290P - Metabolomic profiling and response to neoadjuvant therapy (NAT) in early breast cancer (EBC)

Background

The presence of residual disease (RD) after NAT is correlated with a poor outcome in EBC patients. New strategies are required to early identify patients who will not achieve a pathological complete response (pCR). Metabolomics may represent an innovative technology to identify host-related factors correlated with treatment outcome. In this study, we evaluated the predictive ability of metabolomics combined with machine learning in EBC patients receiving NAT.

Methods

Untargeted metabolomic analysis was performed in EBC patients candidates for NAT at our institution. Plasma samples were collected at baseline, before start of treatment. Small molecules were extracted, derivatized and then analysed using bi-dimensional gas chromatography/mass spectrometer (GCxGC-MS). The metabolomic profiling was then evaluated according to response to therapy (pCR versus RD). A machine learning approach was implemented with Boruta features selection algorithm combined with genetical algorithm as classifier, on a training set of 69 samples and validated on the remaining 27.

Results

From 12/2020 to 05/2023, 109 EBC patients have been enrolled (26 ER+, 45 HER2+, 38 TN); surgery was performed in 96 patients so far, with an overall pCR of 51% (ER+ 19%, HER2+ 51%, TN 55%). 61 small molecules were expressed (p<0.05 and Fold Change FC >1.3) between patients in pCR or RD. By multivariate analysis patients with RD showed higher levels of unsaturated fatty acids, as eicosatrienoate (FC=6), 9-decenoic acid (FC=2.2) and 17-octadecynoic acid (FC=2.2) and lower levels of fatty acids, as hexadecenoic acid (FC >100), doconexent (FC=69.5), 9,11-conjugated linoleic acid (FC=5.2), circulating glucose (FC=71) and glucopyranose (FC=23). Pathway analysis revealed the involvement of amino and fatty acid metabolism. A prediction model, combining metabolomics and machine learning, was implemented on 35 metabolomic probes and correctly identified the type of response with an accuracy of 95%.

Conclusions

Our data indicate that host-related metabolomic signatures may identify patients with EBC failing to achieve a pCR after NAT. We identified several fatty acids, amino acids and small molecules that can be modulated through diet. Possible effects by BC subtype are being investigated.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

A. Gennari.

Funding

AIRC.

Disclosure

A. Gennari: Other, Advisory Board: Roche, AstraZeneca. All other authors have declared no conflicts of interest.