Developmental programmes drive cellular plasticity, disease progression and therapy resistance in lung adenocarcinoma

Article activity feed

1. 

   Review Commons

   Sep 1, 2025

   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

   Note to all Reviewers:

   We would like to thank all the reviewers for their time and insightful feedback. In response to the comments and points raised, we have performed major revisions to our manuscript. We have expanded our analysis on the role of TP53 loss of function in BM activation (Figure 3), investigating human LUAD datasets as well as murine LUAD models. We show that TP53 pathway is significantly negatively correlated with BM, and that loss of TP53 leads to the acquisition of the basal-like phenotype regardless of the type of driver oncogene present (KRAS/EGFR). Furthermore, we added a new figure (Figure 7), where we demonstrate that type I …

   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

   Note to all Reviewers:

   We would like to thank all the reviewers for their time and insightful feedback. In response to the comments and points raised, we have performed major revisions to our manuscript. We have expanded our analysis on the role of TP53 loss of function in BM activation (Figure 3), investigating human LUAD datasets as well as murine LUAD models. We show that TP53 pathway is significantly negatively correlated with BM, and that loss of TP53 leads to the acquisition of the basal-like phenotype regardless of the type of driver oncogene present (KRAS/EGFR). Furthermore, we added a new figure (Figure 7), where we demonstrate that type I interferon can promote BM activation in LUAD harboring TP53 mutations but not in those with wild type TP53. With this, we propose a mechanism of action of how a subset of LUAD tumors (TP53-mut) upregulate BM, become more aggressive and resistant to therapies.

   Finally, we have made the manuscript clearer and transparent by improving the presentation of plots, as well as including source data files and Rmarkdown files for reproducibility.

   Reviewer 1:

   Major comments

   R1-Comment 1: The authors did not submit with the manuscript all the results that they have obtained from their analysis, on which they based their claims. I suggest that the authors submit a SourceData file in Excel format. This file should contain the values and the relevant information for each of the plots presented in the main and supplementary figures. For example, in case of box plots, the five-number summary should be provided. Further, the p-values and the test used for their calculations should be also mentioned. The file could be organized in a way that the data and relevant information for each figure panel are presented in separated data sheets in order that the reader can easily navigate through the file and find the information for each figure panel fast. Similarly, the authors should provide access to the scripts that they have developed or adapted from published scripts to perform the analysis of the datasets and obtain the results presented in the manuscript. The access to the scripts used in the manuscript is important to reproduce analysis. The scripts can be deposited at github, for example.

   Reply: We thank the reviewer for their advice in making the presentation of our results and methods more transparent and reproducible. We have now provided the source data file (supplementary file 2), which contains relevant data for each figure. We have also uploaded Rmarkdown files to github and R Markdown HTML reports are compiled in Supplementary File 3, this shows how the analyses were performed and how each figure generated. All datasets required to reproduce the analyses and figures have also been added to Zenodo (10.5281/zenodo.16964654) and will be published when the article is in press.

   R1-Comment 2: The results and their interpretations are mainly done based on in silico analysis from publicly available transcriptomic datasets. The confirmation of the results obtained by the in silico analysis is limited to the last figure, in which the authors show results obtained by multiplexed immunohistochemistry and histo-cytometry of tissue microarrays from a curated cohort of FFPE samples. The relevance of the results obtained by the in silico analysis may increase if the authors could present results either in a conditional (lung-specific) Kras mutant mouse model, or patient-derived xenograft (PDX) mouse model of lung cancer. The PDX mouse model will be more suitable in case that access to genetically modified mouse models is not given and/or the time for the experiments is limited. In both cases, the hyperactivation of the small GTPase KRAS should expand the BM gene expression signature in the mouse lung in a Sox9-dependent manner, thereby leading to lung tumors. Further, Sox9 loss-of-function experiments should reduce the BM gene expression signature and favor the ALV gene expression signature. These results would strongly support the interpretation of the in silico results by the authors in the present manuscript, and would significantly increase the impact of the manuscript in the scientific field of lung cancer.

   Reply: We thank the reviewer for their insightful feedback on how to improve the impact of our study through further functional validation of in silico findings. To address this comment, we have performed additional analyses, including data and experiments from both murine and human LUAD model systems to elucidate a novel mechanism of BM activation in LUAD. We appreciate the reviewer’s suggestion to pursue analysis of Sox9 involvement in regulating BM activation and agree that both KRAS and SOX9 activation are likely to be involved in at least some elements of the process of disease progression we described in this manuscript. Indeed, previous studies have completed the experiments suggested, demonstrating Sox9 knock-out reduced Kras driven tumour progression and morphological grade in vivo (PMID: 37258742 and 34021911); and was associated with loss of AT2 lineage identity (PMID: 37468622).

   Our analysis of human LUAD using scRNA-seq data has demonstrated that this differentiation spectrum in fact extends beyond loss of lineage fidelity and in a subset of cells leads to transdifferentiation to a basal-like cell state. In our revised manuscript, we have more clearly elucidated the role of KRAS and TP53 in these two events during LUAD progression, demonstrating that while oncogenic KRAS (and likely downstream SOX9 activation) can lead to the loss of lineage commitment in LUAD cells, mutations in TP53 are required for acquisition of the basal-like phenotype. We have also expanded on this mechanism identifying a novel role for type-1 interferon signaling in the presence of TP53 loss-of-function as a mechanism that can lead to BM activation and acquisition of a basal-like cell state in LUAD. The data related to these analyses are now presented in figures 3 and 7.

   In accordance with the 3Rs principles for ethical use of animals in research we have taken advantage of publicly available data from previous experiments analyzing conditional (lung-specific) Kras mutant mouse model to validate our in-silico findings. This confirmed our in silico analysis of human LUAD, demonstrating an important role for TP53 loss of function in regulating BM activation (presented in Figure 3E&F and Figure S3F&G)

   We also showed that the type I interferon signaling is capable of driving BM activation in LUAD but only in the context of TP53 loss-of-function. These experiments were performed using 3D organotypic cultures of H441 cells (human adenocarcinoma cell line with mutant TP53) and A549 cells (human adenocarcinoma cell line with wild-type TP53). These 3D cultures were treated with IFN-alpha, both BM and basal-like marker upregulation (MKI67, CDC20, TOP2A, S100A9, S100A2, SOX9 and KRT17) was observed only in LUAD cells carrying a mutation in TP53. These data are now presented in Figure S7D.

   R1-Comment 3: In general, the description of the results in the corresponding section of the manuscript can be improved to facilitate the understanding of the results presented.

   As an example, the figure 1B is described on page 13 as follows: "...we first used a publicly available microarray dataset [9] to identify genes differentially expressed between epithelial cells engaged in BM (embryonic day 14 [E14]) or ALV (embryonic day 19 [E19]) (Figure 1B and TABLE S1)." By looking at the plot in figure 1B, this description is not sufficient to understand what the authors present in this figure panel, not even after reading the corresponding figure legend.

   Reply: We thank the reviewer for their advice on making our manuscript clearer. Throughout the manuscript we have now edited the result descriptions, we have also provided further detail to the methods sections, figure legends and axes labels to enhance clarity and facilitate understanding of the analyses performed.

   In the example cited we have edited the sections referenced above as follows:

   “To test this hypothesis, we identified genes that were differentially expressed in epithelial cells engaged in active BM (corresponding to embryonic day 14) vs active ALV (corresponding to embryonic day 19), using a publicly available microarray dataset.”

   We have also changed the Y axis label of Figure 1B to: “log2(FC E19 [ALV] – E14[BM])”.

   The description in the figure legend has also been modified to provide more context: “Dot plot showing the identification of genes differentially expressed by epithelial cells during murine developmental-BM (embryonic day 14) and ALV (embryonic day 19) [1]. Genes with the highest Fold Change of expression between day 14 (BM) and 19 (ALV) of murine lung development are coloured green or red, respectively. These genes were used to generate ALV/BM signatures [9]”

   R1-Comment 5: Another example is the description of the figure 3B on page 16: "This showed low levels of BM activation in tumour cells from residual disease (RD) that was significantly increased in samples with recurrent progressive disease (PD) (Figure 3B)." By looking figure 3B and the corresponding figure legend, one cannot find the group "residual disease (RD)".

   __Reply: __We thank the reviewer for their diligent reading and have now corrected the figures to provide clearer labelling of axes and maintain consistency throughout. In the example cited, we have corrected the axis label to Residual disease (RD) and partial response (PR).

   R1-Comment 6: Another example is the description of the figure 3C and 3D on page 16: "Single-cell analysis showed that both ALV-BM- and ALV-BM+ LUAD cells were increased in samples from recurrent progressive disease (Figure 3C,D)." By looking at figure 3D and the corresponding legend, I do not find the explanation of "TRUE" and "FALSE". The same is for figures 3J and 3M.

   Reply: For this example (Figure 3 in the original manuscript is now figure 4), TRUE/FALSE labels have been replaced by PR (partial response) and PD (progressive disease) in panel D; replaced by “Responder (R)” or “Non-responder (NR)” in panels J&M.

   R1-Comment 7: Other figure panels were also poorly described in the results section and in the corresponding legends. Further, the presentation of the results in the main and supplementary figures has to be improved. For example, labeling of the Y-axis in the figures 1H to 1J, 2C, 2D, 2G, 2H, 3B, 3C, 3J, 3L, etc. has to be improved. As a point of reference, I would suggest checking how other authors present similar results in life science journals. These deficiencies in the presentation and description of the results make it difficult for the readers to understand the manuscript.

   Reply: These axes labels have been changed throughout to provide more information. “BM” changed to “BM (ssGSEA score)” or “BM (module score)” and “ALV” changed to “ALV (ssGSEA score)” or “ALV (module score)” for figures 1H, 1I, 1J, S1H-L, 2C, 2D, 3B, 3F, S3E, S3F, S3G, 4B, 4C, 4J, 4L, S4C, S4D, S5A, 6A, S6A, S6B, ssGSEA score was applied to bulk RNAseq samples, and modules scores were calculated for single cells.

   Additionally:

   S2A, S2B – OS label changed to Survival probability/OS probability.

   S4H – y axis label changed to PDL1 (RPPA).

   S3B – y axis label changed to “Tumour mutational burden (mut/mB).

   S3C – y axis label changed to “Tobacco smoking (SBS mutational signature)”.

   4F – y axis label changed to “DFS (proportion)”.

   4H – y axis label changed to “PFS (proportion)”.

   R1-R1-Comment 8: The authors write on page 18 "Despite AT2 cells being well described as the cell of origin for LUAD, this population was significantly less abundant in LUAD samples compared to control, demonstrating a high degree of transcriptomic plasticity within LUAD epithelium (Figure 4D)." How can the authors show that these results are not produced by the process of integration of the four scRNA-seq NSCLC datasets, the implementation of a specific machine learning classifier for the cell type-classification, or the manually filtration to exclude doublets? For example, will the authors achieve the same (or similar) results using a different machine learning classifier? If yes, please include the results in the manuscript.

   Reply: The integration was performed using the method described by Stuart et al. (PMID: 31178118), implemented in the Seurat package. The term “machine learning classifier” has now been replaced by “label transfer” to clarify the method used and avoid confusion. Label transfer was only used to identify major cell types in the datasets used, i.e. the whole epithelial population. Doublet removal was performed as follows (and described in the methods section): epithelial cells were clustered using the shared nearest neighbor (SNN) modularity optimization algorithm implemented by the FindClusters function in the Seurat R package, based on 30 principal components and setting the resolution parameter to 0.1. This clustering solution identified multiple small clusters with divergent expression profiles to the majority of cells that were initially classified as epithelial (in the label transfer analysis). Manual examination of the marker genes for these small clusters showed they were characterized by expression of epithelial genes alongside canonical markers for either B cells (CD79A), macrophages (CD68, SPP1, APOE, CD14, MARCO) or Tcells/NK cells (CD3D, NKG7, CXCR4). These cells were therefore classed as heterotypic doublets and excluded from further analysis. All other cell types from the integrated datasets were analyzed in the same way, and no further epithelial clusters (that were not small clusters of doublets) were identified.

   Further clustering to identify epithelial subpopulations was performed on the integrated dataset and the results presented from this analysis represent the clustering solution that ensures all subpopulations were identified across datasets to mitigate any potential batch-effects not resolved by the integration process. Furthermore, our results showing that LUAD cells exhibit a high degree of transcriptomic plasticity were also confirmed by the lineage fidelity analysis (Figure 5G&I), which demonstrates this observation is not dependent on a single clustering, integration or machine learning algorithm. This observation is also supported by other studies that have described loss of lineage commitment during LUAD tumorigenesis, where tumour cells become transcriptionally and phenotypically distinct from healthy AT2 cells.

   Reviewer 1:

   Minor comments:

   __R1-Comment 9: __Please introduce the abbreviation for alveogenesis the first time that is used in the abstract, as it was done for branching morphogenesis.

   __Reply: __Abbreviation for alveogenesis has now been added to the abstract.

   R1-Comment 10: On page 18 the author write: "Consistent with the analyses presented above, pseudo bulk expression profiles for each sample showed that ALV and BM scores were significantly negatively correlated (r = -0.68, p = 4.1e-09)." Where are these results shown? I was not able to find these results. If they are not in the current version of the manuscript, please include the results

   Reply: Scatter plot showing the negative correlation has now been added as Figure S5A.

   __R1-Comment 11: __The authors should submit a supplementary table containing a list of the different data sets that were used for this manuscript. The table should include accession numbers and links to the different depositories, in which the data sets can be found. This will improve the overview of the datasets used in the study, as well as facilitate the finding of the datasets by the readers.

   Reply: The list of all datasets used in this study, together with accession numbers and links are now in Supplementary file1.

   R1-Comment 12: In figure 1G, change the color for FALSE in the legend.

   Reply: Color for FALSE changed in Figure 1G and Figure S1E.

   R1-Comment 13: Provide the complete list of mutated genes for Figure S2C.

   Reply: Figure S2C has been replaced by figures 3C (top mutated genes in LUAD-BM) and S3A (top mutated genes in LUAD-ALV).

   Reviewer 1 (Significance (required)):

   __R1-Comment 14: __Conceptually, Bienkowska KJ et al. propose that LUAD tumors undergo reversion from an alveogenic to branching morphogenic phenotype during disease progression, generating inflamed or basal-like cell states that are variably persistent following TKI or ICB treatments. This concept is in line with reports using murine models of Kras-driven LUAD. In addition, there are parallels with findings in idiopathic pulmonary fibrosis (IPF, another hyperproliferative lung disease), in which KRT5-/KRT17+ basaloid cells were transiently found, like the basal-like phenotype that Bienkowska KJ identified in human LUAD. In other words, the concept proposed by the authors is novel and in line with previous publication in LC and IPF.

   Response: We are glad the reviewer found our results novel and appreciated how they provide a linkage of previously defined mechanisms seen in murine developmental models to human cancer progression, and how they may be relevant for other diseases such as IPF.

   __R1-Comment 15: __The in silico analysis of publicly available transcriptomic datasets presented by Bienkowska KJ et al. is original and comprehensive. It is an interesting contribution to the cancer research field. However, the impact of their findings to this scientific field will significantly increase if the authors could confirm the interpretation of their results using other experimental systems in addition to the one used in the las figure. For example, the experiments that I suggested in point 2., using either conditional Kras transgenic mice or a PDX mouse model for lung cancer will not only confirm the concpet proposed by the authors, but it will also provide further mechanistic insides related to this model at cellular and molecular level.

   Response: We thank the reviewer for describing our analysis as original and comprehensive and their suggestion to develop the manuscript further with additional mechanistic analyses. We have comprehensively examined the mechanisms responsible for regulating BM activation using a combination of in vivo models and 3D organotypic cultures, elucidating a novel role for type-1 interferon signaling in the presence of TP53 loss-of-function as a mechanism that can lead to BM activation and acquisition of a basal-like cell state in LUAD. For further information regarding these additions to the revised manuscript, we direct the reviewer to the response provided to R1-comment 2 (above).

   __R1-Comment 16: __Overall, the manuscript by Bienkowska KJ et al. addresses topics that are relevant to the field of lung cancer, the leading cause of cancer-related deaths worldwide. The bioinformatic methods implemented are cutting-edge. However, the text of the manuscript and the presentation of the results in the figures have to be improved to better exploit the potential of their findings. In addition, further experiments should be performed to confirm (and perhaps complement) the interpretation of their findings. I hope that my comments support the authors to improve the manuscript to reach the standard of manuscripts recently published at renowned journals in Review COMMONS. I recommend a major revision of the manuscript before publication.

   __Reply: __We are pleased to read that the reviewer found the methods implemented by us to be cutting-edge, and that they recognized the relevance of this topic to the lung cancer field.

   We thank the reviewer for their comments, which have helped us to significantly improve our manuscript.

   We have made changes to how we present our data (as described in responses above) and performed further analyses to support our original findings. We have also now performed further in silico and functional analyses to expand and complement our original findings.

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

   __R2-Comment 1: The study is novel and interesting, but the mechanisms how the dysregulation of developmental program was driven by specific oncogene and how to link these signatures to therapy were also not clear. __

   __Reply: __We are pleased that the reviewer finds our study to be novel and interesting. We appreciate the reviewer pointing out the need to clarify the role of specific oncogenes to BM activation and response to therapies.

   We have now added further analyses and edited the text to examine and explain how the ALV and BM signatures are driven by different oncogenes (Figure 3 and results section “TP53 loss of function is required for BM activation”), which showed that common oncogenic drives (e.g. KRAS and EGFR) can drive reduced ALV signature expression but TP53 mutations (or deletion in murine models) was critical for driving BM activation. Implications for therapy response are shown in (Figure 4). We have shown that BM activation is a key determinant of tyrosine kinase inhibitor (TKI) resistance in LUAD, representing a frequently activated off-target mechanism of resistance that supersedes the presence of an actionable oncogenic driver in terms of response rates; and that the BM signature also identified patients that, although positive for immune checkpoint blockade (ICB) response biomarkers, will likely fail to respond to this treatment. In the manuscript we have now thoroughly revised these sections of the results to clarify the details associated with these conclusions (results sections: “BM activation is associated with targeted-therapy resistance in lung adenocarcinomas” and “BM activation predicts poor response to immune-checkpoint blockade”).

   We have also added further data to the manuscript elucidating the molecular mechanisms regulating BM activation (Figure 7), which has identified an important role for aberrant type-I interferon signaling in the context of mutant TP53.

   Reviewer #2 (Significance (Required)):

   __R2-Comment 2: __The authors in this manuscript aimed to examine the role of developmental programmes, alveogenesis and branching morphogenesis (BM), in regulating phenotypic diversity in NSCLC. They demonstrated that developmental programmes (ALV and BM) frequently become

   dysregulated in NSCLC, with BM activation identifying aggressive LUAD that were resistant

   to multiple therapies, including TKIs and ICB. They found that BM activation in LUAD was associated with TP53 pathway mutations and required AT2 cells to lose their alveolar identity, acquiring a basallike state. The study is very intriguing, and the findings may pave a link to the disease progression and therapy resistance in LUAD.

   __Response: __We are pleased the reviewer found the study intriguing and with the potential to better understand LUAD progression and resistance to therapies.

   __R2-Comment 3: __The current results presented, although comprehensively presented, is still many an association study, the mechanisms how these dysregulations of developmental programmes driven by the driver oncogenes or carcinogens are still unknown.

   Response: We thank the reviewer for challenging us to further examine the molecular mechanisms underpinning our initial observations. As described above (see response to Reviewer #1 comment 2), we have performed additional in silico and mechanistic experimental analyses, which identified a novel role for type-I IFN signaling and TP53 loss-of-function in the activation of the BM program in LUAD. We hope these additions have enhanced the significance of the manuscript presented.

   __R2-Comment 4: __The NSCLC is a heterogeneous disease, LUAD and LUSC are two different diseases in terms of oncogenesis, driver mutations and response to treatment. The manuscript may either just focused on LUAD or describe results carefully to include both LUAD and LUSC. For example, in the result of abstract, only LUAD was described, there was no mention of LUSC.

   __Response: __We agree with the reviewer that NSCLC is a heterogeneous and complex disease. Indeed, this was in part what motivated us to investigate the role played by developmental processes in these distinct oncogenic processes. Our analyses showed that LUSC tumors were generally high for the BM signature (Figure 1I), which likely contributed to why this signature did not stratify survival rates for LUSC (Figure S2B). As a result, we opted to focus on LUAD as we found that BM activation was predictive of disease progression and survival in this NSCLC subtype. However, we did not completely remove LUSC from our manuscript to examine the degree to which LUAD tumors upregulating BM become “LUSC-like” and evaluate whether histological transformations occurred in LUAD cases with BM activation (as described in Figure 5 and the “BM activation in LUAD is associated with a basal-like phenotype” results section).

   We have also now added a description of results from both LUAD and LUSC analyses to the abstract to clarify these points.

   __R2-Comment 5: __The most common driver mutation of LUAD was EGFR, the authors also try to link the BM activation link to TKI resistance. I assumed that the TKIs most of the patients used were EGFR TKI, but the study did not examine the role of EGFR in the dysregulation of developmental programmes.

   __Response: __We would like to thank the reviewer for highlighting an important aspect of how our work fits with current clinical practice in LUAD management. Our analyses were carried out over multiple cohorts that include different patient demographics, which have varied prevalence for specific oncogenic driver mutations (with EGFR mutations typically being more prevalent in Asian cohorts and KRAS mutations generally being the most common oncogenic driver in Western cohorts). To examine these two common oncogenic drivers impact on BM activation, we now include a direct analysis of BM level in cases harboring these mutations (Figure S3D-E). This showed that that irrespective of oncogenic driver mutations TP53 loss of function was associated with increased BM. Our new analysis of KRAS driven mouse models has also showed that KRAS activation is sufficient to induce reduced expression of the ALV signature but failed to elicit increased BM activation. Given our analysis of human tumours showed that EGFR mutant LUAD cases with wild-type TP53 had low levels of BM activation (Figure S3D), we have no reason to suspect that EGFR mutations alone would be sufficient to elicit BM activation.

   __R2-Comment 6: __The TKI resistance was very complicated, not just EGFR T790M, the results and discussion regarding the activation of BM and TKI resistance seems not adequate. The mouse model used by Dr. Chang was mainly KRAS driven mouse lung cancer model (mice carrying RosatdT, Sox2EGFP, ShhCre, Sox9CKO, Fgfr2CKO, RosamTmG, Sox9CreER, Nkx2.1CreER, and KrasLSL-G12D alleles). It is not clear whether the EGFR driven (the most common driver of LUAD) mouse model also has same genetic signature. At least, the authors should describe or discuss these discrepancies.

   __Response: __We thank the reviewer for their comments and advice on making our manuscript clearer. We have now revised the description of BM activation and TKI resistance in the results section (titled “BM activation is associated with targeted-therapy resistance in lung adenocarcinomas”).

   Read the original source
2. 

   Review Commons

   Sep 1, 2025

   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

   The study is novel and interesting, but the mechanisms how the dysregulation of developmental program was driven by specific oncogene and how to link these signatures to therapy were also not clear.

   Significance

   The authors in this manuscript aimed to examine the role of developmental programmes, alveogenesis and branching morphogenesis (BM), in regulating phenotypic diversity in NSCLC. They demonstrated that developmental programmes (ALV and BM) frequently become dysregulated in NSCLC, with BM activation identifying aggressive LUAD that were resistant to multiple therapies, including TKIs and ICB. They found that BM activation in …

   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

   The study is novel and interesting, but the mechanisms how the dysregulation of developmental program was driven by specific oncogene and how to link these signatures to therapy were also not clear.

   Significance

   The authors in this manuscript aimed to examine the role of developmental programmes, alveogenesis and branching morphogenesis (BM), in regulating phenotypic diversity in NSCLC. They demonstrated that developmental programmes (ALV and BM) frequently become dysregulated in NSCLC, with BM activation identifying aggressive LUAD that were resistant to multiple therapies, including TKIs and ICB. They found that BM activation in LUAD was associated with TP53 pathway mutations and required AT2 cells to lose their alveolar identity, acquiring a basallike state.

   The study is very intriguing and the findings may pave a link to the disease progression and therapy resistance in LUAD. The current results presented, although comprehensively presented, is still many an association study, the mechanisms how these dysregulations of developmental programmes driven by the driver oncogenes or carcinogens are still unknown. The NSCLC is a heterogeneous disease, LUAD and LUSC are two different diseases in terms of oncogenesis, driver mutations and response to treatment. The manuscript may either just focused on LUAD or describe results carefully to include both LUAD and LUSC. For example, in the result of abstract, only LUAD was described, there was no mention of LUSC. The most common driver mutation of LUAD was EGFR, the authors also try to link the BM activation link to TKI resistance. I assumed that the TKIs most of the patients used were EGFR TKI, but the study did not examine the role of EGFR in the dysregulation of developmental programmes. The TKI resistance was very complicated, not just EGFR T790M, the results and discussion regarding the activation of BM and TKI resistance seems not adequate. The mouse model used by Dr. Chang was mainly KRAS driven mouse lung cancer model (mice carrying RosatdT, Sox2EGFP, ShhCre, Sox9CKO, Fgfr2CKO, RosamTmG, Sox9CreER, Nkx2.1CreER, and KrasLSL-G12D alleles). It is not clear whether the EGFR driven (the most common driver of LUAD) mouse model also has same genetic signature. At least, the authors should describe or discuss these discrepancies.

   Read the original source
3. 

   Review Commons

   Sep 1, 2025

   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

   Summary:

   Bienkowska KJ et al show in this manuscript a compilation of bioinformatic analysis of publicly available microarray datasets, bulk RNA sequencing (RNA-seq) datasets and single cell RNA sequencing (scRNA-seq) datasets. One transcriptomic data set from mouse (Chang DR et al., Proc Natl Acad Sci USA, 2013) was analyzed in this manuscript to determine the gene expression signatures specific for the developmental processes alveogenesis (ALV) and branching morphogenesis (BM). The rest of the transcriptomic data sets that were analyzed for this manuscript were selected based on different parameters including the involvement of …

   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

   Summary:

   Bienkowska KJ et al show in this manuscript a compilation of bioinformatic analysis of publicly available microarray datasets, bulk RNA sequencing (RNA-seq) datasets and single cell RNA sequencing (scRNA-seq) datasets. One transcriptomic data set from mouse (Chang DR et al., Proc Natl Acad Sci USA, 2013) was analyzed in this manuscript to determine the gene expression signatures specific for the developmental processes alveogenesis (ALV) and branching morphogenesis (BM). The rest of the transcriptomic data sets that were analyzed for this manuscript were selected based on different parameters including the involvement of non-small cell lung cancer (NSCLC), lug adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC) and various cohorts with different characteristics related to lung cancer (LC), such as mutations related to LC (in EGFR, ALK, BRAF, ROS1 and KRAS), and/or treatments/resistance of LC patients with/to tyrosine kinase inhibitors (TKI), anti-PD1 strategies, immune checkpoint blockade, (ICB) among others. In the last figure, the authors present results obtained by multiplexed immunohistochemistry and histo-cytometry of tissue microarrays from a curated cohort of archival formalin-fixed paraffin-embedded (FFPE) samples to confirm their interpretation of the results obtained by the transcriptomic analysis.

   The findings and/or claims of the authors could be summarized in the following bullet points:

   • After defining the ALV and BM gene expression signatures, the authors determined the expression of these signatures in bulk RNA-seq data sets from The Cancer Genome Atlas (TCGA). The ALV signature was significantly downregulated in LUAD and LUSC tumour subtypes compared to control samples, whereas BM was upregulated. These findings were validated using a Laser Capture Micro-Dissected (LCMD) microarray dataset (Lin J et al., AM J Pathol, 2014) comparing the epithelial compartment from non-tumour alveoli, non-tumour bronchi, LUAD and LUSC. Interestingly, ALV was suppressed and BM was activated in all LUSC tumours analyzed; whereas in LUAD tumours were heterogeneous for BM activation with some cases exhibiting ALV expression comparable to control tissue. In summary, the mutually antagonistic regulation of ALV and BM was found to account for a significant proportion of transcriptomic variance in human NSCLC bulk tissue datasets.
   • BM activation was associated with poor overall survival rates in five independent LUAD cohorts (p=2.04e-13); and was significantly prognostic for resistance to TKIs (p=0.003) and ICBs (p=0.014), in pre-treatment biopsies.
   • ScRNA-seq analysis revealed that malignant LUAD cells with loss of alveolar lineage fidelity predominantly acquired inflamed or basal-like cellular states, which were variably persistent in samples from TKI and ICB recurrence.
   • The authors conclude from their analysis that LUAD tumours undergo reversion from an alveogenic to branching morphogenic phenotype during disease progression, generating inflamed or basal-like cell states that are variably persistent following TKI or ICB treatments.

   Major comments:

   1. The authors did not submit with the manuscript all the results that they have obtained from their analysis, on which they based their claims. I suggest that the authors submit a SourceData file in Excel format. This file should contain the values and the relevant information for each of the plots presented in the main and supplementary figures. For example, in case of box plots, the five-number summary should be provided. Further, the p-values and the test used for their calculations should be also mentioned.

   The file could be organized in a way that the data and relevant information for each figure panel are presented in separated data sheets in order that the reader can easily navigate through the file and find the information for each figure panel fast.

   Similarly, the authors should provide access to the scripts that they have developed or adapted from published scripts to perform the analysis of the datasets and obtain the results presented in the manuscript. The access to the scripts used in the manuscript is important to reproduce analysis. The scripts can be deposited at github, for example.

   1. The results and their interpretations are mainly done based on in silico analysis from publicly available transcriptomic datasets. The confirmation of the results obtained by the in silico analysis is limited to the last figure, in which the authors show results obtained by multiplexed immunohistochemistry and histo-cytometry of tissue microarrays from a curated cohort of FFPE samples.

   The relevance of the results obtained by the in silico analysis may increase if the authors could present results either in a conditional (lung-specific) Kras mutant mouse model, or patient-derived xenograft (PDX) mouse model of lung cancer. The PDX mouse model will be more suitable in case that access to genetically modified mouse models is not given and/or the time for the experiments is limited. In both cases, the hyperactivation of the small GTPase KRAS should expand the BM gene expression signature in the mouse lung in a Sox9-dependent manner, thereby leading to lung tumors. Further, Sox9 loss-of-function experiments should reduce the BM gene expression signature and favor the ALV gene expression signature. These results would strongly support the interpretation of the in silico results by the authors in the present manuscript, and would significantly increase the impact of the manuscript in the scientific field of lung cancer.

   1. In general, the description of the results in the corresponding section of the manuscript can be improved to facilitate the understanding of the results presented. As an example, the figure 1B is described on page 13 as follows:

   "...we first used a publicly available microarray dataset [9] to identify genes differentially expressed between epithelial cells engaged in BM (embryonic day 14 [E14]) or ALV (embryonic day 19 [E19]) (Figure 1B and TABLE S1)."

   By looking at the plot in figure 1B, this description is not sufficient to understand what the authors present in this figure panel, not even after reading the corresponding figure legend.

   Another example is the description of the figure 3B on page 16:

   "This showed low levels of BM activation in tumour cells from residual disease (RD) that was significantly increased in samples with recurrent progressive disease (PD) (Figure 3B)."

   By looking figure 3B and the corresponding figure legend, one cannot find the group "residual disease (RD)".

   Another example is the description of the figure 3C and 3D on page 16:

   "Single-cell analysis showed that both ALV-BM- and ALV-BM+ LUAD cells were increased in samples from recurrent progressive disease (Figure 3C,D)."

   By looking at figure 3D and the corresponding legend, I do not find the explanation of "TRUE" and "FALSE". The same is for figures 3J and 3M.

   Other figure panels were also poorly described in the results section and in the corresponding legends.

   Further, the presentation of the results in the main and supplementary figures has to be improved. For example, labeling of the Y-axis in the figures 1H to 1J, 2C, 2D, 2G, 2H, 3B, 3C, 3J, 3L, etc. has to be improved. As a point of reference, I would suggest checking how other authors present similar results in life science journals.

   These deficiencies in the presentation and description of the results make it difficult for the readers to understand the manuscript.

   1. The authors write on page 18

   "Despite AT2 cells being well described as the cell of origin for LUAD, this population was significantly less abundant in LUAD samples compared to control, demonstrating a high degree of transcriptomic plasticity within LUAD epithelium (Figure 4D)."

   How can the authors show that these results are not produced by the process of integration of the four scRNA-seq NSCLC datasets, the implementation of a specific machine learning classifier for the cell type-classification, or the manually filtration to exclude doublets? For example, will the authors achieve the same (or similar) results using a different machine learning classifier? If yes, please include the results in the manuscript.

   Minor comments:

   1. Please introduce the abbreviation for alveogenesis the first time that is used in the abstract, as it was done for branching morphogenesis.
   2. On page 18 the author write:

   "Consistent with the analyses presented above, pseudo bulk expression profiles for each sample showed that ALV and BM scores were significantly negatively correlated (r = -0.68, p = 4.1e-09)."

   Where are these results shown? I was not able to find these results. If they are not in the current version of the manuscript, please include the results

   1. The authors should submit a supplementary table containing a list of the different data sets that were used for this manuscript. The table should include accession numbers and links to the different depositories, in which the data sets can be found. Thiy will improve the overview of the datasets used in the study, as well as facilitate the finding of the datasets by the readers.
   2. In figure 1G, change the color for FALSE in the legend.
   3. Provide the complete list of mutated genes for Figure S2C

   Significance

   Conceptually, Bienkowska KJ et al. propose that LUAD tumors undergo reversion from an alveogenic to branching morphogenic phenotype during disease progression, generating inflamed or basal-like cell states that are variably persistent following TKI or ICB treatments. This concept is in line with reports using murine models of Kras-driven LUAD. In addition, there are parallels with findings in idiopathic pulmonary fibrosis (IPF, another hyperproliferative lung disease), in which KRT5-/KRT17+ basaloid cells were transiently found, like the basal-like phenotype that Bienkowska KJ identified in human LUAD. In other words, the concept proposed by the authors is novel and in line with previous publication in LC and IPF.

   The in silico analysis of publicly available transcriptomic datasets presented by Bienkowska KJ et al. is original and comprehensive. It is an interesting contribution to the cancer research field. However, the impact of their findings to this scientific field will significantly increase if the authors could confirm the interpretation of their results using other experimental systems in addition to the one used in the las figure. For example, the experiments that I suggested in point 2., using either conditional Kras transgenic mice or a PDX mouse model for lung cancer will not only confirm the concpet proposed by the authors, but it will also provide further mechanistic insides related to this model at cellular and molecular level.

   Overall, the manuscript by Bienkowska KJ et al. addresses topics that are relevant to the field of lung cancer, the leading cause of cancer-related deaths worldwide. The bioinformatic methods implemented are cutting-edge. However, the text of the manuscript and the presentation of the results in the figures have to be improved to better exploit the potential of their findings. In addition, further experiments should be performed to confirm (and perhaps complement) the interpretation of their findings. I hope that my comments support the authors to improve the manuscript to reach the standard of manuscripts recently published at renowned journals in Review COMMONS. I recommend a major revision of the manuscript before publication.

   Read the original source
4. 

   Version published to 10.1101/2024.12.03.626626 on bioRxiv

   Dec 6, 2024