COPSAC enrolled 411 high-risk neonates (mother with verified asthma) at age 1 mo; the recruitment was previously described in detail [6–8]. The study was approved by the Ethics Committee for Copenhagen (KF 01–289/96) and The Danish Data Protection Agency (2008-41-1754). Before enrolment, informed consent was obtained from parents. Data were collected on-line and locked after external monitoring. An audit trail was run routinely.

Manchester Asthma and Allergy Study (MAAS) is an unselected population-based birth cohort described in detail elsewhere [13,14]. The study is registered as ISRCTN72673620. Participants were recruited prenatally, and followed prospectively, attending the first review clinic at ages 1, 3, and 5 y (± 4 wk). The study was approved by the Local Research Ethics Committee (SOU/00/258; SOU/00/259). Written informed consent was obtained from all parents.

Genotyping for R501X was performed using a TAQMAN-based allelic discrimination assay (Applied Biosystems). Allelic discrimination was assessed using Applied Biosystems 7700 sequence detection system. Probes and primers were as described [1]. Mutation 2282del4 was genotyped by sizing of a fluorescent-labeled PCR fragment on an Applied Biosystems 3100 or 3730 DNA sequencer [1].

Children were classified according to a two level dominant genetic model (i.e., children were assigned as having a FLG mutation if they were heterozygous or homozygous for the null allele in at least one of the two SNPs).

Effects of environmental factors and their possible modification by genetic mutation were assessed by Kaplan-Meier curves and Cox regression. The children were retained in the analysis from birth until age at diagnosis of eczema, drop-out, or age 1 y, whichever came first. Simultaneous quantification of gene/environment effects was done by multiple Cox regression. Pet exposure at birth and pet allergen exposure were analysed separately. All analyses were carried out using SAS 9.1.3, SPSS 13.0, and the free-ware statistical package R (R Development Core Team, 2006). All estimates are reported with 95% confidence intervals [CI] in brackets.

Of 411 infants, 379 of mixed European ancestry were genotyped for 501x and 2282del4 [1] as well as R2448X and S3247X [15]. The main mutated alleles 501x and 2282del4 were present in 40 children and exhibited a very strong association to the development of eczema as previously reported [1].

Eczema was diagnosed in 28% (105 children) before age 1 y; 265 (75%) did not have pets in their home at birth, 38 (11%) had a cat, 37 (11%) had a dog, and 11 (3%) had both; information on pet ownership was not available in 28 children. Among the 38 children with main mutations and information on pet ownership, 26 (68%) did not have pets, seven (18%) had a dog only, four (11%) had a cat only, and one (3%) had both. Objective measure of allergen exposure was available in 339 children; 205 (60%) had low cat and dog allergen levels, 54 (16%) high cat allergen level, 54 (16%) high dog allergen level, 26 (8%) high cat and dog allergen levels, and 78 (23%) high mite allergen level. Among the 38 children with mutation, 22 (58%) had low cat and dog allergen levels, five (13%) high cat allergen level, nine (24%) high dog allergen level, two (5%) had high cat and dog allergen levels, and ten (26%) high mite allergen level. Six children were lost to follow-up during the first year but were included in the analyses for the available follow-up period.

Kaplan-Meier plots suggested an increased risk of eczema due to FLG mutation, and a further increased risk of eczema due to cat exposure but only in the presence of FLG mutation (Figure 3). Backwards elimination in a multiple Cox regression including FLG status, cat exposure, dog exposure, mite allergen exposure, and interactions between FLG status and cat exposure, dog exposure, and mite allergen exposure showed a significant interaction between cat exposure and FLG status (p = 0.0008) and a significant effect of dog exposure (p = 0.05). An additional exploration of the cat exposure-FLG status interaction demonstrated an increased risk of developing eczema due to mutation (HR 2.26 [95% CI 1.27–4.00], p = 0.005), with an additional increased risk if also exposed to cat (HR 11.11 [3.79–32.60], p < 0.0001). The overall decrease in risk due to exposure to dog was quantified in an HR of 0.49 [0.24–1.01]. Analysis including only children carrying mutant alleles demonstrated a significant effect of cat ownership on development of eczema (HR 7.49 [2.37–23.67], p = 0.0006).

For allergen exposure a similar analysis was made with cat, dog, and mite allergen levels. The model confirmed an increased risk of developing eczema due to mutation (HR 2.53 [1.45–4.41], p = 0.001) with an additional increase in risk if exposed to high cat allergen level (HR 3.77 [1.45–9.81], p = 0.006). In this analysis there was no significant effect of dog or mite allergens. Analysis including only children carrying mutant alleles demonstrated a significant effect of cat allergen exposure on development of eczema (HR 3.09 [1.17–8.19], p = 0.023).

Cat allergen exposure shows high agreement with reported cat exposure at 4 wk (simple Kappa coefficient 0.62 [0.51–0.73]). Similarly dog allergen exposure shows high agreement with reported dog exposure at 4 wk (simple Kappa coefficient 0.59 [0.48–0.70]).

S. aureus was identified from the perineum in one newborn before the age of 3 mo.

Of 940 children with questionnaire data on the age of onset of eczema, 513 provided a blood sample for DNA extraction. Children who provided DNA did not differ from those who did not in terms of family history, maternal age, socioeconomic status, history of eczema, allergic sensitisation, and pet ownership (data available on request).

A total of 503 children of mixed European ancestry had information on FLG main mutations (501x and 2282del4), pet ownership at birth, and the age of onset of eczema. By age 1 y, eczema was reported by 187/503 parents (37%). Mutated alleles were present in 50 children (10%); 324 (64%) did not own any pets at birth, 88 (18%) had a cat at home, 66 (13%) had a dog, and 25 (5%) had both. The frequency of pet ownership did not differ between children with or without FLG mutation (no pets 293/453 [64.7%] and 31/50 [62%], cat 77/453 [17%] and 11/50 [22%], dog 60/453 [13.2%] and 6/50 [12%], and both cat and dog 23/453 [5.1%] and 2/50 [4%], no FLG mutation versus FLG mutation, respectively, p = 0.84). Cat allergen exposure was available in 461 children, mite allergen in 458, and dog allergen in 460.

Kaplan-Meier plots demonstrating the age of onset of eczema in the first year of life related to cat and dog ownership amongst children with and without FLG mutation are presented in Figure 4. The results of a multiple Cox regression that included FLG genotype, cat and dog ownership at birth, the interaction between FLG genotype with cat and dog ownership, and mite exposure and its interaction with FLG genotype indicated an increased risk of developing eczema in the first 12 mo due to FLG mutation (HR 1.95 [95% CI 1.13–3.36], p = 0.017). Furthermore, in the presence of FLG mutation, the risk increased further because of cat ownership at birth, with no significant effect of cat ownership being observed amongst children without FLG mutation (interaction of cat ownership at birth and FLG mutation, HR 3.82 [1.35–10.81], p = 0.011). There was a trend for dog ownership to increase the risk (HR 1.51 [0.96–2.37], p = 0.075), with no significant interaction of FLG genotype with dog ownership (HR 0.59 [0.16-2.20], p = 0.43) or mite allergen exposure (HR 1.15 [0.93–1.43], p = 0.21). There was no significant effect of having both cats and dogs (HR 1.57 [0.82–3.03]), or their interaction with FLG genotype (p = 0.96). Analysis including only children carrying mutant alleles demonstrated a significant effect of cat ownership on development of eczema (HR 2.47 [1.09–5.62], p = 0.03).

In terms of allergen exposure there was no significant correlation between mite, cat, and dog allergen levels. Multiple Cox regression that included FLG genotype, cat, dog, and mite allergen levels, and the interaction between genotype and allergen levels confirmed an increased risk of developing eczema due to mutation (HR 2.22 [1.27–3.89], p = 0.005) and the interaction between cat allergen and FLG genotype (with risk increasing risk with increasing allergen level among children with FLG mutation; HR for interaction 1.31 [1.03–1.67], p = 0.026). Nonsignificant trends were observed for dog allergen exposure (HR 1.07 [1.00–1.15], p = 0.06), the direction of which appeared to inverse in the interaction with FLG mutation (0.84 [0.68–1.05, p = 0.1]).

We have demonstrated a significant interaction between FLG loss-of-function main mutations (501x and 2282del4) and cat ownership at birth on the development of early-life eczema in two independent birth cohorts. Our data suggest that cat but not dog ownership substantially increases the risk of eczema within the first year of life in children with FLG loss-of-function variants, but not amongst those without. The observed effects were independent of allergic sensitisation.

For the primary outcome we selected age of onset of early eczema. The rationale for this is that in both cohorts the increased risk of eczema amongst FLG mutation carriers occurred only in the first year of life and not thereafter (Figures 1 and 2). Therefore development of eczema during the first year of life is the relevant outcome when studying this gene-environment interaction. This also means that the important environmental exposures that may interact with FLG in the development of eczema must occur in the first months of life. This limits the number of relevant exposures, as neonates typically live in a very sheltered environment.

This was a hypothesis-driven analysis. We hypothesized that weakening of physical barrier of skin in FLG deficient individuals may potentiate the effect of relevant environmental exposures. Therefore we investigated whether there is an interaction between FLG loss-of-function mutations with a range of environmental exposures for which the skin could be the relevant route of exposure. Since FLG is not expressed in either lungs or GI tract, we did not test exposures through such routes. We focused on the exposure that was associated with eczema in previous epidemiological studies and that was common for the two populations tested (domestic pets and indoor allergens). In addition, in COPSAC we explored the possibility of testing the effect of skin colonization with S. aureus, but only one child in the mutated group was colonized on the skin.

The different designs of the two cohorts prevented data pooling in favour of replication, and necessitated some differences in data analysis. Clinical outcomes were not identical, resulting in a higher prevalence of infant eczema in MAAS (parentally reported, 37%) compared to COPSAC (investigator-diagnosed, 28%). This suggests that MAAS was capturing milder cases and overreporting of eczema by parents. COPSAC provided closely monitored prospective data on eczema progression. The specificity of diagnosis was high, since the detailed phenotyping and management was carried out solely by the investigators, reducing the risk of misclassification [9], which is of particular importance in the clinical evaluation of eczema where interobserver variation is a problem [16]. Despite the differences in definition of primary outcome, we achieved replication.

We acknowledge the fact that the initial finding is based on only five children with eczema with a cat at home and carrying FLG mutation in COPSAC, but the power of the statistics (p < 0.0001) derives from the longitudinal dataset with the time of onset clearly distinguishing these populations. Cross-sectional analyses would not have such statistical power. Furthermore, it is worth noting that the findings were confirmed in the analysis of allergen levels, that pet ownership did not differ between those with and without the variant, and that the interaction between FLG variants and cat ownership was replicated in an independent cohort that included a larger number of cat exposed FLG mutation carriers.

Allergens levels were measured at age 1 y representing allergen exposure during the first year of life. Dust samples were taken from the child's mattress in COPSAC and from the lounge floor in MAAS, which resulted in different distribution of results (10-fold increase from third to fourth quartile in COPSAC with only 3-fold difference in MAAS, indicating that quartile analysis would not be appropriate in MAAS). Although we used different indices of allergen exposure, we obtained similar results.

In the original discovery in COPSAC of the loss-of-function variants 501x and 2282del4 in the gene encoding filaggrin, we demonstrated a strong association to the development of eczema in the first years of life. Two new SNPs within the filaggrin gene (R2447X and S3247X) have since been reported [15] and were found in eight children in COPSAC. However, these new SNPs are qualitatively different from the old mutations with some residual function as demonstrated by a significantly lower penetrance of eczema. Inclusion of the novel SNPs would therefore weaken our analyses by mixing a well-known strong risk factor with a weak risk factor, which furthermore is qualitatively different and therefore may have a different effect on the gene–environmental interactions studied. As a consequence we did not find it reasonable to formulate a dominant genetic model based on all four SNPs.

Several studies have suggested an effect of early-life pet exposure on the development of atopic disease. However, the data are inconsistent, with some studies showing increased risk [17], and others decreased risk [18], or no effect [19]. These inconsistencies may be due in part to epidemiological artifacts. Recall bias is an obvious risk when exposure assessments are retrospective. Selection for keeping pets is another potential bias. The time from exposure to assessment of the health outcome has been long in some studies, which may lead to confounding by early modification of the exposure and therefore affect the long-term outcome. The causal-effect direction may be uncertain in retrospective assessments, as the pets may have been acquired after the disease onset. Furthermore, cross-sectional studies may report protective effects from pet exposure, when in fact individuals have removed the pet from the household because of the disease or the factors associated with a predisposition to disease. Importantly, in both our studies we documented pet exposure at birth (i.e., before eczema occurrence), thus eliminating uncertainty of the direction of association between exposure and disease. The findings were further supported by objective measures of allergen exposure.

We only assessed animals living in the household, though other exposures outside the home may be relevant, but this would only have biased the risk estimates toward the null, potentially underestimating the true risk. Having pets has been reported to be biased from socioeconomic status, smoking habits, and parental allergic rhinitis [20], but it is unlikely to affect the genetic distribution in the study population and could therefore not act as confounder.

The mechanism by which cat exposure drives the development of eczema is unknown. Our data suggest that it does not act through cat-specific immunoglobulin E (IgE). This is in keeping with previous studies showing that the effect of FLG loss-of-function mutations was equally strong on allergic and nonallergic eczema, demonstrating that the development of eczema is not dependent on allergic sensitisation [2]. Cat exposure may act through other nonallergic mechanisms such as endotoxins. COPSAC recently reported that bacterial colonization of the airways in neonates was associated with later development of recurrent wheeze, asthma, and increased total immunoglobulin E (IgE) [8]. It may be possible that cat exposure mediates its effect in the FLG mutated genotypes via altered bacterial exposure, or that cat exposure may be a surrogate marker for other unknown environmental influences penetrating the imperfect skin barrier because of defective profilaggrin synthesis.

Although there are many new susceptibility genes for complex disorders following genome-wide association studies, there are very few gene–environment studies. Our findings of a consistent effect from early-life cat ownership on the phenotypic expression of eczema in infants with FLG loss-of-function mutations in two independent birth cohorts is an example of gene–environment interaction in a common complex disease, demonstrating that such gene–environment interactions exist and are detectable given appropriately detailed studies of environmental exposures.