Researchers at Addis Ababa University, working at Amanuel Mental Specialized Hospital in Addis Ababa, examined 101 patients with bipolar disorder who had been taking lithium for at least six months. The participants were drawn from a larger cohort recruited for the Neuropsychiatric Genetics of African Populations–Psychosis (NeuroGAP-P-E) project, a multi-country initiative spanning Ethiopia, Kenya, Uganda, and South Africa. By reviewing medical records and applying the widely used Alda scale—a tool that weighs clinical improvement during lithium treatment against factors that might confound the result, such as poor adherence or concurrent medications—the team classified 32.5 percent of patients as good responders and 67.5 percent as insufficient responders.

The genetic work behind the study was ambitious. Using PCR-free whole-genome sequencing performed at the Broad Institute of Harvard and MIT, the investigators genotyped 53 single nucleotide polymorphisms (SNPs) across 22 candidate genes, all chosen because of prior links to lithium’s mechanisms of action or to bipolar disorder itself. Saliva samples were processed with DNA Genotek kits, quantified by spectrophotometry, and sequenced on NovaSeq 6000 flow cells generating 151-base-pair paired-end reads, with variant calling carried out under GATK best practices on the Google Cloud Platform. Rigorous quality control—including checks for Hardy-Weinberg equilibrium, low genotype call rates, sex mismatches, and ancestral population structure—was applied before any association testing began.

The results, published in Annals of General Psychiatry, point to a constellation of genes tied to neuroplasticity, dopamine signaling, and intracellular kinase pathways. The strongest signal came from the brain-derived neurotrophic factor gene, BDNF. Its rs6265 variant, better known as Val66Met, showed the CC genotype in 95.6 percent of insufficient responders compared with 66.7 percent of good responders, a difference that survived correction for multiple testing with a false discovery rate of p = 0.0001. Another BDNF polymorphism, rs2030324, told the opposite story: the A allele and AA genotype were significantly more frequent among good responders, suggesting that variation within this single gene can push patients toward opposite ends of the treatment spectrum.

BDNF encodes a growth factor central to neuronal survival, synaptic plasticity, and mood regulation—all processes thought to underlie lithium’s therapeutic action. Previous studies in European and East Asian populations have produced conflicting results regarding Val66Met, with some linking the Met allele to better outcomes and others finding no association. The Ethiopian findings add an important data point from a population that has been almost entirely absent from psychiatric pharmacogenomics, and they reinforce a 2018 literature review concluding that Val66Met meaningfully influences response to mood stabilizers.

Dopamine receptor genes emerged as a second major theme. The GG genotype and G allele of rs4532 in the DRD1 gene were significantly more common in insufficient responders, a result that remained significant even after false discovery rate adjustment and that mirrors earlier Polish findings connecting the same variant to poor lithium prophylaxis. The rs1800497 variant of DRD2 also showed higher frequencies of the GG genotype and G allele among non-responders before correction. Because dopamine is central to reward processing and mood regulation, these results suggest that lithium’s clinical effects may be partly mediated through dopaminergic circuits whose sensitivity varies with genetic makeup.

Perhaps the most clinically intriguing findings came from the AKT1/GSK-3β signaling pathway, a cascade long considered the molecular bullseye of lithium itself. Glycogen synthase kinase-3 beta is directly inhibited by lithium, an effect discovered in the late 1990s that helped explain the drug’s neuroprotective and circadian influences. In the new study, the AG genotype of the GSK-3β promoter variant rs334558 was associated with significantly reduced treatment response in multivariable analysis, while the TT genotype of AKT1 rs10138227 acted as a powerful positive predictor, carrying an adjusted odds ratio of nearly 12 for good response. Conversely, the GG genotype of BDNF rs962339, the AG and GG genotypes of DRD2 rs1800497, and the GSK-3β AG genotype all predicted poor outcomes after adjustment for age, sex, body mass index, and comorbid psychiatric diagnoses.

Not every candidate gene earned its reputation. Variants in ARRB2, TPH2, DRD3, NR3C1, ANK3, NTRK2, CACNG2, IMPA2, INPP1, CREB1, and the circadian regulators CLOCK, PER3, and NR1D1 showed either no association or only hints that evaporated under statistical correction. Circadian genes such as CLOCK rs534654 and PER3 rs228642 initially appeared to differentiate responders from non-responders, but these signals did not survive false discovery rate adjustment—a result consistent with a 2014 Polish study that also found no link between clock gene polymorphisms and lithium response. The authors caution that limited sample size may have masked genuine effects in these pathways.

The study’s statistical framework was deliberately conservative. Chi-square tests and logistic regression models were complemented by backward stepwise multivariable analysis, multicollinearity diagnostics using variance inflation factors, and post hoc power calculations that exceeded the conventional threshold of 0.8. Even so, the team acknowledges important limitations: 101 patients is modest by genetic association standards, candidate-gene approaches by design ignore variants elsewhere in the genome, and the retrospective Alda scoring of response introduces some subjectivity. The authors call for larger genome-wide association studies in African populations, where genetic diversity is greater than in any other continent and where findings from European cohorts often fail to replicate.

Why does population matter so much? Allele frequencies differ substantially across ancestries, and the Ethiopian cohort revealed patterns that diverge from those reported in East Asian and European samples. The Val66Met association, for instance, contrasts with Japanese studies that found no effect, while the DRD1 finding aligns closely with Polish data. This patchwork of results underscores a growing consensus in psychiatric genetics: lithium response is a polygenic trait, shaped by many variants of small effect whose contributions vary across populations and clinical contexts. No single SNP will ever serve as a crystal ball, but panels of markers—like those flagged here—could eventually inform a probabilistic prediction.

The clinical stakes are considerable. Bipolar disorder affects roughly 40 million people worldwide according to the World Health Organization, carries one of the highest suicide risks of any psychiatric illness—up to 15 percent of patients may die by suicide—and imposes enormous burdens on health systems, particularly in low- and middle-income countries where lithium remains one of the most affordable mood stabilizers. A genetic test that could identify likely responders before treatment begins would spare non-responders months or years of ineffective therapy, accelerate access to alternatives such as valproate or lamotrigine, and reduce hospitalizations.

For now, the Ethiopian findings are a beginning rather than an endpoint. The authors emphasize that functional studies are needed to determine how the identified variants alter gene expression or protein activity, and that replication in independent African cohorts is essential. Yet the study represents a milestone: the first genetic analysis of lithium response ever conducted in Ethiopia, built on whole-genome sequencing rather than targeted arrays, embedded in a continental initiative designed to correct the chronic underrepresentation of African genomes in medical research. As precision psychiatry moves from concept toward clinic, studies like this one ensure that the genetic map of treatment response will not be drawn from Europe and North America alone.