Predicting ADHD in alcohol dependence using polygenic risk scores for ADHD

Attention deficit hyperactivity disorder (ADHD) is a neurodevelopmental disorder with a high degree of comorbidity, including substance misuse. We aimed to assess whether ADHD polygenic risk scores (PRS) could predict ADHD diagnosis in alcohol dependence (AD). ADHD PRS were generated for 1223 AD subjects with ADHD diagnosis information and 1818 healthy controls. ADHD PRS distributions were compared to evaluate the differences between healthy controls and AD cases with and without ADHD. We found increased ADHD PRS means in the AD cohort with ADHD (mean 0.30, standard deviation (SD) 0.92; p = 3.9 × 10−6); and without ADHD (mean − 0.00, SD 1.00; p = 5.2 × 10−5) compared to the healthy control subjects (mean − 0.17, SD 0.99). The ADHD PRS means differed within the AD group with a higher ADHD PRS mean in those with ADHD, odds ratio (OR) 1.34, confidence interval (CI) 1.10 to 1.65; p = 0.002. This study showed a positive relationship between ADHD PRS and risk of ADHD in individuals with co‐occurring AD indicating that ADHD PRS may have utility in identifying individuals that are at a higher or lower risk of ADHD. Further larger studies need to be conducted to confirm the reliability of the results before ADHD PRS can be considered as a robust biomarker for diagnosis.

symptoms into adulthood (persistent ADHD), impacting educational achievement and social relationships (Faraone et al., 2015).Often, ADHD will co-occur with other mental disorders such as depression, autism spectrum disorder and substance use disorders e.g.alcohol dependency (AD) (Thapar, 2018;Wimberley et al., 2020).Those with high impulsivity ADHD are more likely to have an earlier onset of severe alcohol dependency (AD).(American Psychiatric Association, 2013).Furthermore, it has been suggested that persistent or partial remittent ADHD is associated with a more severe AD phenotype compared to those with full remittent ADHD (Huntley & Young, 2014).Genetic and environmental risk factors play a role in the development of ADHD and AD (Faraone et al., 2015).Twin studies have demonstrated that ADHD is a highly heritable disorder with an estimated childhood heritability of approximately 70% (Palladino et al., 2019;Thapar, 2018).AD has a slightly lower estimated heritability of 50% (American Psychiatric Association, 2013) but shares significant genetic risk factors, approximately 64% with ADHD (Capusan et al., 2015).One of the largest ADHD GWAS studies published in 2019 using approximately 20,000 individuals, identified 12 loci that met genome-wide level of significance for ADHD (Demontis et al., 2019).A subsequent larger ADHD GWAS study detected 27 genome-wide significant loci, six of which overlapped with the previous 2019 ADHD GWAS including overlap with other mental disorders such as major depressive disorder (genetic correlation = 0.31, standard error = 0.07) (Demontis et al., 2022).
Polygenic risk scores (PRS) can be used to predict the genetic liability of an individual to a particular trait or disease.PRS is generated from an individual's genotyped data and summary statistics obtained from GWAS study for the phenotype of interest.It has been suggested that in combination with other clinical factors, PRS could be used to aid diagnosis and implement early intervention for common diseases (Lewis & Vassos, 2020;Torkamani et al., 2018).Several studies have investigated whether ADHD PRS generated from the summary statistics from the 2019 ADHD GWAS study (Demontis et al., 2019) could predict ADHD diagnosis but also the risk of other behavioral or psychiatric disorders in independent samples (Du Rietz et al., 2018;Grigoroiu-Serbanescu et al., 2020;Wimberley et al., 2020).Du Reitz et al demonstrated that ADHD PRS positively predicted the risk of ADHD co-occurring traits and disorders including neuroticism, depression, anxiety, risk-taking, alcohol intake, smoking, AD, and increased body mass index.In AD GWAS studies the SNPs reaching genome-wide significance differ from those identified in ADHD GWAS studies however, the AD overall genetic load has shown a strong correlation with ADHD (r g = 0.3669) and other psychiatric disorders (Kranzler et al., 2019).This indicates that there are common risk variants for ADHD and AD as well as other psychiatric disorders described earlier (Demontis et al., 2022;Du Rietz et al., 2018).Another study used ADHD PRS to investigate the relationship with misuse of substances such as alcohol and marijuana.The study demonstrated a positive correlation between the ADHD PRS and the use of all substances regardless of addiction severity (Wimberley et al., 2020).Agnew-Blais and colleagues reviewed the predictability of ADHD longitudinally by assessing ADHD PRS from child to adulthood in a cohort of twins and whether the risk of co-occurring disorders changed over the course of the disorder.The study demonstrated that individuals with a higher ADHD PRS were more likely to meet ADHD diagnostic criteria and for certain symptoms including hyperactivity and impulsivity (Agnew-Blais et al., 2021), and another study evaluating the polygenic architecture of ADHD depending on the age of diagnosis demonstrated persistent ADHD to have a higher genetic correlation with AD than childhood ADHD (Rajagopal et al., 2022).A recent systematic review focused on investigating studies using ADHD PRS to predict ADHD and related traits found that the majority of studies they reviewed demonstrated a positive relationship between ADHD and ADHD PRS.In this study, the ADHD PRS explained 0.7%-3.3% of the variance in the ADHD phenotype (Ronald et al., 2021).
The DSM-5 diagnostic criteria for ADHD include a minimum number of behavioral symptoms and these need to occur independently of another disorder (American Psychiatric Association, 2013).
Assessments are usually made by at least two parties that complete rating scales to determine if similar behaviors are observed in two different settings, which are then considered by the diagnosing physician (National Institute for Health and Care Excellence).Assessments can be subjective and often the results may not be aligned, contributing to misdiagnosis and underdiagnosis in females (Furzer et al., 2022).
Moreover, in some countries that have a national health scheme, patients often experience delays in diagnosis which can be detrimental to the mental well-being of individuals and their families (NHS Digital, 2021;Stafford et al., 2020).The importance of earlier diagnosis has been highlighted with adults diagnosed later in life who agree that earlier diagnosis and support may have changed their outlook, some of which suffer from other mental disorders such as anxiety as a result of being undiagnosed and unsupported (Silny, 2015;Smith, 2020).This highlights the importance of streamlining the diagnosis process and to explore additional objective methods to ensure earlier diagnosis and intervention and to avoid misdiagnosis to improve the social and health outcomes of ADHD.
The overall aim of the study was to investigate if ADHD PRS was predictive of ADHD in the context of AD and had the following specific objectives a) to assess if there was a difference in ADHD PRS distribution and means in healthy controls and AD cases with and without ADHD; b) to determine if there was a difference in the ADHD PRS distribution and means between those with and without ADHD and if ADHD PRS was able to predict ADHD diagnosis, and c) to determine if there was a sex difference in the predictive value of the ADHD in the context of AD.Several studies have investigated the relationship between ADHD PRS and ADHD risk in the general population, autistic syndrome disorder, and other psychological disorders (Ronald et al., 2021;Wimberley et al., 2020), but our study is one of the first, to our knowledge to investigate the relationship of ADHD PRS in an alcohol dependent cohort.Demographics of the individuals are summarized in Table 1.

| Genotyping
416 AD individuals were genotyped as part of the current study.
Genomic DNA was extracted from saliva and blood samples and underwent standard quality control procedures.300 ng of DNA was hybridized to Illumina ® GSA v3 microarrays according to the manufacturer's protocol (Illumina, 2020).Data quality control was performed using PLINK1.9(Chang et al., 2015) according to a published protocol (Guo et al., 2014).Imputation was performed against the Haplotype Reference Consortium panel using the Sanger Imputation Service (McCarthy et al., 2016).
Existing data for an additional 1777 AD individuals that had been genotyped on the Illumina PsychArray was reported previously (Li et al., 2021).Genotyping data from healthy control individuals (n = 1818) were matched for array type and were also genotyped prior to the current study.Genotyping, quality control, and imputation steps for AD samples genotyped on the PsychArray are described elsewhere (Li et al., 2021).The processing of data for the healthy control samples genotyped on the PsychArray is described in Mullins et al. (2021).Genotyping of healthy control subjects' DNA samples on GSA v1 arrays was performed at the Broad Institute and QC was performed using the RICOPILI pipeline (Lam et al., 2020).A summary of the sample ascertainment procedure, sample size, and genotyping platforms utilized for the alcohol-dependent individuals and healthy controls are presented in Figure S2.Imputation was performed as described above.Only non-related individuals with genetic European ancestry were used in the PRS analyses.

| Generation of polygenic risk scores (PRS)
ADHD PRS were generated for the AD cohort and healthy controls separately using PRS-CS, (Ge et al., 2019).PRS-CS requires a linkage disequilibrium reference panel and ADHD GWAS summary statistics, which help infer the posterior effect sizes of SNPs.The European reference panel from the 1000 Genomes Project (phase 3) constrained to HapMap3 reference SNPs was downloaded from url: https:// github.com/getian107/PRScs.The ADHD summary statistics were from a Psychiatric Genomics Consortium meta-analysis including 38,691 ADHD cases (Demontis et al., 2022).A uniform SNP list was generated in R to create a list of SNPs that were overlapping across all datasets (alcohol-dependent and healthy controls).Only the SNPs present in the uniform SNP list were used for subsequent analysis.
Effect sizes generated per SNP for each chromosome were merged to create one set of files per dataset (GSA, PsychArray, and healthy controls).ADHD PRS for GSA and PsychArray datasets were merged to form the final AD dataset and used for downstream analysis.

| Statistical analysis
To include PRS as a continuous variable in regression modes, ADHD PRS scores for the AD cohort and healthy controls were normalized to Z-scores, which were used for downstream statistical analysis.
T A B L E 1 AD and healthy controls demographics summary included in the analysis.A pairwise t-test was carried out to determine if there was a significant difference in ADHD PRS means between the healthy controls, AD individuals with and those without ADHD.Logistic regression models were run for the AD group and healthy controls, adjusted for array type and principal components (for population substructure) to evaluate the relationship between ADHD PRS and ADHD diagnosis.Odds ratios (OR) were also computed from the logistic regression models.To investigate potential confounding due to sex differences, stratified multivariate regression analyses were performed for each sex in the AD group.
ADHD PRS generated for the AD group were categorized into four quantiles for all individuals and separately for each sex with the highest ADHD PRS in quantile 4 and the lowest in quantile 1.A chi-square test was performed to determine if there was a significant difference in ORs between the highest and lowest PRS quantiles.Power calculations were performed using AVENGEME to determine whether the ADHD PRS generated for the current study was sufficiently powered to detect an effect (Dudbridge, 2013;Dudbridge et al., 2018).The area under the curve (AUC) value was computed to determine the selectivity and specificity of the regression model.All statistical analyses were carried out in R (version 1.4.1717).

| Prediction of ADHD diagnosis in AD group
ADHD PRS was generated for 2193 AD individuals of which 970 had missing ADHD diagnosis data and were excluded from the analysis.A total of 1223 AD individuals (with ADHD diagnosis data) and 1818 healthy controls were included in the final analysis.The healthy controls had more females (67%) compared to the AD group (36%).There was no significant difference in sex balance in the AD group between those with and without ADHD.Interestingly, the mean age of AD onset was 6 years younger in those with ADHD than those without ADHD.A demographic summary of the individuals included in the analysis is summarized in Table 1.

| ADHD PRS distributions and means for the AD and healthy volunteer groups
A pairwise t-test showed a significant difference in ADHD PRS distribution between healthy controls and AD cases with (p = 3.9 Â 10 À6 ) and without ADHD ( p = 5.2 Â 10 À5 ) (Table 2, Figure S1).The healthy controls had a lower ADHD PRS mean and distribution compared to those in the AD group.Within the AD group, the mean ADHD PRS was higher in the ADHD group compared to the group without ADHD (p = 0.002) (Table 2, Figure S1).

| Association between ADHD PRS and ADHD diagnosis in AD cohort
There was evidence for a positive relationship between the ADHD PRS and ADHD diagnosis in the AD group (p = 0.002).The direction of effect for these findings was consistent in both individual datasets (PsychArray and GSA) (Table 3).Sex-specific analyses using the combined AD dataset showed evidence for association in males (p = 0.005) but not in females ( p = 0.363) (Table 4).While this study had 93% power to detect an association with ADHD diagnosis, we were underpowered to detect association in males and females (78% and 50%, respectively).
ORs and 95% confidence interval (CI) were computed from the logistic regression model for the AD group and separately for the individual datasets (PsychArray, GSA, and combined datasets) (Figure 1).
The directions of effect were consistent for all datasets.The OR for the AD group was 1.34 (95% CI 1.10, 1.65), that is, for every point increase in the Z-score (PRS normalized score), the relative odds for ADHD increases by 1.34.However, the AUC was calculated to be 0.59.
Sex-specific regression analyses suggested that ADHD PRS was associated with an increased risk of ADHD in males (OR = 1.42,CI 1.11-1.84)but not in females (OR = 1.18,CI 0.82-1.69).
ADHD PRS generated for the AD group were categorized into four quantiles to help with stratification, with the highest ADHD PRS assigned to the fourth quantile and the lowest ADHD PRS assigned to the first quantile as the reference quantile.ORs were computed for all quantiles and a chi-squared test was performed to test if there was a difference between the top and the reference quantile.The fourth quantile had a higher OR (2.19, 95% CI = 1.19, 4.17; p = 0.01) (Figure 2).
ADHD PRS for each sex were categorized into four quantiles and ORs were computed for each using the final AD dataset.A chisquared test was performed for each sex to compare the difference between the OR of the fourth quantile with the reference quantile (Figure 3).In males, the OR for the fourth quantile was higher than the reference quantile (OR = 2.74, 95% CI = 1.27, 6.47; p = 0.0094) (Figure 3).For the females, the direction of effect was the same, but this finding was not significant (OR = 1.47, 95% CI = 0.54, 4.27, p = 0.44) (Figure 3).

| DISCUSSION
To determine if ADHD PRS can predict the risk of ADHD and therefore potentially support the ADHD diagnosis process, ADHD PRS were generated for an AD cohort and healthy controls using the latest ADHD GWAS summary statistics.Regression analysis was conducted to investigate the association between ADHD PRS and ADHD.We report a significant difference in the ADHD PRS distribution between individuals who were not diagnosed with ADHD and those with the diagnosis in the AD cohort, with elevated ADHD PRS in individuals with ADHD compared to those without ADHD.
To further investigate the relationship between ADHD PRS and ADHD diagnosis, a regression analysis was conducted, and ORs were computed for individuals within the AD group.As expected, a strong relationship was observed between ADHD PRS and ADHD diagnosis.
An OR of 1.34 was generated for the AD group, that is, for every unit increase in ADHD PRS, the relative odds of having ADHD increases by 1.34.Similar ranges of ORs were reported in a recent review paper (OR = 1.22-1.76)(Ronald et al., 2021).To compare the risk of ADHD in relation to PRS in the AD group, ADHD PRS was categorized into four quantiles.The OR of the fourth quantile, containing the highest ADHD PRS was significantly higher compared to the lowest quantile, which contained the lowest ADHD PRS.
Statistical power of the ADHD PRS was calculated to determine if ADHD PRS generated in the current study was sufficiently powered to detect association.The ADHD PRS for the AD group was sufficiently powered (93%) but future work should focus on increasing the power between the AD male group and female group (78% and 50% power, respectively, in this study).Furthermore, the AUC was 0.59 which indicates that the ADHD PRS does not meet the criteria for a diagnostic test.

| Sex differences in PRS predictability for ADHD
A relationship between ADHD PRS and ADHD diagnosis was observed in males but not in females.When ADHD PRS were distributed into quantiles for both sexes, males had a significant difference in ORs between the first and the fourth quantile.However, we and others have previously reported similar ADHD by sex outcomes while investigating ADHD PRS in bipolar disorder and other traits and co- Note: Separate logistic regression analyses were performed for each sex from the AD dataset with ADHD diagnosis as the outcome variable.The models were adjusted for principal components and array type.et al., 2020).However, these differences may be due to reduced statistical power for analyses of data from females in the current sample.Indeed Martin et al reported SNP heritability (SNP-h2) differences on the observed scale in males and females of 0.247 (SE = 0.021) and 0.123 (SE = 0.025), respectively.Interestingly, these differences in SNP-h2 were attenuated when these estimates were calculated on a liability scale that allowed for a more balanced male-to-female sex ratio (Martin et al., 2018).
Females diagnosed with ADHD are often underrepresented in ADHD clinical trials (Faraone et al., 2015;Thapar, 2018) and this imbalance is also seen in clinical practice where females often encounter referral bias and are under-recognized for ADHD (Rucklidge, 2010).
ADHD PRS in combination with clinical risk factors could potentially address underdiagnosis of female ADHD.However, this needs to be further investigated in future studies.

| Healthy controls versus alcoholdependent group
ADHD PRS for the healthy controls were compared to the ADHD PRS generated for the AD cohort.A strong statistically significant difference was observed between ADHD PRS distribution of healthy controls and those with ADHD ( p = 3.9 Â 10 À6 ) and without ( p = 5.2 Â 10 À5 ).Furthermore, the ADHD PRS means were higher in the ADHD group compared to those without ADHD, and the healthy controls had a much lower mean compared to both groups.

| Study limitations
The sample size for the current study was a limitation.970 individuals were excluded from the analysis due to incomplete coding of ADHD diagnosis.However, even with the reduced sample size, a strong association was observed between ADHD PRS and ADHD diagnosis.The ADHD PRS was not sufficiently powered for either sex to detect an association, and this was particularly the case for the females.Therefore, the observed differences in the risk of ADHD conferred by the ADHD PRS in males compared to females require further investigation.
While the use of different arrays appeared to have an influence on the analysis (Figure 1), the directional effects were consistent between datasets.SNP imputation and adjustment of the logistic regression analysis for array type were used to account for potential confounders.Furthermore, the ADHD PRS were normalized to Z-scores so that the data set could be merged into a single dataset for the analyses.Additional analyses using a more limited SNP list that included SNPs present after QC in each of the datasets were consistent with the primary results.

| CONCLUSION
In conclusion, the data generated from this study suggests that there is a positive relationship between the genetic liability and risk of ADHD in individuals with co-morbid AD.This was also supported by the analysis comparing the ADHD PRS of healthy controls with those with and without ADHD.ADHD PRS generated for the current study was able to stratify individuals at a higher or lower risk of having ADHD with a high and low risk of ADHD in individuals diagnosed with AD that often co-occurs with ADHD.Even in a relatively small sample, meaningful ORs were observed in the AD group.sex differences were seen, and ADHD PRS was better at predicting ADHD diagnosis in males compared to females, but these analyses had limited power.
Overall our findings lend support to investigate further the potential of ADHD PRS as an adjunctive tool for ADHD diagnosis.Similar findings have been observed in several studies as reported by a recent systematic review where approximately 44 studies utilized ADHD PRS to assess different aspects of ADHD or other measures (Ronald et al., 2021).The authors reported that ADHD PRS was associated with ADHD, ADHD-related traits, and trajectories.They did however note that standardization of the research needs to be considered before it can be used in the clinic (Ronald et al., 2021).Furthermore, it would need to be demonstrated that the addition of the ADHD PRS would increase the diagnostic accuracy over and above the current diagnostic approaches.
Additional studies with larger sample sizes need to be conducted and transferability to other ancestries as well as in a cohort that represents the general population needs to be considered before implementation in the clinic.This study represents a cohort of alcohol-dependent individuals who may have genetic overlap with ADHD and therefore may have contributed to a stronger association observed.Research on sex differences in ADHD PRS predictability for the genetic risk for ADHD should be further investigated to understand if ADHD PRS is sex limited.
Further consideration needs to be taken to avoid introducing inequalities, for example, deprioritizing individuals who may have a low ADHD PRS but later go on to develop ADHD symptoms (false negatives) and vice versa (false positives).Finally, the challenge of implementing these findings from research to the clinic consistently and access to genotyping and expertise needs to be addressed.
AD individuals were recruited from a variety of UK community and hospital-based services to participate in the DNA polymorphisms in mental illness (DPIM) clinical study.AD and ADHD diagnosis was determined by participant interview by National Health System (NHS) clinicians and trained researchers using selected sections of the "semi-structured assessment for genetics of alcoholism (SSAGA-II)."The SSAGA-II renders DSM-IV diagnoses (Collaborative Studies on Genetics of Alcholism (COGA), n.d.; Schuckit et al., 1995).All participants were of English, Scottish, Welsh, or Irish descent with a maximum of one grandparent of non-British but Western European ancestry; none of these individuals were related.Ancestrally matched healthy controls (n = 1338) were recruited from London branches of the National Health Service (NHS) blood transfusion service, from family doctor clinics, and from among university students.Individuals were excluded if screening with the Schedule for Affective Disorders and Schizophrenia (SADS-L) (Endicott & Spitzer, 1978) revealed a lifetime history of neurosis, depression, bipolar disorder, schizophrenia, or alcohol use disorders.The remaining controls (n = 480) were NHS blood donors whose mental health had not been assessed.All procedures involving human subjects were approved by the NHS Metropolitan Multi-centre Research Ethics Committee (now the South Central-Hampshire A Research Ethics Committee) approval number MREC/03/11/090.All participants provided signed informed consent.

a
Significance was set at p < 0.05.U R E 1 ADHD PRS odds ratios for individual datasets for the AD group.Odds ratios (OR) computed for PsychArray (n = 807), GSA (n = 416), and the combined AD datasets (n = 1223).CI, confidence interval.
ORs plot of ADHD PRS quartiles for the final AD group.ORs and 95% CI were computed for the four ADHD PRS quartiles.First quartile (1) contained the lowest, acting as a reference quartile and the fourth quartile contained the highest ADHD PRS.**p ≤ 0.01 with significance set at p < 0.05.U R E 3 ADHD PRS quartiles for AD males and females.Quartile ORs with 95% CI were computed per sex for the final AD group.The ORs for quartile 4 were compared with those for quartile 1 with a chi-squared test to confirm the difference.**p ≤ 0.01.Significance was set at p < 0.05.CI, confidence interval.
Summary of ADHD PRS means, standard deviations, and p values for the healthy controls and AD group.
Abbreviations: AD, alcohol dependence; ADHD, Attention deficit hyperactivity disorder; SD, standard deviation.a Healthy controls as the reference group for the pairwise t-test.b p values when comparing between groups with significance set at p < 0.05.c Comparing AD group with and without ADHD.
Regression analysis for the association between ADHD PRS and ADHD diagnosis for individual datasets.Logistic regression analysis performed with ADHD diagnosis as the outcome variable was adjusted for principal components for PsychArray, GSA, and array type for the combined datasets.Combined dataset, PsychArray, and GSA pooled to make the final AD dataset.Regression analysis to analyze sex differences in the association between ADHD PRS and ADHD diagnosis in the AD group.
occurring disorders (Du Rietz et al., 2018; Grigoroiu-Serbanescu T A B L E 3 a Significance was set at p < 0.05.T A B L E 4