Differential Influence of Maternal Smoking on Infant Birth Weight: Title and subTitle BreakGene-Environment Interaction and Targeted Intervention

Maternal smoking is a significant risk factor for low-birth-weight (LBW) infants, which, in turn, influences infant mortality and the long-term health outcome of surviving infants. Maternal smoking is an ideal target for intervention and the optimal public health outcome would be prevention of all maternal smoking. But given variability in the response to intervention attempts, a secondary strategy is to target more effectively those individuals at highest risk for adverse outcomes.

In this issue of THE JOURNAL, Wang and colleagues¹ provide evidence that the magnitude of the effect of smoking by pregnant women on birth weight depends on their genotype at 2 genes involved in the metabolism of smoking toxins (CYP1A1 and GSTT1). Those with CYP1A1 Aa and aa (heterozygous and homozygous variant types) and GSTT1 absent genotypes had greater reductions in birth weight than CYP1A1 AA (homozygous wild type) or GSTT1 present genotypes (520 g vs 252 g for CYP1A1; 642 g vs 285 g for GSTT1).

Clinicians are well aware that patients are unique individuals with idiosyncratic symptom patterns, clinical courses, prognoses, and responses to treatment. Rather than considering such individual variability as arising from random processes, the "error term" in statistical inference, quantitative genetics provides a framework that views at least some variability as resulting from systematic differences among patients themselves.² The search for evidence of gene-environment interactions arises from this perspective.

Advances in molecular genetics provide the opportunity for more accurate and extensive genotypic assessment than has been previously available. This, coupled with advances in epidemiologic study design³ and statistical methods,⁴ opens up new opportunities to search for gene-environment interactions. Recent studies have reported gene-environment interactions for significant health-related traits as diverse as cancer,⁵ osteoporosis,⁶ obesity,⁷ cardiovascular disease,⁸ Alzheimer disease,⁹ alcohol use,¹⁰ and depression and anxiety.¹¹ Extension of this approach to smoking and pregnancy outcomes is timely and practical.

Familial aggregation has been demonstrated for both smoking¹² and LBW infants,¹³ yet the etiological mechanisms that result in expression of these traits are complex. Quantitative genetic methods for characterizing the nature of genetic and environmental influences, particularly gene-environment interactions, are weak in the absence of good measures of the genotype, environment, or both. The study by Wang et al¹ was designed to study LBW infants in a broader context than a detailed study of the effects of maternal smoking. Indeed, the measure of smoking is simply self-reported smoking (yes or no) at 4 periods throughout the pregnancy. This kind of information about smoking is limited in contrast to measures that provide information about heaviness of smoking (cigarettes/day)¹⁴ or biochemical assessment of smoking using a measure of cotinine.¹⁵ Self-report by pregnant women about their smoking behavior can be inaccurate but the effect of this error may¹⁶ or may not¹⁷ affect inferences about the relationship between smoking and LBW infants. Nevertheless, a dose-response curve between cotinine-assessed exposure and LBW infants has been demonstrated previously in regular smokers^{13 ,18} but possibly not in occasional smokers.¹⁹

Measurement is equally important for assessment of genotypes. The selection of polymorphisms of the CYP1A1 and GSTT1 genes for evaluation by Wang et al¹ is well justified based on the role of these genes in metabolism of toxic metabolites of cigarette smoke.^{20 - 21} Yet, the processes involved in smoking and birth weight involve loci from numerous other systems that affect these traits, such as nicotine metabolism and neurotransmitter systems involved in addictive behavior, other loci involved in detoxification mechanisms, and loci that contribute to variability in fetal growth and development.

Evaluation of a small number of candidate loci is appropiate for studying a targeted mechanism but may or may not be sufficient to describe fully the public health impact of the effects of these loci in the context of a larger, complex causal system. At the level of specific loci, continued advances in molecular genetic technology will permit increasingly refined genotypic assessment so that, for example, heterozygotes and both homozygotes are distinguishable (as with CYP1A1) in contrast to genotypes for which the heterozygote is indistinguishable from one of the homozygotes (as with the GSTT1 polymorphism). Allelic variants with more subtle effects on gene expression, such as quantitative variability in enzymatic activity, will provide increasingly detailed information about genetic influences.

These limitations of measurement do not diminish the significance of the findings of Wang et al¹ but rather should motivate more refined investigation of the interaction in future work. Refined measurement of smoking using cotinine assessment will permit more accurate assessment of smoking as well as investigation of the influence of these loci on the dose-response relationship between smoking and LBW infants. Dietary assessment can be informative about whether this effect is related to systematic differences in dietary habits, such as an inverse relationship between smoking (heavy, light, and never smokers) and consumption of fruit and vegetables.²² More detailed assessment of patterns of alcohol consumption would indicate more effectively how LBW infants are affected by comorbidity between alcohol and tobacco use²² than is possible in the data available to Wang et al.¹ Studies including these more detailed assessments should provide a clearer sense of the mechanism underlying the relationships between maternal smoking, genotype, and LBW infants identified by Wang et al.¹

Despite the promise of molecular genetics for identifying effects of specific loci on complex traits, the literature is full of reports of genetic linkage or association that do not hold up under scientific scrutiny. This phenomenon, which is true for main effects of putative loci, is likely to be exacerbated for reports of gene-environment interactions. Replication of findings remains a critical step to confirming the presence of such effects.

It is increasingly apparent that gene-environment interaction is a significant contribution to individual variability in risk for a variety of health-related outcomes. Effective detection of such effects depends on careful study design, attention to measurement issues in both genetic and environmental assessment, caution about potential bias from confounding factors that are not included in a study, and sensitivity to the need for replication. Once these are satisfied, reliable identification of gene-environment interaction has the potential to alter fundamentally the way in which public health intervention programs are designed. Global interventions can affect the overall population in the desired direction, but individual responses could vary from high to low and even to negative. Understanding of gene-environment interactions provides the opportunity to develop individual-based interventions that use variability in response to improve public health outcomes.

This approach is likely to be most immediately applicable to situations for which identification of positive and potentially negative responders to a therapeutic intervention is important. For example, discovery of a gene-environment interaction that is useful for identifying responders to a particular treatment for hypertension would have substantial clinical utility. While it may not be practical at present to genotype all smoking pregnant women, the CYP1A1 and GSTT1 loci might be considered in the future for inclusion in risk screening strategies as genetic risk assessment becomes more routine and cost-effective.