Resting heart rate and risk of dementia: a Mendelian randomization study in the international genomics of Alzheimer’s Project and UK Biobank

Introduction

Dementia is one of the rising public health challenges affecting 43.8 million people in 2016 worldwide (Nichols et al., 2019, pp. 1990–2016). The number of individuals with dementia is estimated to be 152.8 million by 2050, with nearly two-thirds of them residing countries with low- and middle-incomes (Nichols et al., 2022). Dementia severely affects patients’ life qualities, families, and health-care systems as a whole (Etters, Goodall & Harrison, 2008). Without cures, identification of the modifiable risk factors is important for postponing the development of dementia.

Despite the mounting global effect of dementia, the cause remains largely unknown (Wirdefeldt et al., 2011; O’Gorman, Lucas & Taylor, 2012; Baumgart et al., 2015). Previous evidence has confirmed that cardiovascular diseases (CVDs) along with their modifiable risk factors including diabetes, obesity, smoking, and hypertension, are associated with dementia (Chen et al., 2021; Korologou-Linden et al., 2022). Still, these factors may only recapitulate partial aspects of dementia’s development. Recently, there has been growing recognition of the predictivity of resting heart rate (RHR) beyond CVD, particularly for dementia (Bohm et al., 2012). Two recent prospective studies have shown that elevated RHR is associated with increased risks of cognitive decline and dementia in long-term follow-up (Imahori et al., 2022; Deng et al., 2022). RHR is simple and easily obtained, so it should prove useful to help identify potentially high-risk populations of dementia in a wide variety of settings. RHR is also manageable via exercise and medications. Therefore, a reduction in RHR through exercise or medical treatment may be explored as an intervention target to delay cognitive decline, which might have crucial implications in public health. However, their relationships have courted controversy as null associations are yielded (Swedberg et al., 2010; Haring et al., 2020). Besides, the causal effect of higher RHR on cognitive decline and dementia remains to be investigated. This is important as a higher RHR may be linked to dementia via an indirect pathway of CVDs in observational study, e.g., ischemic heart disease and stroke (Kuźma et al., 2018; Kokkinidis et al., 2020).

These controversy in current literature necessitate the adoption of alternative models, particularly those that emphasize the causal associations and make use of multiple data instead of a single cohort. Mendelian randomization (MR) is a method that can be used to estimate the causal association of the modifiable exposure on the outcome. It utilizes genetic variants to establish a robust link. It also takes the advantage of large summary statistics derived from genome-wide association study (GWAS) (Hemani, Bowden & Davey Smith, 2018).

In the present study, two-sample MR analyses were performed to determine if there is a causal effect of higher genetically predicted RHR on dementia. Data were obtained from a set of multiple sources, including large-scale GWAS analyses on RHR (Eppinga et al., 2016), Alzheimer’s disease (AD) from the International Genomics of Alzheimer’s Project (IGAP) (Lambert et al., 2013), and the UK Biobank (UKBB) (Marioni et al., 2018).

Materials and Methods

Results

A total of 43 SNPs reached the GWAS significance (P < 5 × 10⁻⁸) and were extracted from the RHR GWAS summary data (Eppinga et al., 2016). Detailed genetic instrument filtering process was shown in Table S1. Several sensitivity analyses were conducted to determine the effect of using different types of MR methods, IVW-MR and MR-Egger, as well as the influence of pleiotropy.

A total of 35 SNPs met the criteria for instrument selection and were used for analysis of the IGAP GWAS dataset. Results using IGAP dataset showed no significant causal effect of higher genetically predicted RHR on the risk of AD (β_GSMR = 0.12, P = 0.30) (Fig. 1A).

There was no indication of pleiotropic SNP from MR-PRESSO for all outcome analyses (P = 0.96 for IGAP, P = 0.89 for MA-U, P = 0.85 for FA-U, P = 0.93 for FH-AD by Marioni et al. (2018), and P = 0.81 for FH-AD by Kunkle et al., 2019). No confounding heterogeneity was detected in all analyses (leave-one-SNP-out; Cochran’s Q statistic, P = 0.96) in this case (Fig. S1). Additional analysis was performed by excluding the SNPs with ambiguous strand which had been analyzed in previous analyses. The results from the alternative analysis were not significantly different from the previous results in the IGAP. There is no evidence of a source of pleiotropy through conducting sensitivity analysis (MR-Egger intercept, P = 0.93). Both the MR-Egger and IVW-MR with fixed or random effect verified the results of GSMR approach for RHR ( $\widehat{\beta }$xy-IVW_fixed = 0.12, P = 0.30; $\widehat{\beta }$xy-IVW_random = 0.12, P = 0.18; $\widehat{\beta }$xy-MR-egger = 0.19, P = 0.25) (Fig. S2).

Despite of a protective effect in the UKBB family history GWAS, estimates confirmed that there was no significant causal relationship between RHR and the family history of AD: MA-U (β_GSMR = −0.18, P = 0.13), FA-U (β_GSMR = −0.14, P = 0.39) (Figs. 1B–1C). The results from sensitivity analyses (Cochran’s Q statistic, P_FA-U = 0.72) and pleiotropy analysis (MR-Egger intercept, P_MA-U = 0.83) determined no presence of confounding heterogeneity of effect sizes. In addition, the results of IVW-MR were consistent with the GSMR analysis in this case (MA-U ( $\widehat{\beta }$xy-IVWfixed = −0.18, P = 0.13), FA-U ( $\widehat{\beta }$xy-IVW_fixed = −0.14, P = 0.39), MA-U ( $\widehat{\beta }$xy-IVW_random = −0.18, P = 0.08), FA-U ( $\widehat{\beta }$xy-IVW^random = −0.14, P = 0.33) (Table 1).

Subsequently, we conducted an overall analysis of RHR on the meta-analysis with combination of IGAP and UKBB family history (FH-AD) (Marioni et al., 2018). A total of 36 SNPs overlapped between RHR and the FH-AD dataset. We confirmed that there was no significant causal effect of higher RHR on AD (β_GSMR = −0.03, P = 0.72; $\widehat{\beta }$xy-IVWfixed = −0.03, P = 0.72, $\widehat{\beta }$xy-IVWrandom = −0.03, P = 0.66) (Fig. 1D; Table 1). The pleiotropy and heterogeneity were not detected in this case. (MR-Egger intercept, P = 0.47; Cochran’s Q statistic, P = 0.91). The reverse MR analysis confirmed that there is no significant causal effect of AD on RHR (Fig. S4). Overall, the results of bidirectional MR confirmed the robustness of the primary results (Table S4).

Further analyses adjusting RHR-modifying medication were estimated in the IGAP and UKBB family history datasets. The forest plot shows both a generalized summary of Mendelian randomization (GSMR) and the outcome of the MR-Egger estimates of the relationship between RHR and dementia by adjusting the use of RHR inhibitors (Fig. 2). Participants who were taking RHR-inhibitor medications were excluded to avoid reverse causal effect in these analyses. The results showed that RHR was not associated with AD diagnosis and family history of AD from UKBB (Fig. 2; Supplemental Materials).

The selected instrumental variables accounted for approximately 0.32% to 0.35% of the phenotypic variance (Table 2). The corresponding F-statistics for the datasets from IGAP, MA-U, FA-U, FH-AD by Marioni, and FH-AD by Kunkle et al. (2019) were 850.45, 895.12, 904.63, 870.45, and 920.41, respectively (Table 2). The calculated statistical powers were 10.2%, 40.1%, 16.7%, 6.3%, and 12.8%, respectively (Table 2). These results indicate that, within the sample size utilized in this analysis, the detection of a causal effect of high resting heart rate on the incidence of Alzheimer’s disease was not effectively achievable.

Discussion

In the present study, a two-sample MR analysis was performed using previously published GWAS summary datasets to determine if there is a causal effect of higher genetically inherited RHR on the risk of AD. Additionally, using AD family history as a proxy for the diagnosis of AD, we analyzed the effect of genetically inherited higher RHR on the risk of having maternal or paternal family history of AD. We demonstrated that there was no evidence that supports a causal role of higher genetically predicted RHR in the risk of dementia.

To further substantiate our datasets, an LD Score regression (LDSC) analysis was conducted, revealing a genetic correlation of 0.13 between RHR and AD (standard error = 0.077, P = 0.087, Supplemental Materials). This correlation, however, was not supported by MR results as indicative of a causal relationship. Consequently, we posit that the association between RHR and AD risk may primarily arise from pleiotropy, aligning with conclusions in other cited literature. Our findings, robust to adjustments for RHR modifying medication, suggest no causal role for higher genetically predicted RHR in dementia risk.

There may be several explanations for our findings. Previous studies have demonstrated that higher RHR could more likely result in asymptomatic (paroxysmal) episodes of atrial fibrillation, leading to subclinical brain damage which is strongly associated with dementia (Haring et al., 2020). Alternatively, a higher RHR may occur secondary to CVDs with activation of sympathetic nervous system, where the RHR is a representative marker of performance of the excitability of sympathetic nervous system. Indeed, one previous study on the genetic link between the RHR gene SNP and dementia showed that the T allele of rs17070145 was associated with lower cognitive function, but was not associated with RHR (Wersching et al., 2011). Another study indicated that three SNPs (rs12474609, rs10201482 and rs980286) in the low density lipoprotein receptor-related gene may have a cognition-protective causal effect on AD, although they are not related to genetic susceptibility of RHR (Poduslo, Huang & Spiro, 2010). These results are consistent with our studies showing no evidence of causal effect of RHR on the risk of dementia, and their associations observed in observational studies were probably due to other underlying subclinical disease processes.

Dementia is a broad term used to describe a set of symptoms that impact memory, cognition, and social abilities. AD is one of the most common causes of dementia and may contribute to 60–70% of cases (Iturria-Medina et al., 2016). Although mounting evidence suggests a causal relationship between of CVDs and the risk of cognitive decline, the association between RHR and risk of dementia remains poorly understood. Previous studies have linked heart rate variability (HRV), calculated from RHR signals, with cognitive decline and dementia. The heart rate variability (HRV) is a marker of cardiovascular autonomic control (Thayer et al., 2009). Decreased HRV indexes indicate a low vagal activity that is associated with the development of several diseases such as diabetes, cardiovascular disease and cancers (Thayer, Yamamoto & Brosschot, 2010). Nevertheless, this does not necessarily imply that the results can be replicated in exploring the association between RHR and dementia. Recent analysis of the modified HRV, among a population of almost 1,500 participants, has revealed that the prognostic power RHR and HAV seems to be different for different outcomes (Sacha et al., 2013). With the HRV independent on RHR, the modified HRV rather than RHR proved to be independent risk factors of the dementia (Sacha et al., 2013).

Several existing studies have investigated the association of RHR with dementia and cognitive function (Bohm et al., 2012; Wang et al., 2019; Haring et al., 2020; Singleton et al., 2021). For example, the SNAC-K study (n = 2,147, average age 71 years) suggested that a higher RHR (≥80 bpm) was associated with the risk of dementia (hazard ratio (HR) 1.55). The association is still significant after participants with prevalent and incident CVDs were excluded. Similarly, RHR ≥ 80 bpm was also associated with cognitive decline (Mini-Mental State Examination score, β = −0.13). The ARIC study on middle-aged adults (average age 58 years) without prevalent CVDs showed that RHR ≥ 80 bpm (vs. <60 bpm) was associated with an increased risk of incident dementia and RHR ≥ 70 bpm was associated with a subsequent cognitive decline after the adjustment for various CVDs (β = −0.12), which was consistent with the SNAC-K study. The SPRINT study (n = 5,168, average age 67 years, mean RHR 70.4 beats per minute) indicated that a higher RHR (per ten beat-per-minute increase) was correlated with an increased risk of dementia or mild cognitive impairment (HR 1.09) in adjusted models. Finally, PRoFESS trial on 20,165 patients with post-ischemic strokes showed that an elevated RHR was associated with cognitive decline during the follow-up. Despite of the adjustment of CVDs in all analyses, none of these studies adjusted for the use of RHR medications, e.g., use of beta-blockers and calcium-channel blockers, which could affect the robustness of the results.

Conversely, the results were inconsistent among other two studies (Kuźma et al., 2018; Kokkinidis et al., 2020). The association of higher RHR with cognitive impairment after a 15-year follow-up was not observed from the Women’s Health Initiative Memory Study of participants without prevalent CVDs and cardiovascular risk factors (n = 493, age ≥ 63 years, average age 69 years), although there was an association between higher RHR and ischemic brain lesions on magnetic resonance imaging. RCT also showed that intervention using ivabradine (a specific inhibitor of RHR) to reduce RHR could not improve participants’ cognitive function. These contradictory results could be derived from the facts that RCT studies did not reach statistical power, and observational studies analyze correlation, but not causation.

A major concern we considered when planning this study was the potential confounding from the use of RHR inhibitors. Medication with RHR inhibitor has a modifying effect on the exposure, which may lead to the reverse causal effect between RHR and the risk of AD risk (Eppinga et al., 2016). Our MR results remained unaffected after the adjustment by excluding participants who were taking RHR medications. To the best of our knowledge, this is the first study that explored the causal effect of higher RHR on the risk of dementia. However, this study has several limitations. First, we used self-reported family history of dementia from UKBB participants as a proxy for AD diagnosis. The limitations of using the UKBB include self-reporting bias and use of a proxy phenotype rather than clinical diagnosis of AD, although there is a significant genetic correlation between proxy AD and clinical AD status. Nevertheless, a global meta-analysis demonstrates that self-reported information of family history of AD can indeed accurately reflect proxy for clinical AD diagnosis (Marioni et al., 2018). Second, our analyses did not reach sufficient statistical power to detect a causal effect, and thus further analysis using larger sample sets is needed to provide more conclusive results. Lastly, since there is a lack of enough information to estimate the actual overlap between exposure and outcome, weak instrument bias cannot be ignored. However, the additional R² we provided may partially address this concern.

Conclusions

The present study did not observe evidence that supports causal effect of RHR on the risk of dementia, and the results remained stable after adjusting for RHR-modifying medication use. Association between increased RHR and dementia found in previous studies could be due to the underlying subclinical CVDs processes and cardiovascular risk factors. Further study using updated data from large genetic studies is warranted to verify the results of our MR study. Long-term randomized controlled studies are needed to determine if RHR medications could prevent dementia and be recommended as preventive agents.

Supplemental Information