Back

Wound-like tumor periphery in human breast cancer predicts a convergent drug nonresponse

Li, L.; Liao, X.; Fu, F.; Xi, G.; Kang, D.; He, J.; Guo, W.; Qiu, L.; Han, Z.; Wang, X.; Zhan, Q.; You, S.; Chen, J.; Wang, C.; Boppart, S. A.; Tu, H.

2021-11-04 pathology
10.1101/2021.11.02.467008 bioRxiv
Show abstract

A significant portion of breast cancer patients are nonresponsive to well-established drugs and destined for a poor outcome regardless of molecular subtype. Although several (multiparameter) molecular markers have predicted their resistance to some of these drugs, profound uniparameter markers predictive of a convergent nonresponse to all these drugs remain elusive. We employ co-registered standard-multiphoton histology to representatively sample a few peripheral niches of the primary tumor, so that hundreds of patients can be stratified with either a wound-like or non-wound tumor periphery. With no fitting variable, this simple uniparameter morphological marker is: (a) highly sensitive and specific to predict a multidrug-nonresponsive phenotype that accounts for the majority of recurrence or death, independent of the molecular subtype or related adjuvant drug selection, clinical endpoint (disease-free versus overall survival), and hosting medical center; (b) robust against intratumor heterogeneity and valid at the earliest clinicopathological stage; and (c) dominant in predicting prognosis in the context of routine clinicopathological markers. Considering the mechanistic link between a wound-like extracellular matrix and a microenvironment supporting migratory or mesenchymal tumor cells, we attribute these unusual capabilities to an epithelial-mesenchymal transition nature of the morphological marker long sought after by pathologists.

Matching journals

The top 8 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.