In this report, we sought to evaluate gene–environment correlation and interaction processes for alcohol use behaviors in the context of romantic relationships, and the associations between partner discordance in substance use with emotional well-being and relationship quality. In contrast to our hypothesis, we did not find evidence that polygenic predispositions for alcohol consumption were associated at a statistically significant level with perceived partner substance use behaviors. These null effects were unexpected in view of prior evidence from a Swedish national sample that individuals with a parental history of AUD, who we can infer have a higher genetic (and environmental) loading for the disorder, are more likely to marry AUD-affected spouses (Salvatore et al., 2019), as well as prior evidence that polygenic scores for alcohol consumption were associated with a spouse’s drinking in non-clinically ascertained US and European samples (Clarke et al., 2021; Otten & Mandemakers, 2023). Although we cannot fully explain why our findings diverge from these earlier reports, sample size may be a contributing factor. A post hoc power analysis indicated sufficient (80%) power to detect effects (standardized betas) of 0.062, but the largest effect we observed was b < 0.01. Itis also possible that within the clinically ascertained COGA sample, the associations between participants’ polygenic predispositions and their partners’ substance use were diminished by broader background familial factors that often co-occur with AUD (e.g. lower socioeconomic status or family disruption; Grant et al., 2015) and are also associated with assortment and partner selection (Erola, Härkönen, & Dronkers, 2012; Salvatore et al., 2019; Schuckit, Tipp, & Kelner, 1994). In addition, by focusing on questions about partners’ concurrent alcohol use behaviors, we may not have captured the full range of a partner’s lifetime patterns of alcohol and substance use behaviors. For example, we would expect polygenic associations to be attenuated if a participant with a higher polygenic score reported on concurrent alcohol use behaviors in a partner who was in abstinent remission from AUD.

We found mixed evidence for our hypothesis that perceived partner substance use moderates the associations between polygenic predispositions and alcohol outcomes. Perceived partner alcohol use and heavy drinking moderated the influence of participants’ polygenic risk scores on frequency of alcohol use and drunkenness, respectively. These findings were consistent with our hypothesis that perceived frequent alcohol use or heavy drinking by the partner would strengthen the association between a participant’s polygenic score and their alcohol use or drunkenness. In contrast, perceived partner drunkenness moderated participants’ polygenic risk scores to predict heavy drinking in the opposite direction than anticipated. We observed that individuals with higher genetic predispositions for alcohol consumption engaged in less frequent heavy drinking when they reported that their partner got drunk more frequently. Although this effect runs counter to our original hypothesis, we recently observed a similar pattern of partner GxE effects in a population-based Finnish twin sample (Stephenson et al., 2024). In that study, latent genetic influences on binge drinking were attenuated as partner alcohol use increased.

Interestingly, perceived partner cannabis use was the most pervasive moderator of genetic predispositions on all alcohol outcomes. The associations between the alcohol consumption polygenic score and participant alcohol use, drunkenness, and heavy drinking were weaker when a partner reportedly used cannabis more frequently. This may be the result of substance substitution (Subbaraman, 2016), whereby participants with partners who use cannabis frequently choose to use cannabis instead of alcohol despite a genetic predisposition to alcohol use. As a set, these results underscore the salience of partner characteristics in modifying genetic influences on a variety of drinking patterns, and are consistent with the broader literature that romantic relationships are important social contexts that shape individual health behaviors (Birditt et al., 2022; Holt-Lunstad, Robles, & Sbarra, 2017). Our findings add a genetic perspective to this understanding, demonstrating that genetic factors play only a limited role in alcohol use behaviors when one’s romantic partner is drunk or uses cannabis more frequently. These results, which suggest that individual diatheses such as genetic predispositions vary in salience depending on the couple context, highlight the importance of considering partner alcohol and cannabis use as part of preventive intervention efforts for alcohol use (Epstein & McCrady, 1998; McCrady & Flanagan, 2021).

Our exploratory analyses also revealed that sex and parenthood each modified a two-way gene-by-partner substance use interaction effect. Visualization of these three-way interactions suggested that perceived partner substance use as a moderator of polygenic predispositions tended to be less precisely known in one group compared to the other due to data sparseness. Owing to this, we caution against interpreting these results as evidence that the effects differed between groups (e.g. early midlife versus later life, parents versus non-parents, etc.).

We also found evidence consistent with our hypotheses that partners tended to resemble one another in their substance use, and that having a partner whose substance use is different from one’s own is associated with lower emotional well-being and lower relationship quality. Similar to reports from clinically ascertained and community-based samples (Agrawal et al., 2006; Ask, Rognmo, Torvik, Røysamb, & Tambs, 2012; Low, Cui, & Merikangas, 2007; Treur, Vink, Boomsma, & Middeldorp, 2015), COGA participants and their romantic partners were moderately correlated for alcohol use frequency and had a relatively strong concordance for nicotine use and illegal drug use. Interestingly, partners’ resemblance for frequency of drunkenness was relatively weak, indicating that while participants perceived that they and their partners drink at similar frequencies, they perceived less congruence in their drunkenness. We also found evidence that sex and age group moderated partner resemblance. Compared to male participants, female participants reported that they resembled their partners more in their frequency of alcohol use, heavy drinking, and cannabis use. In opposite-sex relationships, female partners are often considered the ‘barometer’ of the relationship (Floyd & Markman, 1983), though more recent work has challenged this idea (M. D. Johnson et al., 2022), and partners who have concordant substance use patterns tend to be more satisfied and stable (Homish, Leonard, & Cornelius, 2008; Kulak, Heavey, Marsack, & Leonard, 2025; Leonard et al., 2014). In view of this, we speculate that females may be especially motivated to see their partner’s substance use as similar to their own. Our finding that midlife individuals reported greater resemblance between their own and their partner’s cannabis use compared to later-life individuals may reflect the more permissive norms and attitudes that midlife individuals have towards cannabis (Campbell, Twenge, & Carter, 2017).

Consistent with previous studies (Crane, Testa, Schlauch, & Leonard, 2016; Cranford, Floyd, Schulenberg, & Zucker, 2011; Homish & Leonard, 2007), we also found evidence that partner discordance in alcohol and nicotine use was linked to lower relationship quality. Discordance was also associated with several indicators of participants’ individual functioning. Alcohol use and drunkenness discordance were associated with lower self-reported well-being and lower scores on the social relationship scales, which are inclusive of social support, companionship, and social distress. Thus, mismatched substance use patterns appear to be associated with both poorer relationship outcomes and more broadly with poorer emotional well-being. This maps onto a growing literature that demonstrates that couples’ drinking patterns are associated with key individual health outcomes, including blood pressure and mortality (Birditt et al., 2022, 2024), and demonstrates an association with an individual’s emotional health.

The results from this study should be considered in the context of its limitations. First, COGA is a sample enriched for AUD risk, and accordingly, the pattern of effects observed here may not be generalizable to low-risk populations. Second, our measures of partner substance use came from COGA participants. We recognize that such reports may be biased (Kendler et al., 1991). Encouragingly, others have reported moderate associations between partner perceptions of problem drinking and self-reported drinking problems (Rodriguez, DiBello, & Neighbors, 2013). Nonetheless, we recognize that both perceptions of a partner’s substance use and the partner’s actual substance use can have independent and interactive associations with relationship outcomes (Rodriguez et al., 2013). Third, this report is based on cross-sectional data, and for this reason, we cannot make causal claims about the observed associations. Fourth, the definitions for heavy drinking when participants reported on their female partner were four drinks in 24 hours (compared to five drinks for all other participants and partners), which may capture a less intense version of heavy drinking. Fifth, even in our sample enriched for risk, there was very little variation between partners in illegal drug use, which precluded testing hypotheses using these measures. Finally, we recognize that there are different ways to calculate and analyze PGS. Our approach of residualizing PCs within genetic similarity groups and pooling has been used in previous studies (Barr et al., 2022; Choi et al., 2025) and allowed for increased power to detect significant effects, particularly in moderation models.

In summary, the results from this study in a sample enriched for familial risk for AUD offer novel insights into the interplay between genetic predispositions and partner characteristics on alcohol use behaviors. An especially novel aspect of our findings was that perceived partner cannabis use and drunkenness attenuated the associations between genetic predispositions and heavy drinking. These gene-by-environment interaction effects were not confounded by gene–environment correlation, as individuals’ genetic predispositions were not associated with their reports of their partners’ alcohol, cannabis, and nicotine use at a statistically significant level. We also expanded upon earlier insights that partner discordance in substance use is associated with lower relationship quality and found that partner discordance in alcohol use and drunkenness were associated with lower individual well-being and indicators of broader social relationship functioning. Partner discordance in nicotine use was likewise associated with lower social relationship functioning. Taken together, these findings emphasize the importance of considering dyadic partner effects in the etiology of alcohol use behaviors and emotional health.