João Delgado

Multiple long-term conditions (MLTC) – the coexistence of two or more long-term conditions (LTCs) in an individual, is an increasingly prevalent and complex clinical challenge. We aimed to uncover opportunities for therapeutic intervention in MLTC and highlight potential adverse effects using genetics to identify shared biological pathways across conditions. Across 72 heritable LTCs common in adults aged ≥65, we analysed genome-wide association study (GWAS) summary statistics from meta-analyses incorporating up to three data sources (UK Biobank, FinnGen, and condition-specific consortia) from genetically European-ancestry participants. Across 843 condition pairs sharing genome-wide significant variants, we performed statistical colocalisation to identify shared causal genetic variants. Variants showing evidence of causality for ≥3 LTCs were prioritised for investigation of likely causal genes, biological pathways, and druggability. Findings were replicated in Our Future Health (OFH, N=550,000). Colocalisation analyses identified 281 genetic variants causal for specific LTC pairs and 85 variants causal for ≥3 LTCs. These 85 variants were located in 34 genomic regions with strong evidence of shared causality across multiple conditions. Thirteen of the 34 regions fully replicated in OFH, with others having partial replication. Eight variants showed concordant directions of effects across conditions, suggesting potential opportunities for drug development; results include a variant in GDF7 causal for diverticular disease, gastro-oesophageal reflux disease, and prostate cancer, suggesting multisystemic mechanisms and a novel intervention target for this cluster of conditions. We identified six potential drug repurposing opportunities involving clinically targeted proteins, including LPA and IRF5 for cardiovascular and immune-mediated inflammatory conditions, respectively. Fifteen variants showed discordant effects across phenotypes, suggesting possible adverse consequences of pharmacological intervention; for example, a variant in GIPR reduced risk of obesity and sleep apnoea whilst increasing risk of breast cancer. These findings reveal shared mechanisms underlying MLTC and highlight opportunities for therapeutic development and repurposing, whilst also identifying possible adverse effects. Author summary: Many individuals live with multiple long-term health conditions, but relatively little is known about the biological processes linking these conditions together. Where conditions share biology, this could reveal new ways to treat diseases. In this study, we used data from large volunteer cohorts of hundreds of thousands of individuals to find regions of DNA associated with multiple conditions. DNA codes for genes, and we found specific genes and biological pathways that partially explain why some conditions occur together. For example, we found evidence linking diverticular disease, gastro-oesophageal reflux disease, and prostate cancer, suggesting that these seemingly different conditions may share underlying biological mechanisms. We identified several proteins that could be explored as new drug targets and opportunities to repurpose existing medications. We also found evidence suggesting that some treatments could have unintended effects, reducing the risk of one condition but increasing the risk of another. This work may help support the development of treatments for people with multiple long-term conditions.

Thrombolysis has been shown to reduce disability after acute ischaemic stroke in randomised trials. Our study sought to compare the benefit from thrombolysis observed in a comprehensive national stroke registry with that expected from the clinical trials. Data from a total of 168,347 ischaemic stroke patients who attended one of 118 emergency stroke hospitals in England and Wales from 2016 to 2021 were extracted from the Sentinel Stroke National Audit Programme. We constructed a machine learning model, designed to isolate the effect of thrombolysis, using explainable machine learning (XGBoost with SHapley Additive exPlanations [SHAP]). Thrombolysis was found to be associated with a statistically significant improvement in the odds of achieving a better outcome (modified Rankin scale [mRS] threshold). Regression analysis predicted a maximum 2.5-fold improvement in odds of achieving mRS 0-1, with a decline to no treatment effect at 5 h 28 min post-onset. Our results confirm a beneficial effect of thrombolysis in a large prospective national stroke registry, and align closely with meta-analyses of clinical trials of thrombolysis both in terms of magnitude of effect and decline over time. This work also demonstrates the potential to apply explainable machine learning to observational data to assist in understanding how clinical trials are implemented in real-world settings. Thrombolysis, in practice, has the same observed benefit as in the clinical trials.

Background: Hypertension and type 2 diabetes (T2D) are two of the most frequently co-occurring long-term conditions, but their shared mechanisms are not fully understood, often being attributed to adiposity pathways. Here, we aimed to identify shared genetic mechanisms independent of adiposity. Methods: We performed genome-wide association study meta-analyses of T2D and, separately, hypertension. We investigated the bidirectional causal relationship using Mendelian randomisation and quantified genetic correlation before and after accounting for common modifiable risk factors. We then applied a Bayesian GWAS approach to re-estimate SNP-disease effects after accounting for the causal genetic effects of adiposity-related traits. Colocalisation analysis identified shared causal genetic variants, and we investigated the biological pathways involved. Results: We observed a bidirectional causal relationship, and substantial genetic correlation between the two traits (rg = 0.48, 95%CI 0.45-0.52), which persisted after accounting for the genetic contributions of BMI, waist-hip ratio (WHR), and triglycerides (rg = 0.29, 95%CI 0.24-0.34). This indicated shared mechanisms beyond those captured by standard measures of adiposity. We found 98 genetic loci containing variants significantly associated with both hypertension and T2D; colocalisation analysis identified 37 that contained specific shared causal variants. Of these, eight remained statistically significant after adjusting for genetic measures of adiposity, and four were identified only after removing the causal effect of adiposity measures. Shared variants include an allele within PCSK7 associated with risk of both T2D and hypertension and with circulating PCSK7 protein levels, as well as a variant in the 3′ untranslated region of ZNF101, within the TM6SF2 locus, likely reflecting regulatory variation affecting hepatic lipid metabolism and cardiometabolic traits. Conclusions: We identify shared causal mechanisms between hypertension and T2D independent of several measures of adiposity. These findings, particularly PCSK7, could uncover novel interventions or opportunities for prevention of this co-occurring condition pair.

Background: The iron overload disease haemochromatosis is primarily caused by HFE p.C282Y homozygosity, yet penetrance of clinical outcomes (including liver disease/cancer) varies. We estimated the effect of genetic and lifestyle factors on disease penetrance and expressivity in C282Y homozygotes. Methods: We analysed 2,893 C282Y homozygous UK Biobank participants (n=1,295 male). We ascertained haemochromatosis from medical records, liver disease/cancer, osteoarthritis, joint replacement surgeries, and dementia diagnoses. We derived polygenic scores (PGS) for iron biomarkers including hepcidin and transferrin saturation (TSAT). Sex-stratified logistic regression assessed associations with clinical outcomes. We used time-to-event regression estimating effects of age, lifestyle, and PGS. We estimated effects of rare HFE variants using whole genome sequencing data. Results: In male HFE C282Y homozygotes, higher TSAT PGS increased likelihood of diagnosis of haemochromatosis, and separately any clinical consequence (OR _{top-vs-bottom-PGS-quintile} =1.83, 95%CI: 1.26-2.66, p=0.001). Cumulative incidence of assessed haemochromatosis clinical outcomes in men by age 80 was 64.5% (highest quintile) versus 51.6% (lowest) (p-value for difference=0.025). In females, TSAT PGS increased haemochromatosis likelihood (cumulative incidence: 45.3% vs. 23.3% [highest/lower quintile], p=0.00001) but not liver disease. PGS for other iron biomarkers were not significantly associated with clinical outcomes. Rare heterozygous predicted loss-of-function variants in HFE increased haemochromatosis likelihood in non-C282Y homozygotes (aggregate-OR=14.8, 95%CI 4.7-41.1, p=0.003), highlighting the importance of sequencing undiagnosed individuals to find rare causes of haemochromatosis. Conclusion: Higher genetically predicted transferrin saturation significantly increased risk of clinical outcomes in HFE C282Y homozygotes. Combined with modifiable lifestyle factors, genetic information could refine risk stratification and personalise iron monitoring, following validation.

Introduction: HFE genetic variants, especially C282Y homozygosity (C282Y+/+), can increase systemic iron and cause hemochromatosis, though expression varies. Excess iron can lead to liver disease/cancer, yet factors influencing liver iron beyond HFE genotype remain unclear. We investigated genetic/environmental factors influencing liver iron, including HFE genotype and hemochromatosis diagnosis. Methods: We analysed 37,287 European ancestry UK Biobank participants (mean age 64.1, SD 7.6) with HFE genotypes and MRI-estimated liver iron concentrations (MRLIC). Linear regression assessed MRLIC associations with genetic and environmental factors, adjusting for age, sex, and genetic covariates. Results: Mean MRLIC was highest in undiagnosed C282Y+/+ males and females (2.56 and 2.31mg/g) vs. diagnosed (1.23 & 1.51mg/g, p=0.0001 & 0.0004). Other HFE genotypes had nominal increases vs. those without HFE genetic variants. Higher MRLIC was associated with higher alcohol intake (β=0.11, 95%CI: 0.09–0.11, p=6.0×10 ⁻¹²⁸ ; >30 vs. 1–14 units/week), frequent red/processed meat consumption (β=0.08, 95%CI: 0.07–0.09, p=3.7×10 ⁻⁵⁴ ; ≥3 times/week vs. none), high waist-height ratio (β=0.01, 95%CI: 0.006–0.02, p=6.4×10 ⁻⁵ ) and genetically predicted transferrin saturation (β=0.22, 95%CI: 0.19– 0.26, p=3.8×10 ⁻⁴⁶ ). Lower MRLIC was associated with underweight BMI (β=−0.06, 95%CI: −0.09 – −0.03, p=1.1×10 ⁻⁴ ), proton pump inhibitor use (β=−0.03, 95%CI: −0.04 – −0.03, p=3.5×10 ⁻¹⁷ ), and ≥4 cups/day of tea (β=−0.01, 95%CI: −0.02 – −0.003, p=0.005 vs. non-drinkers). Among C282Y+/+ individuals (n=91), tea ≥4 cups/day was associated with lower MRLIC (β=−0.99, 95%CI: −1.89 – −0.08, p=0.03). Conclusion: Undiagnosed C282Y+/+ individuals had excess liver iron vs. diagnosed, likely due to treatment. Genetic and environmental factors influence liver iron beyond C282Y+/+. Tailored lifestyle advice could benefit those at risk of hemochromatosis.

Introduction: HFE genetic variants, especially C282Y homozygosity (C282Y+/+), can increase systemic iron and cause hemochromatosis, though expression varies. Excess iron can lead to liver disease and liver cancer, yet factors influencing liver iron beyond HFE genotype remain unclear. We investigated genetic/environmental factors influencing liver iron, including HFE genotype and hemochromatosis diagnosis. Methods: We analyzed 37,287 European ancestry UK Biobank participants (mean age 64.1, SD: 7.6) with HFE genotypes and MRI-estimated liver iron concentrations (MRLIC). Linear regression assessed MRLIC associations with genetic and environmental factors, adjusting for age, sex, and genetic covariates. Results: Mean MRLIC was highest in undiagnosed C282Y+/+ males and females (2.56 and 2.31 mg/g) versus diagnosed (1.23 and 1.51 mg/g, p =0.0001 and 0.0004). Other HFE genotypes had nominal increases versus those without HFE genetic variants. Higher MRLIC was associated with higher alcohol intake (β=0.11, 95% CI: 0.09–0.11, p =6.0×10 ⁻¹²⁸ ; >30 vs. 1–14 units/wk), frequent red/processed meat consumption (β=0.08, 95% CI: 0.07–0.09, p =3.7×10 ⁻⁵⁴ ; ≥3 times/week vs. none), high waist-height ratio (β=0.01, 95% CI: 0.006–0.02, p =6.4×10 ⁻⁵ ; although magnitude was weak) and genetically predicted transferrin saturation (β=0.22, 95% CI: 0.19–0.26, p =3.8×10 ⁻⁴⁶ ). Lower MRLIC was associated with underweight body mass index (β=−0.06, 95% CI: −0.09 to −0.03, p =1.1×10 ⁻⁴ ) and proton pump inhibitor use (β=−0.03, 95% CI: −0.04 to −0.03, p =3.5×10 ⁻¹⁷ ). Conclusions: Undiagnosed C282Y+/+ individuals had excess liver iron versus diagnosed, likely due to treatment. Genetic and environmental factors influence liver iron beyond C282Y+/+. Tailored lifestyle advice could benefit those at risk of hemochromatosis.

Background: Multimorbidity, the presence of two or more conditions in one person, is common but studies are often limited to observational data and single datasets. We address this gap by integrating large-scale primary-care and genetic data from multiple studies to identify novel multimorbidity patterns and producing digital resources to support future research. Methods: We defined chronic, common, and heritable conditions in individuals aged ≥65 years, using two large primary-care databases [CPRD (UK) N=2,425,014 and SIDIAP (Spain) N=1,053,640], and estimated heritability using the same definitions in UK Biobank (N=451,197). We used logistic regression to estimate the co-occurrence of pairs of conditions in the primary care data. Linkage disequilibrium score regression was used to estimate genetic similarity between pairs of conditions. Meta-analyses were conducted across databases, and up to three sources of genetic data, for each pair of conditions. We classified pairs of conditions as across or within-domain based on the international classification of disease. Findings: We identified 72 chronic conditions, with 43·6% of 2546 pairs showing higher co-occurrence than chance in primary care and evidence of shared genetics. Many across-domain pairs exhibited substantial shared genetics (e.g. iron deficiency anaemia and peripheral arterial disease: genetic correlation R _g =0·45[95% Confidence Intervals 0·27:0·64]). 33 pairs displayed negative genetic correlations, such as skin cancer and rheumatoid arthritis ( R _g =-0·14[-0·21:-0·06]), due to potential adverse drug effects. Discordance between genetic and primary care data was also observed, e.g., abdominal aortic aneurysm and bladder cancer co-occurred in primary care but were not genetically correlated (Odds-Ratio=2·23[2·09:2·37], R _g =0·04[-0·20:0·28]) and schizophrenia and fibromyalgia were less likely to co-occur together in primary care but were positively genetically correlated (OR=0·84[0·75:0·94], R _g =0·20[0·11:0·29]). Interpretation: Most pairs of chronic conditions show evidence of shared genetics, and co-occurrence in primary care, suggesting shared mechanisms. The identified patterns of shared genetics, negative correlations and discordance between genetic and observational data provide a foundation for future multimorbidity research. Funding: UK Medical Research Council [MR/W014548/1]. Research in context: Evidence before this study: We. Searched PUBMED, Embase, Global Health for articles published in English, since 6. Date >2012 (after the definitive Barnett et al. multimorbidity review paper was published) until 01 May 2024., using the search terms: (((multidisease*[tiab] OR multi-disease*[tiab] OR 3ultimorbidity*[tiab] OR multi-morbidit*[tiab] OR multipatholog*[tiab] OR multi-patholog*[tiab] OR pluripatholog*[tiab] OR polypatholog*[tiab] OR poly-pathology*[tiab] OR “multiple long-term conditions”[tiab])) OR (multimorbidity[MeSH Terms])) AND (“genetic*”[tiab] OR “genomic*”[tiab] OR (“genome wide association stud*”[tiab] OR “genome wide association analysis”[tiab] OR “GWAS”[tiab])) NOT (animals[mh] NOT humans[mh]) AND NOT (“retracted publication”[pt]). We excluded studies investigating monogenic mutations and case studies. 243 studies met the primary inclusion criteria, where 68 presented original research addressing multimorbidity and incorporating genetic analysis. Multimorbidity is predominantly characterised by counts or clusters of conditions, with limited exploration of underlying mechanisms. Only 11 investigated multimorbidity beyond the analysis of comorbidities of single conditions, with five reporting systematic investigations of genetics across multiple long-term conditions. None of these studies compare data from more than one genetic database with ‘real-world’ data. This comparison is necessary to identify correlations of diseases and to be able to carry out preventive actions that facilitate better health care. Added value of this study: We describe a large-scale analyses into the genetics of multimorbidity informed by representative primary care data, starting from a systematic standardisation of relevant diagnostic codes, and the ascertainment of conditions harmonised across multiple databases. Importantly, we include the analysis of multiple sources of data from both observational and genetic databases, whereas previous studies have usually been limited to one study, such as UK Biobank. We also produced a detailed interactive interface for 72 common long-term and heritable conditions and 2546 pairs of conditions across multiple sources of data: https://gemini-multimorbidity.shinyapps.io/atlas . Together we describe the parallel analysis of specific co-occurring pairs in primary care and genetic data, involving 100,000s to millions of patients and 2546 pairs of conditions. These comparisons identified novel pairs of conditions, and pairs of conditions that shared genetics but did not co-occur in primary care, or vice versa. Implications of available evidence: Large-scale, digital databases of electronic medical records linked to genetic information involving 100,000s of patients can significantly enhance our comprehension of multiple long-term conditions (multimorbidity). This advancement builds on previous work by providing evidence of potential shared biology between 100s of specific pairs of conditions.

Background: Calcium channel blockers (CCBs) are common antihypertensive medications. Pharmacogenetic variants affect CCB clinical outcomes, although effect sizes are modest in community samples. Variation in patient characteristics may also predict CCB outcomes, and variation attributable to relevant polygenic scores is less prone to confounding. We aimed to test associations between genetically predicted patient characteristics plus pharmacogenetic variants with CCB outcomes in a large community cohort. Methods: We extended our analysis of 32,000 UK Biobank dihydropiridine CCBs treated participants (mean duration 5.9 years) testing 23 variants, where NUMA1 rs10898815 and RYR3 rs877087 showed the most robust associations (for discontinuation and heart failure, respectively). We calculated polygenic scores for systolic and diastolic blood pressures (SBP and DBP), body fat mass, waist hip ratio, lean mass, serum calcium, eGFR, lipoprotein A, urinary sodium, and liver fibrosis. Outcomes were CCB discontinuation, heart failure, coronary heart disease and chronic kidney disease. Results: For heart failure, the highest risk 20% of polygenic scores for fat mass, lean mass and lipoprotein A were associated with increased risks (Hazard-Ratio (HR)Fat-mass 1.46, 95% CI 1.25-1.70, p=1*10-6; HRLean-mass 1.20, 95%CI 1.04-1.38, p=0.01; HRLipoproteinA 1.29, 95% CI 1.12 to 1.48, p= 4*10-4), versus the lowest risk 20% of each score respectively. Across the cohort, RYR3 T-allele modestly increased heart failure risks (HR 1.13: 1.02-1.25) versus non-carriers, but in subsets with high fat mass, lean mass, and lipoprotein A scores, estimates were substantially larger, e.g., in females aged 65-70 the heart failure Relative Risk was 4.4 (95% CI 1.54-12.4) versus no T-alleles and low scores. For CCB discontinuation, high polygenic scores for fat mass and lean mass increased risks versus the lowest 20%, whereas high SBP and DBP scores decreased discontinuation risks. Hazard ratios for discontinuation with the pharmacogenetic NUMA1 rs10898815 A-allele (overall HR 1.07: 1.02-1.12) were higher (HR 1.17: 1.05-1.29) in those with high polygenic scores for fat mass and lean mass. Conclusion: Polygenic scores affecting adiposity and lipoprotein A levels add to known pharmacogenetic variants in predicting key clinical outcomes in CCB treatment. Combining pharmacogenetic variants and relevant individual characteristic polygenic scores may help for personalizing prescribing. What is needed, what do we add?: We previously showed that pharmacogenetic variants in RYR3 and NUMA1 were associated with key clinical outcomes in community CCB patients, although effect sizes were modest. Various patient characteristics reportedly affect CCB outcomes. We therefore tested effects of relevant patient characteristics using polygenic scores. They minimize the effect of unmeasured confounders as genotypes are invariant since conception and reflect lifetime exposure to the risk factor. We showed that combining associated scores with the pharmacogenetic variants improved outcome prediction.