Skip to main content

25-09-2020 | Rheumatology | News | Article

Editor's pick

Molecular-based machine learning tool predicts anti-TNF response in RA

Laura Cowen

medwireNews: Machine learning models based on gene expression and DNA methylation profiles may accurately predict response to adalimumab and etanercept in patients with rheumatoid arthritis (RA), research from the Netherlands shows.

Aridaman Pandit and colleagues from Utrecht University say their findings are “paving the path towards personalized anti‐TNF [tumor necrosis factor] treatment.”

The researchers carried out gene expression and DNA methylation profiling on peripheral blood mononuclear cells (PBMCs), monocytes, and CD4+ T cells from 80 members of the observational BiOCURA cohort before they initiated treatment with either adalimumab (47.5%) or etanercept (52.5%).

After 6 months of treatment, 53% of patients who received adalimumab and 45% of those who received etanercept were classed as responders according to EULAR DAS28 criteria.

When the investigators compared the molecular signatures of the responders with those of the non-responders, they identified 549 and 460 differentially expressed genes in the PBMCs of patients from the adalimumab and etanercept groups, respectively.

These genes were typically involved in DNA and nucleotide binding in both groups, but the overlap between the groups was just 2%, which suggests that the response to each treatment “is defined by distinct gene signatures,” Pandit et al remark.

They also note that TNF expression was not significantly associated with response to either treatment.

Genome-wide DNA methylation analysis of PBMCs identified 16,141 and 17,026 differentially methylated CpG positions associated with adalimumab and etanercept response, respectively. Of these, 46% were hypermethylated in adalimumab responders, whereas 76% were hypermethylated in etanercept responders.

“Thus, on epigenetic level, we observed a distinct hypermethylation pattern between [adalimumab] and [etanercept] responders, suggesting the role of epigenetics in defining response towards [these drugs] in PBMCs,” the authors write in Arthritis & Rheumatology.

The team also found differential gene expression and methylation patterns in monocytes and CD4+ T cells, and they used these, plus the signatures they identified in PMBCs, to build machine learning models to predict treatment response.

They found that the models based on gene expression accurately predicted adalimumab response in 85.9% of cases and etanercept response in 79.0% of cases.

Using the DNA methylation signatures, the accuracy was 84.7% and 88.0% for adalimumab and etanercept, respectively.

“These results suggest that we can accurately predict the clinical response before [adalimumab] and [etanercept] treatment using molecular signatures-based machine learning models, although the prediction accuracy of these molecular signatures differs between cell types and treatments, underlining the need to study more than one drug, cell type or epigenetic layers,” Pandit et al conclude.

medwireNews is an independent medical news service provided by Springer Healthcare Ltd. © 2020 Springer Healthcare Ltd, part of the Springer Nature Group

Arthritis Rheumatol 2020; doi:10.1002/art.41516

See the research in context now

with trial summaries, expert opinion and congress coverage

Image Credits