For the study design, we performed a two-sample MR to investigate the association of LTL with IMIDs, which included SLE, IBD, CD, UC, RA, PSC, SS, AS, AD, IPF, T1D, psoriasis, hyperthyroidism, hypothyroidism, sarcoidosis, and childhood asthma. We utilized most of the summary data of IMIDs, which do not overlap with those of the GWAS of LTL. Then, we conducted a two-sample MR and several sensitivity analyses to identify the causal effect (Figure 1).

Genetic association data for the LTL were retrieved from a GWAS of 472,174 European participants (Codd et al., 2021). All participants in the UK Biobank were aged 40–69 years, with a similar proportion of males (45.8%) and females (54.2%). The measurements of LTL were obtained by using an established quantitative PCR assay and were subjected to several quality checks to control and adjust for ethnicity, gender, age, and technical factors, as described in a previous study.

We used publicly available GWAS summary statistics for different IMIDs acquired from the FinnGen Consortium. The release of R7 GWAS of the FinnGen consortium data was used (https://r7.finngen.fi/). SNPs were analyzed by performing a mixed-model logistic regression model, adjusting for sex, age, 10 principal components (PCs), genetic relatedness, and genotyping batch. Detailed information on it can be found on the official website (https://www.finngen.fi/en). The number of cases and controls is presented in Tables 1, 2. Another SLE summary statistic was acquired from the recent meta-analysis of 5,201 patients and 9,066 controls of European descent in this study (Bentham et al., 2015).

We retrieved SNPs of LTL as instruments, which are associated with LTL at a genome-wide significance threshold of p < 5 × 10⁻⁸. Then, to avoid colinearity between SNPs, we employed the remaining SNPs at a threshold of linkage disequilibrium clumping r ² < 0.001 exceeding a 10 Mb window using the European sample of 1,000 genome data as a reference panel. For the sake of accuracy and consistency in SNPs selected as IVs across different analyses, the SNPs were not replaced with proxy SNPs if they were missing variants for the outcome in the GWAS. In addition, palindromic SNPs with intermediate allele frequencies were excluded when we conducted a harmonization process between effect alleles for SNP exposure and variant alleles for effect estimates from the different data sets of outcome according to the same effect alleles. A total of 153 SNPs were obtained for leukocyte telomere length. We then calculated the F-statistic value of each SNP to evaluate its strength. SNPs with F values greater than 10 indicated strong instruments that could avoid weak instrument bias (Burgess et al., 2016). The F values of 153 SNPs in this study were above 10.

For MR estimates, the IVW method was conducted as the main statistical analysis method. The Wald ratio estimates for each SNP were meta-analyzed under a random-effects model IVW. Given that this method can return unbiased estimates of a causal effect only if all SNPs have no invalid IV and unbalanced horizontal pleiotropy (Hartwig et al., 2017), we performed complementary sensitivity analyses, which included the MR-Egger, MR-RAPS, weighted median, weighted mode, and MR-PRESSO methods, to investigate the sensitivity and examine the robustness of the causal effects of LTL on the risk of each of the IMIDs. If the instrument strength independent of direct effect (INSIDE) assumption is satisfied, the MR-Egger method can examine whether the horizontal pleiotropic effects were presented in the data among all SNPs via the intercept and can also provide a reliable and unbiased assessment of causation (Bowden et al., 2015). An intercept value of the regression line that deviates remarkably from zero (p-value < 0.05) identifies horizontal pleiotropy or failure of the INSIDE. Specific and systematic pleiotropy can be corrected by MR-RAPS, which can provide a robust statistical estimate for our MR study even with several weak IVs (Zhao et al., 2019). The weighted median method uses the median effect of all available genetic instruments. Therefore, if at least half of the weight comes from valid instrumental variables that have no association with confounders, no horizontal pleiotropy, and a robust association with the exposure, the weighted median method could show the consistency of potential causality (Bowden et al., 2016). The weighted mode method can provide an unbiased causal relationship when the largest number of SNPs based on the similarity of causation are valid even though the remaining SNPs are invalid (Hartwig et al., 2017). Moreover, to further identify and avoid bias due to horizontal pleiotropy, we performed the MR-PRESSO method (Verbanck et al., 2018). Three components (the global test, outlier test, and distortion test) constitute the MR-PRESSO, which detects outliers, removes them, and evaluates whether there is a significant difference (p < 0.05) in the causal estimate after adjustment for outliers (Verbanck et al., 2018). Meanwhile, we used the Galbraith radial plot to find outliers as well (Bowden et al., 2018). Cochran’s Q test was executed to evaluate heterogeneity among variant-specific causal estimates, and a p-value of Cochran’s Q test of IVW > 0.05 indicated that there was no presence of heterogeneity (Bowden and Holmes, 2019). Besides, we further performed leave-one-out (LOO) analysis, forest plots, and funnel plots to evaluate the reliability and pleiotropy of the casual estimation. Scatter plots were generated to visualize the estimated effect sizes. We also applied Steiger filtering to test the causal direction of the association (Hemani et al., 2017). After removing pleiotropic SNPs, we retested the MR causal estimation. The effect estimates of IMIDs were converted to odds ratio (OR) with 95% confidence intervals (CIs), which expressed a per unit genetically determined increase in the log OR of telomere length for IMIDs. The statistically significant association of this study was determined primarily by p < 0.05. The statistical tests for MR were performed using the R software (version 4.1.2) with two packages, which included two-sample MR and MR-PRESSO.

All studies only utilized publicly available summary statistics. Ethical approval for these studies can be found in the original publications.

For parts of IMIDs, our findings were similar to those from some previous studies, while for other IMIDs, they were the opposite. The results from the MR study based on data from the FinnGen Biobank corroborate the findings of retrospective studies that show shorter LTL is a risk factor for RA (Blinova et al., 2016; Gamal et al., 2018; Lee and Bae, 2018), SS (Kawashima et al., 2011; Noll et al., 2020), sarcoidosis (Maeda et al., 2009; Afshar et al., 2021), IPF (McDonough et al., 2018; Ley et al., 2020), and psoriasis (Liu et al., 2008). Furthermore, our results are also similar to previously reported MR studies of LTL and the risk of RA (Zeng et al., 2020) and IPF (Duckworth et al., 2021). Meanwhile, our results are in accordance with previous studies, which found that longer LTL is likely to increase the risk of AS (Tamayo et al., 2010), and found a weak or null correlation between childhood asthma (Belsky et al., 2014; Suh et al., 2019a), IBD (which includes UC and CD) (Getliffe et al., 2005), and LTL. Interestingly, Tamayo et al. (2010) implemented a cross-sectional study in 2010 and found longer telomeres in patients with AS (N case = 59) in PBMCs than in 130 controls, but found that a shorter average TL with aging was evident in AS 4 years thereafter (Tamayo et al., 2014). According to a case–control study, the extent of telomere loss due to aging was almost consistent in UC, CD, and control patients (Getliffe et al., 2005). Nevertheless, Risques et al. (2008) found that the TL was shorter in patients with UC than in control participants, and notably, the TL of colon stromal cells was not affected. Thus, the role of LTL in IMIDs requires further study. However, the MR study found different results from previous studies: shorter or longer LTL may increase the risk of T1D (Sanchez et al., 2020; Svikle et al., 2022), AD (Wu et al., 2000), and PSC (Laish et al., 2015). For instance, two cross-sectional studies found that the TL was shorter in T1D patients than in non-diabetic subjects (Sanchez et al., 2020; Syreeni et al., 2022). A longitudinal study suggested that telomeres shorten in 55.3% of subjects with T1D and further reported that shorter TL had an adverse effect on cardiometabolic and renal profiles (Syreeni et al., 2022). Observational studies specializing in LTL with relation to the risk of hypothyroidism or hyperthyroidism are scarce. We retrieved an analysis of this aspect from a previous study (Codd et al., 2021). It investigated differences in LTL and thyroid function (hypothyroidism and hyperthyroidism) between MR and an observational study. It found limited evidence of causality between LTL and hyperthyroidism in the observational study but not in the MR study. However, it evidenced a significant LTL shortening in subjects with hypothyroidism via MR and a previous observational study which supports our findings (Codd et al., 2021). The results of the retrospective studies are mixed, or previous findings are inconsistent with MR, which may be interfered with by confounding factors or reverse causation, whereby changes in telomere length arise as a result of disease.

The mechanisms involving IMIDs are numerous, are complicated, and remain unclear. These diseases, despite their specific characteristics, have certain overlapping aspects. We hypothesized several possible explanations for why the shorter LTL is likely to be a risk factor for multiple IMIDs in our results. Several studies have shown that shortened LTL produces increased secretion of inflammatory cytokines/chemokines such as IL-1, IL-12, IL-17, IFN-γ, TNF-α, and IL-6, which could contribute to the imbalance of pro- and anti-inflammatory cytokines, resulting in corresponding symptoms (Chizzolini et al., 2009; Kunz and Ibrahim, 2009; Rodier et al., 2009; O'Donovan et al., 2011). People with short TL due to telomerase mutation might have severe immunodeficiency of T cells producing pro-inflammatory cytokines, which in turn lead to producing corresponding symptoms (Maly and Schirmer, 2015; Wagner et al., 2018). A study suggested telomerase reverse transcriptase was a previously unrecognized NF-κB target gene. In other words, the inhibition of a master regulator of inflammation (NF-κB) may abolish the induction of telomerase transcription (Gizard et al., 2011) and then cause LTL to shorten.

Our findings revealed that TL is genetically associated with SLE and AS, which may indicate a potentially similar mechanism between the two IMIDs. Notably, a longer TL is not correlated with younger biological age in comparison to controls. A few possibilities can be interpreted as follows. First, Calado et al. (2009)observed that in bone marrow CD34⁺ cells, androgen and estradiol could stimulate telomerase activity, which plays an important role in maintaining the length of telomeres; moreover, androgen also mediated telomerase activity in both normal peripheral blood lymphocytes. Their findings may indicate a significant difference between males and females in the incidence of SLE (Tian et al., 2022) and AS (Tarannum et al., 2022; Zhang et al., 2022). The antigen presentation process stimulates telomerase in normal B cells and T cells, leading to increased telomerase activity (Igarashi and Sakaguchi, 1997; Hathcock et al., 2005). Upregulation of telomerase activity may partially compensate for the shortened telomeres, and this compensation would allow the capacity of replication of the responding peripheral blood lymphocytes to be maintained or even prolonged. Therefore, during the immune response, the function of these lymphocytes is enhanced and mediates the onset of IMIDs (Hathcock et al., 2005). In addition, TRF1 and POT1, two shelterin subunits, are regulators that control telomere elongation (de Lange, 2005; Okamoto et al., 2008), whereby they may be involved in the mechanism by which longer LTL causes SLE and AS.

Numerous studies have reported that many lifestyle habits involved with IMIDs have a suspiciously intimate correlation with the alteration of the TL. For instance, chronic stress (Shalev et al., 2013), dietary habits (Crous-Bou et al., 2014; Buttet et al., 2022; Gong et al., 2022) such as a Mediterranean diet (Crous-Bou et al., 2014), exercise (Latifovic et al., 2016; Dankel et al., 2017; Puterman et al., 2018; Buttet et al., 2022), alcohol consumption (Pavanello et al., 2011; Martins de Carvalho et al., 2019), and smoking (Huzen et al., 2014; Müezzinler et al., 2015; Latifovic et al., 2016) exert significant effects on the pathophysiology of telomere dynamics, but the results of researching their impact on causality is unreliable due to the deficiency of designs. Furthermore, LTL was measured in 1,156 individuals in four longitudinal studies by Benetos et al. (2013). The reality is that over time, LTL is fixed at a specific level. Therefore, lifestyle habits and their modification during adulthood have little impact on LTL ranking.

Our findings can give certain suggestions on the implications of LTL in clinical works. Since LTL is a causal risk factor for IMIDs, it can be used as a predictor of risk of IMIDs. Although changes in LTL show no relationship with IMIDs, it can be used as a target to further discover the mechanism of IMIDs, which may provide new diagnostic biomarkers and potential treatment targets.

The multiple advantages of this MR study are as follows. First, the robust SNPs for LTL are identified from the largest and latest GWAS. Second, to reduce population stratification bias as much as possible, most of the samples in this study were from the FinnGen consortium, and the other GWAS, although not part of the Finnish database, was still from the European population. Third, the estimation of the SNPs associated with LTL and IMIDs was obtained from two independent subjects in order to reduce bias in the evaluation of causality. Moreover, we selected robust SNPs via stringent quality control conditions and employed comprehensive analysis methodologies. Therefore, MR might be a useful tool for investigating the potential causality in a scientific context.

This study is also subject to some limitations. First, we only conducted analyses for peripheral leukocyte TL (Codd et al., 2021) and not for TL of specific tissues directly affected by each IMID, although it is highly correlated with TL in other tissues (Demanelis et al., 2020). Second, the possibility that the LTL-related SNPs affect IMID outcomes through other causal pathways than through LTL exposure cannot be completely ruled out. Third, our results may not fully represent IMIDs (since not all studies of IMIDs provided publicly available data, and our results are underpowered to conclude the causality regarding secondary disease outcomes). Fourth, while the sample size is large in the FinnGen Biobank, too small numbers of some IMID cases lead to unbalanced frequency matching of case and control groups. Then, this may affect the accuracy and precision of the analytical results. Fifth, our MR analysis hypothesized a linear shape of the association between LTL and IMIDs, although the possibility of other associational shapes such as “J” and “U” should be taken into account. Sixth, we failed to explore the statistical differences in gender regarding the causal effect between LTL and IMIDs (Pierce and Burgess, 2013) since details of personal data in the FinnGen biobank and the GWAS data set are not published. Seventh, since all the individuals involved in our study were exclusively of European descent, our findings might not fully represent whole populations. Thus, it is not plausible to extrapolate the findings from our results to other populations. Eighth, our analysis can yield reliable results when three assumptions are satisfied (Figure 1). However, the second and third assumptions are difficult to test in many cases, and they are also difficult to satisfy. Therefore, we need to treat the results of MR with caution.