Is Thyroid Function Associated With Polycystic Ovary Syndrome? A Bidirectional Mendelian Randomization Study.

Abstract


Introduction
Thyroid dysfunction is a common endocrine disorder, which affects many aspects including human metabolism, growth, development and nervous system.Some research does suggest that the risk of hyperthyroidism in women is 5-10 times than that of men, and the risk of hypothyroidism is 4-8 times 1 .Abnormal thyroid function in women can affect the reproductive system, cardiovascular system and many other aspects 2 .Beyond that, a observational study of 305 patients with PCOS found that about 7% patients had coexisting hyperthyroidism 3 .In addition, hypothyroidism may lead to a decrease in Estrogen levels, which may further affect female reproductive function 4 .
Polycystic ovary syndrome (PCOS) is a kind of heterogeneous endocrine disorder characterized by irregular menses, hyperandrogenism and polycystic ovaries.Patients with PCOS usually manifest as reproductive abnormalities, marked insulin resistance, which may increase the incidence of type 2 diabetes mellitus, coronary heart disease and atherogenic dyslipidemia 5 .Some scholars have shown that there are 116 million women affected by PCOS worldwide 6 and PCOS is a highly prevalent endocrine disorder affecting multiple aspects of a women's overall health, with long-term effects that transcend well beyond the reproductive age 7 .However, PCOS is a complex, multifactorial disorder with a strong heritable component 8 .Coi ncidentally, a major focus on thyroid dysfunction concerns is its relationship with PCOS, since both disorders are the major causes of infertility, a medical problem of a growing prevalence that is associated with strong physical, emotional and socioeconomic consequences.Autoimmune thyroid disease (AITD) and hypothyroidism are more common in PCOS patients, and the incidence of HT in PCOS patients increased by 4.42   times in a cross-sectional study in Turkey 9 .The coexistence of both diseases is more dangerous and the exact pathogenesis is still unclear.It is thought that genetic inheritance, immunity and endocrine hormones may play important roles 3,10,11 In spite of lots of epidemiological observation studies, however, a key question whether these associations represent causal relationships remains unproven.Prior studies have given several probable hypothesizes on explaining the mechanisms of comorbidity.However, these hypothesizes remains arguable, as there are no prospective cohort study relevant to this eld.To avoid these pitfalls, Mendelian randomization (MR) is a powerful method for identifying causal relationships between risk factors and diseases using genetic variation as an instrumental variable 12 .MR is a method that uses genetic variants-single nucleotide polymorphisms (SNP) as Instrumental Variables (IV) to infer causality between exposure and outcome method of relationship, which is similar to randomized controlled trials in the natural state.Based on that, it is essential to investigate the casual relationship between PCOS and thyroid function, and a bidirectional MR is more re ective of the causal order.

Study design
At present, with a large number of genome wide association study (GWAS) summary data has been published, two-sample MR has been promoted vigorously, making it an effective meaning to explore the causal relationship between exposure and outcome in observational study.
Here, we performed a bidirectional two-sample MR analysis based on public data from different GWAS to explores the causal relationship between thyroid dysfunction and PCOS, veri ed the robustness of the results with public datasets from different sources and MR methods with different model assumptions.In addition, a reverse causal association was veri ed by another MR study to further verify the results and provide new ideas for the etiology of thyroid dysfunction and PCOS.A schematic overview of the data sources, genetic instrument selection, and statistical analysis in this study is presented in Fig.

Data sources
Female speci c Summary-level data for Thyroid Function of Europeans were obtained from the UK Biobank (http://nealelab.is/uk-biobank.com)and Thyroid Omics Consortium (http://www.thyroidomics.com).To be more speci c, the GWAS of Hyperthyroidism (22,383 cases and 54,288 controls) and Hypothyroidism (27,383 cases and 54,288 controls) were obtained from UK Biobank.The de nition of Hyperthyroidism and Hypothyroidism were International Classi cation of Diseases-10 (ICD-10) and self-reported, which is a population-based survey of UK residents aged 40-69 years recruited between 2006-2010 from 22 assessment centers across the UK.In addition, Normal TSH (54,288 cases and 72,167 controls) and Normal FT4 (49,269 cases and 72,167 controls) were obtained from Thyroid Omics Consortium 13 .It is worth noting that normal TSH and FT4 were analyzed as continuous variables within the cohort-speci c reference range after inverse normal transformation according to the original trait de nition.It is recommended to refer to the original publication for comprehensive information on sample collection, analytical methods, and results.Exclusion criteria for all analyses were non-European ancestry, use of thyroid medication (de ned as Anatomical Therapeutic Chemical code H03), or previous thyroid surgery.Data on PCOS were obtained from a large-scale meta-analysis of PCOS GWAS conducted by Day et al., including 10,074 cases and 103,164 controls of European ancestry, where participants were diagnosed with PCOS according to National Institutes of Health (NIH) criteria, Rotterdam criteria, or self-reported diagnoses 14 .For replication analysis, we used another independent PCOS GWAS from FinnGen R9(https://www.nngen./en, including 3,205 cases and 204,492 controls 15 .In addition, we performed additional analysis by using another PCOS GWAS (2,619 cases and 160, 321 controls) from Estonian Biobank (EstBB) adjusted for BMI and age in order to obtain more accurate conclusions, which found that the association between FTO genotypes and BMI was stronger in the cohorts with PCOS than in the general female populations 16 .we download it from the GWAS Catalog at ebi.ac.uk/gwas/ with accession ID numbers GCST90044903).In this cohort we clarify the mean age of the population with PCOS ranged from 25 to 32 years old 17 .Detailed information regarding the GWAS genetic dataset employed in the present study can be found in Table 1.(3) FinnGen*: FinnGen is a large public-private partnership project aiming to collect and analyze genome and health data from 500,000 Finnish biobank participants (4) UKBiobank*: UK Biobank is a large-scale biomedical database and research resource, containing in-depth genetic and health information from half a million UK participants.The database, which is regularly augmented with additional data, is globally accessible to approved researchers and scientists undertaking vital research into the most common and life-threatening diseases.
(5) THYROIDOMICS CONSORTIUM*: The THYROIDOMICS CONSORTIUM is an organization that studies the determinants and effects of thyroid diseases and function.In order to meet the three hypotheses, We selected SNPs at a genome-wide level (P<5×10 − 8 ) using the clumping algorithm (r 2 threshold = 0.001 and window size = 10,000 kb).But In our primary analysis, limited by the original PCOS GWAS database, we could only adjusted the threshold (P<5×10 − 6 ) to ful ll the criteria of having at least 3 SNPs to analyze 25 .Although 14 genome-wide signi cant SNPs identi ed in the PCOS GWAS conducted by Day et al., 13 SNPs were selected at last as genetic IVs for PCOS after excluding rs853854 (MAF close to 0.5).LD clumping with the same threshold as above.Therefore, in order to make our MR analysis results rigorous, we used PCOS GWAS from FinnGen R9 for replication analysis and 5 genome-wide signi cant (P < 5 × 10 − 8 ) SNPs were identi ed.After adjusting for BMI, we still obtained 4 SNPs in EstBB PCOS GWAS.
We also calculated F-statistic to assess whether the selected SNPs were strongly associated with exposure.In general, F-statistic >10 indicates no weak instrumental bias.We referred to human reference assembly (GRCh37/hg19) to address the issue of rs ID missing.SNPs used to construct the main instrument variable for PCOS in European ancestry individuals have been shown in Table 2.

Statistical Analyses
We performed two sample MR analyses to investigate the association between thyroid function and PCOS.According to the number of SNPs we ltered out above, we selected inverse variance weighting (IVW) as the primary MR analysis.The IVW considers all variants valid or balanced pleiotropy and provides a combined mean effect, which can yield consistent and unbiased causal effect estimates when all instrumental variables are valid 18,19 .However, the IVW method could be in uenced by invalid instrumental variables or horizontal pleiotropic effects.We further conducted several sensitivity analyses based on the MR-Egger and weighted median.Although the MR-Egger is more sensitive to outliers, less e cient potent than IVW analysis 20 .The weighted median estimates the median or mode of the ratio estimate distribution as a causal effect 21- 22,23 .
To assess the heterogeneity among the estimate from each SNP, we conducted a Cochran Q test.It was observed that most of the results showed no heterogeneity with P-value more than 0.05.Next, to further assess causality and investigate the presence of pleiotropy, we performed a set of checks, including MR -Egger Regression and MR-PRESSO 24 .We also used PhenoScanner to examine the potential dimorphic in the individual SNPs assessed to eliminate their potential impact on the results(http://www.phenoscanner.medschl.cam.ac.uk/ ).In this session, rs2068834 and rs429358 were excluded from the sensitivity analysis due to their genome-wide signi cant association with obesity and analyzed again, considering that the potential dimorphic phenotypes of these SNPs may affect our results.
For the interpretation of our MR results, we further used Bonferroni correction to solve the multiple comparison problem 25 .Our MR analysis included 4 exposure factors and 1 outcome, so we set the threshold for statistical signi cance after Bonferroni correction at 0.0125 (0.05/4*1), The p-value between the standard threshold (p = 0.05) and the statistically signi cant one after the Bonferroni correction would be taken as suggestive evidence of a nominal causal association.Most of the above work was performed in R analysis software (version 4.0.3)(R Foundation for Statistical Computing, Vienna, Austria, www.R-project.org ), applying to the related R package including "Two sample MR", "MRPRESSO"Etc.

Results
According to our primary MR analyses via forward MR by the IVW, weighted median and MR-Egger methods, genetically predicted hyperthyroidism, hypothyroidism and normal FT4 didn't have statistical causal association with the risk of PCOS (the P value in each study separately showed > 0.05).The results were similar in the replication studies of FinnGen PCOS GWAS and secondery meta-analysis cohort study of PCOS from EstBB with adjustment for BMI (removing FTO genotypes) and Age (the mean age of the population with PCOS ranged from 25 to 32 years old.).However, we did nd suggestive evidence of an interaction between genetically determined normal TSH and the PCOS cohort after adjusting for age and BMI [P IVW = 0.021; OR = 0.60 (95% CI = 0.61 to 0.96)], which implied normal TSH is a protective factor against the pathogenesis of PCOS.Considering in MR analysis, Bonferroni correction needs to be used to address the multiple comparison problem.In our study, the original hypothesis was rejected only if the p-value 0.0125 (the threshold for statistical signi cance after Bonferroni-correction).Therefore, we considered that normal TSH had a nominal protective effect against PCOS in this study.In the reverse MR analysis (PCOS as exposure and hyperthyroidism, hypothyroidism, normal FT4 and normal TSH as outcomes).The eligible IVs were not causal associated with the outcome (P 0.05).Considering that TSH and FT4 were continuous variables in reverse MR analysis, so when they are used as outcome phenotypes, beta values need to be used Beta (95% CI) instead of OR (95% CI).Table 4 Main results of MR analysis estimates and sensitivity analysis estimates for the association of PCOS with thyroid function.The results of Cochran Q test for IVW to detect heterogeneity also showed P-value more than 0.05, which means no heterogeneity present.In addition to this, in all cases, the MR-Egger intercepts were not different from zero, indicating absence of directional pleiotropy.

Discussion
In this bidirectional MR study, we explored the causal relationship between different thyroid function status and PCOS.In order to make our experimental design more rigorous, we used different PCOS cohorts for repeated validation.Since considering that PCOS only occurs in females, we used gender-strati ed thyroid function data for analysis.In our study examining the causal relationship between thyroid function and PCOS, we found no signi cant association between hyperthyroidism, hypothyroidism, normal TSH and normal FT4 and PCOS.But in our PCOS cohort adjusted for BMI and age, we found that normal TSH was a protective factor for PCOS even though we did not up to the standard of Bonferronicorrection.It was worth noting that this effect was not disturbed by the mixing of reverse causality.This nding at least suggests that thyroid function status in some age and BMI sub-groups is closely related to the occurrence of PCOS.
Our results are not in line with previous studies, previous studies have shown that women with PCOS have a higher prevalence of hypothyroidism and AITD 26 .However, Gawron IM et al recently conducted the research to compare thyroid volume, autoimmunity, and thyroid function with PCOS different subtypes.According to different clinical manifestations, they divided PCOS patients into 4 phenotypes: Phenotype A(hyperandrogenism, HA + oligo-anovulation, OA + polycystic ovarian morphology, PCOM), Phenotype B(HA + OD);phenotype C(HA + PCOM) and phenotype D(OD + PCOM) and reported that none of the PCOS phenotypes were found relating to abnormal thyroid function or AITD 27 .In fact, contrary to some positive studies, it seems that the classic form of PCOS (phenotypes A and B) constitutes approximately two-thirds of the total of PCOS patient.However, it is worth noting that observational studies cannot rule out the possible mediating effect of hyperandrogenism and obesity.
Though our results suggest that PCOS does not cause thyroid dysfunction, several common features such as hyperandrogenism, overweight and insulin resistance (IR)of PCOS do appear to cause, which may explain the epidemiologic association.Coincidentally, another study shows that overweight and hyperandrogenism PCOS patients showed signi cantly higher prevalence of hypothyroidism (TSH >2.5 mIU/L) 28 Primarily, increased BMI is very prevalent in women with PCOS, observed in 54-68% of cases 29 .Although the pathophysiological mechanisms linking thyroid function and obesity have not been clearly established, evidence indicates that overweight and obese patients with PCOS showed a higher tendency toward hypothyroidism were signi cantly more common in patients with BMI > 25 kg/ m 2 (56%) than in those with BMI ≤ 25 (25.8%,p < 0.005) 28 .Interestingly, previous MR studies have repeatedly suggested that obesity is fully causal for the development of PCOS 30 , and at the same time MR evidence continues to suggest that obesity is an inevitable trigger for thyroid nodule [OR = 1.15, 95% CI (1.08-1.22)]and benign nodular thyroid disease, which is one of the common reasons for hypothyroidism 31 .In our study, the positive result was found between normal TSH and PCOS after adjustment for BMI, suggesting that we still need to focus on confounding factors such as BMI on outcome in the future studies.
It is noteworthy that our study has several strengths.On the one hand, we conducted a two-sample MR analysis using robust GWAS data for thyroid function and PCOS.Unlike observational studies, our analyses included a large number of SNPs associated with thyroid function, and we got similar results.In this study, we selected SNPs with genome-wide association and independent inheritance as IVs to detect the causal link between thyroid function and PCOS.To make our conclusions more robust and reliable, the outlier variants identi ed by the MR-PRESSO outlier test were removed step-by-step.We also utilized several robust analytical methods based on different assumptions of two-sample MR analysis with nine groups of summary GWAS data.Since we included many weak instrumental variables in the analysis, the F statistics were used to assess the strength of the association between the genetic variants and exposure.All the F statistics were much greater than 10 in our analysis, hinting the small possibility of weak instrumental variable bias.On the other hand, there is a little research on whether thyroid function can cause PCOS and play a role in the pathogenesis of PCOS.Ideally, prospective studies should compare the incidence of PCOS in patients with thyroid function and in control groups, then estimate whether thyroid function has an impact on the onset of PCOS.In our study, we utilized MR to imitate this process, which can exclude the association of some common confounding factors on genes and phenotype From the perspective of clinical practice, MR could have helped avoid several very expensive late-stage clinical trial failures and might improve prediction of what RCTs will show.
Our study adds to the limited evidence considering that there is no causality between thyroid function and PCOS, although we cannot rule out their potential effects PCOS risk factors.At last, taking advantage of large GWAS and large case-control studies with extensive genotyping enabled us to conduct a well-powered MR study in a cost-e cient way.
Some limitations of our MR analysis need to be considered.First, the summary-level GWAS data merely concern individuals of European descent, and the European ancestry of the samples also limits generalizability to other ancestries.Second, a negative result may re ect a true causal relationship, which means the exposure has no or little effect on the outcome.In this case, MR studies need to increase sample size, select stronger or more instrument variables, perform strati ed or subgroup analysis and other methods to improve statistical power and accuracy.

Conclusion
In conclusion, we found no causal relationship between thyroid function and PCOS among European ancestry by bidirectional MR analysis.
However, our study cannot deny maintaining healthy thyroid function is essential for preventing the occurrence of PCOS among female from 25 to 32 years old, especially those with low BMI.Patients suffering from PCOS without obesity are suggested to take steps in advance to avoid the additional risks associated with weight gain.

Figures
Figure 1 The

Supplementary Files
This is a list of supplementary les associated with this preprint.Click to download.
SupplementaryTable1STROBEMRchecklist llable.docx diagram of Bidirectional Mendelian Randomization Study 1.This study was reported according to Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) MR guidelines.(SupplementaryTable 1 STROBE-MR checklist table)

Table 1 The
Inclusion criteria and data sources of GWAS we used (1) Women diagnosed with ICD-10 code E28.2, ICD-9 code 256.4,or ICD-8 code 256.90 (2) Other women without PCOS in the FinnGen study 3,205 204,492 EstBB PCOS GWAS (1) Women diagnosed with ICD-10 code E28.2 (2) Other women without PCOS in the Estonian Biobank (3) Secondary GWAS metaanalysis adjust the PCOS GWAS for BMI and Age* 2,619 160, 321 Hyperthyroidism UKBiobank* (1) ICD-10 Hyperthyroidism and Hypothyroidism and selfreported: hyperthyroidism/thyrotoxicosis and hypothyroidism/myxoedema (2) In UKB, all participants that were not identi ed as being cases for respective disease were used as controls.(1) Day et al.GWAS*: The Day et al.GWAS we adopted is based on information from 10,074 PCOS cases and 103,164 controls of European ancestry, from 7 cohorts, in which the diagnosis of PCOS is based on National Institute of Health (NIH), Rotterdam criteria, self-reported diagnosis (2) EstBB*: The EstBB is a volunteer-based biobank with over 200 000 participants, currently including approximately 135 ,000 women .
Adjusting the PCOS GWAS for BMI and Age*: In the discovery EstBB PCOS dataset, an additional meta-analysis was conducted adjusting for BMI (removing FTO genotypes) and Age (the mean age of the population with PCOS ranged from 25 to 32 years old.)Abbreviations: GWAS, genome-wide association study; ICD, International Classi cation of Diseases;NIH, National Institutes of Health Ethics All summarized statistics utilized in the MR analyses were generated by previous studies, for which ethical approval and individual consent were obtained for all original studies.
The organization uses various omics technologies (including genomics, epigenomics, transcriptomics, proteomics and metabolomics) to analyze multiple large and well-phenotyped cohort studies .

Table 2
SNPs used to construct the main instrument variable for PCOS in European ancestry individuals Notes: Genomic positions reported in the GWASs refer to human reference assembly (GRCh37/hg19) Abbreviations: CHR chromosome, EA effect allele, EAF effect allele frequency, EstBB Estonian Biobank, NAFLD non-alcoholic fatty liver disease, OA other alleles (reference allele), PCOS polycystic ovary syndrome, SE standard error, SNP single nucleotide polymorphism

Table 3
Main results of MR analysis estimates and sensitivity analysis estimates for thyroid function with PCOS

Table 3
Main results of MR analysis estimates and sensitivity analysis estimates for the association of thyroid function with PCOS

Table 4
Main results of MR analysis estimates and sensitivity analysis estimates for the association of PCOS with thyroid function