Identification of potential causal metabolites associated with atopic dermatitis

Abstract

Atopic dermatitis is a chronically recurrent dermatologic disease affected by complex pathophysiology with limited therapeutic options. To identify promising biomarkers for atopic dermatitis, we conducted a Mendelian randomization (MR) study to systematically screen blood metabolome for potential causal mediators of atopic dermatitis and further predict target-mediated side effects. We selected 128 unique blood metabolites from three European-descent metabolome genome-wide association studies (GWASs) with a total of 147 827 participants. Atopic dermatitis dataset originated from a large-scale GWAS including 10 788 cases and 30 047 controls of European ancestry. MR analyses were performed to estimate the associations of blood metabolites with atopic dermatitis. We then applied a phenome-wide MR analysis to ascertain potential on-target side effects of metabolite intervention. Three metabolites were identified as potential causal mediators for atopic dermatitis, including docosahexaenoic acid (odds ratio [OR], 0.87; 95% confidence interval [CI], 0.81–0.94; P = 3.45 × 10⁻⁴), arachidonate (OR, 0.30; 95% CI, 0.17–0.53; P = 4.09 × 10⁻⁵) and 1-arachidonoylglycerophosphoethanolamine (1-arachidonoyl-GPE) (OR, 0.25; 95% CI, 0.12–0.53; P = 2.58 × 10⁻⁴). In the phenome-wide MR analysis, docosahexaenoic acid and arachidonate were also identified to have beneficial or detrimental effects on multiple diseases beyond atopic dermatitis, respectively. No adverse side effects were found for 1-arachidonoyl-GPE. In this systematic MR study, docosahexaenoic acid, arachidonate and 1-arachidonoyl-GPE were identified as potential causal and beneficial mediators in the development of atopic dermatitis. Side-effect profiles were characterized to help inform drug target prioritization, and 1-arachidonoyl-GPE was a promising target for prevention and treatment of atopic dermatitis with no predicted adverse side effects.

Introduction

Eczema, a chronically relapsing inflammatory skin disease, is characterized by intense itching and recurrent eczematous lesions, and atopic dermatitis is the most common type of eczema. In 2020, atopic dermatitis affected up to 20% of children and 10% of adults in high-income countries, and the number of patients was expected to continue to increase globally (1). In the United States alone, atopic dermatitis results in up to $5.3 billion economic losses annually (2). In view of the high burden of atopic dermatitis, previous epidemiological studies have investigated possible mediators for atopic dermatitis, but no specific biomarkers have been identified so far (3). In addition, the high attrition rates associated with drug development prompt the need for an accurate exploration of potential biomarkers prior to clinical testing (4). Recently, Nelson et al. reported that a metabolite drug target linking with disease supported by genetic association was twice as likely to reach market approval (4). Recent advancements in mass spectrometry allow genome-wide association studies (GWAS) to uncover genetic determinants for more comprehensive interrogation of the metabolome (5–7). Promising biomarkers for atopic dermatitis can be accurately identified by integrating genetic and metabolome data followed by the Mendelian randomization (MR) analysis, an emerging method using genetic variants associated with exposures as instrumental variables to assess the causality for the associations between exposures and outcomes without confounding and reverse-causality biases (8).

MR design has been previously used to explore the associations between certain factors and the risk of atopic dermatitis, including vitamin D (9), interleukin (IL)-18 (10), alcohol consumption (11) and major depressive disorder (12). However, there has been no systematic scan of the human metabolome for promising causal mediators of atopic dermatitis. Additionally, before being tested in clinical trials, a phenome-wide MR (Phe-MR) analysis for the side effects of potential drug targets can be applied to reveal unanticipated adverse effects and opportunities for drug repurposing (13). Herein, we conducted a two-stage MR study to identify promising biomarkers for atopic dermatitis. First, we systematically screened 128 circulating metabolites to identify promising mediators of atopic dermatitis. Subsequently, we conducted the Phe-MR analysis of 679 disease traits to anticipate target-mediated side effects associated with metabolite intervention for a comprehensive appraisal of their clinical safety.

Results

Strength of genetic instruments for blood metabolites

Overall, 128 unique blood metabolites were analyzed in the present study (Fig. 1), and a complete list of specific metabolites was shown in Supplementary Material, Table S2. The F statistics for the genetic instruments of blood metabolites ranged 32–353, suggesting that there is little instrument bias in this study (Supplementary Material, Table S2).

Figure 1

Conceptual framework of two-stage MR study. The study consists of a two-stage design that employs MR at all stages. First, we assessed the causality for the associations between 128 blood metabolites and the risk of atopic dermatitis. Second, we investigated a broad spectrum of side effects associated with targeting identified metabolites in 679 non-atopic dermatitis diseases. Among these, each disease belongs to one of 16 different ICD-9 chapters. At each stage, we adopted a Bonferroni-corrected P-value threshold accounting for both the number of metabolites and diseases analyzed.

Screening the blood metabolome for potential causal mediators of atopic dermatitis

Figure 2 displayed the associations between 128 blood metabolites and the risk of atopic dermatitis in the main inverse-variance weighted (IVW) MR analysis, and detailed results were presented in Supplementary Material, Table S5. Among these metabolites, genetically determined high docosahexaenoic acid [odds ratio (OR) per 1-standard deviation (SD) increase, 0.87; 95% confidence interval (CI), 0.81–0.94; P = 3.45 × 10⁻⁴], arachidonate (OR per 1-SD increase, 0.30; 95% CI, 0.17–0.53; P = 4.09 × 10⁻⁵) and 1-arachidonoylglycerophosphoethanolamine (1-arachidonoyl-GPE) (OR per 1-SD increase, 0.25; 95% CI, 0.12–0.53; P = 2.58 × 10⁻⁴) were associated with low risk of atopic dermatitis (Table 2).

Figure 2

Circular Manhattan plot displaying the associations between blood metabolites and the risk of atopic dermatitis in the IVW MR analysis. The dashed line represents the Bonferroni-corrected significance threshold (P < 3.91 × 10⁻⁴), and the labels are provided for significant metabolites. The 128 blood metabolites are grouped and color coded by super-pathway listed in Supplementary Material, Table S2. The detailed results for the associations between blood metabolites and atopic dermatitis by IVW MR analysis are presented in Supplementary Material, Table S5.

In the sensitivity analyses, results from the weighted median method, the MR-Robust Adjusted Profile Scoring (MR-RAPS) and the maximum likelihood method also indicated significant associations of genetically determined high docosahexaenoic acid, arachidonate and 1-arachidonoyl-GPE with the decreased risk of atopic dermatitis, whereas these associations mildly attenuated in the MR pleiotropy residual sum and outlier (MR-PRESSO) analysis (Table 2; Supplementary Material, Table S6). In light of the intercept for the MR-Egger regression, we did not find evidence against directional pleiotropy for the associations of these three metabolites with risk of atopic dermatitis (all P > 0.05). The leave-one-out analyses showed that there was no individual genetic variant substantially driving the associations of docosahexaenoic acid, arachidonate and 1-arachidonoyl-GPE with atopic dermatitis (Supplementary Material, Figs S1–S3). In the bidirectional MR analysis, we found no significant associations of genetically predicted atopic dermatitis with docosahexaenoic acid, arachidonate and 1-arachidonoyl-GPE [all P > 0.017 (Bonferroni-corrected significance threshold of P = 0.05/3 = 0.017)] (Supplementary Material, Table S7).

Phe-MR analysis for the associations between identified metabolites and 679 diseases

We further performed the Phe-MR analysis to systematically assess the effects of the identified atopic dermatitis metabolites on the risks of 679 non-atopic dermatitis diseases to explore their potential side-effect profiles. Unlike the previous MR analyses, Phe-MR results were standardized to a 10% reduction in atopic dermatitis risk mediated by targeting a given metabolite. As such, resultant associations can be interpreted as concomitant side effects expected to arise if each metabolite is used to prevent or treat atopic dermatitis. In the Phe-MR analysis using IVW method, 47 associations reached a Bonferroni-corrected significance threshold of P = 2.45 × 10⁻⁵ [0.05/2037 (3 metabolites*679 diseases)] (Supplementary Material, Tables S8–S10; Supplementary Material, Figs S4–S6). In the sensitivity analyses with different MR methods, three significant disease associations for docosahexaenoic acid and 10 for arachidonate were identified, whereas none was observed for 1-arachidonoyl-GPE (Supplementary Material, Table S11), and a mild attenuation in several significant associations was observed in the MR-PRESSO analysis. In addition, the MR-Egger regression indicated no directional pleiotropy for these significant associations, except for the association of docosahexaenoic acid with benign neoplasm of colon (Supplementary Material, Table S11).

Collectively, 12 significant associations were identified between two of three metabolites (docosahexaenoic acid and arachidonate) and multiple diseases (Table 3; Fig. 3). Grouping individual phenotypes on the basis of disease category, digestive system was the most commonly affected system for docosahexaenoic acid and arachidonate (Supplementary Material, Table S12). The most significant disease associations for docosahexaenoic acid and arachidonate were cholelithiasis (OR per 10% reduction in atopic dermatitis risk, 0.83; 95% CI, 0.80–0.87; P = 2.57 × 10⁻¹⁵) and benign neoplasm of colon (OR per 10% reduction in atopic dermatitis risk, 1.13; 95% CI, 1.10–1.17; P = 3.74 × 10⁻¹⁶), respectively (Table 3).

Discussion

To our knowledge, this is the first metabolomics study to systematically explore promising dug targets for atopic dermatitis using an MR approach. Among 128 blood metabolites, we identified three potential causal mediators for atopic dermatitis, including docosahexaenoic acid, arachidonate and 1-arachidonoyl-GPE. Phe-MR analysis was further applied to predict on-target side effects associated with potential atopic dermatitis treatment via intervention of identified metabolites. Beyond atopic dermatitis, we found that docosahexaenoic acid had beneficial effects on the risk of two digestive diseases, whereas arachidonate mediated the risk of 10 diseases. In addition, 1-arachidonoyl-GPE appeared to be not implicated in the development of non-atopic dermatitis diseases.

Table 1

Characteristics of GWASs on the metabolome used for genetic instrument selection

Reference	Cohort(s)	Cohort description	Sample size	Number of metabolic traits	Biomarker assay	Blood fraction tested	Ethnicity	Year
Shin et al.	KORA; TwinsUK	Population-based cohort; Population-based twin study	7824	453	Metabolon	Serum; Plasma	German; British	2014
Kettunen et al.	EGCUT; ERF; FTC; FR97; COROGENE; GenMets; HBCS; KORA; LLS; NTR; NFBC 1966; PredictCVD; PROTE; YFS	Population-based cohort; Family-based; Population-based twin study; Population-based cohort; Case–control study (only controls used); Case–control study; Birth cohort; Population-based cohort; Family-based; Population-based twin study; Birth cohort; Cohort study; Population-based; Follow-up study in children	24 925	123	NMR	Serum; Plasma	Estonian; Dutch; Finnish; Finnish; Finnish; Finnish; Finnish; German; Dutch; Dutch; Finnish; Finnish; Estonian; Finnish	2016
Borges et al.	UK Biobank	Population-based cohort	115 078	249	Nightingale Health	Blood	British	2020

Abbreviations: COROGENE, Genetic Predisposition of Coronary Heart Disease in Patients Verified with Coronary Angiogram; EGCUT, Estonian Genome Center of University of Tartu Cohort; ERF, Erasmus Rucphen Family Study; FR97, a subsample of FINRISK 1997; FTC, Finnish Twin Cohort; GenMets, Genetics of METabolic Syndrome; HBCS, Helsinki Birth Cohort Study; KORA, Kooperative Health Research in the Region of Augsburg; LLS, Leiden Longevity Study; NFBC 1966, Northern Finland Birth Cohort 1966; NMR, nuclear magnetic resonance; NTR, Netherlands Twin Register; PredictCVD, FINRISK subsample of incident cardiovascular cases and controls; PROTE, EGCUT sub-cohort; YFS, The Cardiovascular Risk in Young Finns Study.

Table 2

MR analyses for blood metabolites potentially causally associated with atopic dermatitis risk

Metabolite	SNPs	IVW		Weighted median		MR-RAPS		Maximum likelihood		MR-PRESSO			MR-Egger
		OR (95% CI)	P-value	OR (95% CI)	P-value	OR (95% CI)	P-value	OR (95% CI)	P-value	SNPsa	OR (95% CI)	P-value	OR (95% CI)	P-value	Pintercept
Docosahexaenoic acid	161	0.87 (0.81–0.94)	3.45 × 10−4	0.81 (0.73–0.90)	1.47 × 10−4	0.87 (0.81–0.93)	3.50 × 10−5	0.87 (0.82–0.93)	4.26 × 10−5	160	0.88 (0.81–0.94)	5.27 × 10−4	0.83 (0.74–0.94)	2.90 × 10−3	0.36
Arachidonate	5	0.30 (0.17–0.53)	4.09 × 10−5	0.27 (0.14–0.50)	4.79 × 10−5	0.30 (0.16–0.54)	6.04 × 10−5	0.30 (0.17–0.54)	5.25 × 10−5	5	0.30 (0.13–0.68)	4.20 × 10−3	0.15 (0.02–0.96)	0.04	0.44
1-arachidonoyl-GPE	4	0.25 (0.12–0.53)	2.58 × 10−4	0.17 (0.07–0.42)	1.08 × 10−4	0.25 (0.11–0.53)	3.73 × 10−4	0.25 (0.12–0.53)	3.24 × 10−4	4	0.25 (0.07–0.86)	0.03	0.16 (0.01–1.70)	0.13	0.69

ORs with their 95% CIs represent the association estimates with the risk of atopic dermatitis of per 1-SD increase in docosahexaenoic acid, arachidonate and 1-arachidonoyl-GPE levels, respectively.

^aRemaining genetic instruments after excluding outlying SNPs.

Table 3

Phe-MR analyses for potential causal associations of docosahexaenoic acid and arachidonate with the risk of multiple non-atopic dermatitis diseases

Phe Code	Outcome	Disease chapter	SNPs	IVW		Weighted median		MR-RAPS		Maximum likelihood		MR-PRESSO			MR-Egger
				OR (95% CI)	P-value	OR (95% CI)	P-value	OR (95% CI)	P-value	OR (95% CI)	P-value	SNPs	OR (95% CI)	P-value	OR (95% CI)	P-value	Pintercept
Docosahexaenoic acid
574	Cholelithiasis and cholecystitis	digestive	152	0.85 (0.82–0.89)	6.61 × 10−13	0.89 (0.86–0.93)	1.89 × 10−8	0.85 (0.82–0.88)	2.28 × 10−23	0.85 (0.83–0.88)	1.76 × 10−22	150a	0.85 (0.82–0.89)	1.31 × 10−13	0.88 (0.82–0.95)	4.37 × 10−4	0.25
574.1	Cholelithiasis	digestive	152	0.83 (0.80–0.87)	2.57 × 10−15	0.87 (0.84–0.91)	1.49 × 10−10	0.83 (0.80–0.86)	4.35 × 10−26	0.83 (0.81–0.86)	3.78 × 10−25	150a	0.84 (0.80–0.87)	1.36 × 10−15	0.84 (0.78–0.91)	6.07 × 10−6	0.69
Arachidonate
208	Benign neoplasm of colon	neoplasms	5	1.13 (1.10–1.17)	3.74 × 10−16	1.12 (1.08–1.15)	7.70 × 10−11	1.13 (1.10–1.17)	2.24 × 10−14	1.13 (1.10–1.17)	1.02 × 10−14	5	1.13 (1.06–1.21)	2.66 × 10−4	1.08 (0.98–1.19)	0.12	0.30
244	Hypothyroidism	endocrine/ metabolic	5	1.11 (1.07–1.15)	1.51 × 10−8	1.10 (1.06–1.14)	6.00 × 10−7	1.11 (1.07–1.15)	4.62 × 10−8	1.11 (1.07–1.15)	3.35 × 10−8	5	1.11 (1.04–1.18)	2.23 × 10−3	1.14 (1.02–1.28)	0.02	0.56
244.4	Hypothyroidism NOS	endocrine/ metabolic	5	1.11 (1.07–1.15)	2.46 × 10−8	1.10 (1.06–1.15)	6.87 × 10−7	1.11 (1.07–1.15)	7.26 × 10−8	1.11 (1.07–1.15)	5.22 × 10−8	5	1.11 (1.04–1.18)	2.03 × 10−3	1.15 (1.02–1.29)	0.02	0.52
371.3	Inflammation of eyelids	sense organs	5	1.21 (1.11–1.31)	5.28 × 10−6	1.21 (1.11–1.31)	2.35 × 10−5	1.21 (1.11–1.32)	9.12 × 10−6	1.21 (1.11–1.32)	7.72 × 10−6	5	1.21 (1.06–1.38)	4.67 × 10−3	1.34 (1.03–1.74)	0.03	0.43
459	Other disorders of circulatory system	circulatory system	5	1.10 (1.07–1.14)	2.26 × 10−9	1.08 (1.04–1.12)	9.15 × 10−6	1.10 (1.07–1.14)	6.72 × 10−9	1.10 (1.07–1.14)	6.04 × 10−9	5	1.10 (1.02–1.19)	0.01	1.03 (0.89–1.19)	0.67	0.32
459.9	Circulatory disease NEC	circulatory system	5	1.10 (1.07–1.14)	6.60 × 10−9	1.08 (1.04–1.12)	1.34 × 10−5	1.10 (1.07–1.14)	1.72 × 10−8	1.10 (1.07–1.14)	1.57 × 10−8	5	1.10 (1.02–1.19)	0.01	1.03 (0.89–1.19)	0.69	0.34
495	Asthma	respiratory	5	1.09 (1.06–1.12)	3.49 × 10−10	1.08 (1.05–1.11)	2.33 × 10−7	1.09 (1.06–1.12)	1.43 × 10−9	1.09 (1.06–1.12)	1.20 × 10−9	5	1.09 (1.03–1.16)	5.13 × 10−3	1.08 (0.97–1.21)	0.18	0.86
471	Nasal polyps	respiratory	5	1.22 (1.14–1.31)	3.79 × 10−8	1.20 (1.11–1.30)	2.73 × 10−6	1.22 (1.13–1.31)	9.54 × 10−8	1.22 (1.14–1.31)	7.80 × 10−8	5	1.22 (1.06–1.40)	5.28 × 10−3	1.24 (0.95–1.61)	0.11	0.91
574	Cholelithiasis and cholecystitis	digestive	5	0.91 (0.88–0.94)	7.59 × 10−9	0.91 (0.88–0.95)	4.42 × 10−7	0.91 (0.88–0.94)	2.76 × 10−8	0.91 (0.88–0.94)	2.01 × 10−8	5	0.91 (0.86–0.96)	8.03 × 10−4	0.87 (0.79–0.97)	0.01	0.48
574.1	Cholelithiasis	digestive	5	0.89 (0.85–0.92)	2.11 × 10−11	0.89 (0.86–0.93)	7.85 × 10−9	0.89 (0.85–0.92)	1.56 × 10−10	0.88 (0.85–0.92)	1.14 × 10−10	5	0.89 (0.82–0.95)	1.26 × 10−3	0.85 (0.76–0.95)	4.72 × 10−3	0.44

ORs with their 95% CIs represent the effect estimates on the risk of multiple non-atopic dermatitis diseases of per 10% reduction in risk for atopic dermatitis by targeting docosahexaenoic acid and arachidonate, respectively.

^aRemaining genetic instruments after excluding outlying SNPs.

Docosahexaenoic acid, a long-chain n-3 poly-unsaturated fatty acid (PUFA) possessing immunomodulatory and antioxidant properties (14), is effective in reducing pro-inflammatory cytokines (e.g. IL-6 and IL-18) and in turn inhibiting inflammation (15,16). Previous observational studies showed that increased blood docosahexaenoic acid levels were associated with lower morbidity of eczema, including atopic dermatitis (17,18). Several randomized controlled trials also suggested that docosahexaenoic acid supplementation could reduce the risk of atopic dermatitis in infants or children (19–21). On the basis of atopic dermatitis GWAS with 40 835 European children and adult participants, we found that genetically predicted blood docosahexaenoic acid levels were inversely associated with risk of atopic dermatitis, providing an evidence on the causality for the association between them. Additionally, further Phe-MR analysis confirmed previously reported beneficial role of docosahexaenoic acid in the development of cholelithiasis and cholecystitis (22). However, given that our previous MR study with different analytical models suggested strong on-target side effects for interventions of docosahexaenoic acid (23), further studies are needed to determine the safety and feasibility of docosahexaenoic acid supplementation.

1-arachidonoyl-GPE is a novel metabolite related to the endocannabinoid system, and can generate arachidonoylethanolamine (anandamide) by cleavage of glycerophosphoethanolamines with the mediation of phosphodiesterase (24). Of note, anandamide has been shown to able to play an important role in modulating inflammation in the skin of mice by activating cannabinoid 1 and cannabinoid 2 with high potency (25). Recent experimental study showed that anandamide could suppress the production and release of signature T helper cells polarizing cytokines and in turn alleviate T cell-dependent inflammatory skin lesion (26). Furthermore, inhabitation of anandamide hydrolase (fatty acid amides hydrolase) activity was reported to be able to reduce inflammation (27). In the present study, we found that genetically determined 1-arachidonoyl-GPE had a beneficial effect on the risk of atopic dermatitis with no adverse side effects predicted, suggesting that 1-arachidonoyl-GPE can serve as a novel and promising drug target for atopic dermatitis.

Arachidonate, also called arachidonic acid, is a n-6 PUFA implicated in the diminished influx of inflammatory cell (28) and the stimulation of neutrophil clearance (29). Existing evidence suggested that combined supplementation of docosahexaenoic acid and arachidonate could improve the Th2-skewed immune response (30). On the other hand, some sphingolipids metabolites, including arachidonate, that produced by Roseomonas mucosa were natural antimicrobial agents (31,32). However, previous epidemiological studies investigating the role of arachidonate in the development of eczema yielded conflicting results (33–35). Some cross-sectional studies conducted in Japan (33) and Spain (34) suggested an inverse association between arachidonate and risk of eczema in children, whereas an analysis using data of Osaka Maternal and Child Health Study reported that maternal intake of n-6 PUFA was an independent risk factor for eczema in infants (35). Our findings supported an inverse association between arachidonate and risk of atopic dermatitis from the perspective of causality. However, our Phe-MR analysis showed that arachidonate had adverse effects on multiple diseases, which were consistent with findings from previous observational studies (36).

Figure 3

Potential on-target side effects associated with docosahexaenoic acid, arachidonate and 1-arachidonoyl-GPE intervention revealed by phenome-wide MR analysis. ORs with their 95% CIs represent the effect estimates on the risk of multiple non-atopic dermatitis diseases of per 10% reduction in risk for atopic dermatitis by targeting docosahexaenoic acid, arachidonate and 1-arachidonoyl-GPE, respectively. Associations above the horizontal black midline represent deleterious side effects. On the other hand, associations below the horizontal black midline represent beneficial side effects. The horizontal short dash line (OR=1.10) represents the point at which decreased atopic dermatitis risk is counterbalanced by an equal increase in decrease risk.

Interestingly, as an allergic disease that shares the pathobiology with atopic dermatitis (37,38), asthma risk was predicted to increase in reflect to the hypothetical intervention of arachidonate against atopic dermatitis in the Phe-MR analysis. Several possible mechanisms could be referenced to explain the discordance between our findings on atopic dermatitis and asthma. In the self-limited course of inflammation, some arachidonate-derived lipoxins, such as lipoxins A₄ and B₄, were reported to be able to stimulate the non-phlogistic clearance of apoptotic neutrophils through macrophages and attract monocytes for wound healing (28,29). However, some pro-inflammatory mediators (e.g. prostaglandin E2 and leukotriene B4) in the initial phase of inflammation are also derived by arachidonate (28). Of note, the increased atopic dermatitis risk because of such pro-inflammatory effects might be partially counteracted by arachidonate-mediated improvement of skin microbiome (31,32). In addition, several confounding factors or bidirectional relationships across metabolites, atopic dermatitis and asthma can also contribute to this contradictory result. For example, age and genetic factors are well-known influencing factors in the pathologies of atopic dermatitis (1) and asthma (39), whereas there was a discrepancy in age distribution and genetic background among the participants of the GWASs on atopic dermatitis and asthma. Moreover, the metabolic alterations mediated by atopic dermatitis (40) and asthma (41) may also be responsible for our contradictory findings. Further studies are warranted to address the underlying mechanisms. Collectively, our findings suggested that arachidonate supplementation for atopic dermatitis prevention and treatment should be applied under weighting advantages and disadvantage of arachidonate.

Our study has important public health implications in better characterizing the biomarkers for atopic dermatitis and further identifying potential drug target against atopic dermatitis. Atopic dermatitis is a chronically recurrent disease, and its prevention and treatment rely on long-term drug intervention (1). To date, the etiology of atopic dermatitis remains unclear, and so symptomatic treatment is the main therapeutic method for atopic dermatitis (42). Among these, glucocorticosteroids and anti-inflammatory agents are the most commonly used drugs in the treatment of atopic dermatitis (42), whereas continuous regimen for these drugs is not recommended given a high rate of adverse side effects (43,44). On the basis of limited therapeutic options of atopic dermatitis and our findings, it is of clinical interest for dermatologist to consider preventing and treating atopic dermatitis through supplementing 1-arachidonoyl-GPE. Further clinical trials are needed to confirm the effective dose of 1-arachidonoyl-GPE in atopic dermatitis management.

Our study has several strengths. Firstly, this is the first systematic MR study to assess the causality for the associations between human blood metabolome (128 metabolites) and the risk of atopic dermatitis. Secondly, this MR study was conducted on the basis of large-scale GWASs, which enabled us to make a valid causal inference with a high statistical power. Finally, besides identifying potential causal mediators for atopic dermatitis, we also adopted the Phe-MR analysis to estimate the associations of identified metabolites with multiple non-atopic dermatitis diseases for comprehensively predicting the on-target side effects of metabolites.

However, our study also suffers from several limitations. First, it is difficult to completely exclude the bias from invalid instruments and pleiotropy in the MR study. However, in the present study, different MR sensitivity analyses yielded similar results as the main analysis and MR-Egger regression suggested little directional pleiotropy for these associations, indicating that the influence of invalid instruments and pleiotropy was minimal. Second, GWASs on the metabolome included in this MR study varied in metabolite quantification technologies, which may lead to technology-specific bias. Third, the atopic dermatitis GWAS dataset available in the Integrative Epidemiology Unit (IEU) GWAS database (https://gwas.mrcieu.ac.uk/datasets/ieu-a-996/) excluded the participants from the 23andMe study, so there might be potential selection bias in the present study. Forth, the use of lipid-lowering agents and medical history of hyperlipidemia might influence the effects of lipids (including docosahexaenoic acid, arachidonate and 1-arachidonoyl-GPE) on atopic dermatitis. However, the patients with such medication history and medical history were not removed from the UK Biobank data, so there was potential confounding bias in our MR study. Further studies are warranted to examine the associations between blood metabolites, especially lipids and atopic dermatitis among individuals without the aforementioned medication history and medical history. Fifth, the participants included in our study were of European descent, which might limit the reliability when extrapolating our findings to non-European populations. However, this restriction minimized population stratification bias, and further studies in populations with different ethnic background are needed to confirm our findings. Sixth, our Phe-MR analyses on the basis of GWASs with current sample size might not be sufficiently robust to draw conclusions about the causal associations between blood metabolites and non-atopic dermatitis diseases. A more comprehensive large-scale GWAS dataset is needed to address the robustness for causal inference. Further experimental and epidemiological validations are also required to overcome this limitation. Finally, PheCode system was relied on hospital diagnoses, of which disease traits with low rates of hospital admission might be poorly represented.

Conclusion

In summary, in this systematic MR analysis of blood metabolome, docosahexaenoic acid, arachidonate and 1-arachidonoyl-GPE were identified to be potential causal mediators for atopic dermatitis. Phe-MR analysis suggested that 1-arachidonoyl-GPE was a particularly promising drug target for atopic dermatitis without predicted adverse side effects.

Materials and Methods

Study design

We sought to identify promising biomarkers for atopic dermatitis by MR analysis of the blood metabolome (Fig. 1). The summary-level data utilized in the present study were derived from publicly available GWASs of European ancestry (5–7,45,46). The protocol and data collection were approved by the ethics committee of the original GWASs, and written informed consent was obtained from each participant before data collection.

Data source for blood metabolome and atopic dermatitis

We obtained summary-level data of single nucleotide polymorphisms (SNPs) associated with human metabolome as genetic instruments from three large-scale European ancestry-based GWASs with a combined total of 147 827 participants (Table 1) (5–7). Briefly, Shin et al. (5) analyzed 453 metabolic traits in 7824 participants with ⁓3 million SNPs from two cohorts via Metabolon assay; Kettunen et al. (6) analyzed 123 metabolic traits in 24 925 participants with ⁓12 million SNPs from 14 cohorts via nuclear magnetic resonance assay and Borges et al. (7) analyzed 249 metabolic traits in 115 078 participants with ⁓12 million SNPs from UK Biobank via Nightingale Health assay (Table 1). After excluding overlapping metabolites in these three metabolome GWASs, a total of 469 metabolites were retained.

Summary statistics for atopic dermatitis were derived from Early Genetics and Lifecourse Epidemiology (EAGLE) Eczema consortium’s meta-analysis of 20 population-based cohorts (45), involving 10 788 cases and 30 047 controls of European ancestry (available from IEU GWAS database: https://gwas.mrcieu.ac.uk/datasets/ieu-a-996/) (Supplementary Material, Table S1) (47,48). Genotyping and imputation steps of the 20 cohorts were described in detail in previous studies and ⁓11 million SNPs were reserved in the GWAS meta-analysis.

Genetic instruments of blood metabolites

In this MR study, SNPs that were identified to be associated with blood metabolites at genome-wide significance level (P-value < 5 × 10⁻⁸) in the published GWASs and were not in linkage disequilibrium (LD) with other SNPs (r ² < 0.1 within a clumping window of 500 kb) were used as instruments for these metabolites. When we encountered certain SNPs above the LD threshold of 0.1, the SNP with the lowest P-value for association with the metabolite was selected. For the metabolite-associated SNPs that were not available in the outcome dataset (i.e. atopic dermatitis dataset), a proxy SNP (r ² > 0.8) was selected for the MR analysis by default on the basis of a 1000 Genomes European reference panel. Subsequently, we calculated the phenotypic variance of each blood metabolite explained by the corresponding instruments using the package gtx (the ‘grs.summary’ command) in the statistical software R, and the calculation formula has been previously described in detail (49). To ensure sufficient statistical power for a valid causal inference, we removed the metabolites whose phenotypic variance explained by the included genetic variants was ˂0.5% (50). Moreover, some MR sensitivity analyses required at least three SNPs related to exposure as the genetic instrument (48), so the metabolites associated with ˂3 SNPs were excluded.

After excluding metabolites whose phenotypic variance explained by the included genetic variants was ˂0.5% or metabolites with number of associated SNPs ˂3, 341 of 469 metabolites were further excluded. Finally, a total of 128 unique blood metabolites were included in the MR analysis (Fig. 1). A simplified description of the data concerning SNPs used as instruments in this MR study is listed in Supplementary Material, Table S2, and further detailed information is available in Supplementary Material, Table S3. F statistic was used to evaluate the strength of the genetic instruments for blood metabolites. A cutoff of 10 was used to distinguish between strong and weak instruments, and a higher F statistic indicated a stronger instrument (51).

Statistical analysis

In the main analysis, we leveraged the IVW method to estimate the potential causal effects of blood metabolite levels on the risk of atopic dermatitis (52). Cochran’s Q statistic was used to determine the presence of heterogeneity among the genetic instruments used in the main analysis (53). We adopted random-effect IVW model if the heterogeneity was statistically significant; otherwise fixed-effect IVW model was used.

Given that the estimation of associations using the IVW method was subject to the biases from invalid instruments or pleiotropy, we subsequently conducted a series of sensitivity analyses to assess the robustness of our findings. First, we employed the weighted median approach, in which the MR estimates were robust when up to 50% of genetic variants were invalid (53). Second, the MR-RAPS analysis was performed because of its resilience to violations of key MR assumptions, such as horizontal pleiotropy and weak instruments (54). Third, the maximum likelihood method was applied to provide relatively reliable estimates in the presence of measurement error in the SNP-exposure effect (55). Forth, we adopted the MR-Egger regression, which enabled us to ascertain the potential directional pleiotropy via the intercept term (56). Fifth, we applied the leave-one-out analysis to test whether the associations between metabolites and atopic dermatitis were driven by an individual genetic variant (57). Sixth, the bidirectional MR method was utilized to determine whether the identified metabolites can be reversely affected by atopic dermatitis and to identify truly potential causal metabolites (58). Moreover, to identify and correct for the potential horizontal pleiotropic outliers in multi-instrument summary-level MR testing, we further performed the MR-PRESSO analysis to validate our findings (59).

Phe-MR analysis for on-target side effects of atopic dermatitis-related metabolites

We used Phe-MR analysis to evaluate the potential on-target side effects associated with hypothetical interventions that reduced atopic dermatitis burden by targeting identified metabolites. Summary statistics for 1403 disease traits were derived from Zhou et al.’s. GWAS with 28 million SNPs in the UK Biobank cohort (N = 408 961 white British participants) (46). Disease traits were defined using ‘PheCodes’, a system developed to organize International Classification of Disease (ICD) codes into phenotypic outcomes suitable for systematic genetic analysis of numerous disease traits (46,50). In the present study, sex-specific disease traits and disease traits with cases <500 were excluded because of the issues of data availability and statistical power, respectively. In addition, we selected representative phenotypes to minimize the inherent redundancy between PheCodes and in turn to improve the interpretability of results. Finally, a total of 679 non-atopic dermatitis diseases were included in the Phe-MR analysis to further characterize the on-target potential side effects of atopic dermatitis-related metabolites (Fig. 1; Supplementary Material, Table S4). Genetic instruments for atopic dermatitis-related metabolites were derived from the same GWASs as in the primary atopic dermatitis analysis (N < 147 827) (5–7). On the basis of metabolite-atopic dermatitis associations, Phe-MR findings were standardized to a change in metabolite level corresponding to a 10% reduction in atopic dermatitis risk. The results were standardized in this manner to discover the side effect of metabolite-targeted interventions for atopic dermatitis and to directly compare the magnitude and direction of the side effects.

The results were presented as ORs with 95% CIs of outcomes. In stage 1, an observed two-sided P < 3.91 × 10⁻⁴ [Bonferroni-corrected significance threshold calculated as 0.05 divided by 128 (for 128 metabolites)] was considered as statistically significant evidence for a potential causal association. A two-sided P < 0.05 was considered as suggestive evidence for potential directional pleiotropy in the MR-Egger regression method (60). All analyses were performed in R (version 3.4.3; R Development Core Team) with the packages gtx, MendelianRandomization, TwoSampleMR, ggplot2, ggrepel, grid, gridExtra, gtable, qqman, RColorBrewer and RGraphics.

Acknowledgements

We thank the investigators of EAGLE Eczema consortium, UK Biobank and the three European-descent GWASs of blood metabolome for making their results publicly available. Full acknowledgement and funding statements for each of these resources are available via the relevant cited reports.

Conflict of Interest statement. None declared.

Data Availability

All summary statistics used in this two-stage Mendelian randomization are available online from each genome-wide association study. Statistical code is available on the request by directly contacting the corresponding author (email: [email protected]).

Funding

National Natural Science Foundation of China (82103921, 82020108028); Natural Science Research Project of Jiangsu Provincial Higher Education (21KJB330006). The funders had no role in the design and conduct of the study, collection, management, analysis and interpretation of the data; preparation, review or approval of the manuscript; or decision to submit the manuscript for publication.

Author notes

Yiming Jia and Rong Wang authors contributed equally to this work.