesting methodological insights into the use of imaging mass cytometry (IMC) to analyse lung architecture following inflammation. The application of multiplex antibody staining will leverage important information related to how tissues are adapting to chronic immune response. Here the authors rely entirely on mouse models for their studies and compared two lines of WT animals and the same allergen model. This limits the scope of the study, additional timepoints, sex or age would have improved the manuscript.
Whilst we appreciate the reviewers points here, we would like to highlight the time involved in generating such datasets, with a lot of careful optimization and experimental design aspects going into each study. Whilst we have also performed staining and analysis using our described method in human FFPE tissue, we are currently looking to further develop analysis tools to assess ECM-cell interactions. Additionally, data acquisition using IMC takes considerable time, and hence it is not feasible run and analysis the number of samples required to address some of the questions proposed by the reviewer.
We believe our manuscript provides novel methodology to analyse ECM environments within spatial datasets, something that no other spatial datasets have explored to date. Furthermore, we provide numerous new biological findings in relation to how cells are organized within the tissue during allergic pathology and propose immune-fibroblast interactions that may be key for driving ECM remodelling in the lung. Integrating these analyses will be key for further understanding the role of the ECM in disease pathogenesis.
The applicability of this analysis pipeline to human tissue samples is not discussed, which would significantly enhance the impact of the study. Additionally, complementary techniques, such as flow cytometry or immunohistochemistry, could be used to validate the findings and improve reproducibility. A specialised audience of immunology researchers would be interested by the image analysis approach.
As mentioned above, this analysis pipeline is easily applied to human samples or any other species as ECM molecule organization is largely conserved across species. Moreover, we have already explored this using human samples. However, adding human data to this manuscript is beyond the scope of this manuscript which was aiming to build one of the first methodologies for incorporating the ECM into this kind spatial analysis from the start in order to make biological discoveries. Regardless, we will add a discussion point on utilizing these pipelines to other species within the discussion of the manuscript.
Flow cytometry has been published on this model and the exact samples used within this study as mentioned previously (PMID: 33587776), validating some of these findings – we will make this point more clearly in the manuscript. We do appreciate that it would be good to further expand on some findings presented in the manuscript. As such we will expand our immunostaining (as mentioned above) to give more detail on the infiltrating immune cell populations and their interactions with fibroblasts.
__Reviewer #2 (Evidence, reproducibility and clarity (Required): __
Summary Parkinson and colleagues provided a highly intriguing manuscript on spatial resolution of cell-ECM interaction in mouse models of allergic airway inflammation. They used IMH to analyse two common mouse strains for allergic airway inflammation with a human relevant allergen mix. The study implements a novel technique to better segment tissue stainings (DeepThresh) and modifies existing tools to assess non-cellular seqmentation, ECM or fibrous structures that is. The study identifies region-specific ECM and confirms cellular proximity with canonical cell markers. Furthermore, clear ECM and cellular differences between the two mouse strains are found. The study concludes that this IMH approach is superior to existing methods as it provides a high spatial resolution of ECM protein - cell interaction.
Major: ECM Isoform Annotation - The manuscript lacks precise annotation of ECM isoforms, particularly for Collagen I, IV, and VI. This impacts the accuracy of reported associations between ECM environments and cellular interactions.
We thank the reviewer for this excellent comment and pointing this out. We agree that this is very important and will add this data to the manuscript. All information was included by reference of the antibody clones. However, it is an important point to make and we will account for this during interpretation of the results.
Spatial Annotation Consistency- The manuscript inconsistently defines and annotates ECM environments (e.g., adventitial collagen, subepithelial & vascular ECM), leading to confusion in spatial correlation analyses.
We are unsure what the reviewer is exactly referring to here. We have maintained a consistent nomenclature for these annotations throughout the manuscript. If the reviewer has an issue with the names we have provided for the regions; names were chosen these to be more informative than just naming them “region 1, 2, 3…”. Names in the manuscript were based on taking the lung tissue region and the prominent ECM molecules present. Whilst some level of detail will naturally be lost, we considered this the best way to keep data clear and consistent throughout the manuscript. For example, adventitial collagen describes the region predominantly around the adventitial cuff (fig 3c and d; shown in dark blue) that has high levels of Collagen I, III and VI. Yes, HA, laminin and fibronectin are also expressed, but at much lower levels. Regardless, all the information is present within the figures with readers to observe and make their own interpretations. We are happy to consider alternative names if the reviewer were to provide some guidance on what they thought was more appropriate.
Lack of Supplemental Data- Activated cell types and regions are not clearly defined, and no supplemental data is provided to verify classifications. DeepThresh Validation - The method for removing staining artifacts via DeepThresh lacks clear validation. Complexity - Overlapping marker definitions (e.g., CD11b+ cells and infiltrating cells) need clarification for accurate immune cell characterization.
We provide heatmaps in the supplementary data which shows the exact marker expression pattern for all of the clusters we define (Sup Fig 1a). Additionally, we provide graphs showing the cellular contribution and spatial distribution of all the regions we defined with lisaClust (Fig 2h & I; Sup Fig 1d). Most activated cells are a feature of a specific clustered cell type only being present in either PBS or DRA treated animals. However, the features which have led to separation these cell types are available in the heatmaps as mentioned (Sup Fig 1a).
We believe the reviewer may be confused about the purpose of DeepThresh. This algorithm is not for removing staining artifacts. Instead it uses expert annotation of a small training set to generate a method of accurately thresholding images for positive staining in relatively small ROIs which may have diverse structural features with different staining properties. We did not have space in the manuscript to go into this in more detail. However, we appreciate this may not be as clear as needed for readers, and hence, will provide supplementary data showing some example thresholding alongside the original staining in a new edit of the manuscript.
CD11b+ and infiltrating cells are not an overlapping population, they were separately clustered by the algorithm, but we take the reviewers point that further characterisation could be done. As mentioned in comments from reviewer 1, there is a limitation in the number of markers we can use in IMC, especially with the number of ECM markers we included. Additionally, there are limitations in the appropriate antibodies (carrier-free) that work in FFPE mouse tissue with the antigen retrieval that we use for good, reliable staining of ECM components. As such, we will perform additional immunofluorescence staining in 3D precision cut lung slices to better characterize the CD11b+ population to address comments by both reviewers.
Minor: Terminology Inconsistency- The manuscript uses inconsistent terminology for ECM components and anatomical regions (e.g., adventitial collagen, immune foci, inflammatory zone).
This point was directly addressed above in “Major” points and appears to be a duplicate comment.
ROI Mask Coverage - Statistical insignificance in C57BL/6 ROI mask coverage is not addressed.
The increase in C57BL/6 mice upon DRA treatment in panel A is not “significant” in the modern sense of the word. However, we would argue that stating it is “not significant” would also be a mistake. We prefer to use p values as an inferential measure of significance in combination with measures such as effect size and variance (PMID: 8465801). We find this more useful considering the vast number of mistakes made when interpreting p values (PMID: 18582619). The importance of not purely relying on p values for clinical research has been reviewed recently here (PMID: 39909800).
Whilst we appreciate the reviewer’s requirement for significance, we do not want to make sweeping statements based off of a p value of 0.07, especially in only one experiment. Many papers have been published on the pitfalls of stringently adhering to p
Spelling Error - "Immunte foci" in Figure 4h.
We thank the reviewer for pointing this out and will correct this.
Figure 6g Correlation Issue- The matrix environment correlation plot does not align with expected cell-ECM interactions.
We find it hard to comment on this without more detail of the cell-ECM interactions that the reviewer believes should be occurring. We analysed this in an unbiased way, so we have not forced interactions to appear based on our preconceptions. The regions being analysed in Fig 6g are the resting (PBS) and activated (DRA) airways that contain expected cell populations of airway epithelial cells and a low level of fibroblasts, likely from just under the airway epithelial cells. These cell populations align with AEC-associated matrix, laminin and hyaluronan, and adventitial collagen regions. Perhaps the reviewer is questioning why the airways are associated with adventitial collagens? The reason behind this, is due to adventitial cuff residing adjacent to a proportion of all airways, and hence any ECM associated with the adventitial cuff will likely be included in an airway region. However, as mentioned previously there are limitations to this analysis and we are very likely missing finer details due to issues such as resolution which we have discussed within the point-by-point on numerous occasions, and something we will directly address by adding a limitations section to the discussion of the revised manuscript.
Color Issues in Figures - ColI and ColIII have the same color in Fig. 3a, making interpretation difficult.
We agree with the reviewer on this point. The issue we had here was that Col-I and Col-III strongly overlap in these images, whilst one was green and one yellow the effect was to make them look the same in the final images. We will remake these images with clearer colours that better illustrate differences in Col-I and Col-III expression.
Patch Annotation (Fig. 4i) - The method for defining immune cell patches is unclear.
Patches refers to an approach that is used to identify interconnected groups of similar cell types and is a method that is based off published data (PMID: 35363540). We will add further method details that explains this process to the revised manuscript.
Detailed review: Methods: Animal model is suitable for differential analysis of various mouse strain responses to allergic airway inflammation.
We thank the reviewer for this comment and also agree that the mouse models presented in the manuscript can provide insightful and mechanistic data for investigating human disease.
Deepthresh matrix thresholding: IMCDenoise is sensitive to clusters of staining artefacts (specks). Please explain how DeepThresh via manual thresholding enables staining artefacts removal/detection. Manual ground truth mapping is common however it is not clear how your approach is performing against another tool. How was manual thresholding controlled (several analysts thresholded same image)?
As described in a previous comment this is not the function of DeepThresh. Manual annotation for training data was performed by consensus agreement of four independent researchers. In terms of performance against another tool, we are not aware of another tool which performs this function and hence cannot compare. However, we will add additional data showing the validation metrics for the pipeline to make future comparisons easier.
Antibodies Collagen IV, stains col4a1 - please correct, as isoforms vary throughout tissue. Collagen VI, stains col6a1 - isoforms vary in lung tissue, please state correct isoform throughout the document. Heparan sulfate: Molecular weight? Collagen I - isoform not defined, please state in methods.
Figure 3 d As a resultant of the choice of antibodies against some particular isoforms of ECM molecules associations of cells, compartments are correct yet do not comply with all isoforms. Col4a1 is a basal membrane collagen from blood vessels; the adventitial area and vascular area are high in Col4a1. Other Col4 isoforms are found more frequently in the alveolar regions (col4a5,a6) and the subepithelial membrane. It is of utmost importance to clearly label the correct isoforms throughout the document.
This relates to the comment above made by reviewer 2. As mentioned, we agree with this key point and will provide this information from the respective antibody clones.
However, we are unable to provide details on the molecular weight of heparan sulfate as this will vary depending on location/tissue/condition etc. The antibody recognises 10E4 epitope on HS which is found across a wide variety of tissues and species and will recognise many different sizes of HS and even porcine Heparan. Importantly it is relatively specific, not cross reacting with hyaluronan, keratan sulphate, chondroitin sulphate, or dermatan sulphate which is an issue for certain clones. Whilst the size of the HS is an interesting facet, consideration of changes in sulphation patterns would also be of interest, though these currently cannot be accurately assessed via purely immunostaining-based methodologies and would require more targeted biochemical techniques. In addition to this there are multiple nuances in 10E4 antibody binding (PMID: 15044385 and 11278655) which are interesting, but far beyond the scope of this study. Although captured in the antibody clone information, we will also ensure this is clear in the methods.
In relation to Col4 isoforms specifically, often antibodies for the ECM are limited because of their repeating structures it is hard to generate specific antibodies. For collagen IV there many clones for Col4a1, but no specific clones for Col4a3/col4a5 etc, suitable for use in FFPE tissues and metal conjugation required for IMC. Therefore, we were very limited in what was available to detect them at all. We will bring this up in the discussion as this is an important point, not just for our data, but also for people attempting to replicate this kind of analysis.
Figure 2i: The cell-specific marker expression is in part already confounded by region. So vasculature or resting airways show no "resting" fibroblasts as their annotation is linked to activation (indicated by S100A4 expression). Anatomic locations such as airways with remodelling are termed "activated" to explain morphological differences which is acceptable given the model chosen. However, some cell type are not given an anatomical or morpholocial "resting" nomenclature. Only during activation and through location a cell type may aquire e.g. a nomenclature such as "alveolar fibroblast". The correlation blot 2i should provide this basic information. Please explain.
Our staining approach and analysis have only identified certain activated populations as pointed out by the reviewer. Most of the populations that we have identified as “activated” have been identified primarily only in mice administered DRA. The reason that we have not included “resting” and “activated” populations for all cell types is that these clusters were generated using a clustering algorithm based on the cellular markers used within the study. Each cluster was then simply labelled as best we could, using information from marker expression, published biological data, anatomical location, and sample identity (e.g. PBS or DRA).
A caveat to using IMC and other similar imaging techniques is that we will miss certain “flavours” of cell populations because we simply do not have the markers, or scope to include markers, with which to identify these cells. This is partly a problem of appropriate antibody availability, but also for many populations there are no specific markers identified in the literature/databases. Single cell RNAseq has provided deep segmentation of some of these populations, but we (and others) have found that often these make poor antibody choices at the protein immunostaining level.
We are unsure what the reviewer wants adding to plot 2i. This plot shows the cell cluster contribution to different lisaClust defined tissue regions. Hence the presence of alveolar fibroblasts in the resting and activate alveoli region. However, we will include more discussion on the limitations of markers and identification of specific cell populations in the discussion.
Figure 2h: How do you explain subepithelia to "leak" luminally in C57BL/6 DRA animal?
We assume the reviewer is referring to the overlap of some grey circles though/over the red airway epithelial cells in the C57BL/6 DRA panel of figure 2h. This figure represents individual cells as circles with the centroid of the circle at the centroid of the cell. Cells are rarely perfect circles and, in this case, it has made it seem like the cell is coming through the airway epithelium, when likely it is a longer cell that sits directly under it. In addition to this, these are effectively 2-dimensional section (5um thick) that capture as small portion of the lung anatomy, hence occasionally this can result in unusual tissue structures that make no sense in the confines of a 2D section, but instead correlate with the larger 3D structure.
How is an activated airway possible in a Balb/c PBS animal (same for inflammatory adventitia, alveoli)?
Activated airway simply describes a region that is showing some evidence of activation markers such as RELMα and/or Ym2 etc. PBS itself, as with any other liquid administered into the lungs, will drive a very low level of inflammation, which is why it is used as a control in the animal model. Therefore, it is not surprising that we see a low number of these “activated” cells in PBS animals vice versa for their activated counterparts in DRA treated animals. This is similar for the other regions mentioned.
How is subepithelia adjacent to immune foci and inflammatory adventitia (Balb/c DRA).
We are somewhat confused by this question. We have termed the region “subepithelia” because it is mostly found under the airway epithelial cells. We found that this region expands during DRA treatment and covers areas close to the immune foci and inflammatory adventitia, hence they are next to each other.
As described above, the names of these regions were chosen for simplicity and to communicate its general features. These, regions were identified by detection of nearby regions of cells with similar cellular compositions and the names we a “best fit”.
Text for fig 3c: Here it should be mentioned that a cell is used as a proxy locator to the ECM region.
We apologise that this was unclear for the reviewer. Rather than describing it as using the cell as a proxy locator to the ECM region we find it more accurate to think of it as we are characterizing the matrix environment of the cell i.e. what is the cell close to and what is it far away from. We will make this clearer in the results by changing the name to cellular matrix environment, rather than matrix environment.
Again, in UMAP3b location and ECM molecule a mixed a priori which only can be achieved through proxy loction as in fig 3c or correlation matrix analysis. The UMAP shows ECM molecules in various combinations. Fig3c analysis of anatomic location from images with cell proxies would validate morpho-spatial UMAP annotation. Please make this clear in the manuscript or specify why your approach is superior in its presented format.
We struggled to ascertain what the reviewer was referring to here and what edits they were suggesting to the revised manuscript. However, this comment seems to assume that we have used cellular location as an input to the UMAP in figure 3b, which is untrue. This UMAP (and associated clustering) shows each cell as a dot which is organised based on its distance to the different matrix components. Effectively showing us how different cells cluster based on their cellular matrix environment, with no input of cellular based markers. We are unsure what the reviewer is referring to on line 486 – as they seem to be describing exactly what figure 3c already is (a spatial map of the UMAP clusters on representative images, which shows that a cells matrix environment does seem to show patterns that align with the general lung anatomy).
Finally, the reviewer asks us to specify why our approach is superior, but we are unclear what the alternative approach is.
This methodology is effectively a repurposing of the traditional UMAP and clustering methodology used in many single cell techniques, but instead of applying this to cellular markers we are applying it to a cells matrix environment as quantified by the matrix distances. If the reviewer could clarify this comment we would be happy to revisit it. As mentioned in the previous comment, we will more clearly describe cellular matrix environments in the revised manuscript and this may also help with the confusion.
Fig 3d: The Matrix Cluster names are in part not correct. Subepithelial & Vascular ECM does not correlate with Vasculature in LisaClust Regions. Also ColIV is not AEC associated, yet subepithelial.
Respectfully, we completely disagree with the reviewer on this point. In the heatmap (Fig 3d) the Subepithelial & Vascular matrix environment correlates most strongly with the Vasculature and Subepithelial cells as shown by the stronger green-yellow colour in the corresponding cell of the heatmap.
As mentioned previously in response to another comment by reviewer 2, there could be many reasons that we are not detecting collagen-IV in the AEC associate cell matrix environment. One likely explanation is that this is too fine for the resolution of IMC (1-micron2) or it could be that certain subchains are utilised here that are not recognized by the antibody we managed to optimize. Additionally, AEC-associated matrix environment is comprised of both mouse strains and includes higher representation from DRA treated animals. From our previous work (PMID: 33587776), we have shown that Col-IV expression around the AEC is reduced in DRA versus PBS -treated animals.
No ECM molecule is inflammatory zone associated. Does this indicate cellular density does not allow to distinguish ECM?
This is a great point from the reviewer and their explanation is entirely possibly. As mentioned there are huge limitations in the resolution of IMC and so we are likely missing finer matrix structures. There is a huge recruitment of cells within this environment so it could be that we cannot clearly visualise fine ECM structure through this considering we are also looking at a 5-micron thick 2D tissue section. Additionally, cells maybe degrading the ECM in order to infiltrate into the tissue. This is definitely an interesting point to examine in further detail, but would need to be done with a different methodology. We will aim to look at an ECM molecules and its distribution within the inflammatory zone using 3D precision cut lung slices and also immune-staining of tissue sections to see whether we can better resolve this in a revised manuscript.
Also the term "adventitial collagen" is locating to LisaClust Regions Vasculature, Subepithelial Cells, Resting Airways, Infiltrating Cells, Activated Airways. Adventitial per definition of fig. 2g is around blood vessels extending to airways and around it. The adventitial regions are the ECM rich areas after the fibroblasts (as for blood vessels, PMID: 31522963). The definition used in this study therefore generates morphological overlaps between airways and their basolateral regions and blood vessels. Whilst both morphological regions have an adventitia the Matrix cluster assumes from areas to close by this terminology. As a sensitivity analysis I would suggest to reduce the perimeter around blood vessels to the same borderline as seen in airways. If composition remains similar "adventitial collagen" could be a broader term. Alternatively, if adventitia from airway and blood vessel differ these should be separate terms.
Whilst the adventitial cuff does refer to the region immediately around a blood vessel in the lung, these structures are slightly more nuanced as blood vessels normally travel through the lung in close association with an airway. This is true all the way down to the close association with the capillaries and the alveolar spaces where gas exchange occurs. Indeed, previous publications (PMID: 30824323) have shown that these adventitial cuffs extend out from around the contiguous area around the blood vessel and associated airway and these can expand during inflammation (PMID: 24631179). This region is rich in Collagen-I and Collagen-III, as we have shown in this manuscript and previously (PMID: 33587776).
Whilst we agree that there are likely microanatomical niches within this larger structure, our dataset lacks the resolution to study this in more detail. However, as mentioned above we can include matrix markers in our future IF staining to examine this region in more detail. The adventitial collagen environment described in this manuscript and beyond, are vital “meet and greet” spots for immune cell infiltrating into the lungs (PMID: 30824323) as well as being sites of iBALT formation (PMID: 24631179)
We are unsure what the reviewer means by “…reduce the perimeter around blood vessels to the same borderline as seen in airways.” We have not defined a manual threshold for the border of the airways. These regions were all defined by SNN clustering and not manual segmentation. Whilst this methodology could be developed we do not believe that this dataset has the resolution to answer this question, as mentioned previously.
Fig 4c: Balb/c and C57bl/6 labels are incorrect (see a,b)
We thank the reviewer for noticing this incorrect labelling and will update this.
Fig 4h: Cell type "other is highly present in immune foci and inflammatory adventitia but not further classified and not myeloid. This seems either a difficult definition for myeloid or a significant immune population wasn't stained. How is myeloid defined?
We define myeloid broadly as CD11b+ or alveolar macs. There were certain populations that were not stained, notably T cells. We were unable to have suitable or reliable staining in FFPE tissue with CD90, TCRa/b, CD3e antibodies via IMC. The same was true for Eosinophil markers (SiglecF, Ccr3, EPO, MBP). The additional experiments we will perform for a revised manuscript (using 3D precision cut lung slices and/or IF staining) should shed further light on these cells. Additionally, as we are not limited by the processing requirements of IMC, we can use a wider range of markers.
Fig 4l has a vast variety of marker combinations some being very specific within the staining panel, others subsummarise entire groups of cells. It would be very helpful to know if the lables are specific and exclusive or if larger clusters exist, that they then subdivide into specific groups (e.g. Infiltrating cells: any of CD11b, CD44, Ly6C vs. B-cells or CD11b+Lys6C). This graph would profit also from either using markers or cell types only. Your marker set is very distinct and limited so per definition it is either a neutrophil or a Lys6C+. Please decide, explain and provide the other graph as supplement.
We apologise that this was not clear to the reviewer. Labels are exclusive and represent the clusters that were identified in figure 2 and are at the finest level of detail that we felt we were able to biologically infer from the data. In terms of the reviewer’s first point about infiltrating cells, these are completely separate from the other cell types mentioned. As mentioned in the previous comment line 570, we were simply unable to find working antibodies for some of the common lung populations (a common problem for FFPE sections where antigens are often masked or lost due to fixation and processing) and so are limited to general annotations for these. For the reviewer’s second example of Neutrophils vs Ly6C+ cells, neutrophils were classified by expression of Ly6G, CD11b+, and Ym1 whereas there are many other cell types that express Ly6C, including, but not limited to, dendritic cells, monocytes, eosinophils, and even some T cells.
We believe that the graph in combination with data in Fig 1c and supplementary Fig 1a, already shows what the reviewer is asking for.
Fig 5l and sup Fig4i: There is no graph confirming the statement that Ym1 is produced by macrophages. From the graphs in either of the two panels, The AEC are highly associated with Ym1/2 expression or the activated alveoli. Please explain ad amend.
We assume the reviewer means Fig 5l and sup Fig 5i (as there is no figure sup Fig 4i). Whilst we did not include a graph to show that alveolar macrophages produce Ym1, we did include two references in the text and this has been widely shown in the literature for many years (PMID: 11141507 and 15148607). We are somewhat unclear on the reviewers second point. AEC (airway epithelial cells) can definitely also produce Ym1, though this can often be contentious because of issues with cross-reactivity with its highly homologous sister protein Ym2, which is also produced from airway epithelial cells under Type-2 settings. If the reviewer is referring to AEC (alveolar epithelial cells) then this is true. Activated alveoli are lisaClust regions with lots of alveolar macrophages which was the original statement we made and aligns with sup Fig 5i. Activated alveoli II have less alveolar macrophages and also have less Ym1, which would correlate though there are other cell types which can make Ym1 as well.
Fig 6g: The correlation plots again show that the matrix environment labels are somewhat confounded. Whilst AEC associated makes perfect sense, adventitial collagen only weakly correlates, yet was part of the adventitial mapping. Cell types like AEC are expected however fibroblasts, especially in resting airways as large constituent cell populations. There are not other, myeloid or lymphoid cells associated with these airways, which under activated conditions seems rather odd. From fig6a it is appearant that the lisaClust has ascribed subepithelial regions to distal parts of the airway separated by blood vessel or parenchyma (C57BL/6 and Balb/c DRA). Also blood vessels are in part other cell types or epithelium (B6 PBS). Is the annotation here the reason for this rather confusing result? Please explain and/or amend.
We are again somewhat confused by this comment. Adventitial collagen only weakly correlates because it is not within the airway epithelial cells, instead it is adjacent in the subepithelial region which is shown in Fig 6j. We are unsure exactly what the reviewer is referring to in terms of “adventitial mapping” but are happy to comment on this if the reviewer can clarify what they mean.
We agree with the reviewer that it is somewhat surprising to see so many fibroblasts in the resting and activated airway regions. There is a level of ambiguity here in what lisaClust decides to include in one region vs another. However, what it does show is that there are a large population of fibroblasts around the airway, possibly correlating with peribronchial fibroblasts. We did not observe immune cells in between the airway cells or immediately underneath it. We do not believe this is odd, as from our data it appears that these cells are more likely to be found in the adventitial (including peribronchial as mentioned previously) cuff. Cell are most certainly moving into the airways as shown from the BAL in our previous publication (PMID: 33587776). However, we are unlikely to capture this process in the snapshot of our histology across a relatively small section of the airways covered in our 2D sections.
In regards to the reviewers comment about figure 6a we agree that some of the regions between the airways and blood vessels have been characterised as subepithelia. As mentioned previously we are happy to consider alternative names but have been unable to come up with an alternative that encompasses the cells and spatial region more accurately and clearly., Regardless, the main purpose of these names is to provide simple nomenclature to follow throughout the manuscript and make these types of analyses accessible to all readers. We believe that these are accurately labelled and have provided information about the constituent cell populations that are present within them, making the data and subsequent analysis transparent for others to view and explore. Our data suggests that the adventitial cuff may fulfil multiple roles during DRA-induced inflammation, some of which are more focused on immune cell recruitment and others which may correlate more with the fibroblast rich subepithelial region.
The reviewer is entirely correct to point out that some blood vessels were not entirely annotated. We used vWF to manually separate blood vessels from the adjacent smooth muscle layers, which were not separated by the clustering originally. Notably it appears that veins seem to not separate as well as arteries suggesting another marker (e.g. CD31) may help with this, though we were limited in what we could include as mentioned previously. As this is only a small effect, which we do not have a way to correct, and blood vessels were not the focus of this manuscript, we have left the annotation as it is with raw data included.
__Significance __
Strength Innovative ECM-Immune Interaction Approach- The study integrates extracellular matrix (ECM) phenotyping with immune cell spatial mapping, providing novel insights into allergic airway inflammation Multiplex Imaging Technology - The use of Imaging Mass Cytometry (IMC) allows high-resolution spatial characterization of both cellular and ECM components. Strain Analysis - The inclusion of BALB/c and C57BL/6 mice enables differentiation of strain-specific ECM and immune responses. Deep-Learning-Based ECM Quantification - DeepThresh offers an advanced computational approach for ECM analysis, enhancing accuracy in defining ECM-cell associations. Comprehensive Tissue Classification- LisaClust clustering facilitates detailed segmentation of lung microenvironments, improving understanding of localized tissue remodeling.
Limitations ECM Isoform Inconsistencies - The study lacks precise annotation of ECM isoforms, which affects the accuracy of reported ECM-cell interactions. Ambiguous Spatial Correlations- Some ECM clusters, such as "adventitial collagen," overlap inconsistently with anatomical regions, making interpretation challenging. Unvalidated DeepThresh Method - The manuscript does not provide sufficient validation of DeepThresh's ability to remove staining artifacts. Lack of Supplemental Data- Key activated cell types and regions lack supporting data for classification.
__Advance, gap filled __ Clearly the next step to improve organ compendia such as the lung cell atlas, spatial protein analysis is warranted. scRNA-Seq in particular for ECM molecules is challenging as these molecules are produced in small quantities or have a very slow turn-over. This study has the potential to provide novel deep learning algorithms to include not only cellular markers but consider larger panels of ECM molecules and their spatial orientation in the lung.
__Audience __ The manuscript is interdisciplinary located between advanced image analysis with deep learning methods, fundamental lung biology and single cell analysis. The readership would entice molecular biologists, bioinformaticians and basic disease model scientists. The manuscript would appeal to clinician scientists and a broader audience if human tissue pendants could be provided validating the methods and outcomes.
__Own Expertise __ Translational scientist in the field of chronic lung disease, highly familiar with epithelial cells, mouse models, human cohorts and next generation sequencing and imaging of live single cells.
-
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
Summary
Parkinson and colleagues provided a highly intriguing manuscript on spatial resolution of cell-ECM interaction in mouse models of allergic airway inflammation. They used IMH to analyse two common mouse strains for allergic airway inflammation with a human relevant allergen mix. The study implements a novel technique to better segment tissue stainings (DeepThresh) and modifies existing tools to assess non-cellular seqmentation, ECM or fibrous structures that is. The study identifies region-specific ECM and confirms cellular proximity with canonical cell markers. Furthermore, clear ECM and cellular differences between the two …
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
Summary
Parkinson and colleagues provided a highly intriguing manuscript on spatial resolution of cell-ECM interaction in mouse models of allergic airway inflammation. They used IMH to analyse two common mouse strains for allergic airway inflammation with a human relevant allergen mix. The study implements a novel technique to better segment tissue stainings (DeepThresh) and modifies existing tools to assess non-cellular seqmentation, ECM or fibrous structures that is. The study identifies region-specific ECM and confirms cellular proximity with canonical cell markers. Furthermore, clear ECM and cellular differences between the two mouse strains are found. The study concludes that this IMH approach is superior to existing methods as it provides a high spatial resolution of ECM protein - cell interaction.
Major:
ECM Isoform Annotation - The manuscript lacks precise annotation of ECM isoforms, particularly for Collagen I, IV, and VI. This impacts the accuracy of reported associations between ECM environments and cellular interactions. Spatial Annotation Consistency- The manuscript inconsistently defines and annotates ECM environments (e.g., adventitial collagen, subepithelial & vascular ECM), leading to confusion in spatial correlation analyses. Lack of Supplemental Data- Activated cell types and regions are not clearly defined, and no supplemental data is provided to verify classifications. DeepThresh Validation - The method for removing staining artifacts via DeepThresh lacks clear validation. Complexity - Overlapping marker definitions (e.g., CD11b+ cells and infiltrating cells) need clarification for accurate immune cell characterization.
Minor:
Terminology Inconsistency- The manuscript uses inconsistent terminology for ECM components and anatomical regions (e.g., adventitial collagen, immune foci, inflammatory zone). ROI Mask Coverage - Statistical insignificance in C57BL/6 ROI mask coverage is not addressed. Figure 3d Labeling- Matrix cluster names do not always match tissue localization. Spelling Error - "Immunte foci" in Figure 4h. Figure 6g Correlation Issue- The matrix environment correlation plot does not align with expected cell-ECM interactions. Color Issues in Figures - ColI and ColIII have the same color in Fig. 3a, making interpretation difficult. Patch Annotation (Fig. 4i) - The method for defining immune cell patches is unclear.
Detailed review:
Methods: Animal model is suitable for differential analysis of various mouse strain responses to allergic airway inflammation.
Deepthresh matrix thresholding: IMCDenoise is sensitive to clusters of staining artefacts (specks). Please explain how DeepThresh via manual thresholding enables staining artefacts removal/detection. Manual ground truth mapping is common however it is not clear how your approach is performing against another tool. How was manual thresholding controlled (several analysts thresholded same image)?
Antibodies Collagen IV, stains col4a1 - please correct, as isoforms vary throughout tissue. Collagen VI, stains col6a1 - isoforms vary in lung tissue, please state correct isoform throughout the document. Heparan sulfate: Molecular weight? Collagen I - isoform not defined, please state in methods.
Figure 3 d As a resultant of the choice of antibodies against some particular isoforms of ECM molecules associations of cells, compartments are correct yet do not comply with all isoforms. Col4a1 is a basal membrane collagen from blood vessels; the adventitial area and vascular area are high in Col4a1. Other Col4 isoforms are found more frequently in the alveolar regions (col4a5,a6) and the subepithelial membrane. It is of utmost importance to clearly label the correct isoforms throughout the document.
Spelling error in figure 4 h (immunte foci)
ROI mask coverage in C57/6 not significant
Activated cell types/region: This definition is not specified and no supplemental data is provided to see which markers classify such areas/cells. Please provide.
Figure 2i: The cell-specific marker expression is in part already confounded by region. So vasculature or resting airways show no "resting" fibroblasts as their annotation is linked to activation (indicated by S100A4 expression). Anatomic locations such as airways with remodelling are termed "activated" to explain morphological differences which is acceptable given the model chosen. However, some cell type are not given an anatomical or morpholocial "resting" nomenclature. Only during activation and through location a cell type may aquire e.g. a nomenclature such as "alveolar fibroblast". The correlation blot 2i should provide this basic information. Please explain.
Figure 2h: How do you explain subepithelia to "leak" luminally in C57BL/6 DRA animal? How is an activated airway possible in a Balb/c PBS animal (same for inflammatory adventitia, alveoli)? How is subepithelia adjacent to immune foci and inflammatory adventitia (Balb/c DRA).
Fig 3a: ColI and ColIII have same colour, this makes images not easy to understand please change. Text for fig 3c: Here it should be mentioned that a cell is used as a proxy locator to the ECM region. Again, in UMAP3b location and ECM molecule a mixed a priori which only can be achieved through proxy loction as in fig 3c or correlation matrix analysis. The UMAP shows ECM molecules in various combinations. Fig3c analysis of anatomic location from images with cell proxies would validate morpho-spatial UMAP annotation. Please make this clear in the manuscript or specify why your approach is superior in its presented format.
Fig 3d: The Matrix Cluster names are in part not correct. Subepithelial & Vascular ECM does not correlate with Vasculature in LisaClust Regions. Also ColIV is not AEC associated, yet subepithelial. No ECM molecule is inflammatory zone associated. Does this indicate cellular density does not allow to distinguish ECM? Also the term "adventitial collagen" is locating to LisaClust Regions Vasculature, Subepithelial Cells, Resting Airways, Infiltrating Cells, Activated Airways. Adventitial per definition of fig. 2g is around blood vessels extending to airways and around it. The adventitial regions are the ECM rich areas after the fibroblasts (as for blood vessels, PMID: 31522963). The definition used in this study therefore generates morphological overlaps between airways and their basolateral regions and blood vessels. Whilst both morphological regions have an adventitia the Matrix cluster assumes from areas to close by this terminology. As a sensitivity analysis I would suggest to reduce the perimeter around blood vessels to the same borderline as seen in airways. If composition remains similar "adventitial collagen" could be a broader term. Alternatively, if adventitia from airway and blood vessel differ these should be separate terms.
Fig 4c: Balb/c and C57bl/6 labels are incorrect (see a,b) Fig 4h: Cell type "other is highly present in immune foci and inflammatory adventitia but not further classified and not myeloid. This seems either a difficult definition for myeloid or a significant immune population wasn't stained. How is myeloid defined?
Fig 4l has a vast variety of marker combinations some being very specific within the staining panel, others subsummarise entire groups of cells. It would be very helpful to know if the lables are specific and exclusive or if larger clusters exist, that they then subdivide into specific groups (e.g. Infiltrating cells: any of CD11b, CD44, Ly6C vs. B-cells or CD11b+Lys6C). This graph would profit also from either using markers or cell types only. Your marker set is very distinct and limited so per definition it is either a neutrophil or a Lys6C+. Please decide, explain and provide the other graph as supplement.
Fig 5l and sup Fig4i: There is no graph confirming the statement that Ym1 is produced by macrophages. From the graphs in either of the two panels, The AEC are highly associated with Ym1/2 expression or the activated alveoli. Please explain ad amend.
Fig 6g: The correlation plots again show that the matrix environment labels are somewhat confounded. Whilst AEC associated makes perfect sense, adventitial collagen only weakly correlates, yet was part of the adventitial mapping. Cell types like AEC are expected however fibroblasts, especially in resting airways as large constituent cell populations. There are not other, myeloid or lymphoid cells associated with these airways, which under activated conditions seems rather odd. From fig6a it is appearant that the lisaClust has ascribed subepithelial regions to distal parts of the airway separated by blood vessel or parenchyma (C57BL/6 and Balb/c DRA). Also blood vessels are in part other cell types or epithelium (B6 PBS). Is the annotation here the reason for this rather confusing result? Please explain and/or amend.
Significance
Strength
Innovative ECM-Immune Interaction Approach- The study integrates extracellular matrix (ECM) phenotyping with immune cell spatial mapping, providing novel insights into allergic airway inflammation Multiplex Imaging Technology - The use of Imaging Mass Cytometry (IMC) allows high-resolution spatial characterization of both cellular and ECM components.
Strain Analysis - The inclusion of BALB/c and C57BL/6 mice enables differentiation of strain-specific ECM and immune responses.
Deep-Learning-Based ECM Quantification - DeepThresh offers an advanced computational approach for ECM analysis, enhancing accuracy in defining ECM-cell associations.
Comprehensive Tissue Classification- LisaClust clustering facilitates detailed segmentation of lung microenvironments, improving understanding of localized tissue remodeling.Limitations
ECM Isoform Inconsistencies - The study lacks precise annotation of ECM isoforms, which affects the accuracy of reported ECM-cell interactions.
Ambiguous Spatial Correlations- Some ECM clusters, such as "adventitial collagen," overlap inconsistently with anatomical regions, making interpretation challenging.
Unvalidated DeepThresh Method - The manuscript does not provide sufficient validation of DeepThresh's ability to remove staining artifacts.
Lack of Supplemental Data- Key activated cell types and regions lack supporting data for classification.Advance, gap filled
Clearly the next step to improve organ compendia such as the lung cell atlas, spatial protein analysis is warranted. scRNA-Seq in particular for ECM molecules is challenging as these molecules are produced in small quantities or have a very slow turn-over. This study has the potential to provide novel deep learning algorithms to include not only cellular markers but consider larger panels of ECM molecules and their spatial orientation in the lung.
Audience
The manuscript is interdisciplinary located between advanced image analysis with deep learning methods, fundamental lung biology and single cell analysis. The readership would entice molecular biologists, bioinformaticians and basic disease model scientists. The manuscript would appeal to clinician scientists and a broader audience if human tissue pendants could be provided validating the methods and outcomes.
Own Expertise
Translational scientist in the field of chronic lung disease, highly familiar with epithelial cells, mouse models, human cohorts and next generation sequencing and imaging of live single cells.
-
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 study, Parkinson et al. investigated lung extracellular matrix using imaging mass cytometry (IMC) in mouse models. Overall, the paper is well-written, and the data are clear, although major points outlined below need to be addressed. In its current form, the paper appears more like a methods-focused study since, to my understanding, no new biological responses are described. The methods employed are very interesting, particularly the extracellular matrix analysis. However, the scope of the study is quite limited, as all the experiments were performed with mouse samples, which are relatively easy to work with, and the cell …
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 study, Parkinson et al. investigated lung extracellular matrix using imaging mass cytometry (IMC) in mouse models. Overall, the paper is well-written, and the data are clear, although major points outlined below need to be addressed. In its current form, the paper appears more like a methods-focused study since, to my understanding, no new biological responses are described. The methods employed are very interesting, particularly the extracellular matrix analysis. However, the scope of the study is quite limited, as all the experiments were performed with mouse samples, which are relatively easy to work with, and the cell organisation is simple compared to humans. The authors do not discuss how this analysis pipeline could be applied to human samples. Furthermore, the entire paper relies on imaging mass cytometry, and additional techniques could have been used to confirm some of the observations, especially given the availability of mouse samples. The introduction mentions the DRA model without providing an explanation of what it involves. Non-specialist readers may not be familiar with this abbreviation, and further clarification should be provided. In the methods section, it is not mentioned whether the lungs were inflated before tissue collection, which is crucial for preserving normal cellular organization. The authors should clarify whether this was performed. Figure 1 provides a brief summary of the methods employed in the study but could be enriched with additional information. In its current state, it does not provide meaningful insights beyond what is described in the methods section. It would be helpful if the authors clarified whether the mice used were adults and whether both male and female animals were included. Additionally, they could present examples of the cell segmentation approach with zoomed-in images at the cellular level to illustrate the analysis. The first set of data in Figure 2 suggests that C57Bl/6 mice did not respond to allergen treatment, as shown by the non-significant increase in cell numbers. The authors should provide evidence that their model induced inflammation through alternative methods, such as assessing eosinophil counts or pathology. The UMAP representation indicates significant overlap between cell clusters, which raises concerns about the accuracy of cell segmentation. For example, the heatmap in Supplementary Figure 1 shows endothelial cells expressing markers such as VWF, aSMA, Vimentin, and PDGFRα, suggesting that the cell cluster may contain a mixture of endothelial cells, vascular smooth muscle cells, and fibroblasts. Examples of segmented cells should be shown to validate this approach. It is unclear what Figure 2e represents. If it is simply to show that certain clusters can be grouped together, such as AEC, AT1, and AT2 as epithelial cells, this could be conveyed in a simpler way. The analysis of extracellular matrix components presented in Figure 3 provides a novel method for studying these acellular structures, which is a challenge in the field. The authors should be commended for their efforts in this area. However, the parameters investigated in Figures 4-6 do not report any novel findings. While IMC appears to work effectively to analyse these parameters simultaneously, the induction of immune foci and changes in tissue organisation following allergen challenges are already well-documented in both mouse models and human samples.
In Figure 5, the authors show a decrease in neutrophil numbers in challenged mice. This is unexpected, as this model is widely known to induce strong neutrophil recruitment. The authors should clarify this finding and investigate whether neutrophil chemoattractants are increased in these samples. When analysing epithelial cells, the authors separate AT1 and AT2 cells based on podoplanin expression. However, data in Supplementary Figure 4b suggest that both cell types express similar levels of podoplanin. The authors do not provide statistical validation for the claim that AT1 cells express higher levels. Additionally, as podoplanin is expressed by various cell types, such as lymphatic endothelial cells, additional markers are required to confirm the identity of AT1 cells. The last set of data in Figure 7 is interesting and shows that immune cells interact with a population of S100a4 fibroblasts. This finding could be expanded further, as CD11b and Ly6C are expressed by a variety of immune cells. The authors should include additional staining to identify the specific cell types involved, such as monocytes, eosinophils, or airway macrophages. Furthermore, the authors should speculate on why these fibroblast regions attract immune cells. Are these regions enriched in chemokines or other factors? The discussion is engaging but focuses more on methodological aspects than new biological insights. Without mechanistic links, it is challenging to draw meaningful biological conclusions.
Significance
The study of Parkinson et al. provides interesting methodological insights into the use of imaging mass cytometry (IMC) to analyse lung architecture following inflammation. The application of multiplex antibody staining will leverage important information related to how tissues are adapting to chronic immune response. Here the authors rely entirely on mouse models for their studies and compared two lines of WT animals and the same allergen model. This limits the scope of the study, additional timepoints, sex or age would have improved the manuscript.
The applicability of this analysis pipeline to human tissue samples is not discussed, which would significantly enhance the impact of the study. Additionally, complementary techniques, such as flow cytometry or immunohistochemistry, could be used to validate the findings and improve reproducibility. A specialised audience of immunology researchers would be interested by the image analysis approach.
-
-