Decoding and reconstructing disease relations between dry eye and depression: a multimodal investigation comprising meta-analysis, genetic pathways and Mendelian randomization

Graphical abstract

Highlights

• •Our meta-analysis confirmed the positive correlation between DED and DEP.
• •The genetic architecture similarity between DED-DEP was 0.19 in TW biobank and 0.109 in UK biobank.
• •Transcriptome study underlined immune activation at the transaction of DED-DEP concurrence.
• •Bi-direction DED-to-DEP and DEP-to-DED disease Causation was detected by IVW-MR analysis.
• •Disease causation inference remains valid after COJO selected LD-free IVSNPs for MR experiments.

Abstract

Introduction

The clinical presentations of dry eye disease (DED) and depression (DEP) often comanifest. However, the robustness and the mechanisms underlying this association were undetermined.

Objectives

To this end, we set up a three-segment study that employed multimodality results (meta-analysis, genome-wide association study [GWAS] and Mendelian randomization [MR]) to elucidate the association, common pathways and causality between DED and DEP.

Methods

A meta-analysis comprising 26 case−control studies was first conducted to confirm the DED-DEP association. Next, we performed a linkage disequilibrium (LD)-adjusted GWAS and targeted phenotype association study (PheWAS) in East Asian TW Biobank (TWB) and European UK Biobank (UKB) populations. Single-nucleotide polymorphisms (SNPs) were further screened for molecular interactions and common pathways at the functional gene level. To further elucidate the activated pathways in DED and DEP, a systemic transcriptome review was conducted on RNA sequencing samples from the Gene Expression Omnibus. Finally, 48 MR experiments were implemented to examine the bidirectional causation between DED and DEP.

Results

Our meta-analysis showed that DED patients are associated with an increased DEP prevalence (OR = 1.83), while DEP patients have a concurrent higher risk of DED (OR = 2.34). Notably, cross-disease GWAS analysis revealed that similar genetic architecture (r_G = 0.19) and pleiotropic functional genes contributed to phenotypes in both diseases. Through protein−protein interaction and ontology convergence, we summarized the pleiotropic functional genes under the ontology of immune activation, which was further validated by a transcriptome systemic review. Importantly, the inverse variance-weighted (IVW)-MR experiments in both TWB and UKB populations (p value <0.001) supported the bidirectional exposure-outcome causation for DED-to-DEP and DEP-to-DED. Despite stringent LD-corrected instrumental variable re-selection, the bidirectional causation between DED and DEP remained.

Conclusion

With the multi-modal evidence combined, we consolidated the association and causation between DED and DEP.

Introduction

Clinical association between DED and DEP

Dry eye disease (DED) and depression (DEP) are two prevalent health conditions that trouble modern society. The prevalence of DED and DEP varies form 1.5–65.0 % and 2.7–37.4 % respectively across different nations (Table 1 and Supplement 1, Fig. S1) [1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28]. Although distinguished as two independent diseases, DED and DEP are highly comanifested in real-world populations, for which the underlying mechanisms remain uncharted.

Table 1.

Prevalence of Dry Eye Disease and Depression in Various Countries.

Continent		Country	Prevalence
			Dry eye disease	Depression
Africa	North	Egyptian	34.9 % [1]	2.7 % [2]
	South	South Africa	54.9 % [1]	9.7 % [3]
	West	Ghana	38.3 % [1]	25.2 % [4]
Asia	North	Russia	8.3 % [5]	41.5 % [6]
	Northeast	Japan	17.4 % [7]	5.7 % [8]
	South	India	32.0 % [9]	5.3 % [10]
	Southeast	Indonesia	27.5 % [11]	21.8 % [12]
	East	Taiwan	33.7 % [13]	1.9 % [14]
	West	Saudi Arabia	36.7 % [15]	37.4 % [16]
America	North	United State	14.6 % [17]	9.2 % [18]
	South	Brazil	12.8 % [19]	10.8 % [20]
Australia		Australia	1.5–16.3 % [21]	21.3 % [22]
Europe	North	Sweden	65.0 % [23]	17.2 % [24]
	South	Spain	11.0 % [25]	4.7 % [26]
	West	United Kingdom	32.1 % [27]	15.5 % [28]

Notably, a secondary cross-sectional analysis [29] of the DRy Eye Assessment and Management (DREAM) study demonstrated that patients with DEP conditions had poor DED characteristics, including reduced tear break-up time (TBUT), increased corneal epithelium defect, and worse dry-eye composite grading index scores. Significantly, objective DED tear biomarkers were also reduced in newly diagnosed DEP patients who were antidepressant-free [30]. This finding suggested that the association between DED and DEP is valid regardless of the DED-causing drug effects of various anti-depressants. On the other hand, a meta-analysis enrolling three million individuals of multiple ethnicities also concluded that patients with DED have an increased risk (odds ratio (OR) = 2.92) of DEP with a worse DEP severity[31]. These independent but inter-connected evidence indicated that DED-DEP was associated in terms of both disease prevalence and disease severity.

Despite the positive evidence that supports the DED-DEP association, the claim has been challenged by negative results in some cross-sectional[32] and naturalistic observational studies[33], Specifically, the contradictory findings suggested that the DED-DEP association was an epiphenomenon caused by psychiatric subjectiveness and unmeasurable confounding factors, such as chronic inflammation[34] and medication adverse events[35], [36]. While the clinical insights into the DED-DEP association constantly advance our intervention strategies, more avant-garde studies that employ genetic and multimodal data have been conducted to expand the landscape of disease-disease associations[37].

Genetic and pathway similarity between DED and DEP

Measuring cross-disease genetic architecture similarity has emerged as a robust method to evaluate disease-disease associations. Given that both DED and DEP are multifaceted diseases that involve genetic x environmental (GxE) interactions, it is sensible to elucidate the genetic contributions underlying the clinical DED-DEP comanifestation[38]. Genetic pleiotropic analysis has been adopted to confirm the shared genetic architecture between a stress-related disease network including but not limited to the psychiatric and ocular disease entries[39]. Meanwhile the brain and the eye were both densely neuron composite organs, in which a genetic defect of the eye could lead to a parallel brain disease[40], [41]. Hence by decoding the brain-eye intersected mutation profile may uncover valuable molecular hotspots accountable for the pathogenesis of DED and DEP.

According to the UK Adult Twin Registry (TwinsUK) and meta-analysis, the heritability (h²) of DED and DEP were estimated to be approximately 37 %[42] and 41 %[43] respectively. This indicated that genetic predisposition plays an important role in DED and DEP. To date, clinical biomarker discovery, transcriptome analysis and genome-wide association studies (GWASs) have shown overlapping pathways in DED/DEP, including inflammatory cytokine cascade[34] and immune cell activation[44]. Enzyme-linked immunosorbent assays have confirmed that inflammatory mediators such as the cytokine interleukin-1 (IL-1) are highly expressed in the basal tears of DED patients[44] as well as in the plasma of DEP patients[45].

A DED clinical trial demonstrated that IL-1 blockade could improve DED symptoms[46], while another pharmacogenetic study identified single-nucleotide polymorphisms (SNPs: rs16944, rs1143643, and rs1143634) in the IL-1b gene as indicators of DED treatment refractoriness[47]. On the other hand, the InCHIANTI study[45] revealed that increased IL-1 receptor antagonist (IL-1Ra) levels were associated with depressive symptoms in elderly individuals. A concurrent study also showed prominent proinflammatory cytokine (IL-1beta, TNF-alpha, IFN-gamma, IL-6 and IL-17A) elevation in antidepressant nonresponders[48]. In fact, a systemic review indicated that cytokine-blocking biologics such as anti-TNF alpha (adalimumab, etanercept, infliximab) and anti-IL-6 (tocilizumab) demonstrated antidepressant activities in clinical patients[49]. These collective studies indicated a converging interest in targeting the inflammatory axis as a novel treatment for DED and DEP.

Nevertheless, the cytokine association does not explain the cause of the DED-DEP comanifestation; hence, further causative analyses, such as Mendelian randomization (MR) studies, were conducted in this study.

Investigate disease variant causations by MR

Mendelian randomization (MR) is a causation analysis method that employs instrument variables (IVs) to examine the relations between disease exposure and outcome. Among all implemented MR IVs, trait-associated SNPs stood out as a robust disease proxy, given that randomized SNP segregation in meiosis could be employed to eliminate confounding factors in causal inference. The selection of IVs should satisfy three assumptions[50]: (i) relevance assumption: IVs should be related to the exposure disease; (ii) independence assumption: IVs should be unrelated to confounders of both the exposure and the outcome; and (iii) exclusion restriction assumption: IVs do not directly influence the outcome of interest. If any of the assumptions are violated, an extended spectrum of the modified MR can be applied to ensure causative validity[51].

MR-based studies have advanced our knowledge of the causation of DED and DEP; for instance, dysregulation of gut microbiota[52] may cause dry eye symptoms in Sjögren’s syndrome patients, while increased IL-6/C-reactive protein axis activity[53] and less physical activity[54] cause DEP. Through MR studies, investigators identified modifiable risk factors that have the potential to treat DED and DEP patients. However, in terms of the complex DED-DEP clinical comanifestation, whether DED and DEP can directly cause one another is a question yet to be determined.

Albeit a nested body of evidence had affirmed the concurrence between DED and DEP, the overwhelming amount of studies had not brought forth a general consensus, hence it is imperative to digest the hitherto findings with a systemic approach. To this end, we performed an up to date meta-analysis to consolidate the DED-DEP association, then conducted a transcriptome analysis to find the molecular pathways at the transaction of DED-DEP co-axis. Importantly, we employed the multiethnic genetic data from the Taiwan Biobank (TWB) and UK Biobank (UKB) to (i) characterize the disease-associated SNPs and the targeted phenome-wide SNP association; (ii) evaluate the genetic architecture similarity and its contribution to DED-DEP cooccurrence; and (iii) identify genetic causations of a cooccurring disease in a sequential manner (Fig. 1).

Fig. 1.

Study Design Overview. To evaluate the underlying relationship between dry eye disease (DED) and depression (DEP), we performed a three-segment study. First, we reviewed the disease association by literature meta-analysis. Second, we confirmed a two-disease genetic correlation by a genome-wide association study (GWAS) in Taiwan Biobank (TWB) and UK Biobank (UKB), clarifying pathway correlation by gene ontology (GO) and protein−protein interaction (PPI) analysis. Finally, we identified bidirectional causation by regarding single nucleotide polymorphisms (SNPs) as instrumental variables (IVs) in Mendelian randomization (MR).

Material and methods

This is a three-segment study that included meta-analysis, GWAS-ontology and MR analysis, each part addressed the clinical correlation, genetic similarity and causation effects between DED and DEP.

Ethics statement

This study was conducted according to the guidelines of the Declaration of Helsinki and was approved by the Institutional Review Board of Taipei Veterans General Hospital (ID no. 2023–01-006AC). All participants in TWB and UKB signed written informed consent before starting the study. All articles selected for systemic transcriptome review involving animals were conducted according to the ethical policies and procedures approved by the ethics committee of the researchers’ institute.

Meta-analysis for DED-DEPs clinical association

The meta-analysis was completed according to the Meta-analyses Of Observational Studies in Epidemiology (MOOSE) checklist (Supplement 1, Table S1) with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) completeness evaluation (Supplement 1, Table S2) and Newcastle–Ottawa Scale (NOS) literature quality check (Supplement 1, Table S3)[55]. the experimental details are specified in Supplement 1. The final search was conducted on 24 August 2022, and elusive information was clarified by inquiring author responses; the last inquiry was replied to on 31 December 2022. The eligibility criteria for study inclusion were based on the Cochrane Handbook for Systematic Review of Interventions. We included all studies documenting DED (xerophthalmi, xerophthalmia, keratoconjuctivitis sicca or dry eye) and DEP (depression or depressive). We included only original studies. Disease prevalence or graded severity were not documented will be excluded from subsequent analysis (details are specified in Supplement 1, Fig. S2). We gathered the data from studies that fulfilled our inclusion criteria. We utilized a standardized form to document the details of the authors, year of publication, study design, sample size, age, gender, the tool or scale employed to evaluate and rate DED and DEP, the occurrence rate of DED and DEP, and the score of DED and DEP based on the scale or instrument used. The included studies were assessed independently by two reviewers (H-Y Wu and T-H Yu). In case of disagreements, a third party (H-J Dai) would mediate until mutual agreement was reached.

In this study, DED severity was graded by a subjective questionnaire (Ocular Surface Disease Index [OSDI] and 5-Item Dry Eye Questionnaire [DEQS]) or objective measures (ST and TBUT). Meanwhile, DEP severity was evaluated by scales including but not limited to the Patient Health Questionnaire-9 (PHQ-9), Hamilton Depression Rating Scale (HDRS) and Major Depression Inventory (MDI).

Twenty-six articles were finally selected, including 1 cross-sectional study[56], 2 cohort studies[57], [58], and 23 case-control studies[35], [36], [59], [60], [61], [62], [63], [64], [65], [66], [67], [68], [69], [70], [71], [72], [73], [74], [75], [76], [77], [78], [79]. The total number of patients was 515,389 for DED and 571,518 for DEP, and the demographic descriptions are provided in Supplement 1, Table S4.

Review Manager software (RevMan) was used for the analysis, and the primary outcome was the Mantel–Haenszel (M−H) OR between the occurrence of DED and DEP, and the secondary outcome was the disease severity standardized mean difference (SMD) under the effect of disease co-occurrence. A random-effect model was employed if heterogeneity was significant according to Cochran’s Q-statistics χ² test and I² statistic (p value <0.1), and subgroup analysis was conducted to evaluate the DED-DEP association in distinct subpopulations.

Genetic correlation analysis in TWB and UKB

Genetic association results were established based on the genotype-phenotype data from selected subjects from TWB (https://www.twbiobank.org.tw/) and UKB (https://www.ukbiobank.ac.uk/) (Supplement 2, Table S1). Approximately 130,000 Asians and 502,000 European were collected from UKB and TWB, respectively. The disease cases in UKB were curated by International Classification of Disease (ICD) information, whereas the disease labels in TWB relied on self-report questionnaires. The control group was collected by screening individuals above 60 years old who were disease-free.

Genetic data quality control (QC) was processed on PLINK (version 1.9 downloaded from http://zzz.bwh.harvard.edu/plink/), in which we set SNPs and participant QC thresholds by the following conditions: (i) SNPs missing in more than 2 % of participants; (ii) individuals missing more than 5 % SNP data; (iii) SNPs with minor allele frequency (MAF) <0.05; (iv) SNPs deviating from Hardy–Weinberg equilibrium (p value <0.05); and (v) individuals with identity by descent (IBD) over 0.125. The subsequent analysis used 460,9691 SNPs (n = 6066) linked to DEP from TWB1 participants, 268,481 SNPs (n = 24745) linked to DED from TWB2 participants, and 296,724 and 297,473 SNPs linked to DED and DEP, respectively, from UKB. In total, 7,684/27,737 individuals in TWB1, 26,866/68,978 individuals in TWB2, and 215063/215957 individuals in UKB were enrolled for further analysis (Table 2). Age, sex, education, marital status, race, BMI, habits, and addiction were considered risk factors for DED and DEP and were adjusted for in the analysis. For individuals listed in both TWB1 and TWB2, our study only utilized the information from TWB2 to ensure that there was no duplication in the data calculations.

Table 2.

Demographic Information of the Participants.

Characteristic	Taiwan Biobank 1			Taiwan Biobank 2			UK Biobank			p value
Dry Eye Disease	Case (n = 2564)	Control (n = 4479)	p value	Case (n = 8757)	Control (n = 15988)	p value	Case (n = 311)	Control (n = 39223)	p value
Age (years)	53 ± 11	64 ± 3	<0.001+	55 ± 9	64 ± 3	<0.001+	62 ± 6	64 ± 3	<0.001+	<0.001+
Male sex (%)	713 (27.8)	2452 (54.7)	<0.001*	1430 (16.3)	6573 (41.1)	<0.001*	74 (23.8)	18,448 (47.0)	<0.001*	0.905‡
Depression	141 (5.5)	124 (2.8)	<0.001*	657 (7.3)	509 (3.2)	<0.001*	32 (10.3)	1811 (4.6)	<0.001*	<0.001‡
Depression	Case (n = 832)	Control (n = 5234)	p value	Case (n = 2999)	Control (n = 18625)	p value	Case (n = 3449)	Control (n = 11293)	p value
Age (years)	49 ± 11	64 ± 3	<0.001+	52 ± 10	64 ± 3	<0.001+	56 ± 8	64 ± 3	<0.001+	<0.001+
Male sex (%)	308 (37.0)	2646 (50.6)	<0.001*	621 (20.7)	7074 (38.0)	<0.001*	1139 (33.0)	5433 (48.1)	<0.001*	0.994‡
Dry eye disease	142 (17.1)	836 (16.0)	0.4551*	678 (22.6)	3255 (17.3)	<0.001*	8 (0.2)	23 (0.2)	0.9163*	<0.001‡

Note: + Wilcox test, + Chi-square test of independence, ‡ Chi-square goodness of fit test.

Variant-disease association (VDA) studies focus on the linkages between SNPs and disease phenotypes through GWASs, scalable and accurate implementation of generalized mixed model (SAIGE), and conditional and joint association analysis (COJO). As the primary outcome of VDA, the results of the GWAS were presented via a Manhattan plot. Because single-variant score statistics did not follow a Gaussian distribution in DED with an unbalanced case−control ratio of DED in UKB (0.008)[80], SAIGE based on a logistic mixed model was applied to avoid inflation of the type I error rate[81]. COJO was applied to interpret joint effects of multiple SNPs owing to overestimated effects of different SNPs on a given phenotype caused by linkage disequilibrium (LD)[82]. We used the -cojo-cond function in the Genome-wide Complex Trait Analysis (GCTA)—COJO program (version 1.91.1 beta)[83].

Disease-disease association (DDA) studies focus on the genetic correlation among different disease phenotypes according to LD score regression. To calculate the overall genetic correlation of DDA, LD score regression used the product of two Z statistics from GWAS, each taken from a GWAS of both diseases[84], with the advantages of being robust to sample overlap bias and not requiring multiple individual genotypes or LD pruning[85].

Gene Ontology (GO) analysis was conducted in the Enrichr web server (https://maayanlab.cloud/Enrichr/)[86], [87], [88] to identify DED and DEP targeted phenotype association studies (PheWASs) between the databases (TWB and UKB) based on UK Biobank GWAS v1. Protein−protein interaction (PPI) network analysis was evaluated in the Search Tool for Retrieval of Interacting Genes/Proteins (STRING version 11.5 download from http://string-db.org/) with a confidence level of 0.400 (medium) based on experiments, databases, coexpression, neighborhood, gene function, and cooccurrence [88]. The activated pathways of functional genes within the PPI network were further analyzed in the Enrichr web server based on Reactome 2022. To further elucidate the activated pathways in DED and DEP, a systemic transcriptome review was conducted on RNA-seq samples from the Gene Expression Omnibus[89], [90], [91], [92], [93], [94], [95], [96], [97], [98], [99], [100], [101], [102], [103], [104], [105], [106], [107], [108], [109], [110], [111], [112], [113], [114], [115], [116], [117].

Bidirectional causation by MR

MR analysis was completed according to Strengthening the reporting of observational studies in epidemiology using mendelian randomization (STROBE-MR) (Supplement 4, Table S1)[118], [119], the experimental details are specified in Supplement 4.

To evaluate bidirectional causation between DED and DEP, MR analysis was conducted with SNPs from VDA studies using the TwoSampleMR package in R software (version 4.2.0 downloaded from https://github.com/MRCIEU/TwoSampleMR). However, SNPs may influence the expression of different genes owing to LD, and the issue of pleiotropy becomes nonnegligible, including horizontal (multiple independent effects of a given SNP causing multiple phenotypes) and vertical pleiotropy (multiple effects in a cascade causing multiple causally related phenotypes)[120]. To rule out pleiotropy, sensitivity analyses were conducted by reproducing the results through 6 different MR methods, including MR pleiotropy residual sum and outlier (MR-PRESSO)[121], inverse variance-weighted (IVW)[122], MR-Egger[123], two model-based estimation (MBE) methods[124], and weighted median (WM)[125].

In the absence of pleiotropy, the inverse variance weighted (IVW) method was preferred by weighing the effects of SNPs with the reciprocal of the intercept variance to estimate robust causal effects in the absence of directional pleiotropy[126]. In the presence of pleiotropy, the MR−Egger method was conducted by accepting the presence of an intercept under the assumption that the pleiotropic associations are independent[121]. The model-based estimation (MBE) simple mode and weighted mode methods allow the maximum proportion of information from invalid IVs to less and more than half, respectively, under the zero modal pleiotropy assumption (ZEMPA) that the most frequent value of horizontal pleiotropy is zero[124]. Weighted median methods combine data on multiple IVs into a single causal estimate by adopting a median after giving different weights to every IV based on the empirical distribution function[125].

Results

Applying a meta-analysis to confirm the clinical association between DED and DEP

Studies with diagnostic instruments and severity scales were included in this meta-analysis to assess the incidence and severity association between DED and DEP. Nineteen of the 26 included studies reported DEP symptoms in DED cohorts (19 reported the prevalence[8], [29], [30], [31], [32], [33], [34], [35], [36], [37], [38], [39], [40], [41], [42], [43], [44], [45], [46] and 6 graded the severity[56], [61], [63], [69], [70], [71], [73]), while 10 studies reported DED symptoms in DEP cohorts (6 reported the prevalence[35], [56], [59], [64], [74], [75], 7 graded the severity[56], [74], [75], [76], [77], [78], [79], and 3 reported both[56], [74], [75]) (Supplement 1, Table S4). Our meta-analysis confirmed that DED patients, in comparison to healthy controls, had an increased DEP prevalence (summary OR = 1.83, 95 % CI: 1.60–2.10, I2 = 0 %) and DEP severity (SMD = 0.45, 95 % CI: 0.25–0.65, I2 = 0 %) (Fig. 2, Fig. 3). Meanwhile, DEP patients manifested with increased DED prevalence (summary OR = 2.34, 95 % CI: 1.58–3.46, I2 = 40.5 %), but were not linked to significant subjective (OSDI and DEQS) or objective DED severity (ST and TBUT) (Supplement 1, Fig. S3 and S4). Subgroup analyses are presented in Supplement 1, Fig. S5.

Fig. 2.

Forest plots illustrating associations between dry eye disease (DED) and depression (DEP) in terms of prevalence odds ratio (OR). (A) Association between the prevalence of depression in DED patients versus the control group. (B) Association between the prevalence of DED in depression patients versus the control group. CI, confidence interval; M−H, Mantel–Haenszel.

Fig. 3.

Meta-analysis results for the association between dry eye disease (DED) and depression (DEP). (A) Association between the prevalence of depression in DED patients versus the control group. (B) Association between the prevalence of DED in depression patients versus the control group. (C) Association between depression severity in DED patients versus the control group. Since the depression scales varied from study to study, the severity of depression was standardized into a Z score. (D) Association between the Schirmer test (ST) of DED severity in depression patients versus the control group. (E) Association between tear break-up time (TBUT) and DED severity in depression patients versus the control group. NS, not significant (p value >0.05); *, statistically significant with p value <0.05; **, statistically significant with p value <0.01; ****, statistically significant with p value <0.0001.

To validate the meta-analysis findings in an independent population-based cohort, we examined the disease association in TWB (Table 2). The results showed that increased DEP prevalence was observed in DED patients within TWB1 (OR = 2.04, 95 % CI: 1.60–2.61, p value <0.00001) and TWB2 (OR = 2.38, 95 % CI: 2.11–2.68, p value <0.00001). However, increased DED prevalence was observed only in DEP patients in TWB2 (OR = 1.39, 95 % CI: 1.27–1.53, p value <0.00001) but not TWB1 (Table 2). In brief, our meta-analysis and TWB validation supported the positive association between DED and DEP (Table 2).

Genetic similarity and shared disease ontology between DED and DEP

To explore whether genetic factors contributed to the DED and DEP association, we performed PheWAS and a chord-interacting diagram analysis to find the shared functional SNPs in the DED-DEP UKB phoneme. With the input of the top 500 DED and DEP SNPs, we retrieved 87 DED and 62 DEP significant phenotypes (Supplement 2, Table S2) from the UKB phoneme. Among the significant phenotypes, 5 DED-relevant phenotypes, including corneal hysteresis, corneal resistance factor, and whole body water mass (OR = 2.060–3.887), were selected, whereas 6 DEP-relevant phenotypes, including time and frequency of medical experiences (OR = 2.537–9.405), were selected for subsequent analysis. Interestingly, through a chord diagram with the top 200 DED and DEP SNPs, we identified 24 DED and 7 DEP functional SNPs that jointly constitute both DED and DEP phenotypes (Fig. 4A). These results indicate that some functional hotspots were shared between the two diseases.

Fig. 4.

Investigation of dry eye disease (DED) and depression (DEP) association through genetic approaches. (A) Chord diagram showing crosswalks between the phoneme and functional single-nucleotide polymorphisms (SNPs) of DED and DEP. (B) Heatmap of the genetic correlation scores between listed ocular and depressive diseases in Taiwan Biobank (TWB). (C) The protein−protein interaction (PPI) network of the functional SNP_DED and SNP_DEP genes derived from TWB and UK Biobank (UKB). (D) The converging ontology flow chart analyzing the functional SNP_DED and SNP_DEP genes.

We next performed genetic correlation (r_G) score analysis in both UKB and TWB to assess whether global SNP patterns are similar between DED and DEP (Fig. 4B). Notably, in the TWB Asian population, DED shared an r_G = 0.19 genetic correlation with DEP. Cataract disease, which has a less heritable genetic component, was parallel enrolled as a negative control (Cataract-DEP r_G = 0.03) to the DED-DEP genetic correlation. Concordantly, in the UKB, common variant SNP analysis showed a positive genetic correlation (r_G = 0.109) between DED and DEP, indicating that genetic architecture similarity was evident regardless of human ethnicity (Supplement 2, Fig. S1).

Disease-associated variant analysis in this study was performed in tandem with SAIGE or COJO adjustment, thereby mitigating the false-positive SNP findings that were subject to imbalanced case control ratio and LD. After SAIGE and COJO correction, 71 DED SNPs within 38 functional genes and 53 DEP SNPs within 33 genes survived the adjustment (Supplement 2, Fig. S2 and S3) (Supplement 2, Tables S3 - S12). Although there were no identical SNPs found between DED and DEP, some of the top 100 DED and DEP SNPs were found in common functional genes such as MUC16 (DED: rs1862461, p value = 4.5x10⁻⁴; rs2591595, p value = 2.67x10⁻⁴; rs22522675, p value = 2.69x10⁻⁴; rs2216663, p value = 2.77x10⁻⁴; and DEP: rs11707526, p value = 4.5x10⁻⁴) and TENM2 (DED: rs73801227, p value = 2.76x10⁻⁴, and DEP: rs13185961, p value = 7.22x10⁻⁵) in TWB2 (Supplement 2, Tables 3 and 13). Given the overlap of the functional SNP genes, we further investigated whether the top 200 functional gene lists, SNP_DED and SNP_DEP, exhibited interactions at the biochemical level regarding molecular-molecular interactions, protein homologs, gene fusions and gene coexpression (Fig. 4C). From the UKB and TWB combined functional SNP gene dot and line network, we demonstrated crosstalk between DED and DEP (DDX39B-PRPF31; PRKG1- SLC8A1 and MLH1-MSH5). Meanwhile, some protein hotspots (ESR1, SRSF1, DDX39B, RHOC and MYH11) were found to connect functional SNP genes across UKB and TWB. Together, our results indicated that individual DED and DEP functional SNP genes were biochemically connected.

Meanwhile, our protein–protein interaction (PPI) analysis showed that the top 200 SNP_DED and SNP_DEP functional genes formed a functional network that converged on pathway ontologies of immune system, innate immune system and signal transduction (Fig. 4D). The converged pathway was validated on public GEO datasets that consist of clinical and animal transcriptome sequencing samples. By screening 1072 DED accessions and 4212 DEP accessions indexed within the GEO database, 15 DED datasets (blood cells, cornea, conjunctiva and lacrimal glands) and 15 DEP datasets (blood cells, subregion cortexes, hippocampus and amygdala) were included and analyzed over a target-expanded spectrum of Reactome-2022 HSA ontologies (Fig. 5) (Supplement 3, Tables S1-S30).

Fig. 5.

Heatmap of the activated common pathways in published dry eye disease (DED) and depression (DEP) transcriptome databases. R-HAS, Reactome Homo sapiens; GSE, Gene Expression Omnibus Series; DC, dendritic cell; SjS, Sjögren's syndrome; PBMC, peripheral blood mononuclear cells; IL, interleukin; and TLR, Toll-like receptors.

By screening through the detailed entry of the immune ontology, we determined that the immune activations are patterned and asymmetric between DED and DEP samples. Specifically, while the immune system (R-HSA-168256) and cytokine signaling in immune system (R-HSA-1280215) were broadly activated in both DED and DEP samples, the infectious disease (R-HSA-5663205) was primarily activated in the tissues of the ocular surface and in the peripheral blood of DEP samples, but not in the majority brain tissue accessions. Meanwhile, the interferon signaling (R-HSA-913531) and innate immune system (R-HSA-168249) pathways were activated more in DED than in DEP. Together, these differences indicated that the pathogen-elicited immune response may affect DED through direct tissue infection, but influence DEP through an indirect blood stream effect.

Through analyzing the pathway activation patterns, we noticed that the ulcerative cornea of the DED model (GSE195962)[98] and the hippocampal tissue of the stress-induced DEP model (GSE148579)[109] showed pan-activation across immune pathways (cytokine-related, infectious-related and innate immunity pathways). Whereas, the innate immunity category that comprises toll-like receptor cascades and MyD88 cascades was robust in specific DED (dendritic cell, plasmacytoid dendritic cell, monocyte, conjunctiva and cornea) and DEP (hippocampus and pre-frontal cortex) specimen subsets. This highlighted the heterogenous inflammatory etiology of DED and DEP, and echoed to the inconsistent immune modulatory treatment effects observed from the real world DED [127] and DEP [128] trials.

Notably, signal transduction pathways (R-HSA-162582, R-HSA-194315, R-HSA-9716542 and R-HSA-9006934) were also activated in specified DED (cornea Aire knockout GSE208297)[99] and DEP (whole blood GSE182193[102] and dentate gyrus GSE84183[113]) samples; therefore, the significance of signal transduction ontologies requires further validation. In summary, through systemic analysis on transcriptome, we addressed the shared features of immune activation in both DED and DEP pathways.

Bidirectional causation by MR in TWB and UKB

Studying the exposure–outcome causations in DED and DEP is challenging, given the confounding effects of multifactorial environmental and life incidences. Randomized trials for causation investigation remain the gold standard but are often expensive and impossible to carry out. MR is a genetic-based technique that employs SNP IVs to distinguish exposure-outcome causations from observational data. Instead of using summarized statistics, we performed GWASs and confounding IV exclusions to approximate optimal MR assumptions. A total of 48 MR conditions comprising 6 MR methods (IVW, MR-PRESSO, MR−Egger, simple mode, weighted mode, weighted median) with 2 SNP sets (GWAS and COJO) were performed in 2 genetic databases (TWB and UKB), and each MR experiment was carried out in two directions, designated either DED-to-DEP or DEP-to-DED. In the absence of directional pleiotropy (p value: MR-PRESSO >0.05; test of Q >0.05; MR−Egger intercept >0.05) (Supplement 4, Table 2, Table 3), inverse-variant weighted (IVW) modified-MR were selected as the primary instrument to detect disease causal detection in this study.

Table 3.

Mendelian Randomization Results of the Tested Conditions.

Mendelian Randomization		TWB				UKB
		DED-to-DEP		DEP-to-DED		DED-to-DEP		DEP-to-DED
		OR [95 % CI]	p value	OR [95 % CI]	p value	OR [95 % CI]	p value	OR [95 % CI]	p value
Genome-wide Association Study (GWAS)
One-sample	IVW	1.388 [1.144–1.683]	0.0009	1.212 [1.109–1.324]	<0.0001	1.049 [1.020–1.078]	<0.0001	2.547 [1.773–3.661]	<0.0001
	MR-Egger	1.715 [0.820–3.588]	0.1756	0.994 [0.611–1.615]	0.9803	1.013 [0.916–1.121]	0.7922	1.772 [0.400–7.850]	0.4553
	Simple mode	1.460 [0.951–2.242]	0.1054	1.059 [0.844–1.329]	0.6267	1.028 [0.961–1.100]	0.4287	2.285 [0.892–5.856]	0.0930
	Weighted mode	1.463 [0.968–2.212]	0.0925	1.055 [0.861–1.294]	0.6129	1.028 [0.974–1.084]	0.3245	2.367 [0.937–5.982]	0.0760
	Weighted median	1.444 [1.111–1.879]	0.0061	1.099 [0.968–1.247]	0.1432	1.035 [0.999–1.072]	0.0528	2.446 [1.492–4.012]	0.0004
Two-sample	IVW	1.847 [1.428–2.389]	<0.0001	1.089 [1.036–1.144]	0.0008	1.105 [1.048–1.165]	0.0002	1.599 [1.215–2.105]	0.0008
	MR-Egger	2.124 [0.732–6.163]	0.1738	1.127 [0.839–1.513]	0.4444	1.409 [1.021–1.945]	0.0511	1.202 [0.372–3.877]	0.7627
	Simple mode	1.289 [0.613–2.711]	0.5076	1.088 [0.984–1.202]	0.1229	1.069 [0.958–1.194]	0.2461	1.268 [0.704–2.284]	0.4417
	Weighted mode	1.262 [0.620–2.571]	0.5252	1.084 [0.975–1.206]	0.1602	1.072 [0.962–1.195]	0.2204	1.280 [0.724–2.262]	0.4094
	Weighted median	1.593 [1.122–2.262]	0.0093	1.090 [1.017–1.168]	0.0146	1.090 [1.017–1.169]	0.0138	1.417 [0.974–2.060]	0.0678
Conditional and Joint Association Analysis (COJO)
One-sample	IVW	1.362 [1.147–1.616]	0.0004	1.221 [1.114–1.339]	<0.0001	1.050 [1.021–1.080]	<0.0001	2.497 [1.714–3.637]	<0.0001
	MR-Egger	1.712 [0.899–3.259]	0.1189	0.925 [0.572–1.497]	0.756	1.014 [0.912–1.128]	0.7930	1.470 [0.293–7.381]	0.6419
	Simple mode	1.445 [0.967–2.159]	0.088	1.061 [0.840–1.340]	0.6272	1.026 [0.964–1.091]	0.4234	2.136 [0.810–5.633]	0.1330
	Weighted mode	1.445 [0.969–2.156]	0.0869	1.056 [0.855–1.306]	0.6191	1.024 [0.962–1.091]	0.4591	2.080 [0.768–5.628]	0.1573
	Weighted median	1.406 [1.126–1.756]	0.0026	1.097 [0.959–1.256]	0.1758	1.033 [0.994–1.073]	0.0938	2.319 [1.378–3.903]	0.0015
Two-sample	IVW	1.877 [1.447–2.435]	<0.0001	1.104 [1.049–1.162]	0.0001	1.113 [1.055–1.175]	<0.0001	1.659 [1.301–2.116]	<0.0001
	MR-Egger	2.219 [0.773–6.371]	0.1468	1.097 [0.821–1.466]	0.5422	1.285 [0.943–1.751]	0.1286	1.338 [0.460–3.890]	0.5986
	Simple mode	1.347 [0.649–2.793]	0.4289	1.090 [0.975–1.220]	0.1551	1.067 [0.937–1.216]	0.3363	1.275 [0.735–2.212]	0.3965
	Weighted mode	1.300 [0.640–2.639]	0.4728	1.088 [0.979–1.208]	0.1412	1.067 [0.958–1.189]	0.2504	1.281 [0.750–2.188]	0.3742
	Weighted median	1.623 [1.130–2.332]	0.0088	1.093 [1.018–1.175]	0.0145	1.086 [1.008–1.169]	0.0282	1.422 [1.011–1.998]	0.0426

Note: DED, dry eye disease; DEP, depression; IVW, inverse variance-weighted; TWB, Taiwan Biobank; UKB, UK Biobank; OR, odds ratio; and CI, confidence interval.

The adjusted p-value of both the IVW-MR (DED_exposure to DEP_outcome and DEP_exposure to DED_outcome) were <0.05/48≈0.001 after Bonferroni’s correction (Table 3). Notably, the results still stood when we applied COJO correction to select LD-free sanity IV_SNPs. In two-sample MR experiments with no case overlaps, the causative OR of DED-to-DEP was 1.877 (95 % CI = 1.447–2.435) in TWB and 1.113 (95 % CI = 1.055–1.175) in UKB. The ORs of DEP-to-DED were 1.104 (95 % CI = 1.049–1.162) in TWB and 1.659 (95 % CI = 1.301–2.116) in UKB.

To assess the consistency of directional causation, the OR of each MR experiment was plotted and examined (Fig. 6) (Supplement 4, Fig. S1 and S2), confirming the directional causation between DED-to-DEP and DEP-to-DED in both the UKB and TWB populations. IV_SNP selection bias and minimum necessary SNP counts were investigated by replicative MR studies using different SNP numbers (n = 5, 10, 15, 20, 40, 60), whereby a number of approximately 40 can attain statistical significance (p <0.001) with a stable OR.

Fig. 6.

Validation of dry eye disease (DED)-depression (DEP) bidirectional causation using different two-sample Mendelian randomization (TSMR) methods in replicative practices. Scatter plots of (A) TSMR experiments using genome-wide association study (GWAS)-selected SNPs as instrumental variables. Experiments were repeated in both the Taiwan Biobank (TWB) and UK Biobank (UKB). (B) TSMR instrument variables were replaced by conditional and joint association analysis (COJO) and scalable and accurate implementation of generalized mixed model (SAIGE)-corrected GWAS single-nucleotide polymorphisms (SNPs) in UKB. (C) The SNP number needed for TSMR significance and its corresponding effects on the TSMR odds ratio was investigated.

In the heterogeneity analysis, Cochran's Q value implied no substantial heterogeneity with a Q value smaller than a degree of freedom in all models. Sequentially omitting each SNP, leave-one-out (LOO) analysis showed that most of the ORs fell within the 95 % CI of the IVW model, which indicated that the bidirectional causation between DED and DEP is robust without any particular IV influencing the causal inference of GWAS, SAIGE and COJO (Supplement 4, Fig. S3 and S4). Collectively, our MR-LOO sensitivity analysis indicated that a cluster of small effect IV_SNP rather than singular large effect IVSNPs accounted for the bidirectional causal effect underlying DED and DEP.

Discussion

In this study, we sought to investigate the disease relation between dry-eye disease (DED) and depression (DEP). By employing multimodal analysis and genetic data in UK biobank and TW biobank, we conducted meta-analysis, GWAS, transcriptome systematic review and COJO- adjusted MR experiments to reconstitute the interacting relations between DED and DEP. Sequentially, we demonstrated that DED and DEP were tightly linked in terms of clinical co-morbidity and shared a common pathway of inflammation activation. Moreover, our COJO-adjusted MR experiments suggested that a bidirectional genetic causation was detected in the DED-DEP relation.

Clinical association between DED and DEP

This is an updated meta-analysis that confirmed the clinical association between DED and DEP. From the current literature evidence, DED patients were associated with increased DEP prevalence and worse DEP symptoms, whereas DEP patients were associated only with an increased DED prevalence, no increased DED severity was detected. The non-linear discrepancy between the high DED prevalence but no increased DED severity was also reported in DED-associated diseases such as: thyroid dysfunction, osteoarthritis, diabetes, irritable bowel syndrome, hypercholesterolemia, hypertension, and hypertriglyceridemia[129]. Nevertheless, the exact mechanism regarding why certain disease associates do not combine severity association may be context dependent, and requires further cell-biology studies on stratified clinical specimens.

Genetic linkage between DED and DEP

Genetic pleiotropy, defined as one SNP affecting both DED and DEP, was not detected in TWB or UKB. Nevertheless, MUC16 and TENM2 mutations at the functional gene levels were found to contribute to both DED and DEP in TWB.

Our study reported one novel MUC16 SNP_DED (rs7260214) and four synonymous MUC16 SNP_DEP (rs1862461, rs2591595, rs2252675, and rs2216663) in DED and DEP patients in TWB2, respectively. MUC16, a key transmembrane mucin protein that maintains the hydrophilic glycocalyx layer of the ocular surface, exhibits altered glycosylation in DED patients[130]. Concordantly, an observation study also reported that patients with lower MUC16 expression levels at the ocular surface are prone to suffer from DED[131]. MUC16 is a tumor marker also known as CA-125[132], and recent studies have shown that DEP predisposition is associated with higher MUC16 expression in a population with ovarian cancer[133]. Notably, in a hospitalized observation study enrolling young adolescents without underlying ovarian cancer, DEP patients with self-harm behavior had higher MUC16 levels in their peripheral blood. In addition, incorporating MUC16 levels into the Hamilton Depression Rating Scale (HDRS) can improve self-harm behavior prediction in DEP patients[134]. These studies indicated that mutations affecting either the expression level or the posttranslational modification of MUC16 contribute to the cooccurrence of DED and DEP.

Mutations in TENM2, a highly expressed cell–cell adhesion molecule in the central nervous system, were discovered to be accountable for DED (rs73801227) and DEP (rs13185961) in this study. The expression of TENM2 was significantly correlated with DEP gene sets in the postmortem parietal cortex of DEP patients[135] and mutations of TENM2 were also found in the cornea of keratoconus[136] and astigmatism[137] patients. Although the exact functions of the TENM2 intron SNPs are yet unknown, an adaptive evolution GWAS using primate phylogeny and the modern human genome has reported that mutations of TENM2 with 16 other DEP genes were positively selected in East Asian ancestry evolution[138], notably, most of the positive selected variants resided in the untranslated regions (UTRs) and noncoding exons of TENM2. While TENM2 functioned as a presynapse transmembrane protein that guides retinal axon migration[139] and orchestrates synaptogenesis excitatory/inhibitory balance[140], SNPs of TENM2 may disrupt the potential role of TENM2 in the pathogenesis of DED and DEP.

In the PPI network analysis, several interesting hotspots were identified as protein hubs (ESR1 and DDX39B) for DED-DEP transactions. A genetic polymorphism of ESR1, a key estrogen receptor, was associated with both DED[141] and DEP[142]. The expression of ESR1 has been connected to hormone regulation with corneal homeostasis[143]. Meanwhile, estrogen dysregulation and usage of hormone replacement therapies[144] and aromatase inhibitors[145] have all been reported to cause MG dysfunction in the evaporative subtype of DED[146].

Our study recapitulated the canonical disease pathophysiology hotspots and elucidated the extensive molecular networks that account for the co-occurrence of DED and DEP.

Common underlying pathways

For the first time, we proposed and addressed the significance of immune-related pathways at the intersection of DED and DEP pathophysiology. By indexing our converging flow ontology SNPs (Supplement 2, Tables 3-12) in the GWAS catalogue[147], we identified rs17854914 (PLD2), a C-reactive protein (CRP)-associated SNP, that linked DED to the immune ontology. On the other hand, the DEP-associated rs1270942 (CFB) was related to various autoimmune diseases, such as type 1 diabetic mellitus (T1DM), systemic lupus erythematosus (SLE) and autoimmune thyroiditis. Although the immune response and host defense systems have been studied from the respective DED[148] and DEP[149] perspectives, whether immune crosstalk exists between the two diseases has not been fully explored.

In our transcriptome meta-analysis results, we observed that a compelling ontology similarity, mainly immune pathways, was shared across independent DED and DEP GEO datasets. The finding of an elevated cytokine signaling transcriptome was supported by a DED systematic review, which reported elevated cytokine (IL-1β, IL-6, IL-8, IL-10, INF-γ and TNF-α) levels in DED tear samples[150]. The significance of IL-6 in DEP was addressed by another meta-analysis using a longitudinal DEP cohort, which revealed that higher IL-6 levels in the peripheral blood are associated with future DEP[149], and antidepressant treatment can reduce blood IL-6 levels[151]. Although IFN-gamma signaling was not upregulated in the DEP samples, interferon signaling was critical to DED onset. IFN-γ was reported to drive squamous metaplasia in corneal epithelium[152] and to block goblet cell differentiation[153], and both mechanisms may aggravate DED conditions.

These independent but complementary meta-analysis results collectively indicate the significance of cytokine signaling in both DED and DEP pathophysiology.

Causal effects

Compared to using summary statistics, the in-house SNP discovery and pruning process allowed us to combine COJO and SAIGE correction in IV selections, which resulted in better fitness of IVs to the classic MR assumptions.

Our bidirectional IVW-MR results identified both DED and DEP as the risk exposure for the disease. This finding implied that treating either DED or DEP has the potential to resolve the entangled clinical co-occurrence of DED and DEP. Although studies have reported that treating DEP with selective serotonin reuptake inhibitors (SSRIs) may cause DED aggravation through NF-kB inflammation[154], contrasting results have demonstrated that treating DEP with serotonin norepinephrine reuptake inhibitors (SNRIs) does not aggravate DED[155]. Meanwhile, a case−control study reported that effective DED treatment can reduce depressive symptoms[156]. Together, the genetic causation findings of this study provide a theoretical and strategic background to block the DED-DEP vicious cycle in the clinical setting.

Limitations

Meta-analysis

While our results supported the positive correlation between DED and DEP, analytic limitations were noted in this meta-analysis. To be specific, the available published DED-DEP studies scored 15–20 out of 22 in the scale of STROBE checklist (Supplement 1, Table S2), which indicates the majority of current DED-DEP studies had not make full addressment in either the sources of bias, showing sensitivity analysis of the results, adjust confounders or perform adequate subgroups analysis. Therefore, limited the statistical power of the overall meta-analysis. Noteworthily, despite our subgroup meta-analysis and SMD statistics demonstrated that DED-DEP clinical association holds independent to the selected ethnicity, the inevitable heterogeneity (clinical, methodological, and statistical features) of the studies may have restricted the statement, unless further DED-DEP study characterization opted a standardized; uniform protocol.

Genetic correlation

In spite of selecting strongly associated SNPs, common SNPs might not explain the diseases completely since both DED and DEP are multi-genetic-derived. Furthermore, since coding genes only account for 1–––2 % of the whole human genome, SNPs outside coding gene regions were not explained in this research, however, they are likely to affect disease if they are located within functionally important non-coding regions such as enhancers, promoters, silencers, or insulators. Moreover, genetic data were predominantly of Chinese origin, so the results might not apply to other regions.

Causal effects

Albeit we employed COJO and two-sample MR to mitigate LD-inflated instrument bias and prevent over-estimated beta value respectively, the correlated horizontal pleiotropy (CHP) of the underlying immune pathway was not fully considered in current MR experiments. For conditions in which a CHP has violated MR assumptions.

Future prospectives and clinical impacts

The pathognomonic resonance between brain-eye comorbidity were recently proposed and in-depth reviewed from both clinical [157], [158] and biomolecular standpoints [159], [160]. While this study adds DED-DEP relations to the brain-eye comorbidity clusters, we investigated the patternized immune activation in detailed tissue origins, which may guide immune modulatory drug usage on target-specific compartments. Notably, brain atrophy at pre-frontal cortex and hippocampus region were symbolic DEP traits [161], and concordantly our transcriptome analysis showed activated TLR and MyD88 cascades in the pre-frontal and hippocampus segments. These information whereby suggest TLR [162] and MyD88 [163] antagonists may have a role in the DED-DEP blockade.

Conclusion

Collectively, our meta-analysis confirmed the association between DED and DEP; and the GWAS target-PheWAS attributed immune activation under the common ground of DED-DEP co-axis. Last but not the least, we detected DED-DEP bi-directional causation by COJO-adjusted IVW-MR. With the multi-modal evidence combined, we consolidated the association and causation between DED and DEP.

CRediT authorship contribution statement

Kao-Jung Chang: Conceptualization, Methodology, Investigation, Writing – original draft. Hsin-Yu Wu: Conceptualization, Methodology, Investigation, Writing – original draft. Pin-Hsuan Chiang: Software, Formal analysis, Project administration, Data curation. Yu-Tien Hsu: Software, Formal analysis, Project administration, Data curation. Pei-Yu Weng: Software, Formal analysis, Project administration, Data curation. Ting-Han Yu: Validation, Visualization. Cheng-Yi Li: Validation, Visualization. Yu-Hsiang Chen: Validation, Visualization. He-Jhen Dai: Validation, Visualization. Han-Ying Tsai: Formal analysis, Investigation, Project administration. Yu-Jung Chang: Formal analysis, Investigation, Project administration. You-Ren Wu: Writing – review & editing, Supervision. Yi-Ping Yang: Software, Formal analysis, Project administration, Data curation. Cheng-Ta Li: Writing – review & editing, Supervision. Chih-Chien Hsu: Writing – review & editing, Supervision. Shih-Jen Chen: Writing – review & editing, Supervision. Yu-Chun Chen: Writing – review & editing, Supervision. Ching-Yu Cheng: Writing – review & editing, Supervision. Ai-Ru Hsieh: Resources, Writing – review & editing, Supervision, Funding acquisition. Shih-Hwa Chiou: Resources, Writing – review & editing, Supervision, Funding acquisition.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: