Abstract

   Polycystic ovary syndrome (PCOS) affects reproductive and
   cardiometabolic health, yet its pathogenesis remains unclear. Emerging
   evidence links hemoglobin levels to metabolic disorders, suggesting a
   potential role in PCOS development. Here, we integrated a large‐scale
   cohort study, Mendelian randomization (A genetic tool to infer causal
   relationships), bioinformatics analyses, and in vitro experiments to
   investigate the relationship between hemoglobin levels and PCOS. In a
   cohort of 20 602 women, each 10 g L^−1 elevation in hemoglobin levels
   is associated with 22% higher odds of PCOS (adjusted odds ratio: 1.22,
   95% confidence interval: 1.15–1.29, P < 0.001) and PCOS manifestations,
   particularly hyperandrogenism. Mendelian randomization analysis
   confirms that higher hemoglobin levels are associated with increased
   PCOS risk and elevated testosterone levels. The hypoxia‐inducible
   factor 1 (HIF‐1) pathway is enriched, identifying three
   testosterone‐associated genes (nuclear factor kappa B (NFKB1), insulin
   receptor (INSR), protein kinase C alpha. Colocalization and
   druggability analysis supports shared genetic regions and confirmed
   these genes as druggable targets. Upregulation of NFKB1 and INSR are
   confirmed in both blood and ovarian granulosa cells of PCOS patients.
   The findings demonstrate that higher‐end normal hemoglobin levels are
   associated with increased PCOS risk, potentially through a mechanism of
   elevating testosterone levels involving the HIF‐1 pathway.

   Keywords: causality, hemoglobin, hyperandrogenism, hypoxia‐inducible
   factor, polycystic ovary syndrome
     __________________________________________________________________

   High‐end normal hemoglobin levels are associated with an increased odds
   of polycystic ovary syndrome (PCOS), potentially through a mechanism of
   elevating testosterone levels involving the hypoxia‐inducible factor 1
   (HIF‐1) pathway. These findings provide novel insights into the
   pathophysiology of PCOS and highlight that hemoglobin levels may serve
   as a translationally relevant target for understanding and management
   of PCOS.

   graphic file with name ADVS-12-2411679-g004.jpg

1. Introduction

   Polycystic ovary syndrome (PCOS) is one of the most common reproductive
   and endocrine disorders in women of reproductive age, with a global
   prevalence ranging from 5% to 18%.^[ [50]^1 ^] It is characterized by
   three typical features: hyperandrogenism (clinical or biochemical),
   ovulation dysfunction (oligo‐ovulation or anovulation), and polycystic
   ovarian morphology (PCOM).^[ [51]^2 ^] The presence of PCOS has a range
   of implications for women's health. In addition to being one of the
   main causes of infertility disorders, it can also increase the risk of
   other health problems, such as type 2 diabetes, metabolic syndrome, and
   cardiovascular disease.^[ [52]^3 , [53]^4 ^] Despite the high
   prevalence and lifelong consequences, the etiology and pathophysiology
   of PCOS remain largely obscure.

   Hemoglobin (Hb) plays an important role in the transport and storage of
   oxygen and its levels are regulated by genetic and environmental
   factors and vary according to age, gender, race, and living altitude.^[
   [54]^5 ^] It is generally accepted that high‐end Hb levels within the
   normal range are beneficial to health.^[ [55]^6 ^] However, increasing
   evidence suggests that higher Hb levels are associated with adverse
   health effects.^[ [56]^7 ^] A few previous studies have reported
   associations between Hb levels and insulin resistance, hypertension,
   dyslipidemia, and metabolic syndrome.^[ [57]^8 , [58]^9 , [59]^10 ^]
   Although several small‐scale observational studies (each with fewer
   than 100 PCOS patients) have explored the relationship between Hb
   levels and PCOS, the findings have been inconsistent.^[ [60]^11 ,
   [61]^12 ^] The potential for biased results and misinterpreted
   statistical significance might also arise from inadequate adjustment
   for confounding factors, as these studies considered only age and body
   mass index (BMI) as confounders, respectively. Moreover, neither the
   causal association between Hb levels and the onset and progression of
   PCOS nor the underlying mechanisms have yet been explored.

   Mendelian randomization (MR) is a widely used epidemiological approach,
   employing genetic variants as instrumental variables (IVs) for exposure
   to draw causal inferences.^[ [62]^13 ^] By minimizing residual
   confounding and reverse causation, MR analyses are more likely to offer
   causal estimations compared to traditional observational approaches.
   Emerging studies have used MR to identify causal associations between
   PCOS and various factors, but the causal analysis with Hb has not been
   reported.^[ [63]^14 ^] Moreover, the integration of MR with pathway
   enrichment analysis can provide novel insights into the etiology and
   molecular mechanisms of diseases. Furthermore, leveraging genomics
   within MR offers a cost‐effective means to identify potential
   therapeutic targets in the early stages of drug discovery.^[ [64]^15 ^]
   Through the integration of molecular quantitative trait locus, such as
   gene expression quantitative trait loci (eQTL) or protein quantitative
   trait loci (pQTL), with genome‐wide association (GWAS) studies, target
   genes linked to risk variants can be discerned via causal inference.^[
   [65]^16 ^] Colocalization analysis could further identify these
   targets, highlighting genes with robust genetic associations and thus
   higher therapeutic potential.^[ [66]^17 ^]

   In the present study, we designed a systematic research framework to
   investigate whether Hb levels contribute to PCOS. We first performed an
   observational analysis to examine the association between Hb
   concentration within the normal range and PCOS. We then conducted MR
   analyses to assess the causality and pathway enrichment analyses to
   reveal potential mechanistic pathways involved, supplemented by
   colocalization analysis and druggability assessment to detect
   prioritize potential therapeutic targets. Lastly, we validated the
   expressions of the identified key genes through in vitro experiments.

2. Results

2.1. Associations of Hb with PCOS and PCOS Manifestations in the
Observational Study

   Among the 20 602 women (median [interquartile range, IQR] age, 31^[
   [67]^28 , [68]^29 , [69]^30 , [70]^31 , [71]^32 , [72]^33 , [73]^34 ,
   [74]^35 ^] years) included in the retrospective cohort study, the
   median duration of infertility was 3 years (IQR 2–5), with 69.2%
   experiencing infertility due to tubal factor, 9.5% of diagnosed with
   endometriosis, and 6.6% had a history of ovarian surgery. A total of
   3732 (18.1%) were diagnosed with PCOS according to the Rotterdam
   criteria; 964 (4.7%), 412 (2.0%), 511 (2.5%), and 1845 (9.0%) were
   diagnosed with PCOS phenotypes A, B, C, and D, respectively, according
   to the categorization by the National Institutes of Health (NIH); 5198
   (25.2%), 5650 (27.4%), and 3215 (15.6%) presented ovulatory
   dysfunction, PCOM, and hyperandrogenism, respectively. Table [75]S1 in
   the Supporting Information summarizes the baseline characteristics
   overall and across the tertiles of Hb concentrations. In general, women
   with higher Hb concentrations were more likely to present PCOS, all
   PCOS phenotypes and all three separate manifestations of PCOS.

   Logistic regressions demonstrated that elevated Hb levels within the
   normal range were associated with increased odds of all PCOS phenotypes
   and individual clinical manifestations of PCOS (Figures [76]1 and [77]
   2A). We observed higher Hb levels in PCOS patients according to the
   Rotterdam criteria compared with non‐PCOS (mean difference 95%
   confidence interval [CI]: 2.37 [2.09–2.64] g L^−1). After adjustments
   for most confounders, for every 10 g L^−1 increase in Hb levels, there
   was a 22% increase in the odds of developing PCOS (adjusted odds ratio
   [aOR] [95% CI]: 1.22 [1.15–1.29]). As Hb levels increased, the odds of
   hyperandrogenemia increased the most significantly (aOR [95% CI], 1.24
   [1.17–1.31]). Among the four PCOS phenotypes, the greatest increase in
   the odds was observed in phenotype A (aOR [95% CI], 1.38 [1.24–1.54]),
   in which all three clinical manifestations occurred simultaneously.
   Generalized additive model (GAM) partial effect curves confirmed the
   positive associations between Hb levels and the clinical outcomes
   (Figure [78]2B). There was a safe plateau phase in the association with
   testosterone levels when Hb ranged from 120 to 125 g L^−1, while
   associations with antral follicle count (AFC), ovulatory dysfunction,
   PCOS and PCOS phenotypes roughly remained linearity in the adjusted
   models.

Figure 1.

   Figure 1
   [79]Open in a new tab

   Central illustration of this study. Systematic analyses of the
   association between hemoglobin and polycystic ovary syndrome identify
   the potential effect of HIF‐1 pathway. Abbreviations: DEGs,
   differentially expressed genes; eQTL, expression quantitative trait
   loci; HIF, hypoxia‐inducible factor pathway; INSR, insulin receptor;
   NFKB, nuclear factor kappa‐light‐chain‐enhancer of activated B cells;
   PCOS, polycystic ovary syndrome; PRKCA, protein kinase C alpha; SD,
   standard deviation; SNP, single nucleotide polymorphism.

Figure 2.

   Figure 2
   [80]Open in a new tab

   The associations between Hb levels and PCOS and its manifestations. A)
   Univariate and multivariate logistic regression model analysis to
   characterize the associations between Hb levels and PCOS, PCOS
   phenotypes, and separate manifestations. Model 1 was adjusted for
   intervals of appointment date, age, body mass index, and education;
   Model 2 was additionally adjusted for systolic blood pressure,
   endometriosis status, history of ovary surgery, LH/FSH ratio, fasting
   blood glucose, total cholesterol, triglycerides, and high‐ and
   low‐density lipoprotein levels. B) Nonlinear associations between Hb
   levels and PCOS, PCOS phenotypes, and separate manifestations, adjusted
   for confounders in Model 2. Abbreviations: PCOS, polycystic ovary
   syndrome; LH, luteinizing hormone; FSH, follicle‐stimulating hormone.

2.2. MR Analyses Revealed the Causal Association of Hb Levels with PCOS and
Testosterone

   The F‐statistics for IVs in all MR analyses exceeded 10, suggesting
   that the MR estimates were unlikely to be subject to weak instrument
   bias. In the discovery stage, our MR analysis demonstrated that
   genetically predicted elevated Hb levels, per 1‐ standard deviation
   (SD) increase, were associated with increased odds of PCOS according to
   the inverse‐variance weighted (IVW) method (odds ratio [OR] [95% CI]:
   1.26 [1.05–1.51], P = 0.014), along with increased bioavailable
   testosterone (BT) levels (β [standard error, SE]: 0.05 [0.01]
   nmol L^−1, P = 6.110 × 10^−5), and higher total testosterone (TT)
   levels (β [SE]: 0.08 [0.01] nmol L^−1, P = 1.348 × 10^−8) (Figure [81]1
   and Figure [82]3A).

Figure 3.

   Figure 3
   [83]Open in a new tab

   MR results and pathway enrichment analysis. A) Associations of Hb
   levels with PCOS and related traits according to the Mendelian
   randomization analysis. B) Significant results for KEGG pathway
   enrichment analysis of candidate genes from Hb IVs used in MR analysis.
   C) Significant tissue‐specific eQTL MR results in whole blood and
   ovary. D) eQTL and pQTL MR results for INSR, NFKB1, and PRKCA on
   testosterone levels. E) LocusCompare plots comparing genetic signals
   for NFKB1 levels and TT levels in colocalization analysis.
   Abbreviations: BT, bioavailable testosterone; TT, total testosterone;
   SHBG, sex hormone binding globulin; PH4, posterior probability for
   shared causal variants.

   Subsequent replication analysis using another GWAS meta‐analysis
   statistics of Hb levels, with a smaller sample size, confirmed
   significant associations with PCOS risk and TT levels. The analysis
   indicated associations between increased Hb levels and higher risk of
   PCOS (OR 1.42 [95% CI]: [1.14–1.75], P = 0.001) and elevated TT levels
   (β [SE]: 0.05 [0.02] nmol L^−1, P = 0.002) (Figure [84]3A). Other MR
   methods provided statistically consistent estimates of effect
   directions (Table [85]S2, Supporting Information). The type 1 error
   rate caused by sample overlap between Hb and PCOS‐related phenotypes
   was controlled under 0.05, suggesting that the substantial effect of
   partial overlap on causality was acceptable (Table [86]S3, Supporting
   Information).

   In the reverse direction of MR analyses, no significant association was
   found between genetic susceptibility to PCOS and Hb levels in either
   the discovery or replication analysis (Table [87]S4, Supporting
   Information). The multivariate Mendelian randomization (MVMR) analysis
   revealed that the association between genetically predicted Hb levels
   and increased odds of PCOS remained significant, even after adjusting
   for the influences of BT (aOR [95% CI]: 1.25 [1.05–1.49], P = 0.014)
   and TT (aOR [95% CI]: 1.23 [1.03–1.48], P = 0.025), respectively.
   Similar significant associations were observed in the replication
   analysis of Hb levels in PCOS patients after adjustment for BT (aOR
   [95% CI]: 1.36 [1.20–1.68], P = 0.004) and TT (aOR [95% CI]: 1.29
   [1.01–1.65], P = 0.039) (Table [88]S5, Supporting Information).

2.3. Pathway Enrichment Identified the Hypoxia‐Inducible Factor 1 (HIF‐1)
Pathway

   Employing Hb‐associated genetic variants derived from the largest GWAS
   study, three pathways, including HIF‐1 signaling pathway,
   transcriptional misregulation in cancer, and TGF‐beta signaling
   pathway, were identified via Kyoto Encyclopedia of Genes and Genomes
   (KEGG) pathway enrichment analysis, with a false discovery rate (FDR)
   less than 0.05 as the significance threshold (Figure [89]3B and Tables
   [90]S6 and [91]S7, Supporting Information).

2.4. Key Genes Were Identified Using eQTL and pQTL MR Analyses

   Using eQTL data of genes within the HIF‐1 pathway sourced from the
   Genotype‐Tissue Expression (GTEx) v8 database across various tissues
   including whole blood, ovarian, hypothalamic, and pituitary tissues, we
   identified three pairs of genetically proxied HIF‐1 pathway genes
   significantly associated with testosterone levels. Employing a
   Bonferroni‐corrected threshold of P < 4.9 × 10^−4 (Bonferroni‐corrected
   threshold of 0.05 for 102 HIF‐1 pathway eQTLs as exposures) for
   multiple testing, these associations were found between specific tissue
   expressions of INSR (insulin receptor), NFKB1 (nuclear factor kappa B),
   and PRKCA (protein kinase C alpha), and genetically proxied
   testosterone levels (Figures [92]1 and [93]3C,D and Table [94]S8,
   Supporting Information). Consequently, our subsequent in vitro
   experiments focused on validating these three genes.

   Significant associations in the eQTL MR analysis were further
   replicated using pQTL for PRKCA (β [SE]: 0.12 [0.03] nmol L^−1,
   P = 1.276 × 10^−4) and NFKB1 (β [SE]: 0.22 [0.05] nmol L^−1,
   P = 7.263 × 10^−7) (Figure [95]3D). Moreover, we detected strong
   colocalization evidence for NFKB1 and TT (posterior probability for
   hypothesis [PH] 4 = 0.94), indicating a common variant shared between
   NFKB1 and increased TT levels (Figure [96]3E and Table [97]S9,
   Supporting Information).

2.5. Druggability Assessment Suggested Potential Development Prospects

   To assess potential development prospects, we collected data on
   clinical trials from drug databases for the three targets identified in
   the MR analysis to evaluate their druggability. Our findings indicate
   that these three proteins were considered druggable targets albeit at
   varying stages of drug discovery. Clinical trials predominantly focus
   on medications aimed at treating insulin resistance, as evidenced by
   the available data (Table [98]S10, Supporting Information).

2.6. In Vitro Experiments Confirmed the Expressions of the Identified Key
Genes

   The transcription levels of the identified genes, INSR, NFKB1, and
   PRKCA, were examined in both the peripheral blood cells and ovarian
   granulosa cells of women with PCOS and controls. Hb levels were
   significantly greater in PCOS patients than in controls. The gene
   expression levels of INSR and NFKB1 were significantly increased in
   both peripheral blood cells and granulosa cells of women with PCOS than
   in those of control individuals, while no significant difference was
   observed in the PRKCA transcription level between the two groups
   (Figure [99]1 and Figure [100]4A,B). The protein expression levels of
   INSR and NF‐κB were also elevated in the granulosa cells of PCOS
   patients compared with those in the controls (Figure [101]4C). Detailed
   clinical information of the PCOS and control groups was indicated in
   Tables [102]S11 and [103]S12 in the Supporting Information. In
   addition, linear regression analysis indicated a positive correlation
   between Hb levels and INSR expression in both peripheral blood cells
   and granulosa cells after adjusting for BMI (Figure [104]4D and Table
   [105]S13, Supporting Information). Moreover, the expressions of INSR
   and NFKB1 in peripheral blood cells showed a positive correlation to
   testosterone levels, with or without adjustment for BMI
   (Figure [106]4E).

Figure 4.

   Figure 4
   [107]Open in a new tab

   In vitro validation experiments in peripheral blood cells and granulosa
   cells. Hb levels, qPCR analyses of INSR, NFKB1 and PRKCA expression in
   A) peripheral blood cells and B) granulosa cells. The blue dots
   indicate the mRNA levels in the control group, and the red dots
   indicate the mRNA levels in the PCOS patients. P values, two‐tailed
   Student's t‐test or Mann‐Whitney test. C) Representative Western blot
   analysis of INSR and NF‐κB in the granulosa cells of the control and
   PCOS groups in three independent experiments, with the GAPDH level used
   as a control. D) Linear regression analysis of hemoglobin levels and
   INSR and NFKB1 gene expression in peripheral blood cells and granulosa
   cells after adjusting for BMI. E) Linear regression analysis of INSR
   and NFKB1 gene expression and testosterone level in peripheral blood
   cells before and after adjusting for BMI.

3. Discussion

   In the present study, we conducted a systematic investigation into the
   association between Hb levels and PCOS. We found that elevated Hb
   levels, even within the normal range, were associated with increased
   odds of PCOS and its manifestations. Our MR analyses provided robust
   evidence supporting a causal association between Hb levels and both
   PCOS and testosterone levels. Notably, the Hb‐associated genetic
   variants were significantly enriched in the HIF‐1 pathway, and two key
   genes associated with testosterone levels, INSR and NFKB1, were both
   confirmed to be upregulated in PCOS patients. Colocalization analysis
   suggested that NFKB1 shares a causal variant with TT levels,
   highlighting its potential as a therapeutic target for the treatment of
   hyperandrogenic states in PCOS patients. These findings provided new
   insights for the molecular mechanisms underlying PCOS pathogenesis and
   potential therapeutic targets for clinical intervention.

   Few previous studies have explored the association between Hb levels
   and PCOS. One study reported higher Hb levels in both lean and
   overweight/obese subgroups of adolescents with PCOS when compared to
   the age‐matched controls.^[ [108]^12 ^] However, some studies have
   reported no significant differences in hematologic values between PCOS
   patients and controls.^[ [109]^11 ^] In addition, Hb levels are also
   linked to a wide range of disorders, including metabolic syndrome,
   hyperuricemia, and nonalcoholic fatty liver disease, suggesting its
   extensive involvement in metabolic regulation.^[ [110]^18 , [111]^19 ,
   [112]^20 ^] The underlying mechanisms of these associations have not
   yet been thoroughly elucidated, but insights from previous studies may
   provide some potential evidence. In our observational analysis,
   elevated Hb levels were significantly associated with the risk of PCOS
   as well as its inherent indicators, such as hyperandrogenemia,
   ovulatory dysfunction, and polycystic ovarian morphology. The observed
   22% increased odds of PCOS per 10 g L^−1 rise in hemoglobin levels
   suggest that even high‐normal Hb levels may serve as a significant
   biomarker for heightened PCOS risk. Given the multifactorial nature and
   high prevalence of PCOS, these findings highlight the potential utility
   of Hb levels in risk stratification and early identification of
   individuals at greater risk for PCOS. Hb is essential for oxygen
   transport in the body and provides the required oxygen for cellular
   metabolism, energy production, and biological functions.^[ [113]^21 ^]
   Elevated Hb levels may indicate a state of chronic hypoxia or anoxia,
   which has been shown to lead to oxidative stress, insulin resistance,
   and inflammation.^[ [114]^22 , [115]^23 ^] All of these factors are
   associated with the onset and progression of PCOS. The relationship
   between high Hb concentrations and increased blood viscosity, along
   with alterations in plasma volume, endothelial cell dysfunction, or
   changes in the ovarian microenvironment, may also explain this
   association. Elevated Hb levels can lead to insulin resistance by
   increasing blood viscosity, reducing blood flow, and influencing the
   supply of oxygen and metabolites to the tissue.^[ [116]^24 ^]
   Furthermore, an alternative function of Hb may be as a carrier of
   nitric oxide (NO), which can modulate vascular endothelial function via
   the arginine‐NO pathway.^[ [117]^25 , [118]^26 ^] Consequently, higher
   Hb levels may induce endothelial dysfunction, also leading to insulin
   resistance. Moreover, alterations in Hb levels or function could impact
   oxygen delivery to the ovarian microenvironment, potentially leading to
   changes in ovarian autophagy, oxidative stress, and mediating the
   signaling that drives luteinization.^[ [119]^27 , [120]^28 , [121]^29 ,
   [122]^30 ^]

   We found increased Hb levels in women with PCOS compared to controls,
   and the single nucleotide polymorphisms (SNPs) and candidate genes
   closely related to Hb levels were mainly enriched in the HIF‐1
   signaling pathway. A previous study showed that higher Hb levels were
   related to metabolic disorders and the inhibition of the HIF‐1
   pathway.^[ [123]^7 ^] However, there are still inconsistent conclusions
   regarding the changes in the HIF‐1 pathway in PCOS patients. Studies
   have shown that the expression of HIF‐1α messenger RNA (mRNA and
   protein in the granulosa cells of PCOS patients and in the ovaries of
   PCOS‐like rats was decreased.^[ [124]^31 , [125]^32 , [126]^33 ^]
   However, the HIF‐1α protein level in granulosa cells was also not
   significantly different between control and PCOS patients.^[ [127]^31
   ^] Moreover, HIF prolyl hydroxylase activity was increased, which might
   lead to the inhibition of the HIF‐1α signaling pathway.^[ [128]^34 ^]
   There was also inconsistent result showing that the mRNA and protein
   levels of HIF‐1α were upregulated and induced by reactive oxygen
   species.^[ [129]^35 ^]

   Several studies have indicated a relationship between HIF1 and
   androgens, especially testosterone. Under hypoxia conditions, activated
   HIF1 was found to bind to the Star promoter region and repress the
   expression of StAR, thus impairing the synthesis of testosterone.^[
   [130]^36 ^] Research has also suggested that the promoter of Hsd3b1, a
   gene involved in testosterone production, may be a potential target of
   HIF1α.^[ [131]^37 ^] Additionally, adverse environmental conditions,
   such as air pollution and microplastic contamination, may disrupt HIF1α
   signaling pathways, thereby impacting hormone synthesis via sirtuin 1
   (SIRT1), extracellular regulated protein kinases 1/2 (ERK1/2),
   mitogen‐activated protein kinase (MAPK), or protein kinase B (AKT)
   pathways.^[ [132]^38 , [133]^39 ^] However, current research has
   predominantly focused on males and further studies are needed to
   explore this relationship in females to elucidate the role of HIF in
   hormone synthesis and to explore how dysregulation of the HIF1
   signaling pathway contributes to the pathogenesis of diseases. HIF1α
   also affects women's reproductive function by regulating the
   development and excretion of follicles.^[ [134]^40 ^] The luteinization
   of GCs during early corpus luteum formation was found to be regulated
   by HIF1α‐BNIP3 mediated through autophagy.^[ [135]^41 ^] HIF‐1α was
   found to bind to histone 3/4‐acetylated androgen receptor (AR) promoter
   and transcriptionally activate AR via forming a transcriptional
   activation complex with BRD4, then promote ovarian fibrosis in
   PCOS‐like rat.^[ [136]^42 ^] Furthermore, elevated palmitic acid levels
   in obese women could cause the collagen‐crosslinking enzyme lysyl
   oxidase dysregulation via activation of HIF‐1α, resulting in abnormal
   collagen deposition in the ovary and anovulation.^[ [137]^43 ^] In
   addition, activation of the HIF pathway is known to alleviate metabolic
   defects including impaired glucose tolerance, elevated fasting glucose
   levels and insulin resistance in PCOS mouse model.^[ [138]^44 ^] To
   summarize, HIF1 pathway may contribute to pathogenesis of PCOS by
   modulating hormone synthesis, ovulation, and metabolic processes, and
   the underlying mechanisms warrant further in‐depth exploration.

   Our MR analysis and pathway enrichment analysis identified genes
   related to the HIF‐1 pathway, with the expression levels of INSR and
   NFKB1 significantly positively correlated with testosterone levels.
   Previous GWAS have demonstrated associations between INSR variants and
   PCOS in Chinese and European individuals populations, suggesting a
   potential impact on metabolic disruption, such as insulin resistance
   and obesity, in PCOS patients.^[ [139]^45 , [140]^46 ^] In the
   granulosa cells of PCOS patients, the INSR mRNA expression was
   significantly lower than that in the control group.^[ [141]^47 ^] There
   were also some findings that INSR mRNA showed no significant difference
   in the granulosa cells of PCOS patients or in the ovaries of PCOS‐like
   mice.^[ [142]^48 , [143]^49 ^] These findings were not in line with our
   aforementioned higher level of INSR. A previous study showed that
   insulin sensitivity is decreased in PCOS granulosa cells, impairing the
   glucose uptake process.^[ [144]^49 ^] The elevated expression of INSR
   in the present study might be an adaptive compensatory effect. HIF‐1α
   can be induced by the activation of receptor tyrosine kinases. To our
   knowledge, the relationship between INSR, a member of the receptor
   tyrosine kinases (RTK) family, and HIF‐1α in PCOS has not been
   reported, which requires further exploration.

   Inflammation and hyperandrogenism have complex causes and effects on
   each other, and hyperandrogenism can aggravate inflammation, in which
   NF‐κB is a main regulator and orchestrates different steps.^[ [145]^50
   ^] We detected enhanced expression of NFKB1 in the granulosa cells of
   PCOS patients, which was consistent with the chronic inflammatory
   status in PCOS. Previous studies have also indicated the upregulated
   expression or the activation of NF‐κB in women with PCOS.^[ [146]^51 ^]
   A crosstalk between NF‐κB and HIF has been associated with several
   pathological conditions. HIF‐1α can regulate NF‐κB through an
   evolutionarily conserved negative feedback mechanism, and NF‐κB has
   been shown to act as a direct modulator of HIF‐1α expression in
   inflammation and hypoxia.^[ [147]^52 ^] Colocalization analysis
   indicated that NFKB1 shares same causal genetic variants with TT
   levels, suggesting that targeting this way has the potential to
   modulate the hyperandrogenemia state of PCOS. Of the existing drug
   development targeting NFKB1, two major compounds, HE3286 and Triflusal,
   are in clinical trials, but there are no studies related to
   testosterone levels. In the hypoxia‐reoxygenation rat models, Triflusal
   showed neuroprotection by regulating oxidative stress, suggesting that
   the drug may be involved in the metabolic response to hypoxia, but
   whether it is through the HIF‐1 pathway needs to be further explored.^[
   [148]^53 ^]

   This study has several strengths. We systematically investigated, for
   the first time, the nature of the associations between Hb levels and
   PCOS, as well as PCOS‐related traits. Hb levels were positively
   associated with the odds of PCOS and testosterone levels, and the
   genetic causal associations were rigorously confirmed step by step. Our
   findings unveiled the involvement of the HIF‐1 pathway in this process,
   shedding light on key genes such as INSR and NFKB1. The differential
   expression of these genes and their association with Hb levels in PCOS
   patients were further confirmed using in vitro experiments. In
   addition, while the testosterone level has been identified as a type of
   hematopoietic hormone that might confound these associations, we
   conducted multivariable MR analysis to adjust for the potential effects
   of androgens on the association with PCOS.^[ [149]^54 ^] Compared to
   traditional observational studies, MR offers a unique advantage by
   elucidating the causal effects of specific exposures on outcome traits,
   minimizing confounding and reverse causality.

4. Limitation

   The limitations of our study include the observational nature of human
   data and that we did not quantify serum iron or ferritin levels, which
   are both closely related to Hb levels. To address this, we included
   only individuals with normal Hb levels. Although we have also excluded
   conditions known to affect hemoglobin levels (see the Experimental
   Section) and adjusted for BMI, blood pressure, fasting blood glucose,
   and lipid profiles, the residual influence of unmeasured lifestyle
   factors (e.g., diet, physical activity, smoking, and alcohol
   consumption) cannot be excluded due to the nature of our retrospective
   data. Future studies incorporating detailed lifestyle data are
   warranted to more comprehensively evaluate these potential influences.
   Another limitation is that the study population was included from a
   reproductive center, which may limit the generalizability of our
   findings to broader populations, especially those without infertility
   concerns. Therefore, epidemiological research involving a larger, more
   representative sample of the general population is still needed to
   confirm our findings. In addition, the MR analyses were restricted to
   individuals of European ancestry. Although the associations of Hb
   levels with PCOS and related traits were identified in the Han Chinese
   population, whether the results obtained can be extended to other
   ethnicities should be determined with caution.

5. Conclusion

   In conclusion, our comprehensive study unveils a potential link between
   elevated Hb levels and an increased risk of PCOS and its individual
   traits, particularly hyperandrogenemia. Moreover, we identify the HIF‐1
   pathway as a key regulatory mechanism in this association. These
   findings indicate that low‐end Hb levels may be favorable for PCOS
   patients and may be potentially beneficial for the prevention of PCOS.
   Additionally, our results imply that hypoxia, or reduced oxygen levels,
   may hold potential as a preventive measure for PCOS. Further
   exploration of these pathways and their interactions may offer novel
   insights into the pathogenesis of PCOS and potential therapeutic
   strategies.

6. Experimental Section

Source of Data

   For the observational study, 24 757 women underwent reproductive
   examinations and laboratory testing at the First Affiliated Hospital of
   Wenzhou Medical University and finished their first in vitro
   fertilization (IVF) treatments due to infertility between January 2010
   and December 2021. This center is located in the coastal plain region
   of East Asia, where over 95% of the individuals are local residents
   (the specific residential locations were not disclosed due to patient
   privacy concerns). While a small number of individuals originate from
   areas outside the plain region, all have resided and worked in this
   area for an extended period. Among the 22 042 women with normal Hb
   concentrations (ranging from 110 to 150 g L^−1), 1440 were excluded for
   chromosomal abnormalities, thyroid dysfunction, hyperprolactinemia (or
   pituitary microadenoma), current smoking status, kidney (or liver)
   dysfunction, history of hematological disorders or radiotherapy or
   chemotherapy (Figure [150]S1, Supporting Information). All data were
   collected from the health information system customized for
   reproductive centers to store clinical, laboratory, demographic, and
   reproductive‐specific information. This study was approved by the
   institutional review committee of the hospital (Approval No.
   KY2022‐R096), waiving the need for informed consent due to deidentified
   data. This observational study was conducted in accordance with the
   Strengthening the Reporting of Observational Studies in Epidemiology
   (STROBE) statement.^[ [151]^55 ^]

   For the MR analyses, meta‐analysis data from the largest publicly
   available GWAS summary statistics were utilized for Hb concentrations
   in 563 085 participants of European descent for the discovery analysis.
   Genetic associations were adjusted for age, age^2, sex, and top
   principal components.^[ [152]^56 ^] To examine the robustness of the
   association, summary statistics were used from another large‐scale Hb
   GWAS including 173 480 individuals of European ancestry in the
   replication analysis, which made adjustments including age, sex, and
   common factors known to be influential on Hb levels such as BMI,
   alcohol consumption, and smoking status.^[ [153]^57 ^] Regarding PCOS
   and related traits, corresponding summary‐level data were extracted
   from the hitherto largest GWASs.^[ [154]^58 , [155]^59 , [156]^60 ^]
   GWAS statistics for PCOS were derived from a meta‐analysis by Day F
   et al., which included a European population of 10 074 patients and
   103 164 controls, adopting the NIH or Rotterdam diagnostic criteria for
   case definition.^[ [157]^58 ^] The dataset of tissue‐specific eQTL data
   of target genes were obtained from the GTEx web portal, which included
   948 donors and 17 382 samples from 54 tissues.^[ [158]^61 ^] Cis‐pQTLs
   for the corresponding plasma proteins were extracted from the deCODE
   study where aptamers were measured using the SomaScan platform among
   35 559 Icelanders.^[ [159]^62 ^] Details for all phenotypes included in
   MR analysis were taken from the published studies described in Table
   [160]S14 in the Supporting Information.

   For in vitro validation experiments, Chinese women with PCOS and
   healthy women as controls were recruited from the Reproductive Medicine
   Center of Peking University Third Hospital from December 2021 to May
   2023. The diagnosis of PCOS was based on the Rotterdam criteria.^[
   [161]^63 ^] Patients with other etiologies for hyperandrogenemia or
   ovulatory dysfunction (Cushing syndrome, 21‐hydroxylase deficiency,
   thyroid disease, androgen‐secreting tumors, congenital adrenal
   hyperplasia, and hyperprolactinemia) were excluded. The control
   subjects had regular menstrual cycles, normal ovarian morphology, and
   normal levels of hormones. None of the women used medications known to
   affect metabolic function or reproductive function within 3 months
   before enrollment. This part of the study was approved by the Ethics
   Committee of Peking University Third Hospital (No. 2021‐533‐01) and
   informed consent was obtained from all participants.

Outcomes of Interest

   The primary outcome was clinically defined PCOS according to the
   Rotterdam criteria.^[ [162]^63 ^] The secondary outcomes were PCOS
   phenotypes according to the NIH consensus panel,^[ [163]^64 ^] three
   indicators of PCOS, and corresponding continuous forms:
   hyperandrogenism (testosterone concentrations), PCOM (AFC), and
   ovulatory dysfunction. Hyperandrogenism was defined as a serum
   testosterone level > 2 nmol L^−1 and ovulatory dysfunction was defined
   as infrequent menstrual bleeding (>38 d) or amenorrhea.^[ [164]^65 ,
   [165]^66 ^] The testosterone levels and menstrual patterns were
   obtained under the condition that no hormonal treatments were used. AFC
   was defined as the total number of small follicles (with a diameter of
   2–10 mm) in both ovaries counted by a senior or above attending
   reproductive physician on days 2–3 of the menstrual cycle.

Measurements Involved in This Study

   For the observational study, age, BMI, education, systolic (and
   diastolic) blood pressure, infertility type, endometriosis, history of
   ovarian surgery, basal sexual hormones (follicle‐stimulating hormone
   [FSH], luteinizing hormone [LH], LH/FSH ratio, and estradiol), fasting
   blood glucose, total cholesterol, triglyceride, high‐ and low‐density
   lipoprotein as potential covariates were included. Patient
   anthropometric data were collected by nurses with at least 3 years of
   nursing experience. After overnight fasting for at least 8 h,
   peripheral blood samples were collected on days 2–4 of a natural
   menstrual cycle or when amenorrhea for over 40 d with follicular
   diameter not exceeding 10 mm. Basal AFC was taken to be the total
   number of small follicles (diameter of 2–10 mm) in both ovaries.
   Infertility type was classified as primary and secondary determined by
   pregnancy history. Endometriosis was identified according to
   International Classification of Diseases (tenth revision) code N80. Sex
   hormones were analyzed using an ultrasensitive enzyme‐linked
   immunosorbent assay (Unicel Dxl 800, Beckman Coulter, Brea, CA). Serum
   fasting blood glucose, triglyceride, high‐ and low‐density lipoprotein
   were measured using an autoanalyzer (AU 5800, Beckman Coulter, Brea,
   CA). Serum Hb levels were measured no more than one year prior to the
   initiation of IVF treatment, using an automatic blood analyzer (Sysmex
   Corporation). The detection range of Hb was 0–250 g L^−1. For the
   testing method of biochemical parameters, both intra‐assay and
   interassay variations were less than 10%.

   For the in vitro experiments, as patients’ biosamples were collected
   from an independent hospital, measurements of FSH, LH, estradiol, total
   testosterone levels, and androstenedione were performed using a Siemens
   Immulite 2000 immunoassay system (Siemens Healthcare Diagnostics, USA).
   Ultrasensitive two‐site enzymelinked immunosorbent assay (Ansh Labs,
   USA) was used to measure anti‐Müllerian hormone (AMH). The sensitivity
   of the measurement of FSH, LH, estradiol, total testosterone levels,
   androstenedione, AMH were 0.1 U L^−1, 0.05 U L^−1, 15 ng L^−1,
   0.5 nmol L^−1, 0.3 µg L^−1, 23 pg mL^−1, and intra‐assay variation were
   5.57%, 6.00%, 7.63%, 10.23%, 10.53%, and <5%, respectively. The serum
   levels of fasting serum glucose were determined by chemiluminescence
   method using Immulite 1000 system (DPC, USA). Total cholesterol,
   triglyceride, low‐density lipoprotein cholesterol (LDL‐C), and
   high‐density lipoprotein cholesterol (HDL‐C) were measured by a dry
   slide enzymatic colorimetric assay (Johnson, USA). The detection ranges
   of LDL‐C, HDL‐C, and Hb levels were 0.03–11.80 mmol L^−1,
   0.05–3.90 mmol L^−1, and 0–250 g L^−1, respectively, and intra‐assay
   variations were <5%, <5%, and ≤1.5%, respectively. Blood cells were
   collected by centrifugation at 300 g for 10 min for the quantitative
   analyses of target genes.

   All laboratory data were uniformly measured and reviewed by the
   laboratory department of these two tertiary hospitals. All original
   data was recorded in the hospital information systems, had never been
   modified by nonmedical personnel, and only existed within the local
   area network of both hospitals.

Mendelian Randomization Analyses

   In MR, genetic variants, randomly allocated at conception, are used as
   IVs to proxy the effect of the exposure (i.e., Hb levels in this
   study). To construct IVs, SNPs associated with Hb levels were selected
   first at the genome‐wide significance level (P < 5 × 10^−8). Then
   linkage disequilibrium (i.e., genetic correlation matrix) of these SNPs
   based on 1000 Genomes European reference panel was estimated and
   independent variants (linkage disequilibrium r ^2 < 0.001 in a clumping
   window of 10 000 kb) as the IVs for Hb were selected. Data were
   extracted on the IVs from the outcome GWAS data and they were
   harmonized based on effect and noneffect alleles for subsequent MR
   analysis. To assess the bias from weak instruments, F statistics were
   calculated to judge the strength of IVs, with a threshold above ten as
   the conventional rule.

   To investigate the causal association of Hb levels with PCOS
   manifestations, MR approach was used to test the association of
   genetically predicted Hb levels with PCOS and its related traits,
   including irregular menstrual cycle, polycystic ovaries, BT, and TT.
   Considering the possible interplay between androgens and Hb levels, an
   MVMR analysis was employed to account for BT or TT levels in subsequent
   validation stages. This adjustment was crucial to ascertain whether the
   observed MR effects of Hb concentrations on PCOS remained independent
   of testosterone levels.^[ [166]^67 ^] In MVMR analysis, all
   GWAS‐associated SNPs for both traits in the analytical model were
   included to assess whether the effects of Hb concentrations on PCOS
   were independent of BT or TT. The MR part of the study followed the
   Strengthening the Reporting of Observational Studies in Epidemiology
   Using Mendelian Randomization (STROBE‐MR) guideline (Table [167]S15,
   Supporting Information).

Pathway Enrichment Analyses and Pathway‐Specific Mendelian Randomization
Analysis

   Pathway enrichment analysis was conducted to provide insights into
   biological pathways. The gene annotation corresponding to IV was
   implemented in the Ensembl genome browser and then the gene symbol was
   mapped to the Entrez gene identifier.^[ [168]^68 ^] The KEGG database
   was used for pathway definition.^[ [169]^69 ^]

   As follow‐up analyses, MR analysis was conducted using eQTLs of pathway
   candidate genes related to PCOS and related traits. For all genes of
   the HIF‐1 pathway, eQTL data were extracted from the GTEx v8 database
   in whole blood, ovary, hypothalamus, and pituitary tissue, as Hb is
   present in whole blood tissue, and ovary, hypothalamus and pituitary
   tissue are the most relevant to the development of PCOS. The
   corresponding cis‐eQTLs were extracted according to the following
   criteria: variants within ± 1 Mb of each gene transcription start site,
   with P < 5 × 10^−4 and r ^2 < 0.1.^[ [170]^61 ^]

   cis‐pQTLs associated with plasma protein levels (P < 5 × 10^−8) were
   also used as IVs to perform a validation analysis of eQTL MR results.
   Cis‐pQTLs were extracted from the deCODE Health study, where plasma
   aptamers were measured by the SomaScan platform among 35 559
   Icelanders.^[ [171]^62 ^] Cis‐SNPs were defined as SNPs within 1 Mb
   from the gene encoding the protein, and SNPs with high linkage
   disequilibrium (r ^2 > 0.01) were pruned. Since the number of SNPs
   available for colocalization in the eQTL study was limited, association
   analyses on pQTLs and testosterone levels were mainly performed.
   Lastly, to evaluate the druggability of proteins identified, the
   DrugBank, Dependency Map, and ChEMBL databases were searched to
   summarize the current stages of drug development targeting these
   proteins.

Granulosa Cell Isolation

   The individuals received a standard in vitro fertilization antagonist
   stimulation protocol and underwent transvaginal ultrasound‐guided
   follicle aspiration 36 h after human chorionic gonadotropin (hCG)
   administration. Mural granulosa cells were isolated from the follicular
   fluids aspirated during oocyte retrieval by Ficoll density gradient
   centrifugation as previously described.^[ [172]^70 ^]

RNA Extraction and Real‐Time Quantitative Polymerase Chain Reaction (qPCR)
Analysis

   Total RNA was extracted from peripheral blood cells and granulosa cells
   with TRIzol (Thermo Fisher Scientific, USA). RNA purity was determined
   by the absorption ratio (260/280 nm), which was 1.8–2.0 for all
   samples. One microgram of purified RNA was used to generate cDNA with a
   ReverTra Ace qPCR RT Kit (Takara, Dalian, China). Real‐time qPCR
   analysis was performed using SYBR Green PCR master mix (Invitrogen) and
   the QuantStudio 3 real‐time PCR system (Thermo Fisher Scientific). The
   qPCR primers used in this study were as follows:
   5′‐TGGCGATGTTGGGAATGTG‐3′ and 5′‐GATGCGATAGCCCGTGAAGT‐3′ for INSR;
   5′‐GCGGTGACAGGAGACGTGAA‐3′ and 5′‐CAAGTTGAGAATGAAGGTGGATG‐3′ for NFKB1;
   and 5′‐GCCTATGGCGTCCTGTTGTAT‐3′ and 5′‐GCTTGGCTGGGTGTTTGGT‐3′ for
   PRKCA.

Western Blotting

   Protein was extracted from granulosa cells using radio
   immunoprecipitation assay (RIPA) buffer supplemented with protease
   inhibitors and phosphatase inhibitors and quantified by a bicinchoninic
   acid (BCA) assay (BCA Protein Assay Kit; Pierce Thermo Scientific).
   Total protein extracts from granulosa cells were separated via 12%
   sodium dodecyl sulfate ‐ polyacrylamide gel electrophoresis (SDS‒PAGE).
   Western blotting was performed as described previously.^[ [173]^61 ^]
   The following primary antibodies were used according to the provided
   recommendations: anti‐insulin receptor beta (1:1000, ab69508, Abcam),
   anti‐NF‐κB (1:1000, 8242, Cell Signaling Technology), and anti‐GAPDH
   (glyceraldehyde‐3‐phosphate dehydrogenase) (1:1000, 5174, Cell
   Signaling Technology).

Statistical Analysis—The Observational Study

   Categorical variables were reported as number (percentage) and compared
   among Hb tertile groups using the Chi‐squared test.^[ [174]^71 ^]
   Numerical variables were reported as median (IQR) and compared using
   the Kruskall‐Wallis test.^[ [175]^72 ^] No imputation was performed for
   missing data. The associations of Hb concentrations with PCOS, PCOS
   phenotypes, hyperandrogenism, ovulatory dysfunction, and PCOM were
   tested first using logistic regression models. These models were fitted
   respectively adjusting for increasing numbers of covariates and the aOR
   was calculated: Model 1 adjusted for intervals of appointment date,
   age, BMI, and education; Model 2 additionally adjusted for systolic
   blood pressure, endometriosis, history of ovary surgery, LH/FSH ratio,
   fasting blood glucose, total cholesterol, triglyceride, high‐ and
   low‐density lipoprotein. Univariate and multivariate GAM were further
   fitted to account for the nonlinear relationships between Hb levels
   (g L^−1) and PCOS and other indicators and phenotypes of PCOS.^[
   [176]^73 ^] Nonlinear associations were visualized using the GAM
   partial effect plots, which demonstrated that how the smooth component
   of Hb concentrations in the model changed assuming that other variables
   were at their average value. A two‐tailed P‐value of less than 0.05 was
   considered statistically significant.

Statistical Analysis—Mendelian Randomization and Bioinformatics Analyses

   The multiplicative random effects IVW approach was employed as the main
   MR analysis to estimate the associations of genetically predicted Hb
   concentrations with PCOS and related traits. The IVW method provides
   the most accurate estimate under the assumption that all of the
   instruments are valid.^[ [177]^74 ^] Several sensitivity analyses,
   including MR‐Egger regression, weighted‐median method, and MR
   Pleiotropy RESidual Sum and Outlier test (MR‐PRESSO), were performed to
   explore the robustness of the results.^[ [178]^75 , [179]^76 , [180]^77
   ^] MR‒Egger does not force the regression line through the intercept
   and is therefore able to detect the presence of directional
   pleiotropy.^[ [181]^75 ^] The weighted median provides robust causal
   effect estimates with the assumption that at least 50% of the SNPs used
   are valid.^[ [182]^78 ^] MR‐PRESSO was utilized to test the presence of
   outliers and correct pleiotropy via outlier removal.^[ [183]^77 ^] For
   exposure and outcome datasets with partial sample overlap, the type 1
   error rate was additionally assessed to validate the causal
   associations.^[ [184]^79 ^] To confirm the validity of the directional
   assumption, the MR‐Steiger test was also performed. For binary
   outcomes, MR estimates were presented in terms of OR with 95% CIs, and
   for continuous outcomes, β coefficients with SE were used. the F
   statistic was computed to measure the strengths of IVs used in MR by
   the formula F = (N − K − 1/K)(R ^2/1 − R ^2), where N is the sample
   size and K is the number of IVs. R ^2 was calculated as 2EAF(1 − EAF)β
   ^2, where EAF is the effect allele frequency and β is the effect size
   of the effect allele of the IVs.

   In the pathway enrichment analysis, the Benjamini‒Hochberg FDR
   correction was applied for multiple testing. In eQTL MR analysis,
   estimates were assessed using the Wald ratio method for instruments
   with one variant and the IVW method for instruments with two or more
   variants and the Bonferroni correction was used for the significance
   threshold.

   A Bayesian statistical method was used to test colocalization of a
   shared causal genetic variant between protein levels and testosterone
   levels.^[ [185]^80 ^] The prior probabilities for genetic signal were
   set with default values, P1 = P2 = 1 × 10^−4, P12 = 1 × 10^−5, where P1
   and P2 is the probability that a given genetic signal is associated
   with either of the two traits, and P12 is the probability associated
   with both traits. It computes the posterior probability of association
   for alternative hypotheses: H0, no association with either trait; H1,
   signal unique to trait 1; H2, signal unique to trait 2; H3, two
   distinct causal variants in the same locus, one for each trait; and H4,
   presence of a shared causal variant between two traits. It is strong
   evidence of colocalization if the PH4 value of 80% or higher.

Statistical Analysis—In Vitro Validation Experiments

   For the patients’ descriptive statistics, the Shapiro‐Wilk normality
   test was applied. Continuous variables that followed a normal
   distribution are represented as mean ± SD, while those did not follow a
   normal distribution are reported as median (interquartile range). If
   both groups displayed normal distributions and had equal variance,
   unpaired t‐test was used, otherwise nonparametric Mann‐Whitney U‐test
   was used.

Statistical Analysis—Software Availability

   Data manipulation, analyses, and visualization were performed using R
   version 4.2.1. In the observational study and in vitro experiments, the
   R packages “tidyverse” and “data.table” were used for data
   manipulation; “nnet” was used for multinomial logistic regression
   analysis; “mgcv” was used for GAM modelling; and “ggplot2” was used for
   data visualization. In the MR study, the R packages “TwoSampleMR” and
   “MR‐PRESSO” were used for two sample MR and MVMR analysis; type 1 error
   rate was obtained from an online calculator
   ([186]https://sb452.shinyapps.io/overlap/); “coloc” was used for
   colocalization analysis; “clusterProfiler” was used for pathway
   enrichment; and “ggplot2” and “forestplot” were used for data
   visualization.

Conflict of Interest

   The authors declare no conflict of interest.

Ethics Approval Statement

   The ethics approval was obtained from the institutional review
   committee of the First Affiliated Hospital of Wenzhou Medical
   University (No. KY2022‐R096) and the Ethics Committee of Peking
   University Third Hospital (No. 2021‐533‐01). The ethical approvals and
   consent to participate involved in the MR section were available in the
   original GWAS study. The analyses were conducted in accordance with the
   approval and consent principles of the Helsinki Declaration of Ethics.

Supporting information

   Supporting Information
   [187]ADVS-12-2411679-s001.docx^ (230.3KB, docx)

Acknowledgements