Prediction of lithium response using genomic data

William L Stone ,Janusz Rybakowski,Markus M Nöthen,Esther Jiménez,Thomas G Schulze,Francis J Mcmahon,Claire O′Donovan ,Liping Hou,Manuel Mattheisen,Eduard Vieta,Thomas Trappenberg,Alexandre Dayer,Andreas J Forstner,Maria Del Zompo,Vincent Millischer,Alessio Squassina,Claudia Pisanu,Cristiana Cruceanu,Caterina Chillotti,Paul Grof,Maria Grigoroiu-Serbǎnescu ,Pablo Cervantes,Julia Veeh,Giovanni Severino,Catharina Lavebratt,Franziska Degenhardt,Raffaella Ardau,Guy A Rouleau ,Michael Bauer,Nirmala Akula,Jean Michel Aubry ,Mark A Frye ,Po-Hsiu Kuo,Norio Ozaki,Marcella Rietschel,Kazufumi Akiyama,Hsi‐Chung Chen ,Mirko Manchia,John R Kelsoe ,Sarah Kittel‐Schneider ,Peter R Schofield,Martin Schalling,Andreas Reif,Ichiro Kusumi,Lena Backlund,Frank Bellivier,Takeshi Kato ,Philip B Mitchell,Janice M Fullerton,Abraham Nunes,Lina Martinsson,Ryota Hashimoto,Martin Alda

doi:10.1038/s41598-020-80814-z

Abstract

Predicting lithium response prior to treatment could both expedite therapy and avoid exposure to side effects. Since lithium responsiveness may be heritable, its predictability based on genomic data is of interest. We thus evaluate the degree to which lithium response can be predicted with a machine learning (ML) approach using genomic data. Using the largest existing genomic dataset in the lithium response literature (n = 2210 across 14 international sites; 29% responders), we evaluated the degree to which lithium response could be predicted based on 47,465 genotyped single nucleotide polymorphisms using a supervised ML approach. Under appropriate cross-validation procedures, lithium response could be predicted to above-chance levels in two constituent sites (Halifax, Cohen’s kappa 0.15, 95% confidence interval, CI [0.07, 0.24]; and Würzburg, kappa 0.2 [0.1, 0.3]). Variants with shared importance in these models showed over-representation of postsynaptic membrane related genes. Lithium response was not predictable in the pooled dataset (kappa 0.02 [− 0.01, 0.04]), although non-trivial performance was achieved within a restricted dataset including only those patients followed prospectively (kappa 0.09 [0.04, 0.14]). Genomic classification of lithium response remains a promising but difficult task. Classification performance could potentially be improved by further harmonization of data collection procedures.

Highlights

Bipolar disorder (BD) is a neuropsychiatric illness characterized by recurrent episodes of mania and depression
In two subsamples of our genomic dataset, we found non-trivial classification performance informed by variants associated with a cellular component relevant for the pathophysiology of BD
We observed non-trivial classification performance on data pooled across sites whose data were collected prospectively

Summary

Introduction

Bipolar disorder (BD) is a neuropsychiatric illness characterized by recurrent episodes of mania and depression It is associated with a significant risk of suicide1—highest in early course of the illness—with many patients receiving effective treatment only after as many as 10 years[2]. The authors demonstrated that the response-related alleles were associated with lower rates of relapse in an independent sample of 73 patients treated for 2 years of lithium monotherapy, the out of sample predictive power of the genomic data overall remains unknown. This is due to the fact that GWAS is (A) not designed to evaluate predictive capacity, and (B) cannot account for epistatic effects. We sought to explore factors that might limit classification performance in this large multi-site dataset, such as assessment method (prospective followup vs. cross-sectional) and strength of lithium response

Objectives

Methods

Results

Conclusion