Neurobeachin ( NBEA ) is a novel gene associated with GLP‐1 receptor agonist associated weight loss

Abstract

Aims

Nearly 42% of adults in the United States have obesity, a significant risk factor for many cardiometabolic diseases and cancers. Glucagon‐like peptide‐1 receptor agonists (GLP‐1RAs) are promising interventions for weight loss, but their efficacy varies significantly across individuals. This study investigates the role of neurobeachin (NBEA), a gene that encodes a protein kinase A anchor protein, on weight loss response in two large, real‐world cohorts.

Materials and Methods

We utilised data from individuals prescribed a GLP‐1RA in the NIH All of Us (N = 6556) and validated in the UK Biobank (N = 241). The NBEA genetic score for weight loss (12–18 months) was developed using the NIH All of Us cohort and independently validated in the UK Biobank. Logistic regression modelled associations between the score and outcomes, including high responsiveness (top 20th percentile for weight loss) and non‐responsiveness (weight change ≥0%).

Results

Individuals meeting the responsive NBEA score threshold were 82% more likely to be highly responsive (FDR p = 1·8 × 10⁻⁶) on liraglutide and were validated in the UK Biobank (odds ratio (OR) = 2·37; p = ·008). Individuals on semaglutide meeting this threshold for highly responsive had OR = 1·63 and OR = 2·21 in discovery and validation sets respectively (p < ·05). Individuals on liraglutide with a non‐responsive NBEA score were 50% more likely to not lose weight (FDR p = 2.9 × 10⁻⁴) and were validated in the UK Biobank (OR = 1·81; p = ·041), but the non‐response score did not validate for semaglutide.

Conclusion

These findings indicate that NBEA genetic variation is predictive of GLP‐1RA weight loss and may support future efforts to identify individuals likely to experience significant weight loss with GLP‐1RAs, enabling personalised obesity treatment strategies.

1. INTRODUCTION

Obesity is an epidemic in the United States, increasing consistently since 1999 and affecting an estimated 42% of adults. ¹ , ² , ³ Obesity increases the risk of many comorbidities, including cardiometabolic, cerebrovascular conditions and cancers. ⁴ , ⁵ , ⁶ Collectively, this represents an unprecedented burden on the health care system and on the quality of life (QOL) for these individuals. Effective and accessible weight management interventions are necessary to improve the long‐term health and QOL of individuals with obesity. Prior to anti‐obesity medications (AOMs), lifestyle modification and surgery were the primary treatments for obesity. While these interventions are effective, adoption is limited for several reasons. Lifestyle modifications are difficult to maintain long‐term and may also be impacted by systemic disparities. Furthermore, bariatric surgery, another effective intervention, is often considered too invasive, and insurance coverage is often limited to individuals that already have multiple uncontrolled comorbidities or extreme obesity. ⁷ , ⁸

Recent clinical trials demonstrated significant weight loss with glucagon‐like peptide‐1 receptor agonists (GLP‐1RAs) and dual therapies (i.e., GLP‐1RA and gastric inhibitory peptide), which have pushed these AOMs to the forefront of obesity management. ⁹ , ¹⁰ GLP‐1RAs were initially approved for type 2 diabetes (T2D) management, where weight loss was a secondary benefit. Liraglutide 3.0 mg was approved for obesity in 2014, but the superior efficacy of semaglutide has driven global demand for these medications, leading to shortages. ¹¹ In addition to differences in drug‐specific efficacies, the individual efficacy of AOMs on weight loss also varies significantly. For example, during a randomised clinical trial of semaglutide 2.4 mg with lifestyle changes, 30% and 14% of participants did not experience robust weight loss of ≥10% and ≥5% after 68 weeks. ⁹ The high demand (and high cost) of these medications has created a critical need for tests that can identify individuals who are most likely to respond to treatment, thereby ensuring more precise prescribing, relieving accessibility issues and reducing the burden of unnecessary costs on payors.

This study investigated the role of a novel gene, neurobeachin (NBEA), in weight loss responses to GLP‐1RA medications. We leveraged two large real‐world cohorts, that is, the National Institutes of Health (NIH) All of Us and the UK Biobank cohorts, which provide linked genetic and electronic health records (EHR) data. Here, we build on past studies investigating clinical factors ¹² as predictors for GLP‐1RA weight loss, to investigating genetic factors for GLP‐1RA weight loss in individuals with and without T2D. Building on prior pre‐clinical and clinical studies of feeding behaviour, we hypothesised that genetic variation in NBEA may influence GLP‐1RA action. Interestingly, NBEA has been previously implicated in dietary preference in pre‐clinical models, where NBEA^+/− knockout mice consumed larger volumes of food with glucose or fructose than wild‐type mice. We observed a similar finding in humans, where the C allele in rs57081354, a single‐nucleotide variant (SNV) in NBEA, was associated with increased consumption of desserts in a retrospective analysis of the Action to Control Cardiovascular Disease in Diabetes (ACCORD) clinical trial. ¹⁴ In addition, SNVs in NBEA have been associated with increased BMI. ¹³ GLP‐1 stimulates secretion of insulin through activation of protein kinase A (PKA). ¹⁵ NBEA encodes a protein kinase A (PKA) anchor protein, ¹³ , ¹⁵ which helps to drive PKA to specific targets. Although to our knowledge, NBEA has not been investigated in GLP‐1 response, PKA anchor proteins have been implicated in endogenous GLP‐1 response on insulin secretion ¹⁵ and may play a role in GLP‐1 signalling in other tissues, such as the hypothalamus. Collectively, this pointed to the need to investigate the role of NBEA in GLP‐1RA weight loss response.

We developed an NBEA score using data from All of Us and validated the findings in the UK Biobank, an independent cohort, representing the largest assessment of genetic variation in GLP‐1RA weight loss to date (Figure 1).

FIGURE 1.

Study workflow.

2. MATERIALS AND METHODS

2.1. Study Population and Data

Data were obtained from two real‐world cohorts. (1) The NIH All of Us cohort is a real‐world cohort ¹⁶ based in the United States, which contains genetic data linked to EHR up until October 2023 (controlled tier dataset v8), and was used for the discovery analysis and the development of the genetic score. (2) We subsequently validated the findings using the UK Biobank, another large real‐world cohort with EHR data up until 2017, based in the United Kingdom. ¹⁷ In both cohorts, participants were included if they were prescribed GLP‐1 medications and were recorded as overweight (BMI ≥27 kg/m²) or obese (BMI ≥30 kg/m²), with or without T2D at most 6 months before the first GLP‐1RA prescription. The BMI thresholds were chosen based on current prescription guidelines for weight loss medications. ¹⁸

2.2. Weight loss response to GLP‐1RA

Weight loss response to GLP‐1RA was investigated 12–18 months after first GLP‐1RA prescription. Because different GLP‐1RAs result in varying degrees of weight loss, we assessed weight loss in two ways: (1) percentage weight change:

$$\text{Weight Change},\%=\frac{{\text{Weight}}_{12-18\text{months post}\mathrm{Rx}}-{\text{Weight}}_{<6\text{months prior to}\mathrm{Rx}}}{{\text{Weight}}_{<6\text{months prior to}\mathrm{Rx}}}$$

and (2) response type (i.e., non‐responsive and highly responsive). Individuals were categorised as non‐responsive if no weight loss was observed (weight change ≥0%). We tested the 10th, 15th and 20th percentiles of weight loss for each GLP‐1RA. Response type was normalised by weight change distributions specific to each GLP‐1RA based on the assumption that if an individual loses more weight than others on a given GLP‐1RA, they would have also lost more weight than others if they were taking a different GLP‐1RA, even if the overall weight loss between the GLP‐1RAs was not equivalent. Individuals were categorised as highly responsive if they were in the top 20th percentile of weight loss for that specific GLP‐1RA, which represented approximately 11% weight loss for semaglutide (Table S1).

2.3. Development of NBEA score in the NIH All of Us

The NIH All of Us cohort was used for discovery because it has more recent coverage of newer GLP‐1RAs. SNVs from whole‐genome sequencing (WGS), with minor allele frequencies (MAF) >2%, were retrieved for the NBEA gene (GRCh38 positions = 13:34942270‐35673022). ¹⁹ Additional details on the WGS quality control can be found in the Supplemental Material. The MAF threshold was selected based on a conservative estimate from PLINK recommendations of MAF>N^−0.05. The number of subjects (N) across all GLP‐1RAs, after excluding those in the semaglutide withheld test set (30%), was 5953. Associations between the NBEA SNVs and weight change after GLP‐1RA prescription were tested using PLINK v1.9. ²⁰ These associations were adjusted for relevant clinical covariates (i.e., sex, age, baseline BMI, genetic ancestry, GLP‐1RA medication, prescription duration, T2D status). Adherence cannot be directly assessed with EHR data, so we used repeat prescriptions as an approximation for whether the patient was continuing their treatment. All analyses were adjusted for the time between the first and last GLP‐1RA prescriptions to account for differences in treatment duration and to reduce potential confounding from factors related to medication adherence. We stratified the analysis by T2D status as well. The T2D subgroup analysis was adjusted for diabetes management medications that influence weight, that is, insulin, sulfonylureas, thiazolidinedione and SGLT2 inhibitors.

NBEA scores were calculated using the clumping and thresholding method. A null linear regression model was used to predict weight change based on clinical variables. NBEA scores were added to the null model, and the NBEA score with the greatest improvement in R ² values was selected for subsequent testing. Functional Annotation of Variants Online Resource (FAVOR), a tool for WGS, was used for the functional annotation of the SNVs included in the score. Annotations include ClinVar, PolyPhen and SIFT. ²¹

NBEA score thresholds were optimised to identify highly responsive and non‐responsive individuals in the NIH All of Us data. Briefly, the NBEA score threshold with the greatest median estimate for association with GLP‐1RA weight loss responses was selected as the optimal threshold. A score below the threshold was predictive of highly responsive individuals, while the score above the threshold was predictive of no response to GLP‐1RAs. Odds ratios (ORs) for being in groups of GLP‐1RA responsiveness based on NBEA scores were based on logistic regression models and a two‐sided test. P‐values were adjusted for multiple hypothesis testing using the Benjamini Hochberg false discovery rate (FDR) approach, ²² and an FDR p < 0.05 was considered statistically significant. A post hoc power calculation was performed for exenatide using G*Power based on a two‐sided test using logistic regression with an alpha of 0.05 and power of 0.8. ²³ The value of incorporating the NBEA score with clinical variables was evaluated using an analysis of variance (ANOVA) to determine whether there was a significant increase in the variation explained by incorporating the NBEA score with clinical variables to a model of only clinical variables. Additional details are provided in the Supplementary Methods. Response prediction was further stratified by sex, race, T2D status and single GLP‐1RA use.

2.4. Validation of NBEA score

The NBEA score was independently validated in patients in the UK Biobank. To maximise the variant coverage of NBEA in the UK Biobank, the score was calculated from both genotyped and imputed variants. LiftOver was used to convert SNV positions from GRCh38 to GRCh37. ²⁴ Logistic regression was used to test for associations with the type of response (e.g., highly responsive, non‐responsive) and the NBEA score. Models were adjusted for the same covariates as All of Us. A one‐sided statistical significance of p < 0.05 was considered the threshold for successful validation.

2.5. NBEA Cardiometabolic and Neurologic PheWAS in the NIH All of Us cohort

We conducted a PheWAS analysis to test for associations between NBEA SNVs and various health conditions using the PheTK package in All of us. ²⁵ PhecodeX was used for mapping conditions based on International Classification of Diseases (ICD‐9 and ICD‐10) codes. After preprocessing, 480 phenotypes across three chapters were tested: endocrine/metabolism (N = 125), cardiovascular (N = 169) and neurological (N = 186) (Supplementary Methods). Out of 17 114 SNVs in NBEA, 403 independent and common SNVs (R ² ≤ 0.75 and MAF >3%) were tested. Association analyses included the whole cohort and sex‐stratified cohorts (male and female), adjusting for covariates: age, sex at birth (if applicable) and the first 10 genetic principal components. P‐values were adjusted for multiple hypothesis testing using the Benjamini Hochberg false FDR approach. ²²

3. RESULTS

3.1. Study population

The NIH All of Us cohort had 255 052 individuals with genotype data linked to medical records. Out of these individuals, 6556 met inclusion criteria. Out of 198 275 individuals in the UK Biobank, 500 individuals met the same criteria, and approximately 48.2% of these individuals were prescribed liraglutide. In our cohorts, there was greater racial and ethnic diversity in the NIH All of Us, with 1386 (21·1%) and 955 (14·6%) individuals identifying as Black and Hispanic, respectively, compared to 10 (2·0%) and 0 (0%) in the UK Biobank (Table 1). There was also a greater proportion of females in the All of Us cohort (64·0%) than in the UK Biobank (43·6%). BMI at baseline was similar between the two cohorts, with 38.55 kg/m² (SD = 8·1) and 37.0 kg/m² (SD = 5·8) in the All of Us and UK Biobank, respectively.

TABLE 1.

Demographic and clinical characteristics of subjects prescribed GLP‐1RA medications in the NIH All of Us and the UK Biobank cohort.

	NIH All of Us (N = 6556)	UK Biobank (N = 500)
Type 2 diabetes	5526 (84.33%)	500 (100%)
Sex
Male	2296 (35.0)	282 (56.4)
Female	4260 (65.0)	218 (43.6)
Age
<35	382 (5.8)	0 (0)
35–44	835 (12.7)	7 (1.4)
45–54	1626 (24.8)	106 (21.2)
55–64	2249 (34.3)	234 (46.8)
≥65	1464 (22.3)	153 (30.6)
Mean (SD)	56.12 (12.20)	60.49 (7.48)
Median (IQR)	57.09 (48.33–65.02)	60.81 (55.51–66.17)
Race or ethnicity
Asian or Pacific Islander	89 (1.3)	21 (4.2)
Black	1386 (21.1)	10 (2.0)
Hispanic	955 (14.6)	‐
White	3570 (54.5)	457 (91.4)
Mixed race	302 (4.6)	3 (0.6)
Missing	182 (2.7)	3 (0.6)
Other	72 (1.1)	6 (1.2)
Body mass index (kg/m2)
≥27 and <30	708 (10.8)	41 (8.2)
≥30	5848 (89.2)	459 (91.8)
Mean (SD)	38.56 (8.06)	37.03 (5.81)
Median (IQR)	37.10 (32.73–42.62)	35.90 (32.78–40.64)
GLP‐1RA medication
Dulaglutide	1864 (28.4)	‐
Exenatide	542 (8.3)	236 (47.2)
Liraglutide	2064 (31.5)	241 (48.2)
Semaglutide	2012 (30.7)	‐
Discovery	1409 (21.5)
Validation	603 (9.2)
Albiglutide and Lixisenatide a	74 (1.1)	23 (4.6%)

^a The NIH All of Us data do not report exact numbers when fewer than 20 individuals are recorded.

The entire UK Biobank cohort had T2D, whereas 84.3% (N = 5526) of the All of Us cohort had T2D. This difference is primarily due to the earlier approval of GLP‐1RA for T2D management and limited availability of data on these drugs for weight loss, particularly in UK Biobank. Because semaglutide was first approved for T2D in 2017 and the UK Biobank records were limited to before this approval (2017), semaglutide was not available in this cohort. A total of 2012 individuals met our inclusion criteria and were prescribed semaglutide in the All of Us cohort. Since semaglutide was not available in UK Biobank, 30% of the participants in the NIH All of Us cohort with semaglutide prescriptions were randomly assigned to serve as an independent validation cohort (N = 603) and were therefore excluded from model training. Liraglutide was most commonly prescribed with 2064 individuals in the NIH All of Us cohort and 241 in the UK Biobank (Table 1).

3.2. Weight loss response to GLP‐1RAs

Weight loss was not consistent across GLP‐1RAs. Liraglutide resulted in mean weight loss of 2·47% (SD = 7·06) after 12–18 months in All of Us and 3·44% (SD = 6·09) in the UK Biobank (Table S1). Semaglutide resulted in 94% greater weight loss, with a mean of 4·87% (SD = 7·53) in the All of Us cohort. The 20th percentiles of weight loss (i.e., high response) were 7.57%, 6.38%, 7.77% and 10.9% weight on dulaglutide, exenatide, liraglutide and semaglutide, respectively (Table S1, Figure S1). UK Biobank participants in the top 20th percentile lost 8.25% on liraglutide (Table S1).

A meaningful proportion of the cohort did not lose weight on GLP‐1RAs. In the NIH All of Us cohort, 36.30% and 24.30% of patients did not lose any weight after 12–18 months on liraglutide and semaglutide, respectively. Comparatively, fewer individuals did not lose weight on liraglutide in UK Biobank (28·63%) (Table S1). Other clinical variables were also associated with weight loss. In All of Us, individuals were more likely to experience no weight loss if they were younger, had a lower baseline BMI and had shorter prescription regimens (OR <1, Ps <0.05) (Table S2). Individuals were more likely to experience the top 20th percentile of weight loss if they had their prescriptions for longer. Longer prescription durations were also negatively associated with no response in UK Biobank (OR = 0.45 (0.23–0.9), p < 0.05) (Table S3). As expected, no significant differences in clinical variables were observed for the semaglutide discovery and validation sets (Table S4).

3.3. NBEA Score in the discovery cohort

The NBEA score with the highest R ² improvement compared to the null model consisted of 2873 SNVs (Table S5). The majority of these SNVs were intronic (N = 2840, 98.8%). Of the nine exonic SNVs, only one was non‐synonymous. However, SIFT and PolyPhen identified this SNV as benign and tolerated (Table S6), indicating that additional work is needed to understand how these SNVs and surrounding SNVs may impact NBEA function.

The cumulative genetic variation in these SNVs was significantly associated with 12‐ to 18‐month weight change (p = 1·01 × 10⁻¹⁹) (Supplementary Results). The thresholds for the NBEA score are shown in Table S7, and threshold selection details are provided in Figure 1 and Figure S2. The value of incorporating the NBEA score with clinical variables was tested by comparing models using clinical variables only with models using both clinical variables and the NBEA score. The addition of the NBEA score significantly increased the variance explained for the top 20% weight loss (p = 2.35 × 10⁻¹²). Similarly, the addition of the NBEA score improved no response predictions (p = 1.58 × 10⁻⁷).

Notably, the NBEA score was associated with increased odds of being in the top 20% weight loss on any GLP‐1RA (OR = 1·69, 95% CI: 1·46–1·95, FDR p = 1·1 × 10⁻¹²) (Figure 2). When stratified by specific GLP‐1RAs, the NBEA score had an odds ratio of 1.82 (95% CI: 1·42–2·33, FDR p = 1·8 × 10⁻⁶) and 1.63 (95% CI: 1.21–2.20, FDR p = 0.001) for being in the top 20% of weight loss for liraglutide and semaglutide, respectively (Figure 2). Additionally, all GLP‐1RAs, except exenatide, were also significantly associated with being in the top 10% and top 15% of weight loss (FDR p < ·05) (Figure 2). Interestingly, significant associations between weight loss and the NBEA score were only observed for top 20% weight loss in the exenatide subgroup (FDR p < ·05). However, a post hoc power analysis for exenatide revealed that due to the limited number of patients on exenatide (N = 582), we were underpowered to detect the observed effect size (e.g., OR:1·39 for the non‐responsive group) and would have needed a sample size of approximately 1438 individuals, which is closer to the available sample sizes of patients on the other GLP‐1RAs.

FIGURE 2.

NBEA score associations with weight change percentiles in the All of Us cohort. (A) The association between response type and the NBEA score. Logistic regression models were used for estimation while adjusting for sex, genetic ancestry, prescription duration, age and BMI at the time of GLP‐1 prescription. (B) The percentages of the cohort (excluding those taking exenatide) categorised based on their NBEA score. (C) The observed proportion of individuals who were classified correctly as highly responsive (top 20% weight loss) compared to non‐responders (weight change ≥0%) according to the NBEA score.

Individuals meeting the non‐response threshold of the NBEA score were 39% more likely to not lose weight for when all GLP‐1RAs were combined (OR = 1·39; 95% CI: 1·22–1·58, FDR p = 1·0 × 10⁻⁶) (Figure 2). When stratified by specific GLP‐1RAs, the score was associated with increased odds of being non‐responsive for all GLP‐1RAs, except exenatide (Figure 2). The NBEA score displayed a strong association with non‐responsiveness for individuals on semaglutide (OR = 1·44; 95% CI: 1·07–1·94, FDR p = 1·0 x 10).

The NBEA score also displayed similar estimates for GLP‐1RA weight loss response after stratifying the analysis by sex, race and T2D status (Figures S3–S5). The NBEA score remained significantly associated with weight loss responses after adjustment with T2D medications (Figure S3). When analyses were limited to individuals who were prescribed one type of GLP‐1RA for the 12‐month period (N = 4559, 76·5%), those meeting the non‐response threshold were 35% more likely to not lose weight (OR = 1·35; 95% CI: 1·16–1.57, FDR p = 3·9 × 10⁻⁵) and individuals who met the response threshold were 76% more likely to experience high response (OR = 1·76; 95% CI: 1·49–2·08, FDR p = 1·2 × 10⁻¹¹) (Figure S6).

3.4. NBEA score in the validation cohorts

A total of 43 of the 2873 variants in the NBEA score were genotyped in the UK Biobank. For the remaining SNPs, 1640 (69.2%) were imputed with very high confidence (INFO≥·9) and only 140 SNPs (5.9%) had fair imputation confidence (INFO<5). The UK Biobank Affymetrix Axiom array was specifically designed to optimise imputation and is regarded as having high accuracy for imputation of common variants (21), and since our focus is multi‐variant score validation, variants with fair confidence were retained in the score computation in the UK Biobank.

The NBEA score was significantly associated with weight loss on liraglutide, successfully replicating the findings from the Discovery cohort (Figure 3). Validation included liraglutide from UK Biobank and the withheld set of patients prescribed semaglutide from All of Us. Exenatide was not included in the validation because it was not consistently predicted in the Discovery cohort; however, the results of exenatide can be seen in Figure S10.

FIGURE 3.

NBEA score associations with weight change percentiles in the UK Biobank cohort and Semaglutide validation set. (A) The association between weight loss response type and the NBEA score. Liraglutide validation data were collected from the UK Biobank (UKB) and semaglutide validation data consisted of randomly selected 30% of patients from the NIH All of Us Cohort (AoU) that were not used in the discovery set. Logistic regression models were used for estimation while adjusting for sex, genetic ancestry, prescription duration, age and BMI at the time of GLP‐1RA prescription. (B) The percentages of the cohort categorised based on their NBEA score. (C) The observed proportion of individuals who were classified correctly as highly responsive (top 20% weight loss) compared to non‐responders (weight change ≥0%) according to the NBEA score.

The NBEA score was significantly associated with the likelihood of being in the top 15% and 20% of weight loss for patients on liraglutide, with ORs = 2·51 (p = ·009) and 2·37 (p = ·008), but failed to reach significance for the top 10% of weight loss (OR = 1·28, p = ·310). The NBEA score was also significantly associated with non‐responsiveness for patients on liraglutide (OR = 1·81, 90% CI: 1·04–3·21, p = ·040).

For individuals taking semaglutide, in the withheld validation cohort, the NBEA score was associated with top 10% (OR = 2·45, p = ·0002), top 15% (OR = 2·33, p = 9·4 × 10⁻⁵) and top 20% (OR = 2·21, p = ·0001) weight loss, exhibiting a successful validation for the associations observed in the discovery cohort. However, the NBEA score failed to replicate the associations for the non‐responsive group with an OR = 1.07 and p = ·191.

When the cohort was limited to patients who were on the same prescription for the duration of the study, the NBEA score was significantly associated with top 15% and top 20% weight loss (p < ·05) (Figure S9). In those that did not switch medications, the NBEA score remained significantly associated with non‐response for liraglutide (Figure S9).

3.5. NBEA PheWAS in all of Us

We further characterised relationships of NBEA genetic variation across various cardiometabolic and neurologic phenotypes using the NIH All of Us cohort, which included 168 660 individuals with EHR and genotype data (Table S8). A total of 19, 17 and 22 SNVs were significantly associated with at least one phenotype in the whole cohort, female cohort and male cohort, respectively (Figure 4A, Figures S11 and S12, Table S9). Notably, 2 of 19 SNVs in the whole cohort were associated with decreased odds of early satiety (Figure 4B). These two SNVs were chr13:34979920:C>G (OR = 0·54; 95% CI: 0·40–0·72, FDR p = ·018) and chr13:35125567:G>T (OR = 0·54; 95% CI: 0·40–0·74, FDR p = 4·97 × 10⁻²). Endocrine/metabolic disorders, other than satiety, were associated with four and seven SNVs in the males and females, respectively.

FIGURE 4.

PheWAS of NBEA SNVs in the All of Us cohort. (A) Bar plot depicting number of significant phenotypes (FDR p < 0.05) associated with a single‐nucleotide variant (SNV) for each cohort stratified by phenotype category. (B) Manhattan plot depicting PheWAS results for 403 NBEA SNVs and their association with cardiovascular, endocrine/metabolism and neurological phenotypes in the full All of Us cohort (N = 168 660). Similar phenotypes are plotted closer. The table shows significant associations.

4. DISCUSSION

As of June 2024, approximately 12% of the US population (>40 million individuals) have used GLP‐1RAs. ²⁶ Obesity is a major health concern, associated with a wide range of comorbid conditions, including cardiovascular disease and diabetes. ⁴ , ⁵ , ⁶ As indications for GLP‐1RAs in treating obesity continue to grow, ⁹ healthcare payors are expected to increase coverage for these therapies. The long‐term benefits of obesity treatment, including reductions in associated comorbidities, may potentially offset the high costs of GLP‐1RA medications, but the financial burden on the payors remains a concern. Clinical trial data also indicate that approximately 30% of individuals do not achieve robust weight loss with GLP‐1RAs. ⁹ In this study, from real‐world settings in patients with and without T2D, 55% of individuals prescribed semaglutide failed to lose at least 5% body weight, and 24% did not experience any weight loss after 12–18 months. Predictive biomarkers are needed to identify individuals that are most likely to benefit from GLP‐1RA therapy to help optimise clinical decision making and avoid ineffective treatments for non‐responders. In addition to the benefit of weight loss, a lower incidence of microvascular complications (i.e., retinopathy, neuropathy and nephropathy) over the follow‐up period was observed in highly responsive groups compared to the non‐responsive group (Table S10). Here, we investigated a cohort of patients treated with GLP‐1RA for T2D or obesity to determine whether SNVs in NBEA could predict GLP‐1RA weight loss. Although the NBEA score demonstrates predictive ability, additional biomarkers and clinical predictors will be needed before predictions are substantially robust to guide GLP‐1RA weight loss treatment decisions.

The role of GLP‐1 in the regulation of glycaemia via mechanisms in pancreatic ß cells is well established. ¹⁵ However, GLP‐1 also facilitates weight loss through satiety mediation. ²⁷ Satiety is regulated through the central and peripheral nervous system. ²⁸ , ²⁹ One of the most well‐established mechanisms of hunger regulation occurs via modulation of anorexigenic proopiomelanocortin neurons (POMC) and orexigenic Neuropeptide Y/Agouti‐related peptide neurons (NPY/AgRP) in the arcuate nucleus of the hypothalamus (ARC). ²⁹ GLP‐1RAs, liraglutide and semaglutide have been shown to directly activate POMC neurons and inhibit NPY/AgRP neurons. ³⁰ , ³¹ While the exact mechanism of how GLP‐1 receptors (GLP‐1R) activate POMC neurons is unclear, GLP‐1R is known to stimulate the cAMP/PKA signalling pathway in other cells, ³² and this pathway plays a role in neuronal synaptic transmission. ³³ Notably, leptin has been shown to regulate leanness and obesity in mice through PKA in POMC and NPY/AgRP neurons, ³⁴ , ³⁵ supporting the hypothesis that GLP‐1R activation of POMC neurons may also work through the cAMP/PKA signalling pathway. The cAMP/PKA signalling pathway involves NBEA as an anchor protein driving subcellular localisation of PKA. ³⁴ , ³⁶ The data here suggest that additional studies are needed to better understand how NBEA impacts hypothalamic GLP‐1 signalling to regulate satiety. NBEA is expressed in the hypothalamus and linked to extreme obesity. ³⁷ Our PheWAS, which focused on cardiometabolic and neurologic outcomes, confirmed previously identified associations between NBEA and blood pressure, ³⁸ peripheral arterial disease and neurological diseases. ³⁹ Notably, early satiety also emerged as a novel association (Figure 4). These findings and previous literature suggest that mechanistic investigation in NBEA involvement in PKA signalling, satiety and GLP‐1RA weight loss is needed.

Targeting NBEA or its related pathways may enhance therapeutic efficacy or mitigate adverse effects. The A‐kinase anchoring protein (AKAP) domain binds the regulatory subunit of PKA, thereby confining its activity to specific subcellular regions. ⁴⁰ This interaction suggests that NBEA may be modulated through small molecules or peptides that influence its binding capabilities. Its large size (approximately 327.8 kDa) and complex structure may pose challenges for traditional small‐molecule drug design, and targeting specific interactions or domains within NBEA, such as the AKAP domain, could be a feasible strategy worth investigating. ⁴¹

We focused on NBEA due to multiple independent lines of evidence that suggest a role in satiety and feeding behaviour and an involvement in PKA signalling, which is a key mechanism for GLP‐1 signalling. This study used a robust design to develop and validate a novel genetic score for prediction of weight loss response to GLP‐1RA. This NBEA score was robust to the differences in large, geographically, genetically and culturally disparate health systems. However, as with any study, there are important limitations that should be considered in the interpretation of these findings. The findings here are based on real‐world cohorts, which increase the risk of uncontrolled confounding compared to clinical trial settings, and usage of GLP‐1RAs varied. Although adherence cannot be assessed directly from the EHR, we attempted to mitigate this by adjusting for GLP‐1RA prescription duration and included multiple GLP‐1RAs, with liraglutide being the most used GLP‐1RA in these cohorts, due to data availability. The approach of using prescription durations as evidence of ongoing GLP‐1RA is similar to approaches used previously when prescription fill data are unavailable. ⁴² , ⁴³ Notably, the NBEA score was not significantly associated with exenatide weight loss (Figure 2); however, as demonstrated with our post hoc power analysis, the small number of patients prescribed exenatide in this cohort limited the statistical power available to observe this relationship, and larger cohorts are needed to better determine whether the NBEA score is associated with weight loss on exenatide. The strongest associations between the NBEA score and GLP‐1RA weight loss were observed for semaglutide, which was not available in UK Biobank, but was evaluated using a withheld test set from the NIH All of Us cohort. Although its mechanism of action is the same as liraglutide, semaglutide has a modification allowing for a longer half‐life, which is the primary driver of its improved efficacy for weight loss. ⁴⁴ Furthermore, the cohorts consisted primarily of individuals who were prescribed these medications for T2D management instead of weight management. Individuals treated with GLP‐1RA for T2D receive a lower dose and subsequently experience less weight loss than those treated with higher doses for obesity. ⁴⁵ To address this potential source of confounding, we adjusted for T2D in our models and other T2D medications known to impact weight in the T2D‐stratified analysis. This stratified analysis showed similar results among individuals without a diagnosis of T2D, supporting the NBEA score associations in those being treated for obesity (Figure S3). It will be important to determine whether NBEA variation is predictive as the data for the newer, such as GIP/GLP‐1RA, become available. Additional human and mechanistic studies are needed to better understand how genetic variation in NBEA impacts NBEA expression, function and downstream signalling of pathways, such as PKA in response to GLP‐1RA. Better insight into which variants are deleterious for NBEA function is needed, and future work should focus on incorporating variants from other genes in the GLP‐1RA signalling pathway until sufficient sample sizes are available to perform a genome‐wide association study of GLP‐1RA weight loss. Although not available with the data repositories used for this analysis, Sanger sequencing or other methods are useful to validate the individual variants detected from genotyping arrays or next‐generation sequencing. This study highlights the role of PKA anchor proteins, such as NBEA, in GLP‐1 signalling for weight loss. However, genes in other signalling pathways need to be evaluated. For example, CNDP1, which modulates carnosinase metabolism and regulates the breakdown of carnosine, a histidine‐containing dipeptide, has been implicated in satiety, insulin sensitivity and overall weight management. ⁴⁶ , ⁴⁷ Although additional predictors will need to be identified and incorporated before we can sufficiently predict who will respond to GLP‐1RA weight loss treatments, a score derived from NBEA SNVs demonstrates potential to support advancements in precision obesity management.

CONFLICT OF INTEREST STATEMENT

D.M.R. has received research funding and consulting honoraria from Novo Nordisk, has an equity stake in Clarified Precision Medicine, Genovation Health, LLC, and has intellectual property related to treatment decision making in the context of type 2 diabetes and liver cancer. K.M.P. has received research support from Bayer AG, Merck & Co., Inc, Novo Nordisk Inc, and Twinhealth, consulting honoraria from AstraZeneca, Bayer AG, Boehringer Ingelheim, Corcept Therapeutics Inc, Diasome, Eli Lilly and Company, Merck & Co., Inc, Novo Nordisk Inc, and Sanofi, speaker honoraria from AstraZeneca, Corcept Therapeutics Inc, and Novo Nordisk Inc in the past 12 months. A.M.S has intellectual property related to type 2 diabetes treatment decision making. M.L.G has received research funding from Novo Nordisk in the past 12 months.

PEER REVIEW

The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer‐review/10.1111/dom.16612.

Supporting information

Data S1. Supporting information.

Mariam‐Smith A, Breeyear JH, Daniels NJ, et al. Neurobeachin ( NBEA ) is a novel gene associated with GLP‐1 receptor agonist associated weight loss. Diabetes Obes Metab. 2025;27(10):5632‐5642. doi: 10.1111/dom.16612

DATA AVAILABILITY STATEMENT

The data that support the findings of this study were retrieved from the NIH All of Us and UK Biobank. Restrictions apply to the availability of these data, which were used under license for this study. Data can be made available upon approval from these programs.