Organic Solvents as Risk Factor for Autoimmune Diseases: A Systematic Review and Meta-Analysis

Background Genetic and epigenetic factors interacting with the environment over time are the main causes of complex diseases such as autoimmune diseases (ADs). Among the environmental factors are organic solvents (OSs), which are chemical compounds used routinely in commercial industries. Since controversy exists over whether ADs are caused by OSs, a systematic review and meta-analysis were performed to assess the association between OSs and ADs. Methods and Findings The systematic search was done in the PubMed, SCOPUS, SciELO and LILACS databases up to February 2012. Any type of study that used accepted classification criteria for ADs and had information about exposure to OSs was selected. Out of a total of 103 articles retrieved, 33 were finally included in the meta-analysis. The final odds ratios (ORs) and 95% confidence intervals (CIs) were obtained by the random effect model. A sensitivity analysis confirmed results were not sensitive to restrictions on the data included. Publication bias was trivial. Exposure to OSs was associated to systemic sclerosis, primary systemic vasculitis and multiple sclerosis individually and also to all the ADs evaluated and taken together as a single trait (OR: 1.54; 95% CI: 1.25–1.92; p-value<0.001). Conclusion Exposure to OSs is a risk factor for developing ADs. As a corollary, individuals with non-modifiable risk factors (i.e., familial autoimmunity or carrying genetic factors) should avoid any exposure to OSs in order to avoid increasing their risk of ADs.


Introduction
Autoimmune diseases (ADs) are initiated by the loss of immune tolerance and mediated through T or B cell activation leading to tissue damage. ADs share clinical signs and symptoms, physiopathological mechanisms, and genetic factors [1]. They are complex diseases caused by the interaction between genetic, epigenetic, and environmental factors over time [2,3].
Despite the difficulties in defining environmental risk factors that lead to immunopathology, the number of candidates proposed for specific ADs is continuously growing as new evidence is reported for infectious agents, chemicals, physical factors, adjuvants, and hormones [4][5][6][7][8][9][10][11][12][13][14][15]. A significant body of research has pointed out that, for autoimmunity to occur, the genetic background warrants to be combined with environmental injuries and novel associations has been described as the case of the air pollution [5,16]. However these environmental factors often explain only a small number of cases, and, on their own, they are not sufficient to cause the disease [5].
Solvents are liquids that dissolve a solid, liquid or gas. They can be broadly classified into two categories: organic and inorganic. Organic solvents (OSs) are compounds whose molecules contain carbon. They may be broken down further into aliphatic-chain compounds, such as n-hexane, and aromatic compounds with a 6carbon ring, such as benzene or xylene. OSs arose in the latter half of the 19th century from the coal-tar industry. Common uses for OSs are: dry cleaning (e.g., tetrachloroethylene), paint thinner (e.g., toluene, turpentine), nail polish removers and glue solvents (acetone, methyl acetate, ethyl acetate), spot removers (e.g., hexane, petrol ether), detergents (citrus turpenes), perfumes (ethanol), nail polish and chemical synthesis [17]. In contrast, the use of inorganic solvents (other than water) is typically limited to research in chemistry and some technological processes.
The applications of OSs became more diversified in both developed and developing countries. Research in this area began in 1957 when the first patients developing a scleroderma-like syndrome after exposure to vinyl chloride, epoxy resins, trichloroethylene (TCE), perchloroethylene and other mixed solvents were reported [18,19]. Nevertheless, few published studies have analyzed the wide spectrum of ADs in subjects exposed to OSs. Therefore, we aimed to analyze the evidence of an association between the exposure to OSs and the development of AD through a systematic literature review and a meta-analysis. In addition, a comprehensive review concerning the mechanisms by which OSs exposure induces immunological alterations is presented.

Literature Search
The search was done using the following databases: PubMed, SCOPUS, SciELO and LILACS and took into account articles published up to February 2012. We followed the PRISMA guidelines for meta-analysis of observational studies [20] in our data extraction, analysis, and reporting (Text S1).

Study Selection, Data Extraction, and Quality Assessment
Inclusion criteria for the systematic review were the following: any types of study that used accepted classification criteria for ADs and had information about exposure to OSs explicitly listed as a category.
Articles were excluded from the analysis if they included the same data that were published in another study.
Abstracts and full text articles were reviewed in the search for eligible studies. Two reviewers did the search independently while applying the same selection criteria. The two resulting databases were compared and disagreements resolved by consensus. For articles in languages other than English or Spanish, translations of abstracts or full text articles were reviewed to determine eligibility.
Each eligible study was classified as: review, case report, case series, cohort, or case-control. Inclusion criteria for the metaanalysis were applied to publications that provided epidemiologic data on risk factors [relative risks (RR) and odds ratios (OR) with confidence intervals (CI)] or that provided information that let us calculate these data. For cohort studies, the requirements were the number of subjects exposed, the number unexposed, and the number of subjects who developed the disease in each of the two cases. For case-control studies, the requirements were the number of subjects with AD that were exposed and not exposed, and the number of controls that were exposed and not exposed. In those instances where the study had not reported the number of subjects in each group, either the RR or the OR with the CI, at least, must have been reported in order for them to be included in the metaanalysis calculations.
Studies were excluded from the meta-analysis if the groups were not clearly defined, e.g. case-controls studies with likely AD diagnosis in control subjects or exposed cohorts with low specificity for OS.
For each eligible study, the type of exposure and exposure assessment was analyzed regarding the source of information (census, database, interview, mailed questionnaires, etc.) and classified as follows: ''qualitative'' if it was stated by the subject or interviewer on questionnaires measured by the quality of exposure rather than its quantity, ''quantitative'' if it was related to a number or quantity, and ''semi-quantitative'' if it was expressed as a quantity susceptible of measurement but was not related to a number. Quantitative assessment was sub-classified in ''indirect quantitative'' if it was defined by an estimate from a register of specific jobs at risk or calculated using a job-exposure matrix formula, and ''direct quantitative'' if the OS was directly measured in the environment or as a biomarker in the subject. Furthermore it was extracted the information that described the condition of exposure (e.g occupation, living characteristics.) The quality and strength of scientific evidence was evaluated supporting an etiologic relationship between ADs and the proposed risk factor. In this investigation, a quantitative scoring system based on the Bradford Hill criteria was used [21]. The quantitative Bradford Hill score (qBHs) is divided into categorical ratings of the overall strength of causal association as follows: 0 to 6 points was considered poor or no causal association; 7 to 14 points was considered moderate or inconclusive causal association, and 15 to 21 points was considered a strong causal association. No study was excluded from the review based on this assessment.

Meta-analysis
Data were analyzed using the Comprehensive Meta-Analysis version 2 program (Biostat, Englewood, NJ, 2004). Calculations were carried out for the whole group of articles depending on the binary data available for any AD: number of subjects and risk data (OR and RR with the corresponding 95% CI). Effect size was calculated based on studies that only showed the OR and respective 95% CI and the raw data from case-control and cohort studies. A second effect size was calculated independently with studies that only showed the RR and the respective 95% CI and the raw data from cohort studies. Different study designs were used to compute the same effect size since the effect size had the same meaning in all studies and were comparable in relevant aspects. Thus, this study was able to transform all values to log values (log odds ratio and standard error), which were used in the pooled analysis. This approach prevented the omission of studies that used an alternative measure.
A sensitivity analysis was done in which the meta-analysis results of the studies as a whole was compared to the same meta-analysis with one study excluded in each round to determine how robust the findings were. It was also done to evaluate the impact of decisions that lead to different data being used in the analysis and whether the conclusions reached might differ substantially if a single study or a number of studies were omitted.
Additional meta-analyses were done for studies with complex data structure and non-cumulative results if the information for the different effects was not totally independent. Thus, articles showing multiple independent subgroups within a study were considered in these analyses (i.e. different definitions of the disease, gender differences, toxic exposure or more than one comparison group within a study). To compare effects across subgroups we typically use subgroup as the unit of analysis in an independent meta-analysis.
Supplementary analyses were done for the association between each specific AD and OSs exposure. Additional analyses were also done grouping the data according to the exposure assessment category.
ORs were grouped by weighing individual ORs by the inverse of their variance. For each analysis, the final effect OR and 95% CI were obtained by means of both random and fixed effect models. The selection of the computational model was done based on the expectation that the studies shared a common effect size. The random effect model was preferred because it accepts that there is a distribution of true effect sizes rather than one true effect and assigns a more balanced weight to each study. It was also used because all the studies were considered to be unequal in terms of specific ADs.
Heterogeneity was calculated by means of Cochran's (Q) and Higgins's (I2) tests. The I2 test showed the proportion of observed dispersion that was real rather than spurious and was expressed as a ratio ranging from 0% to 100%. I2 values of 25%, 50%, and 75% were qualitatively classified as low, moderate, and high respectively. A significant Q-statistic (p,0.10) indicated heterogeneity across studies. Publication bias was determined using Funnel plots and Egger's regression asymmetry tests, and additional tests were applied if it was found.

Results
The search with the defined MeSH Terms in PubMed, SCOPUS, SciELO, and LILACS [DeCS Terms] retrieved 531 articles. Using text words, 794 articles were found in PubMed, SCOPUS, SciELO and LILACS. Nine additional records were identified through references ( Figure 1).
Types of exposure and exposure assessments are described in table 1 for each study. The average qBHs for the total publications included in the meta-analysis was 14.25 points (SD, 1.586; range, 11-17 points; 99% CI, 13.528-14.972) reflecting a categorical rating of moderate relationship.
We found a significant association between OSs exposure and the increased risk of developing an autoimmune trait by evaluating all ADs as a single group. Figure 2 shows the forest plot corresponding to the meta-analysis including the most relevant outcome per author where the final common effect size based on a random model was statistically significant (OR: 1.54; 95% CI: 1.25-1.92; p-value,0.001). The results of different measures for heterogeneity calculated for the analysis showed in Figure 2 were as follows: Q-value: 132.1; degree of freedom (Q):30; pvalue,0.0001; I-squared: 77.3%; Tau-Squared 0.19. The relative weight of each study is included in the forest plot ( Figure 2) There were 5 studies showing complex data structure with different and non-cumulative results where the information for the different effects was not totally independent [52,57,70,76,78]. Then, 22 additional meta-analyses including 30 articles and the different outcomes of four of the above mentioned studies were calculated independently [57,70,76,78]. These analyses included five from Diot et al. 2002 [76] (different toxic exposure measured: chlorinate, ketones, aromatic, toluene, TCE), one from Nelson et al. 1994 [57] (control not disabled population) and Purdie et al. 2011 [70] (a different cutoff point to disease criteria) and fifteen from Thompson et al. 2002 [78] (different toxic exposure measured: toluene, benzene, white spirit, perchlorethylene, TCE, trichlorethane, vinyl chloride, urea formaldehyde, meta-phenylenediamene, bicromade, aromatic hydrocarbons, aliphatic hydrocarbons, fenfluramine, diethylpropion, L5 OH-tryptophan). In these meta-analyses, the studies that provided uniquely RR data were not included [52,54] for statistical reasons. All these additional meta-analyses showed a significant association between the exposure to OSs and ADs as a trait (Figures S1, S2, S3, S4, S5, S6, S7, S8, S9, S10, S11, S12, S13, S14, S15, S16, S17, S18, S19, S20, S21, S22). After doing a sensitivity analysis excluding one study at a time, the results were similar to the cumulative analysis ( Figures S23 and S24).
Additional analysis limited to the association between each specific AD and OSs exposure presented significant associations in the random model. For MS, the OR was 1.53 with 95& CI 1.03-2.29 and p value: 0.035, with fifteen studies included. For primary systemic vasculitis (PSV), the OR was 3.15 with 95% CI: 1.56-6.36 and p-value: 0.001, with one study included in the cumulative analysis for this disease. Systemic sclerosis (SSc) showed these  Self-report OS exposure (qualitative) was defined as 1 hour or more weekly for 3 consecutive months or longer. Increased risk was detected in exposed to fuels.     Evidence of significant publication bias was identified using the Egger test (p-value 2-tailed: 0.002; intercept: 1.09) for the metaanalysis which included studies that report OR with its respective 95% CI and raw data from case control and cohort designed studies. The Funnel plot showing the standard error on the Y axis is shown in Figure S27. Therefore, a second analysis was run in a search for publication bias. The classic fail-safe analysis indicated that 279 missing studies would give a p-value of .0.05. Begg and Mazumdar rank correlation was not significant (p-value 2-tailed: 0.16) and the trim and fill adjustment did not suggest a lower risk than the original analysis [adjusted values (11 studies trimmed) point estimate 1.03 (0.83-1.28), Q value: 227]. Based on all the analyses for publication bias, we consider the impact of bias in the present meta-analyses trivial.
Since 1977, 20 publications including case-reports and case series (Table S2) have reported 37 cases of AD possibly being triggered by OSs. We also found 3 previous meta-analyses. The first was published in 1996 by Landtblom AM et al [29] and concerned OSs exposure as a cause of Multiple Sclerosis (MS). They found 13 studies and reported an overall RR of 1.7 (with a  [31] published the most recent meta-analysis about occupational exposure to solvents and gender-related risk of SSc, they found a statistically significant association of SSc with OS exposure (OR 2.4; 95% CI 1.7-3.4: P = 0.002) and concluded that whereas SSc affects women predominantly, among subjects with occupational exposure to OS, men are at a higher risk of developing the disease than women. All the studies included in these publications were examined in our analysis. No meta-analysis evaluating ADs as a trait was found.
Regarding the systematic search for the OSs molecular mechanisms related to responses of immune system and ADs, with defined MeSH Terms and text words, retrieved 893 articles. After duplicates were removed, we obtained 827 articles of which 86 were included according to the inclusion criteria. The results are described in detail in Tables S3 and S4 and inclusion/ exclusion criteria are described on Figure S28. Table 2 shows selected articles, representing main molecular processes related with OSs exposure and their potential implication on immune system or autoimmune pathologies. We found that the effects of OSs on the immune system include lymphoproliferation, autoantibody production, Th1 and Th17 responses, oxidative stress, protein modification as well as effects on gene expression.

Discussion
Our results indicate that OS exposure is a risk factor for developing ADs. Even though the individual meta-analyses (i.e. each AD considered separately) disclosed significant association for MS, PSV and SSc (Figure 3), the direction and significance of this association did not change when all ADs, considered as a single trait, were analyzed ( Figure 2).Different combinations of factors involved in the generation of autoimmunity produce diverse clinical pictures within the wide spectrum of ADs (mosaic of autoimmunity) [2]. Our study, which takes into account both OSs as a whole and each solvent separately, reinforces this as well as the fact that ADs might share several common mechanisms(i.e., the autoimmune tautology) [98]. However, the term ''separately,'' which is used to refer to the studies that analyze only one solvent, is not the most biologically appropriate because most of the solvents are a mixture [99].
Our meta-analysis with ORs as the measure of association including 31 articles regarding 8 ADs showed a significant relationship of OSs exposure with ADs (OR: 1.54; 95% CI: 1.25-1.92; p-value,0.001) and that with RRs as the measure of association including 10 articles and 5 ADs showed a near significant relationship (OR: 1.62; 95% CI: 0.99-2.65; pvalue:0.051). When each AD was considered individually, there were also significant results with MS, PSV, SSc and PBC, although the latter was positively associated but not statistically significant.
A systematic and comprehensive review of the effects of OSs on the immune system is shown in Table 2. OSs are capable of altering cellular proliferation, apoptosis and tissue-specific function . Both the amount and duration of OSs exposure are Organic Solvents and Autoimmune Diseases PLOS ONE | www.plosone.org essential in pathology causation. Chronic exposure to OSs might lead to deposits in an organ and consequently to immune infiltration, similar to what is observed in ADs. The self-proteins that are modified by OSs may become immunogenic, recognized as foreign, and then initiate an inflammatory response and tissue injury. In this regard and according with our results, there are similar pathways operating on the incidence of ADs, but there are also specific mechanism that could lead to the particular manifestations of each AD; for instance, lymphocyte infiltration and immunoglobulin's deposits in SLE, and enzymatic alteration and scleroderma-specific antibody subsets in SSc [86,106,109].
Ketones are the most common OS used by the general population. Acetone, the simplest example of the ketones, is a commonly used solvent and is the active ingredient in nail polish remover and some paint thinners. It has been suggested that nail polish use may be associated with PBC [67]. These data are intriguing in view of the xenobiotic hypothesis proposed for the development of PBC with specific halogenated compounds. These compounds could increase the immunogenicity of mitochondrial proteins and induce anti-mitochondrial antibodies in animal models [127]. In fact, only one clinical study was included in the meta-analysis regarding PBC and nail polish exposure disclosing a positive associated but not statistically significant [67]. More studies involving PBC patients searching for this association could be useful.
Long term exposure to OSs seems to foster massive hepatic mononuclear infiltration leading to autoimmune hepatitis although it is important to highlight that this infiltration is the first step in the immunopathogenesis of not only autoimmune hepatitis but also the rest of the ADs [108]. As shown by Cai et al [109], lymphocyte infiltration was found in the pancreas, lungs, and kidneys in addition to the liver.
In autoimmune thyroid disease, it is probable that solvents may interfere with iodine transportation and induce oxidative stress that leads to an inflammatory response to the thyroid gland [128].
The relevance of our results rely in the fact that relation between SSc and environmental exposure, especially involving OSs is significant. Mice MRL+/+, an autoimmunity susceptible strain, when exposed to TCE increase the total IgG serum concentration, antinuclear antibodies (ANAs) and anticardiolipin autoantibodies [100]. On the other hand, in an in vitro model of human epidermal keratinocytes, was possible to determinate that TCE not only stimulates reactive oxygen species release, but also it stimulates nitric oxide synthesis by nitric oxide synthase. These cellular changes may contribute to the physiopathological process that lead to skin injury such as shown in SSc [106]. The biological mechanisms by which OSs may induce the development of ADs support the results observed trough the meta-analysis.
Concerning MS, when an independent analysis was done for each disease, MS show a significant association with OSs exposure. These results are like those reported by Landtblom et al [29], in their 1996 meta-analysis. Landtblom et al implemented a Mantel-Haenszel RR calculation. The main differences between their analysis and ours rely on the statistical approach because the Mantel-Haenszel method for combining OR is an alternative to the fixed-effect inverse variance method and we developed a random effect model. Our meta-analysis included 15 MS studies, 7 new to the previous meta-analysis [26,55,59,[63][64][65][66] published between 1994 and 2012.
The precise mechanisms responsible for the development of environmentally-induced autoimmune disorders are unknown. Although many hypotheses for the occurrence of autoimmune phenomena after various environmental exposures have been proposed, none of the hypotheses is completely supported by direct causal evidence. Also, mechanisms thought to be involved in the initiation of the disease process might differ from the mechanisms believed to exacerbate an established illness. However, the experimental approaches have been able to identify different environmental factors that use the same toxicity paths and mechanisms and either individually or jointly can have strong effects on molecular signaling pathways, immune responses or regulation mechanisms actively involved in health and disease ( Figure 4 and Table 2).
It could be suggested that, as described for autoimmune/ inflammatory syndrome induced by adjuvants, the toxic effect TCE Inhibition of cellular apoptosis of naive CD4+ and CD8+ T cells subset; anti-histone autoantibodies production (Mouse). [111] TCE-TCAH Inhibition of lymphocyte apoptosis in the thymus through decrease FasL or peripheral lymphocyte (Mouse). [112] TCE + HgCl2 Anti-Hsp90 autoantibodies and antibodies against liver proteins production (Mouse). [114] TCE2DCVC High-doses cause cellular necrosis. Low-doses produces changes in the transcription of several genes involved in apoptosis and cellular proliferation (Human). [115] TCE Low-doses cause DNA-hypermetylation on cardiac myoblasts (Rat). [116] Benzene Decrease of T cells population (cellular immunity) (American kestrels Birds). [117] Benzene Decrease in number of CD4+ and CD8+ T cells, B cells, granulocyte and platelets (Human). [118] Benzene Increase of ROS production and induction of DNA-fragmentation (Human). [119] Benzene Changes in gene transcription involved in apoptosis, oxidative stress, cellular cycle and cytokine production (Mouse). [121] Toluene High-doses affect the IFN-c, IL.-4 and IL-13 production by T cells and increases TNF-a expression (Human PBMCs). [122] Toluene-hexane-Methyl ethyl Ketone Produce oxidative stress (Human Jurkat Cells). [123] TCE -Benzene-HgCl2 Changes in gene expression with effects on cellular proliferation, apoptosis and tissue-specific function (Rat).
[ 124] In parenthesis is shown the model where the effect was studied. influences the appearance of these conditions only in subjects who are genetically susceptible [13].

Study Limitations
Significant differences between case-control and cohort models were found. This fact can be explained by the limitations of each of these methodologies [129]. The following are the limitations in case-control studies. (I) the information about exposure is primarily based on interviews and may be subject to recall bias.
(II) Validating the information on exposure can be difficult or even impossible. (III) By definition, case-control studies evaluate a single disease. (IV) The selection of an appropriate control group might be difficult. Most of the studies ignored the common origin of ADs and this generates the possibility of including patients with an underlying autoimmune condition as controls.
Most of the cohort studies included in our meta-analysis were retrospective. This implied that: (I) data was collected before the research hypothesis was defined leading to inaccurate data for the research. (II) The crude information was taken from databases or census. Therefore, the report on the exposure is not a direct quantification of the exposure. (III) The outcome information came from databases or medical records, but the subjects were not examined and this can lead to misdiagnosis. The explanation for why the result in the meta-analysis that included studies that reported the RR and raw data from cohort studies was not significant could be based on the abovementioned information as well as the low power due to a small sample size.
Exposure misclassification is a major problem when assessing the roll of environmental factors in complex diseases. Most individuals are not aware of the specific agents to which they have been exposed, and databases do not provide further information. None of the studies included in this meta-analysis employed an objective method of exposure assessment. Only two studies retrieved in this search [83,85] reported a direct-quantitative measure of exposure (i.e. TCE in concentrations from 6 to over 500 parts-per-billion [83]). After performing the analyses according to the exposure assessment category (figure S26) the final common effect size remained significantly associated as risk factor. A significant effort is necessary to determine the proper way to test the causal factors for autoimmunity. Nevertheless, we believe that identifying the causal pathways of toxics already known to be associated with generating autoimmunity is a breakthrough. Standardizing the pathways as validated biomarkers would lead to more accurate studies. Future research on environmental exposure will enhance our knowledge of the common mechanisms associated with ADs.
In conclusion, an association between OSs exposure and ADs was observed. This approach could be applied to any study of the association between exposure to other toxics and ADs. Although OSs exposure has not yet been sufficiently investigated, in order to clarify their roles in ADs pathogenesis, there is a need to study their relationship with genes associated, whether involved in protection or susceptibility to each AD and their effects on development of the autoimmune process. Figure S1 Forest plot of supplementary meta-analyses. Footnote: final common effect size based on a random model. Odds Ratio (95%CI) with raw data from case control and cohort designed studies were included. Studies that provided uniquely RR data were not included for statistical reasons. Each different outcome of the studies with complex data structure was