Serum metabolomic profiling for predicting therapeutic response and toxicity in breast cancer neoadjuvant chemotherapy: a retrospective longitudinal study

Abstract

Background: Neoadjuvant chemotherapy (NACT) is the standard-of-care treatment for patients with locally advanced breast cancer (LABC), providing crucial benefits in tumor downstaging. Clinical parameters, such as molecular subtypes, influence the therapeutic impact of NACT. Moreover, severe adverse events delay the treatment process and reduce the effectiveness of therapy. Although metabolic changes during cancer treatment are crucial determinant factors in therapeutic responses and toxicities, related clinical research remains limited.

Methods: One hundred paired blood samples were collected from 50 patients with LABC before and after a complete NACT treatment cycle. Untargeted metabolomics was used by liquid chromatography-mass spectrometry (LC-MS) to investigate the relationship between dynamically changing metabolites in serum and the responses and toxicities of NACT.

Results: Firstly, we observed significant alterations in serum metabolite levels pre- and post-NACT, with a predominant enrichment in the sphingolipid and amino acid metabolism pathways. Second, pre-treatment serum metabolites successfully predicted the therapeutic response and hematotoxicities during NACT. In particular, molecular subtype variations in favorable treatment responses are linked to acyl carnitine levels. Finally, we discovered that the therapeutic effects of NACT could be attributed to essential amino acid metabolism.

Conclusion: This study elucidated the dynamic changes in metabolism during NACT treatment, providing a possibility for developing responsive metabolic signatures for personalized NACT treatment.

Keywords: Acylcarnitine; Breast cancer; Essential amino acid; Metabolomics; Neoadjuvant chemotherapy; Serum.

Fig. 1

Serum metabolic characteristics of LABC patients in response to NACT. a Clinical design overview of metabolomics study. b Number of identified metabolites in the serum sample and the distribution of their chemical categories. c Principal component analysis (PCA) reveals significant changes in metabolic profiles before and after NACT treatment. d Volcano plot of metabolites significantly associated with NACT treatment (P value < 0.05, FC > 1.5 or < 0.66). Aquamarine dots (n = 24) and orange dots (n = 36) represent metabolites that were increased and decreased, respectively, before and after NACT treatment. The grey dots represent unchanged metabolites. e Pathway enrichment analysis of metabolites (n = 60) significantly altered before and after NCAT (Hypergeometric test; P < 0.05). The pathway enrichment was performed using MetaboAnalyst. f, g Heatmaps show the top 20 decreased (f) and top 20 increased (g) metabolites associated with NACT treatment

Fig. 2

Pre-therapeutic metabolic traits predict the hematologic toxicities caused by NACT. a Significant correlation between the lowest counts of the four cell types associated with hematological toxicity and the pre-treatment serum metabolites. (Pearson correlation; two-sided Student’s t test P < 0.05). Hb, hemoglobin; WBC, white blood cells; PLT, platelets; NEUT, neutrophils. b Pathway enrichment analyses of hematologic toxicity associated metabolites (Hypergeometric test; P < 0.05). c Four examples of metabolites and their relevance to hematological toxicity. (Pearson correlation; two-sided Student’s t test P < 0.05). Neu5Ac, N-Acetylneuraminic acid. Error bands represent 95% confidence intervals. d Multiple linear regression models constructed from different serum baseline metabolites to predict nadir cell counts and hematological toxicity during NACT. (Pearson correlation; two-sided Student’s t test P < 0.05). Error bands represent 95% confidence intervals. Supplementary Tables 3–6 provide details of the prediction models

Fig. 3

Predictive model for therapeutic efficacy based on key serum metabolites. a Heatmap revealed baseline metabolites significantly correlated with response (R) to NACT (P < 0.05; Two-sided Student T test). b Pathway enrichment analysis (P < 0.05; Hypergeometric test). c Lasso regression for the selection of key metabolites. d Selected key metabolites and breast cancer subtype for the logistic regression model to predict the chances of R. e The sensitivity and specificity of the prediction model with a risk score of 0.5 (R: n = 18; NR: n = 32). f Receiver operating characteristic curve of the logistic regression model

Fig. 4

Acyl carnitines are associated with subtype differences in R status following NACT. a Non-Luminal patients had a significantly higher chance of achieving R than Luminal (P = 0.08; χ.² test). b Venn plot showed the number of metabolites associated with the subtype in R patients (red) and all patients (blue). c, d All patients (Luminal = 19, non-Luminal = 31) and R patients (Luminal = 4, non-Luminal = 14) total carnitine and Car (18:0) levels. e Subtype disparity in carnitine levels associated with R status after NACT. (nd, not detected; **, P < 0.01; *, P < 0.05; ns, not significant; Two-sided Wilcoxon test)

Fig. 5

Achieving R to NACT may be attributed to elevated essential amino acids. a Differentially changed metabolites associated with NACT treatment and R status, and the combination of both using a two-way ANOVA analysis (P < 0.05). b Pathway enrichment analysis of metabolites related to both NACT treatment and R status (Hypergeometric test; P < 0.05). c Metabolic correlation between two amino acids in R and NR patients (Pearson correlation; r > 0.50 and two-sided Student’s t test P < 0.05). d, e Amino acid levels in serum samples from R (n = 18) and NR (n = 32) patients before and after NACT treatment: c, before NACT at baseline; d, after NACT (Two-sided Wilcoxon test; ns, not significant; ***, P < 0.001; **, P < 0.01). f Amino acid levels in serum samples from R (n = 18) and NR (n = 32) patients after NACT treatment. Data normalized to baseline levels (Two-sided Wilcoxon test). g Odd ratios of amino acids related to R status after NACT (two-sided z-test of logistic regression; ns, not significant). R (n = 18) and NR (n = 32). Error bars represent 95% confidence intervals