Overview of examine recruitment
Participants on this examine have been recruited from the client base of 23andMe. Participants supplied knowledgeable consent and volunteered to take part within the analysis on-line, below a protocol authorised by the exterior Association for the Accreditation of Human Research Protection Programs-accredited Salus Institutional Review Board (https://www.versiticlinicaltrials.org/salusirb). Participants have been included within the evaluation on the idea of consent standing verified on the time knowledge analyses have been initiated.
The 23andMe GLP1 survey was launched to analysis members in August 2024. The survey aimed to seize members’ experiences with GLP1 receptor agonist remedy, and was focused to 23andMe members who had beforehand responded within the affirmative to the query, ‘Have you ever taken prescription medications to help you lose weight?’. The survey included questions concerning drug model, dosing routine, time on remedy, efficacy (together with pre-treatment weight and weight on remedy), and side effects, in addition to causes for pursuing or stopping GLP1 remedy. We centered the survey and subsequent evaluation on primarily six drug varieties; Ozempic, Wegovy, compounded semaglutide, Mounjaro, Zepbound and compounded tirzepatide, the primary three of which symbolize variations of semaglutide, and the final three symbolize variations in tirzepatide. A full record of survey questions might be present in Supplementary Table 24.
Phenotype definitions
Using the knowledge derived from the surveys, we outlined phenotypes that aimed to seize facets of drug efficacy and side effects. We outlined our efficacy phenotype because the distinction between pre-treatment BMI to post-treatment BMI (or present BMI, if remedy is on-going). In common, for examine members who reported taking a couple of GLP1 remedy, we chosen the GLP1 remedy that they reported taking for the longest interval of time. Specifically, we outlined a proportion BMI change phenotype as:
$${{rm{Delta BMI}}}_{ % }=100({mathrm{BMI}}_{2}-{mathrm{BMI}}_{1})/{mathrm{BMI}}_{1}$$
the place BMI1 and BMI2 symbolize pre-treatment and post-treatment BMI, respectively, measured in weight in kilograms per top in metres squared. We utilized high quality management filters to individuals with weight lower than 36 kg or higher than 181 kg, top lower than 1.39 m or higher than 2.06 m, BMI lower than 14 kg m−2 or higher than 70 kg m−2, or age lower than 18 years. In combination, these preliminary filters eliminated 80 individuals (0.29%). Inspection of the ΔBMI% phenotype revealed a heavy tailed distribution, so we additional high quality managed the ΔBMI% phenotype to take away outlier members with BMI adjustments above 20% or beneath −45% (Extended Data Fig. 10). The ΔBMI% estimates have been set to lacking for members who didn’t go high quality management.
To allow genetic associations to be interpreted in models of weight quite than ΔBMI%, we additionally outlined a corresponding Δweight phenotype, outlined because the change in weight from baseline in kilograms. We word that, as a result of grownup top is handled as fixed through the remedy window, the share change in BMI (ΔBMI%) is mathematically similar to the share change in weight (Δweight%).
For the side impact phenotypes, we outlined separate case–management phenotypes for every side impact recorded within the survey, contrasting those that self-rated their side effects as reasonable or extreme (instances) to those that self-rated their side effects as delicate or non-existent (controls). As earlier than, for examine members who reported taking a couple of GLP1 remedy, we chosen the GLP1 remedy that they reported taking for the longest interval of time.
We additional outlined phenotypes to symbolize covariates, particularly for drug kind (semaglutide = 1 versus tirzepatide = 0), dosage and days on remedy. For the dosage phenotype, we used the reported most up-to-date weekly dosage in milligrams; this was both the ultimate dose or the present dose for individuals nonetheless taking remedy.
Comparison of self-report and EHR knowledge
As half of the 23andMe expertise, analysis members are supplied the chance to share EHR data collected on their Apple iPhone units. Specifically, the Apple Health utility allows connection to healthcare suppliers for the needs of sharing EHR data with third events by Apple HealthPackage (https://developer.apple.com/documentation/healthkit). 23andMe analysis members can elect to share their EHR data for analysis functions. We used these knowledge to carry out comparisons with the self-report survey knowledge. Full particulars of comparability analyses are supplied in Supplementary Information.
Non-genetic predictors of BMI loss
To analyse the dependence of achieved BMI loss on non-genetic components comparable to drug kind, dosage and time on remedy, we match the next mannequin:
$$start{array}{c}{{rm{Delta BMI}}}_{ % }sim mathrm{age}+mathrm{intercourse}+{mathrm{BMI}}_{1}+mathrm{drugType}+mathrm{dose}+{mathrm{days}}_{mathrm{deal with}} ,+,mathrm{drugType}:mathrm{dose}+mathrm{drugType}:{mathrm{days}}_{mathrm{deal with}} ,+,mathrm{dose}:{mathrm{days}}_{mathrm{deal with}}+mathrm{dose}:{mathrm{days}}_{mathrm{deal with}}:mathrm{drugType}finish{array}$$
(1)
the place ‘drugType’ is an indicator variable that equals 1 for people utilizing semaglutide and 0 for tirzepatide, ‘dose’ represents the dose in milligrams, daysdeal with represents the full days on the related drug and ‘:’ represents an interplay time period between two or extra variables. Note that semaglutide and tirzepatide sometimes have completely different commonplace dosing ranges, which is dealt with within the regression mannequin by the ‘drugType:dose’ interplay time period.
Genotyping and SNP imputation
DNA extraction and genotyping have been carried out on saliva samples by Clinical Laboratory Improvement Amendments-certified and College of American Pathologists-accredited scientific laboratories of Laboratory Corporation of America. Samples have been genotyped on one of 5 genotyping platforms. The V1 and V2 platforms have been variants of the Illumina HumanHap550 BeadChip and contained a complete of about 560,000 SNPs, together with about 25,000 customized SNPs chosen by 23andMe. The V3 platform was primarily based on the Illumina OmniExpress BeadChip and contained a complete of about 950,000 SNPs and customized content material to enhance the overlap with our V2 array. The V4 platform was a completely customized array of about 950,000 SNPs and included a decrease redundancy subset of V2 and V3 SNPs with extra protection of lower-frequency coding variation. The V5 platform was primarily based on the Illumina Global Screening Array, consisting of roughly 654,000 preselected SNPs and roughly 50,000 customized content material variants. Participant genotype knowledge have been imputed in opposition to a reference panel composed of knowledge from the Haplotype Reference Consortium34 and augmented with extra sequences to spice up imputation efficiency (Supplementary Information).
Association testing
We carried out a GWAS of ΔBMI% in individuals of European ancestry utilizing strategies which were described beforehand35. In transient, unrelated members have been included within the GWAS analyses on the idea of European ancestry as decided by a genetic ancestry classification algorithm36. The GWAS was carried out together with covariates as described in equation 1 above, with the addition of 5 genetic principal elements to account for fine-scale genetic ancestry, and indicator variables to account for variation within the genotyping platform. Among 21,822 individuals of European ancestry, we required members to have full knowledge wanted to assemble the goal phenotype and GWAS covariates (that’s, knowledge out there for pre-treatment weight, post-treatment weight, drug kind, dosage, time on remedy and components comparable to age, intercourse and top), leading to 18,488 members. Finally, members have been filtered on relatedness such that no two individuals shared greater than 700 cM identification by descent37, which corresponds roughly to the minimal anticipated sharing between first cousins in an outbred inhabitants, leading to a remaining GWAS pattern measurement of 15,237. An equal process was used for a GWAS of side impact phenotypes. For the needs of testing drug-specific associations, we repeated the GWAS process for the semaglutide and tirzepatide-treated populations individually, eradicating the drug-type covariate and interplay phrases as acceptable. All GWASs have been adjusted for inflation utilizing genomic management, with the inflation issue being not more than 1.035 in all phenotypes.
Given the smaller pattern sizes out there in non-European populations, we didn’t carry out genome-wide affiliation testing in these populations, and as a substitute centered analyses on variants found as related within the European GWAS. For these variants, we examined for affiliation in non-European populations following an analogous method to that described above.
Replication
We carried out replication of the recognized efficacy affiliation within the All of Us cohort38, utilizing Controlled Tier Dataset v.8. We extracted genomic knowledge, EHR knowledge and a drug code referring to both semaglutide or tirzepatide from 9,579 members. After filtering to retain members with data concerning pre-treatment and post-treatment BMI and genotype knowledge passing high quality management, we obtained 4,889 members, of which 3,948 had full knowledge when incorporating covariates akin to these used within the GWAS. For the replication evaluation, we examined for affiliation between the EHR-derived ΔBMI% and the genotype, together with covariates. We repeated the replication evaluation having carried out mean-imputation of lacking drug dose knowledge, permitting a bigger pattern measurement of 4,855 to be analysed.
We additionally tried replication evaluation within the UK Biobank cohort, though the out there knowledge predate the supply of semaglutide or tirzepatide, and therefore relied on earlier variants of GLP1 receptor agonists. Full particulars of the replication evaluation methodology is supplied in Supplementary Information.
Genetic and non-genetic threat modelling
To assemble mixed genetic and non-genetic fashions of ΔBMI% and threat of treatment-related side effects, we chosen remedy, scientific, demographic, illness analysis and genetic variables as predictors. In addition to the covariates included within the GWAS, we additionally included years of schooling as a proxy for socio-economic standing, and binary indicators of earlier illness analysis for T2D, hypertension and non-alcoholic fatty liver illness. All steady predictor variables have been standardized earlier than modelling to permit for the comparability of impact sizes.
We used a linear multi-variable mannequin to suit ΔBMI%. Given the binary nature of side impact phenotype definitions, we fitted multi-variable logistic regression fashions (equal to a generalized linear mannequin with a binomial household and logit hyperlink perform). The dataset was partitioned randomly into coaching (70% of the pattern) and held out take a look at (30%) units, with the take a look at set getting used to evaluate mannequin efficiency. Further particulars are outlined in Supplementary Information.
Model efficiency of efficacy was additional assessed by making use of the mannequin derived from our self-report knowledge in a pattern of 642 individuals who had supplied HealthPackage EHR knowledge however had not accomplished the GLP1 survey, and therefore weren’t used within the building of the mannequin. To replicate the state of affairs the place efficacy predictions are made earlier than remedy, we assumed the dose, remedy length and drug kind variables have been unknown, and imputed these values within the mannequin to an arbitrary fixed worth for all members.
Reporting abstract
Further data on analysis design is on the market within the Nature Portfolio Reporting Summary linked to this text.
