Introduction

Clinical background of antipsychotic metabolic liability

Antipsychotic medications are essential treatments for schizophrenia, bipolar disorder, and other severe psychiatric conditions. Their use, however, is associated with a range of metabolic complications, including weight gain, obesity, glucose dysregulation, dyslipidemia, metabolic syndrome, and increased cardiovascular risk. These complications are clinically important because people with severe mental illness already experience higher rates of cardiometabolic disease and premature mortality than the general population. The metabolic consequences of treatment therefore occur in a population that may already have elevated baseline vulnerability.

The magnitude of metabolic liability is not uniform across antipsychotic drugs. Clinical trials, observational studies, systematic reviews, and network meta-analyses have consistently shown differences between individual agents. Clozapine and olanzapine are generally placed toward the less favorable end of the metabolic spectrum, with greater average adverse effects on weight, glucose, or selected lipid measures. In contrast, aripiprazole, ziprasidone, and lurasidone are usually associated with smaller average effects on weight or selected metabolic outcomes. These differences are not identical across all endpoints. A drug may have a relatively modest effect on weight but a different pattern of lipid or glucose changes, and estimates can vary according to treatment duration, patient population, baseline body mass index, comparator, and whether patients are antipsychotic-naive or switching from another agent [1-5].

The network meta-analysis [2] is particularly relevant to the present work because it included glucose, glycated hemoglobin, insulin resistance, triglycerides, total cholesterol, high-density lipoprotein (HDL) cholesterol, and LDL cholesterol. In that analysis, olanzapine and clozapine were associated with greater glyco-metabolic deterioration, whereas aripiprazole and ziprasidone were associated with smaller average deteriorations. The regression analysis also examined receptor occupancies and found that higher H1, M1, and M3 receptor occupancy was associated with larger glyco-metabolic changes. The earlier meta-analysis [1] similarly showed that the magnitude of antipsychotic-associated weight gain varies between agents and increases with longer exposure, although its analysis focused on weight and body mass index rather than LDL specifically.

The clinical importance of these effects has led to recommendations for baseline and follow-up monitoring of weight, glucose-related measures, and lipid abnormalities in patients receiving antipsychotic treatment. The consensus statement from the American Diabetes Association and collaborating organizations emphasized the need to recognize obesity and diabetes risk during antipsychotic treatment [6], while later clinical guidance addressed the management of weight gain, metabolic disturbances, and cardiovascular risk [7,8]. These recommendations provide the clinical context for computational prioritization, but they do not establish that a receptor-based score can substitute for patient-level monitoring or outcome-based risk prediction.

Receptor mechanisms

Receptor pharmacology has long been used to explain why antipsychotic drugs differ in their adverse-effect profiles. Histamine H1 receptor affinity is one of the best-known pharmacological correlates of antipsychotic-associated weight gain. A receptor-binding analysis reported that H1-histamine receptor affinity predicted short-term weight gain across typical and atypical antipsychotics [9]. The analysis also identified associations involving α1A-adrenergic, 5-HT2C, and 5-HT6 receptor affinities. These findings helped establish the rationale for assigning greater prior importance to HRH1 and HTR2C in metabolic-risk models.

The underlying biology is nevertheless multireceptor and context-dependent. Antipsychotics often bind to several receptor systems simultaneously, and their downstream effects depend on receptor occupancy, tissue distribution, intrinsic activity, dose, treatment duration, and individual susceptibility. Serotonergic, histaminergic, muscarinic, adrenergic, and dopaminergic pathways can influence appetite, energy balance, insulin secretion, glucose transport, lipid metabolism, and central reward processing. Reviews of antipsychotic metabolic effects have therefore emphasized that no single receptor fully explains the clinical phenotype [10-12]. The multiplicity of serotonin receptors provides an additional reason to treat receptor-level attribution as context-dependent rather than as a direct statement of causal importance [13].

The contribution of a receptor in a computational model should also be distinguished from its causal importance in patients. A receptor may receive a high prior weight but contribute little to a particular drug score if the available binding measurements, dose proxy, or transcriptome-wide association study (TWAS) signal are weak. Conversely, a receptor with a low or absent prior weight may contribute substantially under a uniform model because of its affinity profile or the transformation applied to the genetic association signal. Quantifying this distinction is the central purpose of the present analysis.

Genomic context and multi-ancestry considerations

LDL cholesterol is a clinically relevant lipid trait with a substantial genetic component. Multi-ancestry lipid studies have demonstrated both shared and ancestry-dependent features of lipid-associated genetic architecture. Differences in allele frequencies, linkage disequilibrium, effect sizes, imputation quality, and fine-mapping resolution can affect the apparent strength and location of genetic associations. The Global Lipids Genetics Consortium has shown that genetic diversity can improve the discovery and localization of lipid-associated loci, while also highlighting the importance of including populations that have historically been underrepresented in genome-wide association studies [14].

Transcriptome-wide association studies provide a complementary way to connect genetic variation with gene expression and complex traits. By estimating genetically regulated expression and testing its association with a phenotype, TWAS methods can prioritize genes and tissues that may lie on relevant biological pathways. However, a TWAS association does not automatically establish that the implicated gene is causal. The association may arise because the expression-prediction model is correlated with a causal variant acting through another mechanism. After all, several genes have correlated prediction models, or because of linkage disequilibrium between the predicted-expression signal and an independent causal locus. Tissue mismatch, prediction uncertainty, ancestry mismatch, and differences in linkage-disequilibrium structure also complicate interpretation [15-18].

These considerations are especially important when ancestry-specific TWAS results are integrated into a pharmacological score. Similar rankings across ancestry datasets may indicate shared biological signal, but they may also reflect the common pharmacological component of the model. Conversely, ancestry-specific candidate genes may reflect genuine population-specific genetic architecture, differences in prediction accuracy, or stochastic variation in the underlying summary statistics. The present study therefore treats ancestry comparisons as an analysis of model stability and sensitivity rather than as a direct test of ancestry-specific clinical risk.

Prior work in the present series

The present analysis forms part of a broader series examining the behavior of an integrated Ki-defined daily dose-transcriptome-wide association study (Ki-DDD-TWAS) framework. The first study compared four transformations for incorporating TWAS evidence into antipsychotic metabolic-risk scores. It found that changing the TWAS transformation primarily altered score magnitude while preserving much of the drug ordering [19]. The second study removed literature-derived receptor weights and showed that the broad ranking could remain relatively stable even when the receptor-level explanation changed substantially. That analysis also identified high-TWAS/low-weight candidates, including ADRB1, as hypothesis-generating signals [20]. The third study examined receptor-prior sensitivity in HDL-related analyses across multiple ancestry datasets and described both transancestral stability and clinical discordance [21].

In this paper, “literature-derived receptor priors” means predefined numerical weights assigned to selected metabolic receptors on the basis of prior pharmacological literature. “Ki-DDD-TWAS” denotes a composite score combining inverse Ki, a log-transformed defined-daily-dose term, a receptor weight, and a TWAS-derived scaling factor.

The present paper focuses on LDL and extends the Stage 3 analysis to a second major lipid trait. The central question is whether the prior-sensitivity pattern observed in the earlier work is reproducible when ancestry-stratified LDL TWAS outputs are substituted into the same pharmacological framework. The LDL-specific prior-sensitivity comparison and Stage 3 synthesis are the analyses added here; the LDL TWAS files and previously generated receptor-level and drug-level outputs were reused. The analysis is deliberately downstream. It uses saved long-format and drug-level outputs and does not introduce a new genome-wide association study (GWAS), TWAS, receptor-binding, or causal-inference method.

Aim

The primary objective was to quantify the sensitivity of an integrated Ki-DDD-TWAS prioritization framework to the removal of literature-derived receptor priors in ancestry-stratified LDL analyses. The secondary aims were to quantify changes in receptor-class and gene-level contribution when literature-derived metabolic weights were removed; evaluate the transancestral robustness of drug rankings and gene-direction changes across five ancestry-defined LDL datasets; place the weighted ranking in clinical context and characterize rank movements produced by uniform weighting without treating those movements as clinical validation; identify high-TWAS/low-weight candidates and determine whether they were supported across more than one ancestry; and assess whether the evidence justified dual reporting of a literature-weighted reference model and a uniform sensitivity model.