ABSTRACT
Precision medicine is critically dependent on better methods for diagnosing and staging disease and predicting drug response. Histopathology using Hematoxylin and Eosin (H&E) stained tissue - not genomics – remains the primary diagnostic method in cancer. Recently developed highly-multiplexed tissue imaging methods promise to enhance research studies and clinical practice with precise, spatially-resolved, single-cell data. Here we describe the “Orion” platform for collecting and analyzing H&E and high-plex immunofluorescence (IF) images from the same cells in a whole-slide format suitable for diagnosis. Using a retrospective cohort of 74 colorectal cancer resections, we show that IF and H&E images provide human experts and machine learning algorithms with complementary information that can be used to generate interpretable, multiplexed image-based models predictive of progression-free survival. Combining models of immune infiltration and tumor-intrinsic features achieves a hazard ratio of ∼0.05, demonstrating the ability of multi-modal Orion imaging to generate high-performance biomarkers.
Competing Interest Statement
PKS is a co-founder and member of the BOD of Glencoe Software, a member of the BOD for Applied Biomath, and a member of the SAB for RareCyte, NanoString, and Montai Health; he holds equity in Glencoe, Applied Biomath, and RareCyte. PKS is a consultant for Merck and the Sorger lab has received research funding from Novartis and Merck in the past five years. YC is a consultant for RareCyte. DC, JC, EM, SR, and TG are employees of RareCyte. The DFCI receives funding for KLL research from the following entities: Amgen, Travera, and X4. DFCI and KLL have patents related to molecular diagnostics of cancer. SJR receives research support from Bristol-Myers-Squibb and KITE/Gilead. SJR is on the Scientific Advisory Board for Immunitas Therapeutics. The other authors declare no outside interests.
Footnotes
Human Tissue Atlas Center
We have collected a substantial amount of new data and made extensive changes to the text to address the reviewers' concerns. Most importantly, we have collected data from 34 additional human CRC specimens so that a classic split could be performed between training and test data. I am happy to report that the image feature models (biomarkers) we describe performed as well on the new test data as on the training data, substantially strengthening the manuscript. We also demonstrate a cyclic approach to Orion data collection, increasing the number of molecular markers from 18 to 32, and address a range of other technical concerns raised by the reviewers.