Triangulating evidence from observational and Mendelian randomization studies of ketone bodies for cognitive performance

Ketone bodies (KBs), including 3-β-hydroxybutyrate (BOHBUT) and acetoacetate (ACACE), can cross the blood-brain barrier and serve as alternative energy substrates for the brain when glucose is limited. BOHBUT has been implicated in brain energy supply under low-glucose conditions, neurotransmitter regulation, reduction of oxidative stress, and protection against pathogenic proteins such as β-amyloid and phosphorylated tau, which contribute to neurodegenerative diseases. Prior small experimental and randomized controlled trials suggest that elevating KBs may improve memory and cognitive function in older adults, patients with type 2 diabetes, mild cognitive impairment, and Alzheimer’s disease. Animal studies indicate BOHBUT can increase hippocampal BDNF, supporting neuronal survival and synaptic plasticity. However, causal evidence in humans remains unclear. This study aims to triangulate observational and Mendelian randomization (MR) evidence to assess associations and potential causal effects of BOHBUT (and ACACE) on general cognitive performance and risk of Alzheimer’s disease.

The background summarizes prior evidence: Small clinical and experimental studies report that BOHBUT supplementation or ketogenic interventions can enhance memory in type 2 diabetes, older adults without dementia, and patients with mild cognitive impairment or mild-to-moderate Alzheimer’s disease. Ketone bodies have shown neuroprotective effects in animal models, including increased hippocampal BDNF expression, improved mitochondrial energy metabolism, reduced oxidative stress, and potential protection against β-amyloid and phosphorylated tau accumulation. Mouse studies also show improved learning and memory with BOHBUT via enhanced neuronal mitochondrial energy. Despite these findings, robust human causal evidence was lacking prior to this work.

Study design: Triangulation combining observational analyses and two-sample Mendelian randomization (MR), with additional cis-MR focused on OXCT1. Observational study: Data were from 5506 European participants aged 44–70 years in the Whitehall II (WHII) study within the UCLEB Consortium. After QC and complete-case analysis, 2214 (BOHBUT) and 2224 (ACACE) participants were included. Fasting (≥8 h) serum BOHBUT and ACACE were quantified by NMR spectroscopy with QC exclusions for degradation/contamination. Cognitive assessments included word memory (N=3115), verbal fluency (N=3122), and verbal meaning (N=3142). A general cognitive performance score was derived as the first unrotated principal component of the three tests, and individual test scores were also analyzed. Observational analyses: Primary exposures were categorized KBs (high vs normal) with varying thresholds across the KB range to reflect potential therapeutic levels (BOHBUT cutoffs 0.07–0.52 mmol/L; ACACE 0.02–0.20 mmol/L), ensuring ≥90% power for two-sample t-tests. Multiple linear regression estimated associations with general cognitive score and each test, adjusted for age, sex, diabetes (yes/no), smoking (ever/never), alcohol (heavy/other), waist-to-hip ratio, and occupational position (low/intermediate/high). Effect units: SD of cognitive scores per KB category above cutoff. Secondary analyses treated KBs as continuous variables (effects per 1 mmol/L). Bonferroni correction for three cognitive outcomes set significance at 0.0167; P<0.05 considered suggestive. Mendelian randomization: Exposure instruments for KBs were obtained from an in-house meta-analysis (UCLEB + Kettunen et al.) totaling N=45,031. Two IV sets were defined: strong (P<5×10−8) and suggestive (P<1×10−5), clumped at r2<0.001, 10,000 kb, using 1000 Genomes European LD reference. Outcomes: cognitive performance (SSGAC; N=257,841) and Alzheimer’s disease (IGAP; N=54,162). Primary MR method was IVW (fixed/random per Cochran’s Q P<0.1). Sensitivity methods included weighted median, weighted mode, MR-Egger (for directional pleiotropy), and MR-PRESSO (outlier detection/correction). Instrument strength assessed by R2 and F-statistics. Instruments were checked against potential confounders (alcohol, diabetes, waist-to-hip ratio, occupational position, smoking) with Bonferroni threshold 3.8×10−3. cis-MR: Focused on OXCT1 (key enzyme in ketone catabolism). Selected SNPs within 100 kb of OXCT1 with P<1×10−5 and r2<0.4; applied IVW and MR-Egger accounting for residual correlation. Software: TwoSampleMR and MendelianRandomization packages in R.

Observational analyses: - BOHBUT categorized at higher levels showed a positive association with general cognitive function, strongest around cutoff 0.35 mmol/L (β=0.257 SD, P=9.74×10−3), with significant evidence when BOHBUT ≥0.32 mmol/L (P<0.0167). Suggestive association with verbal meaning (βmax=0.136, P=3.27×10−2 at 0.32 mmol/L); no clear associations with verbal fluency or word memory. - ACACE showed a strong association with general cognitive function when ACACE ≥0.15 mmol/L (βmax=3.40, P=6.69×10−3 at 0.165 mmol/L) and suggestive association with verbal meaning (βmax=0.119, P=1.77×10−2 at 0.1 mmol/L); no clear associations with fluency or memory. - Continuous analyses: BOHBUT showed little evidence with general cognitive function (β=0.318, P=0.15) and verbal meaning (β=0.292, P=7.28×10−2). ACACE was positively associated with verbal meaning (β=1.152, P=4.59×10−3) and showed suggestive evidence for general cognitive function (β=1.285, P=1.97×10−2). MR analyses (BOHBUT): - Cognitive performance: Using strong instruments (5 SNPs; R2≈0.6%; F=54.6), genetically predicted BOHBUT was positively associated with cognitive performance (IVW β=7.89×10−2, P=1.03×10−2; weighted median β=8.65×10−2, P=9.60×10−3; weighted mode β=0.126, P=8.56×10−2). Results were consistent using suggestive instruments (IVW β≈6.11×10−2; details referenced). No heterogeneity (Cochran’s Q P>0.1), no directional pleiotropy (MR-Egger intercept P>0.05), and no outliers (MR-PRESSO P>1×10−6). - Alzheimer’s disease: Evidence for a protective effect (IVW β=−0.308; OR=0.735; P=3.06×10−2). Weighted median similar but less precise (β=−0.319; OR=0.727; P=5.43×10−2); weighted mode β=−0.327; OR=0.721; P=0.212. Using suggestive instruments, evidence was not observed. - Sensitivity excluding rs6982503 (associated with smoking): BOHBUT still positively associated with cognitive performance (IVW β=9.26×10−2, P=3.43×10−3; weighted median β=0.115, P=9.99×10−4). AD protective effect similar but less precise (IVW β=−0.274; OR=0.76; P=7.22×10−2). No pleiotropy/outliers detected. MR analyses (ACACE): No evidence of causal effects on cognitive performance or AD; no pleiotropy/outliers detected. cis-MR (OXCT1): Three cis-SNPs (R2=0.19%; F=28.4). Little evidence for effects on cognitive performance (IVW β=0.038; P=0.312) or AD (IVW β=−0.331; P=0.290), though directions were consistent with main MR; no pleiotropy (P=0.869).

The triangulation strategy indicates that higher BOHBUT levels are linked to better general cognitive performance and may reduce AD risk. Observational analyses identified beneficial associations at BOHBUT concentrations ≥0.32 mmol/L, suggesting that sub-ketogenic increases might benefit early cognitive function. MR findings support a causal interpretation, reducing confounding and reverse causation concerns. The nominal protective effect on AD aligns with the observed positive effect on cognitive performance and supports hypotheses from prior clinical and preclinical research on ketone neuroprotection. Potential non-linearity is suggested by attenuated associations at higher categorical cutoffs and known adverse effects of very high ketone levels (ketoacidosis), indicating an optimal therapeutic range may exist. The cognitive tests used primarily reflect functions in prefrontal/frontal cortices and hippocampus; effects on other brain regions (e.g., white matter affecting processing speed) warrant study. ACACE, despite instability, also showed positive observational associations, supporting the overall ketone-cognition link. The MR results imply lifelong genetically elevated BOHBUT improves cognition, which may conceptually align with dietary patterns that intermittently elevate KBs (e.g., low-carbohydrate diets, intermittent fasting), though immediate intervention effects require RCTs. Strengths include precise NMR metabolite quantification, concordant observational and genetic evidence, and biologically plausible instruments (e.g., proximity to OXCT1).

Triangulated observational and MR evidence indicates that higher levels of 3-β-hydroxybutyrate are beneficial for general cognitive performance, with some evidence suggesting a protective effect against Alzheimer’s disease. These findings support the hypothesis that increasing BOHBUT may improve cognitive function and potentially slow cognitive decline and AD risk. Further randomized controlled trials, larger GWAS to improve instrument strength, evaluations of non-linear effects, and studies in diverse populations are recommended.

- Potential measurement error in ketone body levels due to variable fasting duration before blood sampling. - Limited instrument strength: genetic instruments explained a small variance in KBs (BOHBUT ~0.6%, ACACE ~0.5%). Larger GWAS are needed to identify stronger instruments. - Assumption of linearity between KBs and cognitive performance may be violated; potential non-linear relationships were not modeled. - Generalizability is limited as analyses were conducted in predominantly European (Caucasian) populations. - Observational analyses constrained by the KB range in WHII (mostly <0.5 mmol/L), limiting exploration of broader dose–response and potential non-linearity.