Introduction: Intrinsic breast cancer subtypes (e.g., Normal-like, Luminal A/B, HER2-enriched, Triple Negative) vary in their responsiveness to neoadjuvant chemotherapy (NACT) and are associated with differing long-term prognosis. Expression dis-regulation of critical regulatory pathways partially accounts for variance in cancer phenotype responsiveness which has allowed development of targeted therapies to improve prognosis for some subtypes. Expression dis-regulation also plays a role in regulating cancer metabolism—the Warburg Effect being the most notable example. Metabolic dis-regulation opens avenues for cancer cells to exploit environmental niches arising in the tumor microenvironment (TME) (e.g., lactate cross-feeding). Focusing on the role TME plays in defining cancer behavior, we undertook a theoretical investigation to uncover how metabolic cross-feeding affects subtype behavior.
Methods: An in silico analysis of cross-feeding variability arising from metabolic differences in Luminal A/B (LA, LB), HER2-enriched (HER2+) and Triple Negative (TN) breast cancers was performed. Community models of the TME were created by coupling metabolic models of adipocytes, glandular, and cancer tissues derived from a comprehensive reconstruction of human metabolism (Pornputtapong, Nookaew, Jensen, Database, 2015). A parsimonious flux balance analysis technique was used to elucidate the consumption, production, and cross-feeding of metabolites by the three tissues (Lewis et al. Mol. Syst. Biol. 2010). Constraints were applied to the model to limit the uptake of 701 metabolites to levels measured in the blood (Wishart et al., Nucleic Acids Res. 2018). In total, 1222 community models were generated by personalizing cancer models using expression levels of metabolic enzyme transcripts for breast cancer patients in the TCGA BRCA study (The Cancer Genome Atlas Network, Nature, 2019).
Results: Previously characterized gene expression features were recapitulated via metabolic consumption/production in the models: TNBC/HER2+ had higher asparagine, glutamine (Glu), tryptophan, phenylalanine, consumption than LA/LB (van Geldermolsen et al. Oncogone 2016), Glu was cross-fed from cancer to fat in TNBC (Cao et al. BMC Cancer, 2014), TNBC produced methionine and proline (Kanaan et al. Can. Gen. Prot. 2014), and TNBC/HER2+ cancers consumed high density lipoprotein produced by cancer (Balabum et al. Cancer Met. 2017).
The community models also revealed novel findings. TNBC/HER2+ relied on glucose as the primary energy source, while LA/LB relied on amino acids alanine, glycine and Glu. TNBC/HER2+ produced high levels of lactate which was consumed by adipose tissues. Uniquely, glycine produced by cancer was consumed by fat in TNBC. Several environmental niches were induced in the healthy tissue by the presence of the cancer; for example, ornithine was predicted to be cross-fed from fatty to glandular tissues in TNBC/HER2+ cancers, and glutamate and inosine from glandular to fatty tissues in HER2+ cancers.
To identify metabolic differences that could explain variability in survival, hierarchical clustering was performed resulting in 13 distinct metabolic categories. As one illustrative example, two clusters enriched in LA/LB cancers had different 10-year survival rates. The poorer prognosis cluster lacked valine, leucine, isoleucine biosynthesis, and arginine, proline and pyruvate metabolism, suggesting resource starvation. Since many NACT therapies act by killing proliferating cells and pathological complete response is correlated to better prognosis, we posit that such starving tumors will be less susceptible to treatment.
Conclusion: While preliminary, these results suggest that metabolic pathway usage can lead to difference in growth, cross-feeding and potentially drug efficacy.
