Some patients treated for ductal carcinoma in situ (DCIS) of the breast will experience cancer recurrences, whereas other patients will not. Unfortunately, current techniques cannot identify which preinvasive lesions will lead to recurrent cancer. Because the mechanism of cancer recurrence is unknown, it is difficult to design a test that detects its activity. We propose that certain pentose phosphate pathway enzymes, glutathione synthesis enzymes, and RhoA cluster at the epithelial cell periphery before cancer recurrences. Enzyme clustering enhances metabolic flux. Using fluorescence microscopy, we show that phosphophorylated glucose transporter type-1, transketolase-like protein-1, glutathione synthetase, GTP-loaded RhoA, and RhoA accumulate as a peripheral layer near the epithelial cell surface in surgical biopsies of women who will suffer recurrences, but not in samples from women who will not experience recurrences as judged using 2×2 contingency tables. Machine-learning studies of phospho-glucose transporter type 1-labeled tissue sections of patients with DCIS demonstrated strong cross-validation and holdout performance. A machine study of individual cribriform, papillary, micropapillary, and comedo forms of DCIS demonstrated 97% precision and 95% recall in the detection of samples from women who will not experience a recurrence and 90% precision and 94% recall in the detection of lesions that will become recurrent. A holdout study of these patients showed 73% true negatives, 18% true positives, 4% false positives, and 4% false negatives at a 50% threshold. This work suggests mechanistic features of cancer recurrences that may contribute to a new clinical test distinguishing high from low-recurrence risk in patients with DCIS.
Keywords: RhoA; fluorescence microscopy; glucose transporter; intracellular location/trafficking; machine learning.