Investigating the gut microbiota's influence on psoriasis and psoriatic arthritis risk: a Mendelian randomization analysis

Abstract Background Numerous investigations have revealed the interplay between gut microbiota (GM) and psoriasis (Ps) and psoriatic arthritis (PsA). However, the causal relationship between them remains unknown. Methods We curated a collection of genetic variants (P < 1 × 10−5) associated with GM (n = 18 340) derived from the MiBioGen study. To explore the intricate relationship between GM and Ps as well as PsA, we harnessed the comprehensive resources of the FinnGen database, encompassing a vast cohort of individuals, including 4510 Ps cases and 212 242 controls and 1637 PsA cases and 212 242 controls. Mendelian randomization (MR) was used, including an inverse variance weighting method, followed by a sensitivity analysis to verify the robustness of the results. Results For Ps, some bacterial taxa, including Lactococcus, Ruminiclostridium 5, and Eubacterium fissicatena, were identified as risk factors; but Odoribacter demonstrated a protective effect against Ps. In the case of PsA, Lactococcus, Verrucomicrobiales, Akkermansia, Coprococcus 1, and Verrucomicrobiaceae were identified as risk factors; Odoribacter and Rikenellaceae exhibited a protective effect against the development of PsA. Conclusion Our study establishes a causal link between the GM and Ps and PsA. These findings provide insights into the underlying mechanisms and suggest potential therapeutic targets.


Introduction
Psoriasis ( Ps ) , a c hr onic inflammatory skin disorder, is c har acterized by aberrant proliferation of keratinocytes and infiltration of immune cells into the epidermis. 1 , 2The condition affects a significant portion of the population, with v arying pr e v alence among differ ent ethnic gr oups.Psoriatic arthritis ( PsA ) , a se v er e comorbidity of Ps, manifests as joint pain, swelling, and rigidity. 2Both Ps and PsA exhibit a strong genetic predisposition, with heritability estimates ranging from 60% to 100%. 1 , 3 Furthermore, the prevalence of these conditions is still on the rise. 3s and PsA have garnered significant research attention to investigate their intricate relationship and potential implications in disease pathogenesis, pr ogr ession, and mana gement.4 Observ ations hav e r e v ealed distinct alter ations in the gut micr obiome ( GM ) of individuals with Ps and PsA, presenting a unique dysbiosis pattern.5 , 6 Se v er al hypotheses hav e been pr oposed to elucidate the role of the GM in the pathogenesis of Ps and PsA, encompassing factors such as intestinal permeability, perturbed immune homeostasis, and imbalances in specific bacteria producing short-and medium-chain fatty acids.5 , 6 Notably, interventions aimed at restoring the microbiome have exhibited promise as pr e v entiv e and ther a peutic str ategies for Ps and PsA.7 For in-stance, in murine models displaying Ps-like skin inflammation, or al administr ation of a br oad-spectrum antibiotic effectiv el y mitigated the se v erity of skin inflammation by downregulating the Th17 immune response.][6][7] Mendelian randomization ( MR ) is a statistical method used in genetic epidemiology r esearc h to investigate causal relationships between exposures and outcomes by utilizing genetic variants as instrumental variables.9 MR analyses provide valuable insights into the potential causal relationships between the GM and various health conditions.By utilizing genetic variants as instrumental v ariables, r esearc hers can explor e the r ole of the GM in disease de v elopment and identify specific microbial taxa that may be causally linked to certain conditions .T hese findings contribute to a better understanding of the complex interactions between the GM and human health.
In this study, we conducted a two-sample MR investigation using large-scale genome-wide association study ( GWAS ) data of GM, Ps , and PsA.T he objecti ve of our stud y was to uncover the potential causal effects of 196 GM taxa on Ps and PsA.

Data sources and filter instrumental variables
The study design fr ame work is depicted in the gr a phical abstr act, illustr ating the structur e of our inv estigation.The MiBioGen gr oup conducted an extensive genome-wide meta-analysis of GM composition, incor por ating genetic variation data pertaining to the gut microbiota. 10This remarkable research endeavor encompassed a cohort of 18 340 individuals hailing from diverse regions, such as the USA, the UK, Finland, Sweden, Denmark, T he Netherlands , and other countries .T he compr ehensiv e dataset emplo y ed in this study encompassed 16S rRNA gene sequencing profiles and genotyping information.Through meticulous analysis, we identified and classified bacteria at various taxonomic levels, including 9 phyla, 16 classes, 20 orders , 35 families , and 131 genera.Subsequently, 3 unidentified families and 12 unknown genera were excluded from the dataset, resulting in the inclusion of 9 phyla, 16 classes , 20 orders , 32 families , and 119 genera for further analysis in the subsequent MR investigation ( supplementary Table 1, see online supplementary material ) .
To ensure the selection of high-quality genetic data, we chose the GWAS dataset with the most compr ehensiv e cov er a ge of single nucleotide pol ymor phisms ( SNPs ) published in 2021, sourced from the FinnGen project. 11This specific GWAS dataset focused on the phenotypes "psoriasis" and "psoriatic arthritis" and incor por ated Finnish adult subjects, consisting of 4510 cases and 212 242 controls for Ps and 1637 cases and 212 242 controls for PsA.
To investigate potential causal links and associations between GM and Ps and PsA, it is essential to select valid instrumental variables ( IVs ) that satisfy three k e y assumptions: ( 1 ) the correlation hypothesis, ( 2 ) the exclusivity hypothesis, and ( 3 ) the independence assumption. 9Due to the limited number of SNPs available for MR analysis, we set a lenient threshold of P < 1 × 10 −5 .To ensure the independence of each SNP, we applied a linkage disequilibrium ( LD ) factor ( r 2 ) of 0.01 and a clumping window width ( kb ) of 10 000. 12 Subsequently, we extracted information on SNPs associated with the aforementioned intestinal flor a fr om the summary GWAS data on Ps and PsA.We eliminated missing SNPs and set the minor allele frequency ( MAF ) at 0.01. 13Additionally, we excluded all SNPs with palindr omic structur es to mitigate the influence of alleles on the results.To examine the presence of bias in the causal relationship between intestinal flora and Ps and PsA due to w eak IVs, w e emplo y ed the F value.When the F -statistic w as < 10, w e considered the used SNP a w eak IV and excluded it fr om the anal ysis. 14To e v aluate the potential influence of confounding factors, we utilized the PhenoScanner V2 online tool 15 and r eferr ed to the Eur opean Dermatological Association Guidelines on Ps and PsA. 16SNPs that sho w ed associations with known confounders were subsequently excluded from the analysis.

MR analysis
In this study, we utilized four different approaches, namely, MR-Egger, weighted median, random-effect inverse variance weighted ( IVW ) , and weighted mode, to perform the MR analysis and calculate causal estimates between GM composition and the risk of Ps and PsA.Each approach has its specific requirements and assumptions. 17In the MR analysis conducted in this study, we adopted the IVW method as the cornerstone of the analytical fr ame work for probing the association among bacterial taxa, Ps, and PsA.The IVW method, a widel y embr aced tool, serv es as a r obust means to derive causal estimates for discr ete v ariables. 18By employing these a ppr oac hes, the study aimed to assess the potential causal effects of GM composition on Ps and PsA and provide insights into their relationship.

Sensitivity analysis
The sensitivity analysis encompassed a heterogeneity test and a m ultiplicity of v alidity test.To confirm IV heter ogeneity, Coc hr an's Q-test was employed, and a P value < 0.05 was considered indicative of the absence of heterogeneity. 19To ensure that the selected IV does not affect the outcome variable of Ps and PsA risk through biological pathways other than the gut microbiota, we conducted an e v aluation of the pleiotropic associations between the IV and other potential confounding factors utilizing the MR-Egger intercept test. 17In this r egr ession methodology, the slope assumes the role of an estimate for the causal effect, while the intercept signifies the mean le v el of the genetic variant, serving as a metric for the assessment of pleiotropic effects. 20MR-PRESSO also aggregated the residuals for each SNP to evaluate the magnitude of horizontal pleiotropy.The MR-PRESSO outlier test facilitated the identification of outlier SNPs that contributed to pleiotropy at the ov er all le v el. 21T he impact of individual outliers on the o v er all r esults was assessed using a leave-one-out analysis, calculating the remaining SNP effects after iteratively removing each SNP.The MR-Egger intercept test, MR-PRESSO, and leave-one-out analysis methods w ere emplo y ed to identify and eliminate SNPs exhibiting pleiotropy or heterogeneity. 21Furthermore, the causal direction was investigated using the MR Steiger test. 22Additionally, a r e v erse MR anal ysis was performed.The MR anal yses wer e conducted using the R ( version 4.3.0 ) computational environment, utilizing the "TwoSampleMR" and "MR-PRESSO" pac ka ges .T he R pac ka ge 'for estploter' w as emplo y ed to gener ate some figur es.Statistical significance for causal effects was determined using a P value threshold of < 0.05.

Results
By a ppl ying genome-wide significance thr eshold scr eening ( P < 1 × 10 −5 ) , LD tests, harmonization, and verification of F -statistics, multiple SNPs were identified as IVs for each of the 196 bacterial taxa.The F -statistics of all the retained SNPs demonstrate a correlation strength > 10, indicating a sufficient association between the instrumental variables and their corresponding bacterial taxa ( supplementary Tables S2 and S3, see online supplementary material ) .Consequently, our study had negligible instrument bias.

Effect of gut microbiome on psoriasis and psoriatic arthritis
We screened 2033 SNPs as instrumental variables from 196 gut micr obiota.The r esults of the MR anal ysis for IVs ar e shown in circus plots ( Figs. 1 and 2 ) and detailed in supplementary Tables S4 and S5 ( see online supplementary material ) .Our study demonstr ated the pr esence of 8 taxa of GM that accelerate the onset of Ps and PsA, while 2 taxa of GM have a protective effect, reducing the risk of these conditions.In Ps, the results of IVW indicated suggestive causal effects of genetically predicted increased abundance of Odoribacter ( OR, 0.71; 95% CI, 0.53-0.96;P = 0.024 ) , which exhibited pr otectiv e effects a gainst Ps risk ( Fig. 3 ) .Conv ersel y, Eubacterium fissicatena at the group level [ ( odds ratio ( OR ) , 1.14; 95% F igure 1. Cir cus plot showing analysis results of all gut microbiota on Ps .T he circular representation depicts the estimates obtained through the IVW, weighted media, and MR-Egger methods , mo ving from the outer to the inner circle .T he classification of GM was based on order, phylum, class, family, and genus .T he v arying shades of color in the circle r epr esent the ma gnitude of the P v alues, with the corr esponding label inside the circle .Ps , Psoriasis; MR, Mendelian randomization; IVW, inverse variance-weighted; WM, weighted median.P < 0.05.confidence interval ( CI ) , 0.99-1.30;P = 0.001 ) , Ruminiclostridium 5 ( OR, 1.31; 95% CI, 1.01-1.70;P = 0.043 ) , and Lactococcus ( OR, 1.25; 95% CI, 1.08-1.44;P = 0.003 ) were associated with a higher risk of Ps.For PsA, the IVW anal yses demonstr ated suggestiv e causal effects of genetically predicted increased abundance of Rikenellaceae at the famil y le v el ( OR, 0.70; 95% CI, 0.50-0.97;P = 0.034 ) , whic h exhibited pr otectiv e effects a gainst PsA risk.Conv ersel y, Verrucomicrobiales at the order le v el ( OR, 1.60; 95% CI, 1.14-1.24;P = 0.006 ) , Coprococcus1 ( OR, 1.51; 95% CI, 1.03-2.19;P = 0.03 ) , Akkermansia ( OR, 1.60; 95% CI, 1.14-1.24;P = 0.006 ) , Verrucomicrobiae at the class le v el ( OR, 1.60; 95% CI, 1.14-1.24;P = 0.006 ) , and Verrucomicrobiaceae at the famil y le v el ( OR, 1.60; 95% CI, 1.14-1.24;P = 0.006 ) were associated with a higher risk of PsA ( Fig. 4 ) .It is worth noting that the order Verrucomicrobiales , along with the fam-ilies Verrucomicrobiaceae and Rikenellaceae , belongs to the subclass Verrucomicrobiae .As a r esult, ther e may be significant ov erla p in the SNPs present in these four sets, as detailed in supplementary Table S3.Figures 3 and 4 show the results of the four analytical methods ( MR-Egger, weighted median, IVW, and weighted mode ) , including P values, ORs and 95% CIs.

Sensitivity analysis
The forest plots sho w ed that several IV intercepts in the MR-Egger analysis deviated from zero.Furthermore, several IVs in the MR leave-one-out sensitivity analysis had effect values that span zero within the 95% CI, suggesting potential instability in the findings ( supplementary Figs.S1-S4, see online supplementary material ) .Ho w e v er, further statistical anal-F igure 2. Cir cus plot showing analysis results of all gut microbiota on PsA.The circular representation depicts the estimates obtained through the IVW, weighted media, and MR-Egger methods, moving from the outer to the inner circle.The classification of GM was based on order, phylum, class, family, and genus .T he varying shades of color in the circle r epr esent the ma gnitude of the P v alues, with the corr esponding label inside the circle.PsA, Psoriatic arthritis; MR, Mendelian randomization; IVW, inverse variance-weighted; WM, weighted median.P < 0.05.ysis r e v ealed that all P v alues indicating heter ogeneity among the bacterial taxa mentioned above were > 0.05 ( supplementary Tables S6 and S7, see online supplementary material ) .MR-Egger's test found no evidence of horizontal pleiotropy, indicating that the genetic variants used as instruments for MR did not have pleiotropic effects ( supplementary Tables S4  and S5 ) .Additionally, the reliability of the results was assessed using MR-PRESSO, which identified no outliers; and the MR Presso Global Test results all exceeded the threshold of 0.05 ( supplementary Tables S6 and S7 ) .We recognized that ov erl y stringent filtering criteria may inadv ertentl y exclude some valid positive findings .Hence , we continued to affirm the causal relationships of the aforementioned microbial taxa as valid.

Reverse MR analysis
To mitigate the influence of r e v erse causality on the aforementioned findings, we conducted a r e v erse MR anal ysis with significant gut flora as the outcome and Ps and PsA as the exposure v ariables.Our anal ysis did not pr ovide an y additional e vidence supporting a causal effect of Ps and PsA ( supplementary Figs.S5 and S6, see online supplementary material ) .

Discussion
Ps, a c hr onic autoimm une disease c har acterized by arthritis, often leads to the de v elopment of PsA as a common complication.PsA is estimated to occur in 7% to 42% of individuals with Ps, and its pr e v alence gr aduall y incr eases as the dur ation of Ps persists. 23Pr e vious studies have suggested a potential association between the GM and the de v elopment of Ps and PsA. 5 Ho w e v er, dir ect e vidence establishing a causal correlation is curr entl y lac king.In this study, we conducted a compr ehensiv e inv estigation using MR analysis to explore the relationship between 196 gut bacterial taxa and the occurrence of Ps and PsA.Regar ding Ps, w e found that certain bacterial taxa, such as Lactococcus , Ruminiclostridium 5 , and E. fissicatena , were identified as risk factors .Con versely, Odoribacter demonstrated a protective effect against Ps.In the case of PsA, our r esults r e v ealed a distinct set of risk factors and pr otectiv e factors among gut bacterial taxa.Lactococcus , Verrucomicrobiales , Akkermansia, Coprococcus 1, and Verrucomicrobiaceae were identified as risk factors for PsA.On the other hand, Odoribacter and Rikenellaceae exhibited a pr otectiv e effect against the development of PsA.
Inter estingl y, among these risk factors, only one species of Lactococcus was shared between Ps and PsA, while the remaining bacteria differed.These findings suggest that while PsA is a complication of Ps, its pathogenesis does not completely align with that of Ps.Se v er al of the risky bacteria identified in our study align with pr e vious r esearc h findings .For instance , the abundance of Lactococcus has been shown to increase in the GM of patients with Ps. 24 , 25 Ruminiclostridium 5 , another risky bacterium, exhibited increased abundance in an experimental model of Ps induced by imiquimod. 26Odoribacter , a copr otectiv e bacterium for both Ps and PsA, has been reported to be more abundant in healthy individuals than in patients . 27Furthermore , studies ha v e demonstr ated that the compound r esv er atr ol can increase the abundance of Odoribacter gr oups, ther eby r estoring intestinal ecology in mice. 28nter estingl y, both or al and topical administr ation of r esv er atr ol have shown potential in alleviating imiquimod-induced Ps-like dermatitis, 29 , 30 suggesting that Odoribacter might serve as a probiotic for Ps.
Additionally, among the protective flora associated with PsA, Rikenellaceae has been found to decrease in abundance in the GM of Ps patients. 31This finding underscores the shar ed c haracteristics and differences between Ps and PsA.The results concerning Verrucomicrobiae , Akkermansia , and Coprococcus in our study further support this notion.Pr e vious r eports hav e shown a decreased abundance of these bacterial types in the intestinal flora of Ps patients, [32][33][34][35] indicating their potential as protective factors against Ps.Ho w ever, our findings indicate that they act as risky flora in PsA, further emphasizing the distinct disease c har acteristics and pathogenesis of PsA compared to Ps.
Curr entl y, ther e is a pr e v ailing belief that Ps and PsA share common pathogenic factors, including genetic risk alleles , en vironmental triggers, and cytokine pathwa ys .Howe v er, it is important to note that the resident cells in the skin and joints differ significantly, and the clinical manifestations of musculoskeletal disorders and skin lesions exhibit substantial variation among individuals. 33 , 36 , 37Despite the common involvement of tumor necrosis factor ( TNF ) and the interleukin ( IL ) -23-IL-17 axis in the pathogenesis of both Ps and PsA, monother a py tar geting IL-17 or IL-23 has demonstrated high efficacy in Ps but not in PsA.Although the effectiveness in PsA is less pr onounced, these observ ations further underscore the distinct pathogenic mechanisms underlying skin and joint diseases.One potential mechanism contributing to these differences lies in the microbiome and mucosal immunity.
9][40] Moreover, compared to Ps patients, PsA patients exhibit lo w er ov er all intestinal diversity, 35 suggesting that alterations in intestinal immune dynamics may contribute to synovial enthesis inflammation.Notably, a specific subset of osteoclast precursors, CD14 + CD16 + , has been identified in PsA patients but not in Ps patients. 41Typically, patients with Ps experience skin lesions first, follo w ed b y the onset of PsA.Ho w e v er, it is worth noting that ∼15% of cases present with arthritis and skin lesions occurring sim ultaneousl y or with arthritis preceding the skin lesions. 42A recent study r e v ealed that patients who de v elop PsA as the initial symptom often delay seeking medical attention and initiating tr eatment, whic h can significantl y impact long-term pr ognosis.Pr e vious r esearc h has identified potential predictors of Ps progression to PsA.For instance, C-X-C motif chemokine ligand 10 ( CXCL10 ) has been proposed as a predictive marker for the de v elopment of PsA. 43Another case −control study demonstrated independent associations between PsA and serum le v els of integrin beta 5 ( ITGB5 ) , Mac-2 binding protein ( M2BP ) , and C-r eactiv e pr otein ( CRP ) . 44Furthermor e, e v aluation of the skin proteome and serum samples has r e v ealed the pr esence of ITGB5 and periostin in PsA patients, distinguishing them from those with Ps alone. 45Our study suggests that specific gut flora analysis may aid in the early diagnosis of PsA among patients presenting with joint inflammation.Moreover, targeting shared pathogenic bacteria, such as Lactococcus , or considering probiotic supplementation with Odoribacter could potentiall y serv e as treatment options for individuals with Ps and concomitant PsA.
In this study, we conducted a compr ehensiv e inv estigation to explore the causal relationship between GM and Ps and PsA, utilizing publicly available GWAS summary statistics.We uncovered specific bacterial groups that hold the potential to influence the de v elopment of Ps and PsA.Certain intestinal flor a wer e implicated in the pathogenesis of PsA, suggesting their potential role as early diagnostic indicators .Furthermore , we identified se v er al GMs that exhibit a potential pr otectiv e effect a gainst the occurrence of Ps and PsA.These discoveries lay a solid foundation for future endeavors in the prevention and treatment of these conditions.One of the k e y strengths of our study lies in the rigorous utilization of the MR method, which effectively mitigates the impact of r e v erse causal associations and confounding factors.This methodological a ppr oac h adds consider able r obustness to our findings and enhances the validity of our causal inferences.Notably, our MR study encompassed a remarkably broad population, le v er a ging publicl y av ailable data at a minimal cost.This extensiv e cov er a ge not onl y enhances the gener alizability of our results but also augments the practicality and persuasiveness of our findings when compared to conv entional observ ational studies.Natur all y, it is essential to acknowledge the limitations of our r esearc h.First, the MiBioGen study includes gut microbiota GWAS data fr om v arious countries, including the USA and Eur ope, while FinnGen focuses solely on the Finnish population.Although MR analysis can combine individual-level and summary data, 46 we should inter pr et our r esults cautiousl y due to potential genetic v ariation acr oss these populations.Mor eov er, the causal r elationships identified in our MR study r el y on IVs meeting a rigorous genome-wide significance threshold ( P < 1 × 10 −5 ) .It is crucial to recognize that this stringent criterion could potentially affect the precision of our results.Second, inconsistencies were observed in the analysis of certain bacterial populations, which could potentially be attributed to the utilization of MR-Egger's method for estimating causality.It is plausible that this method introduces bias by altering the Type 1 error rate, leading to inflated rates of Type 1 errors and subsequently influencing the OR. 21Thir d, w e focused exclusiv el y on the Eur opean population, whic h r estricts the generalizability of our findings to other ethnicities or regions.
Ther efor e, caution should be exercised in extr a polating our results to populations beyond the scope of our study.Mor eov er, we emphasize the importance of conducting further observational studies and laboratory-based investigations to validate and expand upon our current findings.By consolidating evidence from m ultiple r esearc h a ppr oac hes, we can advance the knowledge base and provide a more robust understanding of the intricate relationship between the GM and the development of Ps and PsA.

Figure 3 .
Figure 3. Forest plots for the association of GM and genetic susceptibility to Ps, analysed with MR.Ps, Psoriasis; GM, gut microbiota; OR, odds ratio; CI, confidence interval.P < 0.05.

Figure 4 .
Figure 4. Forest plots for the association of GM and genetic susceptibility to PsA, analysed with MR.PsA, Psoriatic arthritis, GM, gut microbiota, OR, odds ratio; CI, confidence interval.P < 0.05.