Can we apply the Mendelian randomization methodology without considering epigenetic effects?
© Ogbuanu et al; licensee BioMed Central Ltd. 2009
Received: 21 February 2008
Accepted: 11 May 2009
Published: 11 May 2009
Instrumental variable (IV) methods have been used in econometrics for several decades now, but have only recently been introduced into the epidemiologic research frameworks. Similarly, Mendelian randomization studies, which use the IV methodology for analysis and inference in epidemiology, were introduced into the epidemiologist's toolbox only in the last decade.
Mendelian randomization studies using instrumental variables (IVs) have the potential to avoid some of the limitations of observational epidemiology (confounding, reverse causality, regression dilution bias) for making causal inferences. Certain limitations of randomized controlled trials, such as problems with generalizability, feasibility and ethics for some exposures, and high costs, also make the use of Mendelian randomization in observational studies attractive. Unlike conventional randomized controlled trials (RCTs), Mendelian randomization studies can be conducted in a representative sample without imposing any exclusion criteria or requiring volunteers to be amenable to random treatment allocation.
Within the last decade, epigenetics has gained recognition as an independent field of study, and appears to be the new direction for future research into the genetics of complex diseases. Although previous articles have addressed some of the limitations of Mendelian randomization (such as the lack of suitable genetic variants, unreliable associations, population stratification, linkage disequilibrium (LD), pleiotropy, developmental canalization, the need for large sample sizes and some potential problems with binary outcomes), none has directly characterized the impact of epigenetics on Mendelian randomization. The possibility of epigenetic effects (non-Mendelian, heritable changes in gene expression not accompanied by alterations in DNA sequence) could alter the core instrumental variable assumptions of Mendelian randomization.
This paper applies conceptual considerations, algebraic derivations and data simulations to question the appropriateness of Mendelian randomization methods when epigenetic modifications are present.
Given an inheritance of gene expression from parents, Mendelian randomization studies not only need to assume a random distribution of alleles in the offspring, but also a random distribution of epigenetic changes (e.g. gene expression) at conception, in order for the core assumptions of the Mendelian randomization methodology to remain valid. As an increasing number of epidemiologists employ Mendelian randomization methods in their research, caution is therefore needed in drawing conclusions from these studies if these assumptions are not met.
The use of genotypes that affect modifiable risk factors to make causal inferences falls under the umbrella of Mendelian Randomization (MR) studies [1, 2]. Instrumental variable (IV) methods – the statistical methods that underlie such inferences – have been widely used in econometrics, but not in epidemiology [1, 3]. Mendelian randomization refers to the random assortment of alleles inherited by offspring from their parents at conception [4, 5]. This random assortment of inherited alleles has been likened to a randomized clinical trial (RCT), in which the research subjects are randomly allocated to different genotypes rather than to medical interventions . Mendelian randomization studies include any study that uses genetic variation as a robust proxy for a potential disease risk (which cannot be assessed without biases) for the purpose of making causal inferences about the outcomes of the modifiable exposure .
To date, the potential impact of epigenetics on the core assumptions that underlie the use of genes as instrumental variables has not been addressed. This paper opens up this inquiry by assessing the appropriateness of the use of Mendelian randomization as an instrumental variable in the presence of epigenetic modifications of gene expression, and cautions investigators to, at the least, recognize the existence of these limitations. We will delineate the major rationale and the core assumptions of the Mendelian randomization methodology, explore the current understanding of epigenetics, and discuss the methodological challenges that arise from the use of genotypes as instrumental variables for modifiable exposures when epigenetic modifications of gene expression are present. The goal of this paper is to emphasize that effect sizes will be biased when the presence of epigenetic phenomena violate the implicit fundamental assumptions in Mendelian randomization studies (and are not compensated for in the analytic models). We will illustrate the occurrence of the epigenetic bias both algebraically and with a data simulation.
What is currently known
Mendelian randomization and its shortfalls
Mendelian randomization studies exploit the idea that the genotype only affects the disease status indirectly and is assigned randomly at meiosis, independent of measured and unmeasured (or measured-with-error) confounders [1, 5]. These properties define an instrumental variable (IV), which is a variable associated with the outcome only through its robust association with an intermediary variable – the exposure of interest . If the levels but not the function of a potential disease risk factor is determined by a genetic polymorphism, then the levels of the risk factor will effectively be assumed to have been randomly assigned at conception. This 'randomization' will potentially obviate the effect of confounding in the test of genotype-disease associations .
Core Assumptions Implied in Mendelian Randomization Studies
In order to understand the bias introduced by epigenetic modifications of gene expression, the underlying statistical assumptions in MR studies are outlined next. For this and subsequent sections, let the causal effect of X (intermediate phenotype) on Y be the relationship of primary interest, and let G (genotype) be the variable we want to use as the instrumental variable (IV). Also, let U be an unobservable variable that confounds the effect of X on Y.
G is independent of the confounding factors U that confound the association of X and the outcome Y. This assumption accrues from the random allocation of alleles at conception. Any event that alters this allocation could lead to nullification of this assumption (Figure 2A).
The instrumental variable G is associated with the exposure of interest X (i.e. must not be (marginally) independent of X). The stronger this association, the "better" an instrument G is, providing more information on the causal association between X and Y (small standard errors and narrow confidence intervals) . It is important to state at this point that G does not need to be causal for X, i.e. to be useful as an IV, G does not have to be the "right" gene. The association could instead be due to a mediator variable or through another unobserved variable that affects both G and X . This may occur in situations where there is linkage disequilibrium (LD) between G (G1) and another genotype (G2) (Figure 2B).
Conditional on X and U, the instrument and the response are independent.
There are several methodological advantages of MR studies. Unlike conventional RCTs, MR studies can be conducted in a representative sample without imposing any exclusion criteria or requiring volunteers to be amenable to random treatment allocation . Secondly, "randomization" in MR studies occurs at conception while RCT studies randomize patients in adulthood. This minimizes biases due to canalization and developmental adaptation [2, 4]. Mendelian randomization studies may also counteract some key shortcomings of randomized controlled trials (RCTs), such as high cost, unfeasibility with some exposures, and problems with generalizeability [1, 4].
Previous studies have addressed some of the key potential shortfalls of MR studies [2, 4, 10–14]. These include the paucity of suitable genetic variants, unreliable (genetic) associations, genetic confounding by population stratification, linkage disequilibrium (LD), pleiotropy, functional genomic confounding due to developmental canalization and gene imprinting, the need for large sample sizes and some potential problems with binary outcomes . Other factors that may adversely impact MR studies include selective survival, gene-covariate (environment) interactions and gene-gene interactions. In addition to these potential shortfalls is the potentially distorting effect of epigenetics. This "epigenetic bias" is the focus of this paper.
What is Epigenetics?
Classical Mendelian inheritance of traits from parents to offspring follows the DNA pairing and transmission patterns as illustrated by the Watson-Crick model of DNA. However, there is an emerging consensus among experts in genetic epidemiology that DNA is not the sole unit of heredity and that genotype and environment are not the only determinants of phenotype . A growing body of evidence suggests that the impact of environmental influences may extend beyond the DNA sequence . The emerging field of epigenetics studies heritable changes in gene expression that occur without directly altering the DNA sequence . Thus, epigenetics is dedicated to the study of non-Mendelian (meiotically and mitotically), heritable changes in gene expression not accompanied by a change in genotype/DNA sequences [14, 18–20]. These changes generally involve DNA modification (without change in nucleotide sequence), histone protein modifications, and regulation of gene expression by microRNAs. MicroRNAs (miRNAs) are small (approximately 22 nucleotides long) RNA molecules that may be involved in post-transcriptional control of gene expression . These regulatory mechanisms specify which regions of the genome are active in any given cell at any one time [16, 20].
The cells of eukaryotic organisms contain an additional level of information superimposed on the DNA double helix, a nucleoprotein entity known as "chromatin". This "DNA packaging" (or "epigenome") has recently been implicated in the regulation of the complex interactions of the enzymatic processes of replication, transcription, recombination, and DNA repair . The molecular mechanisms of epigenetic modifications include DNA methylation and chromatin structure and histone modifications . Acetylation, methylation, phosphorylation, and ubiquitylation are implicated in activation; while methylation, ubiquitylation, sumoylation, deimination and proline isomerization are involved in gene repression . However, depending on the location, any given modification is capable of either activation or repression . While acetylation and phosphorylation are thought to be responsible for short-term reversible changes in gene expression, methylation is generally more stable and involved in the long term maintenance of expression status . Inherited changes in the "epigenome" have been postulated as a possible pathway explaining the differences in gene expression seen in individuals with identical "genomes" .
A major difference between genetic and epigenetic outcomes is that, while DNA sequence is static, the epigenome is dynamic and changes with cell type, during the cell cycle, in response to biologic signals, and with the environment . Epigenetic effects have been shown to occur not just in utero, but over the life course . Certain features of complex diseases that have defied classical Mendelian genetics may be explained, at least in part, by the inheritability, partial stability and reversibility of epigenetic regulation. Gluckman et al. recently proposed that maternally-mediated changes in gene expression may be more relevant in elucidating the etiology of disease in the offspring than the inheritance pattern of the genetic code . In agreement with a possible epigenetic inheritance pattern, a recent study in mice showed that experimentally-induced contact dermatitis in the mother prior to conception resulted in an increased asthma incidence in the offspring . Further strengthening this evidence in humans, Li et al. demonstrated that grandmaternal smoking was associated with an increased risk of asthma in grandchildren, independent of maternal smoking status, suggesting a transmission of epigenetic effects inherited from the grandmother to the grandchild .
Note that in the above derivation, the intercepts are not related to the relationship between βITT, βXG, and βIV. From Figure 3, estimation of βIV depends heavily on assumptions of compliance, i.e. blinding to allocation (G associated with X and independent of U) and the absence of any other pathway from G to Y (i.e. G being independent of Y given X and U) . An intuitive motivation for equation (5) is that the change in outcome for a unit change in the instrument, βITT, is the product of the change in the exposure for a unit change in the instrument, βXG, and the change in the outcome per unit change in exposure, βIV . Estimation of the causal effect βIV by this equation makes use of the variation in Y that is due to the IV, G, and the core assumptions above. Based on our derivation, we can see that this estimate avoids the contaminating effect of the variation in Y due to the confounders of the X-Y association .
In Mendelian randomization studies, βIV would be equivalent to the effect of the intermediate phenotype on the outcome (disease) and βITT to the (direct) genetic effect on disease, while the denominator in equation (5) would capture the observed, or presumed, relation between genetic allocation, G, and its gene product, the intermediate phenotype X .
The Impact of Epigenetics on Mendelian Randomization
We use E(X)' to denote the expected gene expression when there is possible gene-environment interaction. Here "E × G" denotes the interaction between the environmental factor E and the genotype G. βEG measures the strength of the interaction and is assumed to be non-zero. We now show that the environmental factor at the maternal level changes the relationship stated in equation (5).
Let βXG* = βXG+ βEG.
βXG* may be greater or less than βXG depending on the predominant mechanism (activation vs. repression, for instance). Note that βXG* includes the regular effect of G on X in addition to the effect of gene-environmental interaction that altered the gene expression of the mother's gene. This gene-environment interaction by definition is the epigenetic effect only if it is heritable. As indicated in Figure 4, where such epigenetic effect is inherited by the child, the epigenetically modified gene expression in the child is a result of the G × E interaction inherited from the mother. Thus, the association between X and G in the child includes the effect of G on X and the gene-environment interaction inherited from the mother. (For simplicity in notation, we use the same X and βXG* to respectively denote the gene expression in the child and the inherited epigenetic effect, although their values may be different from those at the maternal level).
which gives βIV' = βITT/(βXG+ βEG) = βITT/βXG*, where βXG* = βXG+ βEG.
This new relationship does not agree with that given by equation (5), which is derived based on the MR assumptions. The additional effect (βEG) is what we have termed the "epigenetic bias."
It is noteworthy that a gene-environment interaction in parents may lead to an epigenetic effect only when it is inherited by the offspring. Certain gene-environment interactions (in ancestral parents) are thought to be mediated through molecular changes in the epigenome. The non-random nature  of such alterations may necessitate caution in the interpretation of the estimates from the Mendelian randomization (instrumental variable) analyses. In recognition of the presence of this "epigenetic bias," Bjornsson et al. have suggested that including epigenotypes in models of disease causation might act as a surrogate of parental environmental exposure and thus increase the power of epidemiological studies .
Further illustrating conceptually with the randomized clinical trial (RCT) analogy, if an RCT is blinded, intention-to-treat (ITT) cannot be influenced by external factors such as motivation or education (non-differential compliance). However, in a non-blinded trial, participants in the control or treatment arm may choose to alter treatment or compliance (differential compliance, see Figure 1). For instance, control participants may take additional steps to improve their outcome status. This results in an interaction between the allocated treatment and extraneous variables (see Figure 4) and violates the assumptions of ITT. In non-blinded randomized trials, opportunities for extraneous variables to interact with ITT increase with the length of the trial . Similar to this finding in RCTs, in Mendelian randomization studies specific genotypes may interact with a range of extraneous factors and establish epigenetic changes that are inherited by subsequent generations. Thus, the setting of Mendelian randomization is comparable to the setting of a non-blinded randomized clinical trial. Given that the time window between "randomized allocation" of genes at conception and initiation of a MR study is likely to span decades, MR studies are prone to violate the ITT assumption. For instance, if the gene expression of IL13 is higher in a child if the mother smoked, then the distribution of the epigenetically modified gene expression is no longer random at birth. Consequently, the child may or may not develop asthma depending on the inception of gene expression in the mother. In essence, since the assumption of MR is to detect an unbiased association between genetic markers and health outcomes due to randomized genes, if studies cannot demonstrate that the gene expression is also distributed randomly, the MR model is not justified and will introduce an "epigenetic bias."
Data Simulation Example
We use simulations to illustrate our findings given in Analysis section. The goal of this simulation is to demonstrate the bias in the estimates of the β coefficients if epigenetic effects transmitted from mother to child are ignored. For this simulation, we assume that a mother smoked in adolescence, stopped smoking as an adult, and then conceived the index study child. Thus, non-epigenetic effects of tobacco smoke exposure in utero did not occur. However, gene expression (epigenetic) changes in the mother prior to the index pregnancy were inherited by the child as discussed in Analysis section.
Statistical findings of data simulation using log-linear models (scenario outlined in Figure 5)
Gene → asthma
Gene → gene expression of CC
Gene expression of CC → asthma
Effect of genetic polymorphism on asthma
Effect of genetic polymorphisms on gene expression:
Effect of genetic polymorphism, maternal and offspring smoking, and gene-"maternal smoking in adolescence" interaction on gene expression:
"Maternal smoking in adolescence" × CC-genotype
Effect of expression of genetic olymorphism on asthma:
Gene expression of CC
Effect of genetic polymorphism, maternal and offspring smoking on asthma
Gene expression of CC
"Maternal smoking in adolescence"
Of 10,000 mothers in the simulated data, 30.5% (n = 3,050) smoked in adolescence, while 69.5% (n = 6,950) did not.
The genotypes of mothers who smoked in adolescence and of those who did not were distributed at random: 7.2% of the never-smoking mothers had the GG genotype compared to 9.8% of ex-smoking mothers; 41.7% of never-smoking mothers had the CG genotype compared to 39.3% in ex-smoking mothers; and about 51% in both groups were CC. These percentages were selected using the probabilities that each genotype will be present in a mother.
We also assumed that the gene × smoking (G × E) interaction resulted in changes in gene expression in the mother that were inherited by the offspring, leading to asthma manifestation in the offspring.
The inheritance of gene expression is assumed to be complete and therefore G and X at the maternal level are consistent with the G and X at the child's level; also paternal genes were assumed not to play any role.
Next, we assumed that smoking among the offspring varied with history of maternal smoking: 36.1% of the offspring whose mother smoked in adolescence also smoked, while only 28.2% of the offspring whose mother never smoked also smoked. The overall smoking prevalence for both groups of children was 30.6%. These percentage values were generated from a normal distribution with mean 0.30 and standard deviation 0.10.
The data were also simulated such that there was altered gene expression in 25% of the children of ex-smoking mothers compared to 7.4% of the children with never-smoking mothers.
Next, 3.6% of the children with ex-smoking mothers had childhood asthma, compared to 1.4% of the children with never-smoking mothers. These percentages were derived from the assumed probabilities of asthma depending on smoking exposure.
This simplification is to enable demonstration of the epigenetic effect illustrated in Figure 5. Using the above data scenario, we applied a log-linear model with a dominant effect of the G allele (see SAS program). Hence, the genotypes GG and CG combined served as the reference. We estimated the effects (the values of β) of the CC genotype, maternal smoking, offspring smoking, and the "G × E interaction" for maternal smoking.
the effect of the genetic polymorphism on outcome Y (in this case, asthma) estimating βITT;
the effect of the genetic polymorphism on X (in this case, gene expression) estimating βXG*
the effect of X (gene expression) on Y (asthma) estimating βIV'
The significance level was selected as α = 0.05.
If we ignore the inherited altered gene expression due to maternal smoking in adolescence, we will find a statistically significant crude effect of the CC-genotype on the gene expression (βXG* = 0.367, p-value = 0.0001). However, this effect actually includes contributions from both the CC-genotype and from the gene-smoking interaction (G × E interaction) in the mother leading to an epigenetic effect. The contribution from each of these two can be estimated by including a "gene by smoking" interaction term in the model. Inclusion of this term showed a significant gene-smoking interaction effect (βEG = 0.269, p-value= 0.015). Regarding the outcome, asthma (Y), we found a statistically significant effect of the CC-genotype (βITT = 0.458, p-value < 0.0012).
The above observations imply two important cautionary findings. Firstly, this data simulation demonstrates that the gene effect, even if randomly distributed on the exposure, may be overestimated when the gene × "maternal smoking in adolescence" interaction is ignored. Secondly, we show in this scenario that when the epigenetic effect is ignored, a significant epigenetic effect (gene × smoking interaction) tends to violate the core assumptions of Mendelian randomization. This interaction results in variations in gene-expression and subsequently in increased or reduced risk of the outcome (i.e. an epigenetic effect)
Hence, βIV' = βITT/(βXG+ βEG) = βITT/βXG* as demonstrated earlier, but βIV' is not equal to βITT/βXG as given by equation (4). This is due to the inclusion of gene-smoking interaction, which causes βXG* to deviate from βXG. As seen in our example, βITT = 0.458, βXG* = 0.367, and βXG = 0.215, so we have βIV' = 0.458/0.367 = 1.248 while βIV = 0.458/0.215 = 2.130. This implies a tendency for βIV' and βIVto differ systematically, if evaluated from a study where the gene by "prior maternal smoking" interaction is not taken into account. We refer to this bias as the "epigenetic bias".
As an increasing number of epidemiologic researchers employ Mendelian randomization in their research, the impact of epigenetics on the core assumptions that drive the use of genotypes as instrumental variables deserves attention. Our work demonstrates that epigenetic bias may distort the effects detected in MR studies.
It is important to differentiate between gene-environment interactions that affect MR studies in the same generation, and epigenetic effects inherited through subsequent generations and that may invalidate MR studies in the second or third generation. Additional tools, such as molecular methods to assess the level of DNA and histone modifications may soon be widely available. Future research in the field of Mendelian randomization needs to collect information on epigenetic changes (e.g. gene expression) of the various genes tested. These additional data would enable adjustment for "epigenetic bias" as described in this paper. Such tools may make it possible to apply certain "correction factors" or to stratify future analyses by "methylation status" (or "expression status"), in the assessment of gene effect . Until such a time, caution is needed in the interpretation of Mendelian randomization studies.
randomized controlled trial
Intention to treat analysis
Sources of Funding: This study was funded in part by the National Institutes of Health R01 AI061471.
- Lawlor DA, Harbord RM, Sterne JA, Timpson N, Davey Smith G: Mendelian randomization: Using genes as instruments for making causal inferences in epidemiology. Stat Med. 2007, 27 (8): 1133-63. 10.1002/sim.3034. 10.1002/sim.3034View ArticleGoogle Scholar
- Davey Smith G, Ebrahim S: 'Mendelian randomization': can genetic epidemiology contribute to understanding environmental determinants of disease?. Int J Epidemiol. 2003, 32: 1-22. 10.1093/ije/dyg070View ArticleGoogle Scholar
- Greenland S: An introduction to instrumental variables for epidemiologists. Int J Epidemiol. 2000, 29: 722-729. 10.1093/ije/29.4.722View ArticlePubMedGoogle Scholar
- Nitsch D, Molokhia M, Smeeth L, DeStavola BL, Whittaker JC, Leon DA: Limits to causal inference based on Mendelian randomization: a comparison with randomized controlled trials. Am J Epidemiol. 2006, 163: 397-403. 10.1093/aje/kwj062View ArticlePubMedGoogle Scholar
- Didelez V, Sheehan N: Mendelian randomization as an instrumental variable approach to causal inference. Stat Methods Med Res. 2007, 16: 309-330. 10.1177/0962280206077743View ArticlePubMedGoogle Scholar
- Keavney B, Danesh J, Parish S, Palmer A, Clark S, Youngman L, Delepine M, Lathrop M, Peto R, Collins R: Fibrinogen and coronary heart disease: test of causality by 'Mendelian randomization'. Int J Epidemiol. 2006, 35: 935-943. 10.1093/ije/dyl114View ArticlePubMedGoogle Scholar
- Peduzzi P, Henderson W, Hartigan P, Lavori P: Analysis of randomized controlled trials. Epidemiol Rev. 2002, 24: 26-38. 10.1093/epirev/24.1.26View ArticlePubMedGoogle Scholar
- Goetghebeur E, Loeys T: Beyond intention to treat. Epidemiol Rev. 2002, 24: 85-90. 10.1093/epirev/24.1.85View ArticlePubMedGoogle Scholar
- Davey Smith G, Harbord R, Ebrahim S: Fibrinogen, C-reactive protein and coronary heart disease: does Mendelian randomization suggest the associations are non-causal?. QJM. 2004, 97: 163-166. 10.1093/qjmed/hch025View ArticleGoogle Scholar
- Knowler WC, Williams RC, Pettitt DJ, Steinberg AG: Gm3;5, 13, 14 and type 2 diabetes mellitus: an association in American Indians with genetic admixture. Am J Hum Genet. 1988, 43: 520-526.PubMed CentralPubMedGoogle Scholar
- Ebrahim S, Smith GD: Mendelian randomization: can genetic epidemiology help redress the failures of observational epidemiology?. Human Genetics. 2007, 123: 15-33. 10.1007/s00439-007-0448-6View ArticlePubMedGoogle Scholar
- Bergman A, Siegal ML: Evolutionary capacitance as a general feature of complex gene networks. Nature. 2003, 424: 549-552. 10.1038/nature01765View ArticlePubMedGoogle Scholar
- Siegal ML, Bergman A: Waddington's canalization revisited: developmental stability and evolution. Proc Natl Acad Sci USA. 2002, 99: 10528-10532. 10.1073/pnas.102303999PubMed CentralView ArticlePubMedGoogle Scholar
- Weinhold B: Epigenetics: the science of change. Environ Health Perspect. 2006, 114: A160-167.PubMed CentralView ArticlePubMedGoogle Scholar
- Rakyan V, Whitelaw E: Transgenerational epigenetic inheritance. Curr Biol. 2003, 13: R6 10.1016/S0960-9822(02)01377-5View ArticlePubMedGoogle Scholar
- Wade PA, Archer TK: Epigenetics: environmental instructions for the genome. Environ Health Perspect. 2006, 114: A140-141.PubMed CentralView ArticlePubMedGoogle Scholar
- Miller RL, Ho SM: Environmental epigenetics and asthma: current concepts and call for studies. Am J Respir Crit Care Med. 2008, 177: 567-573. 10.1164/rccm.200710-1511PPPubMed CentralView ArticlePubMedGoogle Scholar
- Devaskar SU, Raychaudhuri S: Epigenetics–a science of heritable biological adaptation. Pediatr Res. 2007, 61: 1R-4R. 10.1203/pdr.0b013e31805cdbd8View ArticlePubMedGoogle Scholar
- Adcock IM, Ford P, Barnes PJ, Ito K: Epigenetics and airways disease. Respir Res. 2006, 7: 21 10.1186/1465-9921-7-21PubMed CentralView ArticlePubMedGoogle Scholar
- Bjornsson HT, Fallin MD, Feinberg AP: An integrated epigenetic and genetic approach to common human disease. Trends Genet. 2004, 20: 350-358. 10.1016/j.tig.2004.06.009View ArticlePubMedGoogle Scholar
- Chuang JC, Jones PA: Epigenetics and microRNAs. Pediatr Res. 2007, 61: 24R-29R. 10.1203/pdr.0b013e3180457684View ArticlePubMedGoogle Scholar
- Gosden RG, Feinberg AP: Genetics and epigenetics–nature's pen-and-pencil set. N Engl J Med. 2007, 356: 731-733. 10.1056/NEJMe068284View ArticlePubMedGoogle Scholar
- Gluckman PD, Hanson MA, Cooper C, Thornburg KL: Effect of in utero and early-life conditions on adult health and disease. N Engl J Med. 2008, 359: 61-73. 10.1056/NEJMra0708473PubMed CentralView ArticlePubMedGoogle Scholar
- Lim RH, Arredouani MS, Fedulov A, Kobzik L, Hubeau C: Maternal allergic contact dermatitis causes increased asthma risk in offspring. Respir Res. 2007, 8: 56 10.1186/1465-9921-8-56PubMed CentralView ArticlePubMedGoogle Scholar
- Li YF, Langholz B, Salam MT, Gilliland FD: Maternal and grandmaternal smoking patterns are associated with early childhood asthma. Chest. 2005, 127: 1232-1241. 10.1378/chest.127.4.1232View ArticlePubMedGoogle Scholar
- Greenland S, Brumback B: An overview of relations among causal modelling methods. Int J Epidemiol. 2002, 31: 1030-1037. 10.1093/ije/31.5.1030View ArticlePubMedGoogle Scholar
- Sadeghnejad A, Karmaus W, Arshad SH, Kurukulaaratchy R, Huebner M, Ewart S: IL13 gene polymorphisms modify the effect of exposure to tobacco smoke on persistent wheeze and asthma of childhood, a longitudinal study. Respir Res. 2008, 9: 2 10.1186/1465-9921-9-2PubMed CentralView ArticlePubMedGoogle Scholar
- van Gool CH, Penninx BW, Kempen GI, Miller GD, van Eijk JT, Pahor M, Messier SP: Determinants of high and low attendance to diet and exercise interventions among overweight and obese older adults. Results from the arthritis, diet, and activity promotion trial. Contemp Clin Trials. 2006, 27: 227-237. 10.1016/j.cct.2005.11.002View ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.