What is the evidence for interactions between filaggrin null mutations and environmental exposures in the aetiology of atopic dermatitis? A systematic review†

Summary Background Epidemiological studies indicate that gene–environment interactions play a role in atopic dermatitis (AD). Objectives To review the evidence for gene–environment interactions in AD aetiology, focusing on filaggrin (FLG) loss‐of‐function mutations. Methods A systematic search from inception to September 2018 in Embase, MEDLINE and BIOSIS was performed. Search terms included all synonyms for AD and filaggrin/FLG; any genetic or epidemiological study design using any statistical methods were included. Quality assessment using criteria modified from guidance (ROBINS‐I and Human Genome Epidemiology Network) for nonrandomized and genetic studies was completed, including consideration of power. Heterogeneity of study design and analyses precluded the use of meta‐analysis. Results Of 1817 papers identified, 12 studies fulfilled the inclusion criteria required and performed formal interaction testing. There was some evidence for FLG–environment interactions in six of the studies (P‐value for interaction ≤ 0·05), including early‐life cat ownership, older siblings, water hardness, phthalate exposure, higher urinary phthalate metabolite levels (which all increased AD risk additional to FLG null genotype) and prolonged breastfeeding (which decreased AD risk in the context of FLG null genotype). Major limitations of published studies were the low numbers of individuals (ranging from five to 94) with AD and FLG loss‐of‐function mutations and exposure to specific environmental factors, and variation in exposure definitions. Conclusions Evidence on FLG–environment interactions in AD aetiology is limited. However, many of the studies lacked large enough sample sizes to assess these interactions fully. Further research is needed with larger sample sizes and clearly defined exposure assessment. Linked Comment: Park and Seo. Br J Dermatol 2020; 183:411.

decreased AD risk in the context of FLG null genotype). Major limitations of published studies were the low numbers of individuals (ranging from five to 94) with AD and FLG loss-of-function mutations and exposure to specific environmental factors, and variation in exposure definitions. Conclusions Evidence on FLG-environment interactions in AD aetiology is limited. However, many of the studies lacked large enough sample sizes to assess these interactions fully. Further research is needed with larger sample sizes and clearly defined exposure assessment.
What's already known about this topic?
• Gene-environment interactions are considered important in the aetiology of atopic dermatitis.
• Loss-of-function mutations in the gene coding filaggrin (FLG) are the most consistently reported genetic variants for atopic dermatitis. What does this study add?
• There is some evidence for FLG-environment interactions in the aetiology of atopic dermatitis; however, the evidence is limited.
• Studies lack large enough sample sizes to achieve adequate power in order to assess these interactions fully.
Atopic dermatitis (AD), also known as eczema or atopic eczema, is a complex, multifactorial, often debilitating disease. 1 The prevalence of AD has risen rapidly, suggesting that environmental factors might be responsible for such changes. 2 It is estimated that up to 20% of children and 3% of adults in high-income countries are affected by AD. 3 In order to discover ways to reduce the personal and public health burden, it is necessary to gain a better understanding of the aetiology of AD.
Considerable phenotypic heterogeneity, evidence for multiple genetic risk mechanisms 4 and incomplete penetrance have led to complexities in understanding the genetic basis of AD. 5 There have been 31 risk loci identified for AD to date. 4 Lossof-function mutations in the gene encoding filaggrin (FLG) are the strongest and most significantly associated genetic variants for AD. 6 Profilaggrin is an insoluble protein found in the outer epidermis; monomeric filaggrin has multiple functions including aggregation of keratin filaments. 7 FLG is essential for normal epidermal barrier function and formation, 8 contributing to the skin water-holding capacity and pH balance. 9 The two most prevalent loss-of-function mutations in FLG in white European populations are R510X and 2282del4, present in approximately 9% of healthy people in Northern European populations. These mutations are strongly associated with AD risk, particularly early-onset and severe disease. 6 Other, less prevalent, loss-of-function mutations in FLG have been identified. [10][11][12] Despite increased understanding of the importance of genetic factors, the rising AD prevalence has been too substantial and rapid to be explained purely by genetic factors. 13 Environmental factors have been implicated in the rising AD prevalence; exposure to such factors in utero or in later life may play a role in AD aetiology. 14 Gene-environment interaction (GEI) may be defined as occurring when individuals with different genotypes respond to an environmental exposure in different ways; this interaction contributes to many common phenotypes and complex genetic traits. There is evidence that GEI plays a role in atopic diseases, such as asthma, in which genotype interacts with environmental factors, including maternal smoking and house dust mite. 15,16 A recent review on allergic diseases highlighted that several studies exploring GEI in AD exist, yet findings have not been synthesized (e.g. in a systematic review). 17 Our objective was to perform a systematic review of the evidence for GEI in AD, focusing on interactions with the FLG null genotype, because this is the strongest and most widely replicated AD genetic risk factor, and because the role of filaggrin in skin barrier function provides a priori support for a hypothetical GEI effect.

Materials and methods
This systematic review was prospectively registered (PROS-PERO ID CRD42017057818). A detailed electronic search of MEDLINE and Embase via Ovid, and BIOSIS via Web of Science was undertaken from inception of each database to September 2018, identifying manuscripts in any language. To define AD, the search terms 'atopic dermatitis', 'atopic eczema' and 'eczema' were used, and to define FLG mutations, 'filaggrin', 'FLG', possible misspellings and previously reported FLG mutation names were included in the strategy (Methods S1; see Supporting Information). We focused on incident AD cases, including studies that examined interactions for AD development rather than interactions for established AD. To avoid defining an exhaustive list of environmental factors a priori, the inclusion of any environmental factor was assessed during the title and abstract screening. We defined environmental exposures as proposed by Rothman. 18 The primary outcome measure was evidence of a statistically significant (defined as P < 0Á05) FLG-environment interaction in the aetiology of AD and the secondary outcome was the strength of the association of the interaction (evidence of a dose-response relationship) and AD severity. Details of abstract screening, inclusion and exclusion criteria and data extraction are provided in Methods S2 and Table S1 (see Supporting Information).
Quality and bias assessment was performed using criteria modified from guidance for nonrandomized studies to determine quality of studies; ROBINS-I to assess risk of bias in nonrandomized studies of interventions, and HuGENet for genetic studies, including assessing whether confounders were considered. 19,20 These two tools were combined to determine bias in genetic and environmental studies, as neither tool was designed for GEI studies.
Post hoc sample-size calculations were undertaken to estimate the sample size required to detect a GEI effect varying between 1Á2 and 2 6 in a case-control/cohort study for a binary single-nucleotide polymorphism (SNP) and binary exposure under a series of assumptions for model parameters (Table S2; see Supporting Information), using R 3Á5Á0 (R package powerGWASinteraction).

Results
The search identified 1817 papers of possible relevance (Fig. 1); 12 met our inclusion criteria (Table S1; see Supporting Information). Papers tested various environmental exposures ( Table 1).
The study designs of these 12 papers included 10 cohort studies, one case-control study and one family-based study (Table S3; see Supporting Information). Study populations ranged from 296 to 5188 individuals and participant ages ranged from 1 month to 69 years. The number of participants in each study with FLG loss-of-function mutations ranged from 27 (9Á1%) to 459 (10Á2%). The AD definition and method of ascertainment varied between studies (Table S4; see Supporting Information). [21][22][23] None of the included studies investigated the strength of interaction or AD severity.
Of the 12 publications, including 15 studies (Table S4; see Supporting Information), six studies showed evidence for GEI (P < 0Á05) ( Table 1). Most studies used regression models to calculate P-values and some presented hazard ratios (HRs). Heterogeneity in study design and exposures precluded formal meta-analysis.

Cat exposure
Two studies assessed the FLG-cat interaction. 24,25 Bisgaard et al. tested for an interaction in the Copenhagen Prospective Study on Asthma in Childhood (COPSAC) (n = 379) and reported an increased risk related to cat exposure at birth among children aged 0-5 years who had FLG null mutation [HR 11Á11, 95% confidence interval (CI) 3Á79-32Á60; P interaction = 0Á0008]; findings were replicated in the Manchester Asthma and Allergy Study (MAAS) (n = 503), with an increased risk owing to interaction of cat exposure at birth and FLG null genotype (HR 3Á82, 95% CI 1Á35-10Á81; P interaction = 0Á011). Schuttelaar et al. (n = 934) reported no overall interaction (P = 0Á85) between FLG null genotype (one/two FLG loss-of-function mutations) and cat exposure at home [odds ratio (OR) with FLG loss-of-function mutation(s) and cat exposure = 1Á9; OR for FLG wild-type individuals and cat exposure = 2Á1]. However, Schuttelaar et al.
reported an interaction when examining the 2282del4 mutation only (P = 0Á003), with a stronger effect in children aged 0-8 years with a cat at home (OR 6Á0, 95% CI 3Á2-11Á3) compared with those without (OR 2Á2, 95% CI 1Á4-3Á7). 25 All those with 2282del4 mutations were heterozygous. 25 As all FLG loss-offunction mutations have biological equivalence on filaggrin protein expression, there is no clear biological plausibility for an interaction with one mutation and not another. Evidence for GEI comes from small numbers of individuals with FLG mutation, cat exposure and development of AD. In Bisgaard et al. five such individuals were reported. Schuttelaar et al. did not provide the number of individuals, but it can be inferred that n < 84 for the overall interaction and n < 50 for the 2282del4 interaction. 24,25 Dog exposure Bisgaard et al. tested for an interaction between FLG loss-offunction mutations and dog ownership in the first year of life. There was no evidence for an interaction in COPSAC (n = 379) (result statistics not reported) or MAAS (n = 503) (HR 0Á59, 95% CI 0Á16-2Á20; P = 0Á43). 24

Siblings
One study reported an interaction between FLG genotype and presence of older siblings among both children attending and those not attending daycare at 2 years of age in two separate studies -LISAplus cohort (aged 6-72 months, n = 1037, interaction OR 3Á27, 95% CI 1Á14-9Á36; P < 0Á05) and GINIplus cohort (12-72 months, n = 1838, interaction OR 2Á41, 95% CI 1Á06-5Á48; P < 0Á05). 26 This interaction increased the risk of AD. 27 Parity, maternal atopy and child's sex Henderson et al. found no evidence of an interaction between FLG genotype with parity (n = 4463, P = 0Á802), maternal asthma or AD (n = 5188, P = 0Á486 and P = 0Á884, respectively) or the child's sex (P = 0Á959) 28 in children aged 6 months to 11 years who were part of a prospective cohort study.

Maternal IgE sensitization
Esparza-Gordillo et al. conducted a parent-of-origin analysis investigating the effect of a child's FLG genotype and maternal and paternal FLG genotypes on the child's AD risk (n = 1209 families). 29 Although interactions were not the focus of this study, results were stratified by maternal IgE sensitization status, allowing us to compare the effect of the child's FLG genotype in those whose mothers were or were not IgE sensitized. The child's FLG genotype had a stronger risk effect when mothers had normal IgE levels [relative risk (RR) for one FLG loss-of-function mutation = 2Á30 (95% CI 1Á64-3Á22); RR for two FLG loss-of-function mutations = 7Á19 (95% CI 3Á77-13Á7)] compared with those who had sensitized mothers [RR1 1Á37 (95% CI 0Á97-1Á94); RR2 2Á98 (95% CI 1Á19-7Á45)]. However, CIs were wide and overlapping. Esparza-Gordillo et al. reported the opposite for effect of maternal genotype, so it is unclear whether this evidence points to a true interaction with exposure to mothers with elevated IgE or whether the observation results from maternal genetic effects in utero or imprinting effects of FLG genotype. 29

Smoking
One study tested for possible interactions with FLG mutations and maternal smoking during pregnancy (n = 5140) or childhood environmental tobacco smoke exposure (n = 4874). Their results showed no evidence for either interaction 43 full-text articles excluded for the following reasons: 12 were conference proceedings or abstract only 10 had no interaction testing 5 were duplicates not identified above 6 had an outcome that was not atopic dermatitis 2 did not have a control group 2 were literature reviews 2 were letters 2 included no environmental factors 1 presented no statistical analysis results 1 is follow-on work informed by the preliminary findings of this systematic review   (maternal smoking P = 0Á362, child environmental tobacco smoke exposure P = 0Á742). 23 Breastfeeding Ziyab et al. found evidence for a protective association between breastfeeding duration and AD in children aged 1 or 2 years (n = 885) carrying at least one FLG loss-of-function mutation (P = 0Á02), with no evidence in those without an FLG null mutation (P = 0Á64). 30 However, this was a stratified analysis and formal interaction testing was not undertaken. Henderson et al. found no evidence for this interaction (P = 0Á952) in their earlier, larger study (n = 5158). 28

Birth year
Thyssen et al. investigated FLG and year of birth in adults aged 18-69 years (n = 3202) but did not report any evidence of interaction (P = 0Á19) on AD risk. 31

Water hardness
Interaction of FLG genotype and water hardness has been investigated in two papers within the same cohort. Perkin et al. investigated the association between calcium and chlorine levels in water and AD during the first 3 months of life in the Enquiring About Tolerance study, a U.K. population-based cohort of 1303 infants. Interaction tests did not show a statistically significant interaction between FLG genotype and high calcium-low chlorine concentration in water, low calciumhigh chlorine or high calcium-high chlorine concentrations. 26 They subsequently reported evidence of an interaction between FLG loss-of-function mutations and water hardness increasing AD risk, when studying high calcium concentrations in water (> 256 mg L À1 CaCO 3 ) from 3 months of age in this cohort (N = 1303; n = 75 with FLG null mutation exposed to high CaCO 3 levels, P = 0Á008). 32

Vitamin D
A possible interaction between serum vitamin D levels and FLG genotype was investigated by Berents et al. They measured serum vitamin D levels in 558 participants at age 1-13 months and 2 years alongside interviews assessing vitamin D intake. They did not find any evidence for an interaction (P > 0Á13). 33

Urine phthalate metabolites and household dust phthalate
Phthalates are added to plastic to increase flexibility. They have been reported to be associated with childhood AD. 34 Wang and Karmaus investigated whether there was an interaction between urine phthalate metabolite levels and FLG genotype in children aged 3 years in the aetiology of AD (n = 453). They studied four phthalates, monoethyl phthalate, monobutyl phthalate (MBP), monobenzyl pthathlate (MBzP) and mono(2ethyl-5-hydroxyhexyl) phthalate, which they classified as lower or higher levels in relation to the median. They reported evidence for an interaction between the P478S genotype TT and phthalates MBP (P = 0Á015) and MBzP (P = 0Á018) 35 and increased AD risk; however, they did not replicate their findings or perform corrections for multiple testing. 35 A similar interaction was investigated by Ait Bamai et al. who assessed seven phthalates found in household dust, and 11 phosphorus flame retardants. They found evidence for an interaction between FLG loss-of-functions and diisononyl phthalate (P = 0Á039). 36 The group also reported a nonsignificant negative dose-response relationship among children with FLG loss-of-function mutation(s) in a categorical model (first quartile compared with fourth quartile, P for trend = 0Á087). This analysis was undertaken on a sample size of five children with AD and FLG null genotypes and the researchers did not correct for multiple testing. 36

Quality of studies
Most studies included unselected cases from the general population or cohorts, and controls were selected from the same population as those with AD. FLG genotype was assessed using accepted methodology (Table S3; see Supporting Information). The timing and method of assessment of environmental exposures was variable; studies may be vulnerable to reverse causality owing to exposure status being assessed after AD onset. Participants in each study were of homogeneous ethnicity and analyses were adjusted for age. Details of confounder adjustment were missing in five of 12 included studies (Table S3; see Supporting Information). Studies varied in their presentation of interaction results, with some providing only a P-value or statement of statistical significance and others also providing effect estimates and CIs across strata and a P-value for the interaction term or statement of statistical significance (Table S4; see Supporting Information). None of the included studies adjusted for multiple testing and only Berents et al. reported power calculations. 33 The results of the studies supported their conclusions; however, they were underpowered and therefore the results of these studies must be interpreted with caution. Under reasonable assumptions on the magnitude of main effects, prevalence of AD, environmental exposure and SNP allele frequency, post hoc power calculations (Table S2; see Supporting Information) indicate that the sample size required to detect an interaction with OR~2Á0 is approximately 5000 individuals, whereas a sample size of approximately 63 000 individuals is required to detect a more modest GEI effect with OR 1Á2.

Discussion
Our results highlight important challenges when studying GEIs in the aetiology of AD. We identified only 12 articles that reported FLG-environment interactions in AD. Our initial search strategy returned many results, but the majority were excluded because they lacked information essential to the review. Evidence was found for interactions between FLG genotype and breastfeeding duration, older siblings, phthalate exposure in household dust and urine phthalate metabolite levels, early-life exposure to cats, and water hardness. All interactions increased the risk of AD apart from prolonged breastfeeding, which decreased the risk. Owing to very limited evidence in support of these interactions, small numbers and lack of replication (one study undertook replication) it is difficult to interpret the results, and findings must be interpreted with caution. Table 2 shows our suggestions of the components required for a rigorous GEI study, which may improve conclusions in future studies.
Our review has several strengths. A detailed search strategy was used to identify all relevant papers. Screening and data extraction were carried out in duplicate, with secondary resolution of conflicts, reducing the possibility of introducing bias by systematically selecting certain papers. The majority (10 of 12, 83%) of studies used data from cohort studies, thus they were able to consider temporality, and researchers mostly measured the outcome 'AD' using validated criteria.
Our findings should be considered in light of some limitations. Many studies were excluded from the review as they did not specifically test for GEI in their analyses. Studies were also excluded if they measured indirect outcomes of AD by examining IgE levels, transepidermal water loss, or skin-prick tests, which are not measures of the outcome (AD), but responses to exposures. Many of the included studies performed GEI analysis as a secondary analysis, e.g. Berents et al., 33 meaning they did not aim to have sufficient power to assess GEIs; hence, the importance of measuring GEIs as primary outcome ( Table 2). We were unable to evaluate the risk of reporting biases formally, so we cannot rule out the possibility that studies which found nonsignificant interactions yet failed to report such results are missing from our review. Many of the studies included in this review relied on population data and there may be heterogeneity in outcome definition (Table 2). 38 The predefined scope of this review was to investigate FLG-environment interactions, which by definition excluded the study of effects within populations where FLG null mutations are not prevalent or have not been identified. 7,39 Many large genetic studies have been conducted within populations of European ancestry, where FLG null mutations are prevalent. Ongoing work to increase diversity in genetic research 40 will allow future investigations of GEI in populations of all ethnicities (Table 2).
Heterogeneity in methodology between published studies and the limited number of studies assessing the same exposure precluded meta-analysis or formal assessment of publication bias. None of the included studies reported correcting for multiple testing; hence, interaction effects could be a result of chance. It is unclear how many studies had predefined hypotheses, which risks introducing reporting bias. Replication of findings was limited and in some cases where two studies investigated the same interaction, discordant findings were seen, such as Ziyab et al. (FLG genotype and breastfeeding) and Henderson et al. 28,30 One reason for the limited evidence for GEI and lack of replication is lack of statistical power. Detailed review showed that the number of individuals on which the interaction analysis was based (i.e. cases with both exposure and FLG null genotype) was small; hence, included studies were likely to be underpowered. The number of individuals with both a FLG loss-of-function mutation and exposure to the specific environmental factor was not always specified, but in those studies that did specify this number, it ranged from N = 5 to N = 167. 24,28 In complex diseases such as AD, where the main genetic effect sizes are small, a large sample size is necessary to detect small interaction effects. 41 Researchers need to utilize sufficiently large sample sizes to detect GEIs, and generally investigators should demonstrate that their sample has adequate power to detect an interaction effect. 42 Even in cases where meta-analysis across studies is possible, results are not always meaningful owing to variable measurement of environmental exposures. 41 Studies of GEI face inherent challenges in attempting to gain a full understanding of interactions because of the difficulty in uniformly measuring the environmental parameter, which in turn limits the understanding of the underlying disease mechanism. 43 The difficulty in measuring exposure in GEI studies in AD is shown in the Wang et al. study, which tested for an interaction between phthalate exposure and FLG genotype. To measure exposure, phthalate metabolite levels were measured using urine samples. 36 This is not a direct measure of the exposure; therefore, we questioned whether it should be included in the review. It provides only a moderate prediction of exposure owing to the short half-life and rapid excretion of phthalates leading to considerable day-to-day variation. 36 Other studies used different methods, such as questionnaires, to derive environmental exposure data retrospectively; this could introduce recall bias. 27 Variation in the timing of environmental exposure is important in terms of influencing subsequent disease risk, as timing of exposure may not be accurately measured with methods such as infrequent questionnaires. 44 Using robust validated measures of exposure reduces variation and aids reproducibility of results (Table 2). For some of these exposures it is easy to hypothesize a biological explanation as to why people with FLG haploinsufficiency might have different responses; for example, pet exposure and older siblings could act via microbial exposure, as proposed by the hygiene hypothesis. 45 With other possible interactions such as urine phthalate metabolites, it is harder to hypothesize plausible mechanisms. Cohort studies may be vulnerable to reverse causality when assessing early-life exposures, as, although outcomes were measured after the exposure in the majority of studies, there remains a possibility that early signs of AD, or the presence of older siblings with AD, influenced the behaviour of parents who subsequently modified the exposure.
GEIs are widely viewed as important in the aetiology of AD. However, the limited evidence and lack of power of published studies to detect GEI effects, as indicated by the sample-size calculations we carried out, highlights the importance of further research. Such research is needed to test for replication of interactions reported to date (Table 1) using larger sample sizes. Furthermore, unexplored GEIs may also warrant investigation, including genetic risk variants in addition to FLG lossof-function mutations. The Early Genetics and Lifecourse Epidemiology consortium is investigating possible GEIs with selected SNPs associated with AD. Our recommendations for future studies of GEIs can be shown in Table 2, which would improve the quality of evidence and enable us to draw more robust conclusions about the nature of GEIs. Together this work will improve understanding of GEI in the aetiology of AD, which will help to inform both public health and individual lifestyle decisions. manuscript. L. Palla contributed to the design of the study, interpretation of analysed data and statistical calculations, and critically reviewed the manuscript. C.F. contributed to the design of the study and critically reviewed the manuscript. R.B.W. contributed to the design of the study and critically reviewed the manuscript. A.D.I. refereed the data extraction and critically reviewed the manuscript. T.M., A.R., H.C.W. and N.R. critically reviewed the manuscript. S.J.B., L. Paternoster and S.M.L. contributed to the design of the study, developed the search strategy and the PROSPERO protocol, performed, supervised and refereed data extraction, reviewed data analysis, contributed to drafting the manuscript and critically reviewed the final manuscript. All authors approved the final manuscript as submitted. Medicine, London, WC1E 7HT, U.K.) had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. The corresponding author and guarantor (H.B.) affirms that this manuscript is an honest, accurate and transparent account of the study being reported; that no important aspects of the study have been omitted and that any discrepancies from the study as planned (and, if relevant, registered) have been explained.

Data sharing
Our OVID MEDLINE search strategy has been registered on PROSPERO (https://www.crd.york.ac.uk/prospero/) and all data are publicly available.

Supporting Information
Additional Supporting Information may be found in the online version of this article at the publisher's website: Methods S1 Search strategies. Methods S2 Method of screening and data extraction of relevant studies.
Table S1 Inclusion and exclusion criteria. Table S2 Sample size required to detect gene-environment interaction. 46,47 Table S3 Characteristics of studies of loss-of-function mutations in FLG and gene-environment interaction in atopic dermatitis.
Table S4 Studies examining the interactions between FLG mutations and environmental exposure and risk of atopic dermatitis.
Powerpoint S1 Journal Club Slide Set.