Multi-omics staging of locally advanced rectal cancer predicts treatment response: a pilot study

Abstract

Treatment response assessment of rectal cancer patients is a critical component of personalized cancer care and it allows to identify suitable candidates for organ-preserving strategies. This pilot study employed a novel multi-omics approach combining MRI-based radiomic features and untargeted metabolomics to infer treatment response at staging. The metabolic signature highlighted how tumor cell viability is predictively down-regulated, while the response to oxidative stress was up-regulated in responder patients, showing significantly reduced oxoproline values at baseline compared to non-responder patients (p-value < 10^-4). Tumors with a high degree of texture homogeneity, as assessed by radiomics, were more likely to achieve a major pathological response (p-value < 10^-3). A machine learning classifier was implemented to summarize the multi-omics information and discriminate responders and non-responders. Combining all available radiomic and metabolomic features, the classifier delivered an AUC of 0.864 (± 0.083, p-value < 10^-3) with a best-point sensitivity of 90.9% and a specificity of 81.8%. Our results suggest that a multi-omics approach, integrating radiomics and metabolomic data, can enhance the predictive value of standard MRI and could help to avoid unnecessary surgical treatments and their associated long-term complications.

Keywords: Magnetic resonance imaging; Metabolomics; Multi-omics; Radiomics; Rectal cancer; Treatment response.

Fig. 1

Workflow of untargeted metabolomics approach. The number of features taken into consideration in the various steps, up to the final data matrix, is shown in bold. Each processing step was described detailing the inclusion and exclusion criteria. Exclusion list includes compounds present in the blank sample that are considered potential contaminants. Inclusion list includes molecules present in the sample excluding contaminants. Chromatographic retention times (RT) and the minimum signal-to-noise ratio (S/N) were the inclusion criteria for the quantification step. The databases and the mass tolerance (expressed in ppm) were the inclusion criteria in the identification step

Fig. 2

A sPLS-DA based on 4000 features in the plasma of responder (R) and non-responder (NR) patients. B Volcano plot of 4000 metabolomic features classifying them in not significant (gray), significantly down-regulated in RP (blue dots) and significantly up-regulated in RP (red dots). C Prediction of down-regulated cellular functions “Cell viability” and “Cell viability of cancer cells” in R patients. D Predicted of up-regulated cellular function “Mobilization of Ca2 + ” in R patients. E Predicted up-regulation of cellular function “Oxidative stress response of cells” and “Release of l-glutamic acid” in R patients. F Legend of color code for increased and decreased measurement of metabolites, and predicted activation and inhibition of disease and cellular function

Fig. 3

ROC curve and best cutoff point confusion matrix delivered by the classification performed relying on A sMRI data, B ADC metrics, C T2w metrics, D MD data and E sMRI + ADC + T2w + MD data

Fig. 4

A, B Violin plot showing the distribution of oxoproline (Panel A) and gamma-glutamyl-tirosine (Panel B) for non-responder (in red) and responder (in green) patients. C Heatmap of the p-value for the correlation of 14 radiometabolomics features described in the legend on the right. D Pearson correlation between oxoproline levels and radiomics features called (original shape maximum 2D diameter slice). E Pearson correlation between gamma-glutamyl-leucine and glutamate levels. F Pearson correlation between glutamate and proline level. *** means p-value of Mann–Whitney test < 0.001, * means p-value of Mann–Whitney test < 0.05

Fig. 5

Study workflow. In the first step (I), rectal cancer was manually segmented on MR images (T2w, ADC), followed by radiomics features extraction. In the second step (II), plasma untargeted metabolomics analysis and metabolites identification and quantification were performed. In the final step (III), a machine learning algorithm (k-nearest neighbors—kNN) was used to select radiometabolomics features and ROC analysis delivered the classification model accuracy for treatment response prediction (created with BioRender.com)