Identifying biomarkers of lithium responsiveness from a large structural and functional retinal dataset

Background

Lithium remains the therapy of choice for bipolar disorder due to its potency as a mood stabiliser. The primary clinical measure of responsiveness to lithium is the Alda scale. It captures treatment response in terms of beneficial changes to mood in its first subscale, with a separate subscale capturing potential confounding factors (e.g., concurrent medications, duration of treatment). The total score of 7+ is taken to indicate ‘good responsiveness’ (Nunes et al., 2020). While the Alda score has been validated and is in wide clinical use, reliable biomarkers that would predict responsiveness to lithium prior to administration of the drug are not yet known. However, the search for such biomarkers is complicated by the fact that understanding of the mechanisms of lithium’s action is still lacking. Recent identification of the melanopic light-sensitive pathway in the retina, which feeds directly into circadian rhythm control (Berson et al., 2002), has led to a promising hypothesis that lithium acts by reducing retinal hypersensitivity to light. This hypothesis is being investigated in the ongoing, Wellcome Trust funded Helios-BD project (for research protocols, see Roguski et al., 2024). Helios-BD workstreams that characterize bipolar disorder retinal structure and function are conducted in Edinburgh and are led by the proposed PhD project’s supervisory team.

While Helios-BD intends to evaluate the light hypersensitivity hypothesis by comparing data from 60 healthy controls and 120 bipolar disorder patients – 60 of whom take lithium, and 60 of whom take other medications – the proposed PhD project would focus on the retinal and retinogeniculate structural and functional parameters from the Helios-BD patient data, to assess if any of them could act as reliable predictors of the Alda score. Helios-BD will provide by far the largest dataset on bipolar disorder and retinal function. As retina is an outgrowth of the brain with a complex structure including major excitatory and inhibitory neurotransmitters and glial cells, it represents a highly promising simplified model for brain research. The Helios-BD dataset includes psychophysical thresholds that characterize sensitivity of retinal cones, pupillary responses to different types of light stimuli, visual evoked potential (VEP) data elicited by contrast-modulated stimuli targeting specific cone-opponent and cone-additive retinogeniculate pathways, as well as various other parameters of retinal structure obtained by optical coherence tomography (OCT). With the data collection in its second year, the amount of data is sufficient to commence biomarker identification. Being able to predict responsiveness to lithium in individuals with bipolar disorder using data provided by a relatively easy/quick to administer visual test or scan would lead to increased precision in treatment provision and avoidance of potential negative outcomes in those patients for whom lithium responsiveness would be lower. 

Training outcomes

In the course of their training, the PhD student will gain a thorough understanding of structural and functional retinal assessment techniques, as well as an expertise in the state-of-the-art advanced data analysis techniques. They will become comfortable in performing analyses in a range of software environments such as Matlab and/or Python (for signal processing) and R (for statistics), and in the application of explainable AI techniques for identifying suitable biomarkers. For the latter, Dr Onken (Informatics) will advise on the use of contemporary computational techniques to validate and extend the standard statistical approaches. This should provide the successful candidate with a skill-set that would be highly useful in developing and deploying state-of-the-art data analysis methods in both basic and clinical research.