A multi‐omic biomarker signature in pre‐treatment rectal tumours stratifies patients with different pathological responses to neoadjuvant treatment

Dear Editor, Rectal cancer (RC) incidence is rising, particularly in individuals < 50 years, who present with aggressive, treatment-refractory tumours.1 Resistance to neoadjuvant treatment (neo-tx) is a significant problem, with no biomarkers of response currently in use. Tumours of similar clinical characteristics can have vastly different responses to neo-tx, suggesting the dichotomy in response is due to differences in the tumour molecular environment. Alterations in mitochondrial function and energy metabolism play a role in the pathogenesis of gastrointestinal cancers,2, 3 implicating the metabolome as a potential untapped source of predictive biomarkers. To address this unmet need, we performed multi-omic analysis of metabolomic and transcriptomic profiles from normal, non-cancer rectal tissue and pre-treatment RC biopsies (Supporting Information) to identify alterations associated with the pathogenesis of RC. Liquid chromatography-mass spectrometry revealed 29 metabolites significantly altered in RC tissue (n = 32) compared to non-cancer rectal tissue (n = 20) (Figure 1A). Pathway analysis uncovered 65 upregulated and four downregulated pathways significantly associated with altered metabolites (Figure 1B). Most altered metabolites were lipid molecules or mediators of lipid metabolism, suggesting that remodelling of lipid metabolism is a feature of RC. Diacyl phosphatidylcholines (PCs) are important mediators of lipid metabolism, supporting other studies highlighting a role for choline metabolism and lipid remodelling in tumourigenesis.4 SM C18:0 and SM (OH) C22:1 are sphingolipids, important structural lipid components of biological membranes, which support the physiological function of the colon and are deregulated in RC.5 Interestingly, cancer cells hydrolyse sphingomyelin to maintain production of PCs,4 suggesting a mechanism for the concomitant increase in PCs and decrease in sphingomyelins demonstrated in RC here. Real-time metabolic analysis demonstrated that (Figure 1C)OCR rates and OCR/ECAR ratios were significantly decreased in RC compared to non-cancer rectal tissue, highlighting metabolic remodelling in RC. Inhibition of mitochondrial metabolism results in accelerated turnover of PCs in neuronal cells,6 suggesting a mechanism underlying the altered choline metabolism demonstrated in RC. Transcriptomics revealed 2337 genes differentially expressed between RC (n = 31) and non-cancer rectal tissue (n = 28) (Figure 1D). Pathway analysis revealed 41 upregulated and seven downregulated pathways significantly associated with altered genes (Figure 1E). Interestingly, several of the most altered genes play roles in mitochondrial respiration. ND2, ND3 and ND5 encode subunits of the NADH dehydrogenase enzyme, a crucial component in the electron transport chain, supporting the altered OCR demonstrated in RC tissue. In pre-treatment RC biopsies, altered metabolites had significant correlations with OCR, ECAR, OCR/ECAR ratios, and several clinical variables, including the modified Ryan Tumour Response Score (TRS) (TRS0 = complete response; TRS1 = near complete response; TRS2 = partial response) (Figure 2A). Two metabolites correlated significantly with TRS: Serotonin and lysoPC a C16:1 (Figure 2B). Unsupervised hierarchical clustering showed decent grouping in TRS2, with little distinction between TRS0 and TRS1, suggesting similar expression profiles between the two (Figure 2C). Only lysoPC a C16:1 had significant correlations with clinical variables (Figure 2D). SLC6A4 expression in CRC tissue, which transcribes serotonin, had no significant effect on relapse-free survival (Figure 2E), while low SLC6A4 resulted in significantly worse post-progression survival (Figure 2F). Serotonin is demonstrated to enhance radiosensitivity in colon cancer,7 suggesting that the decreased serotonin in TRS2 patients is a mechanism underlying neo-tx resistance. Altered genes had significant correlations with several clinical variables (Figure 3A). RPL30 and CXCL14 were significantly upregulated, and SNORA81, SNORD50A, LCN2 and SNORA64 were significantly downregulated in TRS2 compared to TRS0 (Figure 3B). Increased expression of RPL30 is associated with amplification of the oncogene MYC, which promotes cytotoxic therapy resistance.8 Deletion of SNORD50A is associated with poorer survival outcomes in several cancers.9 Interestingly, SNORD50A binds and inhibits the oncogene KRAS, with depletion of SNORD50A causing activation of the MAPK cascade,9 which is involved in tumour resistance to therapy,10 suggesting a potential role in neo-tx resistance in RC. Separation of TRS based on gene expression is improved, with a clear distinction between TRS0 and TRS2 (Figure 3C). Several genes had significant correlations with clinical variables (Figure 3D). Low CXCL14 expression was associated with significantly worse overall (Figure 3E) and post-progression survival (Figure 3F). High RPL30 was associated with significantly longer overall survival (Figure 3G); however, it resulted in significantly worse relapse-free survival (Figure 3H) and post-progression survival (Figure 3I), aligning with the high RPL30 in TRS2 patients. Similarly, low LCN2 expression was associated with worse overall survival (Figure 3J). Altered genes and metabolites were integrated into an 8-feature multi-omic biomarker panel. Only TRS0 (n = 3), TRS1 (n = 6) and TRS2 (n = 5) groups could be included in the final analysis as matched transcriptomic data were not available for the TRS3 patient. TRS0 and TRS2 cluster separately using these eight features, with TRS1 interspersed between them (Figure 4A). Principal component analysis showed similar patterns, with Serotonin, SNORA64, SNORD50A, SNORA81, RPL30 and CXCL14 contributing most to group separation (Figure 4B). Examining expression of each feature individually, TRS0 and TRS2 have higher expression levels of separate features, while TRS1 sits within the range of TRS0 and TRS2 (Figure 4C). Leave-one-out cross-validation demonstrated poor distinction of TRS0 from TRS1/TRS2 (area under the curve [AUC] = 0.273, Sensitivity = 0%, Specificity = 54.5%), likely due to an imbalance for samples between the groups (TRS0 = 3, TRS1/TRS2 = 11) and thus not accurately representing the panel's performance between these groups (Figure 4D). Classification accuracy for TRS0 versus TRS2 was far superior (AUC = 0.933, Sensitivity = 100%, Specificity = 80%), aligning with the distinct expression profiles observed (Figure 4B,C). Lastly, in this pilot cohort, TRS0/TRS1 versus TRS2 demonstrated perfect classification (AUC = 1, Sensitivity = 100%, Specificity = 100%), suggesting that overall, the expression profiles of TRS2 patients are distinct from both TRS0 and TRS1, which are more closely related to each other (Figure 4D). In a pilot cohort, we demonstrate that rectal tumours have metabolic and transcriptomic remodelling, highlighting altered lipid metabolism as a common feature in rectal tumours, suggesting a novel therapeutic targeting approach. We also identify a novel multi-omic 8-biomarker panel, with potential, following validation in an independent cohort representing the full TRS spectrum, as a predictive signature of pathological response to neo-tx for improved patient stratification. Laura E. Kane: Data curation, Formal analysis and Writing - original draft. Croí E. Buckley: Investigation, Data curation and Formal analysis. Rebecca M. O.Brien: Data curation and Resources. Meghana S. Menon: Data curation and Resources. Aisling B. Heeran: Data curation and Resources. Xiaofei Yin: Investigation and Formal analysis. Timothy S. Nugent: Clinical samples and Clinical data. Noel E. Donlon: Clinical samples and Clinical data. John V. Reynolds: Resources. Adnan Hafeez: Clinical samples. Diarmuid S. O.R-ordáin: Clinical samples. Robert A. Hannon: Clinical samples. Paul Neary: Clinical samples. Reza Kalbassi: Clinical samples. Brian J. Mehigan: Clinical samples. Paul H. McCormick: Clinical samples. Cara Dunne: Clinical samples. John O. Larkin: Clinical samples. Lorraine Brennan: Formal analysis and Writing - review and editing. Michael E. Kelly: Clinical samples, Clinical data and Writing - review and editing. Jacintha O. Sullivan: Conceptualization, Supervision and Writing - review and editing. NLL: Conceptualization, Supervision, Funding acquisition, Writing. original draft and Writing - review and editing. We are deeply grateful to the patients who consented to the use of their tissue samples in this study. This research was funded by the Health Research Board (Ireland), grant number: EIA-2017-020. We also acknowledge funding from The Comprehensive Molecular Analytical Platform (CMAP) under the SFI Research Infrastructure Programme, reference number: 18/RI/5702. Laura E. Kane is supported by a Research Ireland fellowship, grant number: GOIPD/2024/331. The authors declare no conflict of interest. This research was funded by the Health Research Board (Ireland), grant number: EIA-2017-020. We also acknowledge funding from The Comprehensive Molecular Analytical Platform (CMAP) under the SFI Research Infrastructure Programme, reference number: 18/RI/5702. LE. Kane is supported by a Research Ireland fellowship, grant number: GOIPD/2024/331 This research was conducted in compliance with the Declaration of Helsinki and was approved by the joint St James's Hospital/AMNCH ethical review board (Ref: 2011/43/02) and the Beacon Hospital Research Ethics Committee (Reference BEA0139). Informed consent was obtained from all subjects involved in the study. Detailed data are available from the corresponding author on reasonable request. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.