Transcriptional pattern enriched for synaptic signaling is associated with shorter survival of patients with high-grade serous ovarian cancer

  1. Department of Medical Oncology, University Medical Center Groningen, University of Groningen, Groningen, the Netherlands
  2. Cancer Biology and Immunotherapies Group, Sanford Research, Sioux Falls, United States
  3. Penn Ovarian Cancer Research Center and Basser Center for BRCA, University of Pennsylvania, Perelman School of Medicine, Philadelphia, United States
  4. Department of Obstetrics and Gynecology, University Medical Center Groningen, University of Groningen, Groningen, the Netherlands

Peer review process

Not revised: This Reviewed Preprint includes the authors’ original preprint (without revision), an eLife assessment, public reviews, and a provisional response from the authors.

Read more about eLife’s peer review process.

Editors

  • Reviewing Editor
    Katherine Lawler
    University of Cambridge, Cambridge, United Kingdom
  • Senior Editor
    Tony Ng
    King's College London, London, United Kingdom

Reviewer #1 (Public review):

Summary:

This manuscript explores the transcriptional landscape of high-grade serous ovarian cancer (HGSOC) using consensus-independent component analysis (c-ICA) to identify transcriptional components (TCs) associated with patient outcomes. The study analyzes 678 HGSOC transcriptomes, supplemented with 447 transcriptomes from other ovarian cancer types and noncancerous tissues. By identifying 374 TCs, the authors aim to uncover subtle transcriptional patterns that could serve as novel drug targets. Notably, a transcriptional component linked to synaptic signaling was associated with shorter overall survival (OS) in patients, suggesting a potential role for neuronal interactions in the tumor microenvironment. Given notable weaknesses like lack of validation cohort or validation using another platform (other than the 11 samples with ST), the data is considered highly descriptive and preliminary.

Strengths:

(1) Innovative Methodology:
The use of c-ICA to dissect bulk transcriptomes into independent components is a novel approach that allows for the identification of subtle transcriptional patterns that may be overshadowed in traditional analyses.

(2) Comprehensive Data Integration:
The study integrates a large dataset from multiple public repositories, enhancing the robustness of the findings. The inclusion of spatially resolved transcriptomes adds a valuable dimension to the analysis.

(3) Clinical Relevance:
The identification of a synaptic signaling-related TC associated with poor prognosis highlights a potential new avenue for therapeutic intervention, emphasizing the role of the tumor microenvironment in cancer progression.

Weaknesses:

(1) Mechanistic Insights:
While the study identifies TCs associated with survival, it provides limited mechanistic insights into how these components influence cancer progression. Further experimental validation is necessary to elucidate the underlying biological processes.

(2) Generalizability:
The findings are primarily based on transcriptomic data from HGSOC. It remains unclear how these results apply to other subtypes of ovarian cancer or different cancer types.

(3) Innovative Methodology:
Requires more validation using different platforms (IHC) to validate the performance of this bulk-derived data. Also, the lack of control over data quality is a concern.

(4) Clinical Application:
Although the study suggests potential drug targets, the translation of these findings into clinical practice is not addressed. Probably given the lack of some QA/QC procedures it'll be hard to translate these results. Future studies should focus on validating these targets in clinical settings.

Reviewer #2 (Public review):

Summary:

Consensus-independent component analysis and closely related methods have previously been used to reveal components of transcriptomic data that are not captured by principal component or gene-gene coexpression analyses.

Here, the authors asked whether applying consensus-independent component analysis (c-ICA) to published high-grade serous ovarian cancer (HGSOC) microarray-based transcriptomes would reveal subtle transcriptional patterns that are not captured by existing molecular omics classifications of HGSOC.

Statistical associations of these (hitherto masked) transcriptional components with prognostic outcomes in HGSOC could lead to additional insights into underlying mechanisms and, coupled with corroborating evidence from spatial transcriptomics, are proposed for further investigation.

This approach is complementary to existing transcriptomics classifications of HGSOC.

The authors have previously applied the same approach in colorectal carcinoma (Knapen et al. (2024) Commun. Med).

Strengths:

Overall, this study describes a solid data-driven description of c-ICA-derived transcriptional components that the authors identified in HGSOC microarray transcriptomics data, supported by detailed methods and supplementary documentation.

The biological interpretation of transcriptional components is convincing based on (data-driven) permutation analysis and a suite of analyses of association with copy-number, gene sets, and prognostic outcomes.

The resulting annotated transcriptional components have been made available in a searchable online format.

For the highlighted transcriptional component which has been annotated as related to synaptic signalling, the detection of the transcriptional component among 11 published spatial transcriptomics samples from ovarian cancers appears to support this preliminary finding and requires further mechanistic follow-up.

Weaknesses:

This study has not explicitly compared the c-ICA transcriptional components to the existing reported transcriptional landscape and classifications for ovarian cancers (e.g. Smith et al Nat Comms 2023; TCGA Nature 2011; Engqvist et al Sci Rep 2020) which would enable a further assessment of the additional contribution of c-ICA -- whether the cICA approach captured entirely complementary components, or whether some components are correlated with the existing reported ovarian transcriptomic classifications.

Here, the authors primarily interpret the c-ICA transcriptional components as a deconvolution of bulk transcriptomics due to the presence of cells from tumour cells and the tumour microenvironment.

However, c-ICA is not explicitly a deconvolution method with respect to cell types: the transcriptional components do not necessarily correspond to distinct cell types, and may reflect differential dysregulation within a cell type. This application of c-ICA for the purpose of data-driven deconvolution of cell populations is distinct from other deconvolution methods that explicitly use a prior cell signature matrix.

Author response:

eLife Assessment

This valuable study uses consensus-independent component analysis to highlight transcriptional components (TC) in high-grade serous ovarian cancers (HGSOC). The study presents a convincing preliminary finding by identifying a TC linked to synaptic signaling that is associated with shorter overall survival in HGSOC patients, highlighting the potential role of neuronal interactions in the tumor microenvironment. This finding is corroborated by comparing spatially resolved transcriptomics in a small-scale study; a weakness is in being descriptive, non-mechanistic, and requiring experimental validation.

We sincerely thank the editors for the valuable and constructive feedback. We appreciate the recognition of our findings and the significance of identifying transcriptional components in high-grade serous ovarian cancers. We acknowledge the insightful point on our study's descriptive nature and limited mechanistic depth. While further experimental validation would indeed enhance our conclusions, such work extends beyond the current scope of this manuscript. However, we would like to highlight that mechanistic studies demonstrating the impact of tumor-infiltrating nerves on disease progression are emerging (Zahalka et al., 2017; Allen et al., 2018; Balood et al., 2022; Jin et al., 2022; Globig et al., 2023; Restaino et al., 2023; Darragh et al., 2024). Importantly, members of our group have contributed to these findings. These studies, including in vitro and in vivo work in head and neck squamous cell carcinoma as well as high-grade serous ovarian carcinoma, demonstrate that substance P released from tumor-infiltrating nociceptors potentiates MAP kinase signaling in cancer cells, thereby influencing disease progression. This effect can be mitigated in vivo by blocking the substance P receptor (Restaino et al., 2023). Our present work identifies a transcriptional component that aligns with the presence of functional nerves within malignancies. These published mechanistic studies support our findings and suggest that this transcriptional component could serve as a potential screening tool to identify innervated tumors. Such information is clinically relevant, as patients with innervated tumors may benefit from more aggressive therapy.

Reviewer #1 (Public review):

This manuscript explores the transcriptional landscape of high-grade serous ovarian cancer (HGSOC) using consensus-independent component analysis (c-ICA) to identify transcriptional components (TCs) associated with patient outcomes. The study analyzes 678 HGSOC transcriptomes, supplemented with 447 transcriptomes from other ovarian cancer types and noncancerous tissues. By identifying 374 TCs, the authors aim to uncover subtle transcriptional patterns that could serve as novel drug targets. Notably, a transcriptional component linked to synaptic signaling was associated with shorter overall survival (OS) in patients, suggesting a potential role for neuronal interactions in the tumor microenvironment. Given notable weaknesses like lack of validation cohort or validation using another platform (other than the 11 samples with ST), the data is considered highly descriptive and preliminary.

Strengths:

(1) Innovative Methodology:

The use of c-ICA to dissect bulk transcriptomes into independent components is a novel approach that allows for the identification of subtle transcriptional patterns that may be overshadowed in traditional analyses.

We sincerely thank the reviewer for recognizing the strengths and novelty of our study. We appreciate the positive feedback on our use of consensus-independent component analysis (c-ICA) to decompose bulk transcriptomes, which we believe allowed us to detect subtle transcriptional signals often overlooked in traditional analyses.

(2) Comprehensive Data Integration:

The study integrates a large dataset from multiple public repositories, enhancing the robustness of the findings. The inclusion of spatially resolved transcriptomes adds a valuable dimension to the analysis.

Thank you for recognizing the robustness of our study through comprehensive data integration. We appreciate the acknowledgment of our efforts to leverage a large, multi-source dataset, as well as the additional insights gained from spatially resolved transcriptomes. We believe this integrative approach enhances the depth of our analysis and contributes to a more nuanced understanding of the tumor microenvironment.

(3) Clinical Relevance:

The identification of a synaptic signaling-related TC associated with poor prognosis highlights a potential new avenue for therapeutic intervention, emphasizing the role of the tumor microenvironment in cancer progression.

We appreciate the reviewer’s recognition of the clinical implications of our findings. The identification of a synaptic signaling-related transcriptional component associated with poor prognosis underscores the potential for novel therapeutic targets within the tumor microenvironment. We agree that this insight could open new avenues for intervention and further highlights the role of neuronal interactions in cancer progression.

Weaknesses:

(1) Mechanistic Insights:

While the study identifies TCs associated with survival, it provides limited mechanistic insights into how these components influence cancer progression. Further experimental validation is necessary to elucidate the underlying biological processes.

We appreciate the reviewer’s point regarding the limited mechanistic insights provided in our study. We agree that further experimental validation would enhance our understanding of how the biology captured by these transcriptional components influence cancer progression. However, we respectfully note that such validation is beyond the current scope of this article. Our current analyses are done on publicly available expression array and spatial transcriptomic array datasets. For future studies, we therefore intend to combine spatial transcriptomic data with immunohistochemical analysis of the same tumors for validation purposes. We have started with setting up in vitro cocultures of neurons and ovarian cancer cells to obtain mechanistic insight in how genes with a large weight in TC121 regulate synaptic signaling and how that affects ovarian cancer cells.

(2) Generalizability:

The findings are primarily based on transcriptomic data from HGSOC. It remains unclear how these results apply to other subtypes of ovarian cancer or different cancer types.

In Figure 5, we present the activity of TC121 across various cancer types, demonstrating broader applicability. However, due to limited treatment response data, we were unable to assess associations between TC activity scores and patient response. Additionally, transcriptomic and survival data specific to other ovarian cancer subtypes beyond HGSOC are currently not available, limiting our ability to generalize these findings to those groups. We intend to leverage survival data from TCGA to explore associations between TC activity scores and overall survival of patients with other cancer types. Nonetheless, we recognize limitations with TCGA survival data, as outlined in this article: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8726696/.

(3) Innovative Methodology:

Requires more validation using different platforms (IHC) to validate the performance of this bulk-derived data. Also, the lack of control over data quality is a concern.

We acknowledge the reviewer’s suggestion to validate our results with alternative platforms, such as IHC; however, we regret that such validation is beyond the scope of this article. Regarding data quality control, we implemented a series of checks:

  • Bulk Transcriptional Profiles: We applied principal component analysis (PCA) on the sample Pearson product-moment correlation matrix, focusing on the first principal component (PCqc), which accounted for approximately 80-90% of the variance, primarily reflecting technical rather than biological variability (Bhattacharya et al., 2020). Samples with a correlation below 0.8 with PCqc were removed as outliers. Additionally, we generated unique MD5 hashes for each CEL file to identify and exclude duplicate samples. Per gene, expression values were standardized to a mean of zero and a variance of one across the GEO, CCLE, GDSC, and TCGA datasets to minimize probeset- or gene-specific variability.

  • Spatial Transcriptional Profiles: We used PCA for quality control here as well, retained samples only if their loading factors for the first principal component showed consistent signs across all profiles (i.e., all profiles had either positive or negative loading factors for the first PC) from that individual spatial transcriptomic sample. Samples that did not meet this criterion were excluded from analyses.

(4) Clinical Application:

Although the study suggests potential drug targets, the translation of these findings into clinical practice is not addressed. Probably given the lack of some QA/QC procedures it'll be hard to translate these results. Future studies should focus on validating these targets in clinical settings.

While this study is exploratory in nature, we agree that future studies should focus on validating these potential drug targets in clinical settings. As suggested, QA/QC procedures were integral to our analyses. We applied rigorous quality control, including PCA-based checks and duplicate removal across datasets, to ensure data integrity (detailed in our previous response).

In terms of clinical application, which we partially discussed in the manuscript, we will discuss additional strategies to prevent synaptic signaling and neurotransmitter release in the tumor microenvironment (TME). Drugs such as ifenprodil and lamotrigine are used in treating neuronal disorders to block glutamate release responsible for subsequent synaptic signaling, whereas the vesicular monoamine transporter (VMAT) inhibitor reserpine can block the formation of synaptic vesicles (Reid et al., 2013; Williams et al., 2001). Previous in vitro studies with HGSOC cell lines showed a significant effect of ifenprodil alone on cancer cell proliferation, whereas reserpine seemed to trigger apoptosis in cancer cells (North et al., 2015; Ramamoorthy et al., 2019). Such strategies could potentially be used to inhibit synaptic neurotransmission in the TME.

Reviewer #2 (Public review):

Summary:

Consensus-independent component analysis and closely related methods have previously been used to reveal components of transcriptomic data that are not captured by principal component or gene-gene coexpression analyses.

Here, the authors asked whether applying consensus-independent component analysis (c-ICA) to published high-grade serous ovarian cancer (HGSOC) microarray-based transcriptomes would reveal subtle transcriptional patterns that are not captured by existing molecular omics classifications of HGSOC.

Statistical associations of these (hitherto masked) transcriptional components with prognostic outcomes in HGSOC could lead to additional insights into underlying mechanisms and, coupled with corroborating evidence from spatial transcriptomics, are proposed for further investigation.

This approach is complementary to existing transcriptomics classifications of HGSOC.

The authors have previously applied the same approach in colorectal carcinoma (Knapen et al. (2024) Commun. Med).

Strengths:

(1) Overall, this study describes a solid data-driven description of c-ICA-derived transcriptional components that the authors identified in HGSOC microarray transcriptomics data, supported by detailed methods and supplementary documentation.

We thank the reviewer for acknowledging the strength of our data-driven approach and the use of consensus-independent component analysis (c-ICA) to identify transcriptional components within HGSOC microarray data. We aimed to provide comprehensive methodological detail and supplementary documentation to support the reproducibility and robustness of our findings. We believe this approach allows for the identification of subtle transcriptional signals that might be overlooked by traditional analysis methods.

(2) The biological interpretation of transcriptional components is convincing based on (data-driven) permutation analysis and a suite of analyses of association with copy-number, gene sets, and prognostic outcomes.

We appreciate the reviewer’s positive feedback on the biological interpretation of our transcriptional components. We are pleased that our approach, which includes data-driven permutation testing and analyses of associations with copy-number alterations, gene sets, and prognostic outcomes, was found convincing. These analyses were integral to enhancing the robustness and biological relevance of our findings.

(3) The resulting annotated transcriptional components have been made available in a searchable online format.

Thank you for acknowledging the availability of our annotated transcriptional components in a searchable online format.

(4) For the highlighted transcriptional component which has been annotated as related to synaptic signalling, the detection of the transcriptional component among 11 published spatial transcriptomics samples from ovarian cancers appears to support this preliminary finding and requires further mechanistic follow-up.

Thank you for acknowledging the accessibility of our annotated transcriptional components. We prioritized making these data available in a searchable online format to facilitate further research and enable the community to explore and validate our findings.

Weaknesses:

(1) This study has not explicitly compared the c-ICA transcriptional components to the existing reported transcriptional landscape and classifications for ovarian cancers (e.g. Smith et al Nat Comms 2023; TCGA Nature 2011; Engqvist et al Sci Rep 2020) which would enable a further assessment of the additional contribution of c-ICA - whether the c-ICA approach captured entirely complementary components, or whether some components are correlated with the existing reported ovarian transcriptomic classifications.

We appreciate the reviewer’s insightful suggestion to compare our c-ICA-derived transcriptional components with previously reported ovarian cancer classifications, such as those from Smith et al. (2023), TCGA (2011), and Engqvist et al. (2020). To address this, we will incorporate analyses comparing the activity scores of our transcriptional components with these published landscapes and classifications, particularly focusing on any associations with overall survival. Additionally, we plan to evaluate correlations between gene signatures from these studies and our identified TCs, enhancing our understanding of the unique contributions of the c-ICA approach.

(2) Here, the authors primarily interpret the c-ICA transcriptional components as a deconvolution of bulk transcriptomics due to the presence of cells from tumour cells and the tumour microenvironment. However, c-ICA is not explicitly a deconvolution method with respect to cell types: the transcriptional components do not necessarily correspond to distinct cell types, and may reflect differential dysregulation within a cell type. This application of c-ICA for the purpose of data-driven deconvolution of cell populations is distinct from other deconvolution methods that explicitly use a prior cell signature matrix.

Thank you for highlighting this nuanced aspect of c-ICA interpretation. We acknowledge that c-ICA, unlike traditional deconvolution methods, is not specifically designed for cell-type deconvolution and does not rely on a predefined cell signature matrix. While we explored the transcriptional components in the context of tumor and microenvironmental interactions, we agree that these components may not correspond directly to distinct cell types but rather reflect complex patterns of dysregulation, potentially within individual cell populations.

Our goal with c-ICA was to uncover hidden transcriptional patterns possibly influenced by cellular heterogeneity. However, we recognize these patterns may also arise from regulatory processes within a single cell type. To investigate further, we plan to use single-cell transcriptional data (~60,000 cell-types annotated profiles from GSE158722) and project our transcriptional components onto these profiles to obtain activity scores, allowing us to assess each TC’s behavior across diverse cellular contexts after removing the first principal component to minimize background effects.

References

Allen JK, Armaiz-Pena GN, Nagaraja AS, Sadaoui NC, Ortiz T, Dood R, Ozcan M, Herder DM, Haemerrle M, Gharpure KM, Rupaimoole R, Previs R, Wu SY, Pradeep S, Xu X, Han HD, Zand B, Dalton HJ, Taylor M, Hu W, Bottsford-Miller J, Moreno-Smith M, Kang Y, Mangala LS, Rodriguez-Aguayo C, Sehgal V, Spaeth EL, Ram PT, Wong ST, Marini FC, Lopez-Berestein G, Cole SW, Lutgendorf SK, diBiasi M, Sood AK. 2018. Sustained adrenergic signaling promotes intratumoral innervation through BDNF induction. Cancer Res 78:canres.1701.2016.

Balood M, Ahmadi M, Eichwald T, Ahmadi A, Majdoubi A, Roversi Karine, Roversi Katiane, Lucido CT, Restaino AC, Huang S, Ji L, Huang K-C, Semerena E, Thomas SC, Trevino AE, Merrison H, Parrin A, Doyle B, Vermeer DW, Spanos WC, Williamson CS, Seehus CR, Foster SL, Dai H, Shu CJ, Rangachari M, Thibodeau J, Rincon SVD, Drapkin R, Rafei M, Ghasemlou N, Vermeer PD, Woolf CJ, Talbot S. 2022. Nociceptor neurons affect cancer immunosurveillance. Nature 611:405–412.

Bhattacharya A, Bense RD, Urzúa-Traslaviña CG, Vries EGE de, Vugt MATM van, Fehrmann RSN. 2020. Transcriptional effects of copy number alterations in a large set of human cancers. Nat Commun 11:715.

Darragh LB, Nguyen A, Pham TT, Idlett-Ali S, Knitz MW, Gadwa J, Bukkapatnam S, Corbo S, Olimpo NA, Nguyen D, Court BV, Neupert B, Yu J, Ross RB, Corbisiero M, Abdelazeem KNM, Maroney SP, Galindo DC, Mukdad L, Saviola A, Joshi M, White R, Alhiyari Y, Samedi V, Bokhoven AV, John MSt, Karam SD. 2024. Sensory nerve release of CGRP increases tumor growth in HNSCC by suppressing TILs. Med 5:254-270.e8.

Globig A-M, Zhao S, Roginsky J, Maltez VI, Guiza J, Avina-Ochoa N, Heeg M, Hoffmann FA, Chaudhary O, Wang J, Senturk G, Chen D, O’Connor C, Pfaff S, Germain RN, Schalper KA, Emu B, Kaech SM. 2023. The β1-adrenergic receptor links sympathetic nerves to T cell exhaustion. Nature 622:383–392.

Jin M, Wang Y, Zhou T, Li W, Wen Q. 2022. Norepinephrine/β2-adrenergic receptor pathway promotes the cell proliferation and nerve growth factor production in triple-negative breast cancer. J Breast Cancer 26:268–285.

  1. Howard Hughes Medical Institute
  2. Wellcome Trust
  3. Max-Planck-Gesellschaft
  4. Knut and Alice Wallenberg Foundation