The abuse of different psychoactive substances frequently co-occurs—if not as diagnosable disorders, then at the symptom level. The symptoms of abuse and dependence for different substances may develop sequentially alongside changes in an individual’s drug use repertoire or may coexist, reflecting concurrent use. Comorbidity for any two disorders has been mathematically interpreted in several association models that are grounded in the concept of multifactorial liability, a latent (unobserved) trait resulting from all factors influencing the probability of the disorder (1, 2). The models differ in the directions of causality and in the thresholds of liability at which one disorder (or liability) may influence risk for another. The relationships forming these models can be drawn from several sources that are not necessarily mutually exclusive: (a) both disorders are alternative manifestations of, or located on, the same liability continuum; (b) a third disorder, or another source of shared liability, increases the risk for or causes both disorders; and (c) one disorder increases the risk for or causes another, including reciprocal effects. Tests of these models to account for comorbidity between alcohol and drug dependence suggest common liability as the best explanation (3).

This explanation is not inconsistent with genetic data. Twin research suggests that although liabilities to addictions form two distinct groups that correspond to licit and illicit substances, each with a single additive genetic component of variance (representing variation in many genes) that accounts for most of the covariation between symptoms of abuse or dependence across respective substances, these two genetic factors are highly correlated (4). The distinctiveness of these factors is obviously not related to anything specific to the substances per se, different as they are within the groups. Rather, it is a reflection of social norms dividing the use of psychoactive substances into two categories: use of legal substances and use of illegal substances—which involves substantial norm violation. This clearly environmental factor induces different patterns of genetic relationships between and within the two substance groups. From the standpoint of this general (non-drug-specific) liability to addiction model, the pattern of drug use initiation—frequently from licit to illicit, or from “softer” to “harder” substances (but in many cases the reverse)—is variable and opportunistic, rather than uniform and developmentally deterministic (5), as posited by the “gateway hypothesis” (6). Despite that, and even though the gateway hypothesis addresses only sequential drug use initiation and does not extend to addiction disorders, the gateway hypothesis has been the foundation of many drug-related policy and research decisions. Obviously, the gateway hypothesis, which is focused on drugs per se, particularly the inconsistently defined “gateway drug”—alcohol and tobacco, or marijuana—calls for controlling the supply of that substance and the substance use behavior as such. In contrast, the general liability to addiction theory focuses on social behavior and behavior regulation precursors and concomitants of drug abuse, which is one possible developmental manifestation of socialization failure.

A study by Vrieze et al. (7) in this issue of the Journal contributes new information pertaining to comorbidity of addictions by exploring developmental changes—from age 14 to age 29—in the relationships among the symptoms of abuse of and dependence on alcohol, tobacco, and marijuana. The study is conducted in a population-based twin sample, enabling tests of biometric (behavior genetic) models. The results show that the relationships among symptoms across substances decrease over the years, as does the genetic commonality among them. This is consistent with the general liability theory, inasmuch as earlier experimentation, use, and symptoms are reflective of the general norm violation associated with any substance use before legal drinking or smoking age, whereas at a later age alcohol and tobacco become legal while marijuana remains illegal. Indeed, it seems that the contribution of the common factor in the symptom variance drops by adulthood (age 20 in this study) but remains relatively stable thereafter.

Vrieze et al. note that the decrease over time in associations among symptoms for different substances contradicts the gateway hypothesis because “the initial drug use would cause them to use other drugs” over time. A counterargument may be that it is difficult to interpret these data unequivocally as supporting or contradicting the gateway hypothesis because the study does not analyze information on “hard” drugs. It is transition to the latter from marijuana—the gateway drug in the current applications of the gateway hypothesis—that is considered the critical stage in the gateway progression. According to the original formulation of the gateway hypothesis (6), however, the first stage is alcohol or tobacco use, which is followed by transition to marijuana. Thus, as the use of these licit substances grows over time, its correlations with marijuana use should grow as more alcohol users transition to marijuana. This pattern is different from those observed in the study, however—contrary to the gateway hypothesis.

It should also be emphasized that the Vrieze et al. study operated with symptoms of the disorder rather than with substance use initiation, the subject matter of the gateway hypothesis. Although likely correlated, the sequences of substance use initiation and development of symptoms are not necessarily exactly aligned. It is the symptom development, however, that is of medical importance. Accordingly, it is the fact that the study’s data are consistent with the general liability theory, this time with its developmental aspect, that is an important conclusion of the article. This conclusion is consistent with other reports indicating incongruity of the gateway hypothesis with empirical data (5).

The correlations across substances seen in the Vrieze et al. (7) results are similar to those in the recent study by Palmer et al. (8) (whose participants were younger than the last assessment age in the Vrieze et al. study), but somewhat lower than those estimated in research by Kendler et al. (4) (whose sample was older on average). If the age trend reported by Vrieze et al. was maintained, the Kendler et al. sample should have had even lower correlations. The highest correlations in the Kendler et al. study were observed between symptoms across illicit substances, cannabis and cocaine, of which only the former was examined in the studies by Vrieze et al. and Palmer et al. It is possible that longitudinal trends for commonalities among liabilities for abusing licit and illicit drugs are different. That, however, cannot explain the discrepancy between the observations of a decline in heritability for nicotine dependence symptoms in the study by Vrieze et al. and a growth in this measure in a previous report from the same group (9). Similarly, heritability growth with age has also been observed for externalizing behaviors (10) that are strongly associated with general liability to addiction. Additional research seems necessary to resolve these discrepancies.

It should also be taken into account that comorbidity may not be optimally addressed when using symptoms that do not reach the diagnostic threshold. On the other hand, although these studies examined symptoms rather than diagnoses, substantial proportions of the samples did qualify for a diagnosis, and dichotomizing the phenotypes by using diagnoses would have reduced analytic power. The dimensional symptom count is also more consistent than the categorical diagnosis with the concept of continuous liability. Developing quantitative indices that would cover the entire range of liability phenotypes in the population, including nonsymptomatic ones (11), would be a logical extension of findings that identify general liability to addiction. Drug-specific risks notwithstanding, addressing general liability to addiction in both research and intervention could be an efficient approach to the addiction problem.