Background: The definitive diagnostic test for Nonalcoholic steatohepatitis (NASH) is liver biopsy, which carries risks and cannot be used for frequent monitoring. There is no single noninvasive method that can accurately and simultaneously capture steatosis, inflammation, hepatocyte ballooning and fibrosis, the four major pathologic components assessed by biopsy. We show that large scale proteomics has promise as an alternative to liver biopsies in clinical trials or longitudinal studies of NASH.
Methods: Using modified-aptamer proteomics, we scanned ~5000 proteins in each of 2852 serum samples from the NASH CRN*, including 636 participants from a natural history cohort and longitudinal samples from the PIVENS (pioglitazone, vitamin E and placebo) and the FLINT (obeticholic acid and placebo) clinical trials, for a total of ~15 million protein measurements. Liver biopsy results were modeled with measured proteins using machine learning methods independently for each biopsy component.
Results: Results for the 4 protein models in training/paired validation were: fibrosis (AUC 0.92/0.85); steatosis (AUC 0.95/0.79), inflammation (AUC 0.83/0.72), and ballooning (AUC 0.87/0.83). A concurrent positive score for steatosis, inflammation and ballooning predicted the biopsy diagnosis of NASH with an accuracy of 73%. When applied longitudinally, model scores predicted decreasing biopsy scores in the active groups vs. stable for placebo and differential pharmacodynamic effects were evident on each model component.
Conclusions: Serum protein scanning is the first technique to capture four components of the liver biopsy individually and noninvasively. The four models were sufficiently sensitive and precise to characterize the time-course and extent of three drug mechanisms. Concurrent positive results from the protein models had performance characteristics of “rule-out” tests for diagnosis of NASH. These tests may assist in new drug development and medical intervention decisions.