Causal effect of insulin resistance on small vessel stroke and Alzheimer's disease: A Mendelian randomization analysis

The causal effect of insulin resistance on small vessel stroke and Alzheimer’s disease (AD) was controversial in previous studies. We therefore applied Mendelian randomization (MR) analyses to identify the causal effect of insulin resistance on small vessel stroke and AD.


EFFECT OF INSULIN RESISTANCE ON SMALL VESSEL STROKE AND ALZHEIMER'S DISEASE
Insulin resistance is a pathological condition resulting from decreased insulin sensitivity both in the periphery and brain [7]. It has been demonstrated that insulin resistance is associated with the risk of ischemic stroke and poor outcomes of ischemic stroke [8][9][10].
Recent observational studies have reported that insulin resistance is associated with an increased risk of CSVD [11][12][13][14]. Moreover, brain insulin resistance was recently demonstrated to play an important role in AD [15]. However, the causal effect of insulin resistance on small vessel stroke and AD was controversial in previous Mendelian randomization (MR) analyses [16,17].
Mendelian randomization is an analytic technique, simulating the design of randomized controlled trials, that uses genetic variants associated with exposure as instrumental variables to infer causality between such exposure and risk of diseases [18]. Because genetic variants are randomly allocated at meiosis and independent of many other confounders, MR analysis can avoid potential biases of conventional observational studies and reverse causality. In the present study, we aimed to determine the causal associations of insulin resistance with the risk of small vessel stroke and AD based on MR analysis.

MATERIAL S AND ME THODS Study Design and Data Sources
We performed this MR analysis in accordance with the recommendations of the STROBE-MR statement [19]. A two-sample MR analysis was applied to evaluate the causal effect of insulin resistance on the risk of small vessel stroke and AD ( Figure 1). MR design is based on the theory that genotypes are randomly assorted at meiosis and independent of confounding factors, so that potential confounders and reverse causation can be controlled for and more reliable causal inferences can be obtained [18]. MR relies on three assumptions: (i) the instrument is associated with the exposure; (ii) the instrument influences the outcome only through the exposure; and (iii) the instrument is not associated with other confounders.
Single nucleotide polymorphisms (SNPs) that are associated with insulin resistance and satisfy the MR assumptions were extracted as instrumental variables. SNPs for insulin resistance based on fasting insulin were obtained from the Meta-Analyses of Glucose and Insulin-related traits Consortium (MAGIC) [20]. SNPs for insulin resistance based on "gold standard" measures were obtained from the GENEticS of Insulin Sensitivity consortium (GENESIS) [21]. Data on the association of SNPs with small vessel disease were obtained from the Multi-ancestry Genome-Wide Association Study of Stroke consortium (MEGASTROKE) [22]. Data on the association of SNPs with AD were obtained from the Psychiatric Genomics Consortium-Alzheimer Disease Workgroup (PGC-ALZ) [23]. All genome-wide association studies (GWASs) of these consortia are based on individuals of European ancestry. The characteristics of the GWASs used in the present study are shown in Table S1. All the original studies had obtained ethical approval, and all participants had provided informed consent.

Selection of genetic variants
We used two methods to measure insulin resistance: (i) a common proxy of insulin resistance measured by fasting insulin and (ii) gold standard measures of insulin resistance based on the euglycemichyperinsulinemic clamp and the insulin suppression test. We selected SNPs associated with insulin resistance from two GWAS consortia: MAGIC [20] and GENESIS [21]. The MAGIC assessed potential SNPs associated with fasting insulin in 108,557 non-diabetic individuals of European ancestry [20]. The consortium using the Illumina CardioMetabochip containing ~66,000 SNPs for a range of cardiovascular and metabolic traits, contributed ~1000 SNPs for fasting insulin. The estimates of associations were adjusted for body mass index and fasting insulin was natural logarithm transformed. The consortium identified 12 loci that achieved significance (p < 5 × 10 −8 ) for insulin resistance. The F statistic of these index variants was 1016, indicating sufficient strength of the instruments. We extracted independent genetic variants without linkage disequilibrium with other SNPs for insulin resistance. The GENESIS analyzed GWAS data on 2764 individuals and replication in 2860 individuals of European ancestry with direct, standard measures of insulin resistance from four cohort GWASs (the RISC consortium, the Uppsala F I G U R E 1 Conceptual framework for the Mendelian randomization analysis of insulin resistance and risk of small vessel stroke and Alzheimer's disease (AD). The design is based on the assumption that the genetic variants are associated with insulin resistance, but not with confounders, and affect small vessel stroke and AD only through insulin resistance. SNP, single-nucleotide polymorphism Longitudinal Study of Adult Men, the EUGENE2 consortium, and the Stanford Insulin Suppression Test [21]). Since no SNPs reached GWAS significance levels of p < 5 × 10 −8 in the initial meta-analysis, the investigators took forward variants representing four of the top signals into follow-up studies, from which they identified that five SNPs most strongly associated with insulin resistance reached significance levels of p < 6 × 10 −6 . The F statistic of these index variants was 72, indicating sufficient strength of the instruments. The associations of the 17 individual SNPs with insulin resistance are shown in Table 1. These insulin resistance-associated SNPs were at different loci and there was no linkage disequilibrium (r 2 < 0.2). Furthermore, they were not associated with other potential risk factors related to CSVD or AD at a genome-wide significance level (p < 5 × 10 −8 ) after performing a search in the PhenoScanner database [24].

Outcomes
Summary statistics for the associations of individual SNPs with small vessel stroke, defined as stroke caused by small vessel disease, were acquired from previously published GWASs conducted by the MEGASTROKE consortium [22]. The MEGASTROKE consortium is a multi-ancestry genome-wide association meta-analysis of stroke and stroke subtypes that tests ~8 million SNPs and indels with minor al-  Table 2.

Statistics analysis
We performed two-sample MR analyses on summary statistics to evaluate the impact of insulin resistance-associated genetic variants on small vessel stroke and AD. The SNP-insulin resistance and SNP-outcome associations were used to calculate estimates of insulin resistance-outcome (small vessel stroke and AD) associations using inverse-variance-weighted (IVW) MR analysis. In sensitivity analyses, we performed complementary analyses using robust IVW, MR-Egger, simple median, weighted median, weighted mode-based estimator (MBE), and MR pleiotropy residual sum and outlier (MR-PRESSO) methods. These methods include pleiotropic or invalid instruments that are more robust to potential violations of the standard instrumental variable assumption. The MR-Egger method can identify and control for bias due to directional pleiotropy (SNPs influence the outcome through different biological pathways other than exposure) [25]. The weighted median method allows stronger SNPs to contribute more toward the estimate and, therefore, allows them to contribute more toward stronger causal inferences [26]. The MR-PRESSO method was applied to identify and correct potential outliers in multi-instrument summary-level MR testing [27]. Potential pleiotropic effects of these SNPs were evaluated via MR-Egger regression, in which the slope represents causal estimates corrected by pleiotropy and the intercept represents the average pleiotropic effects of all SNPs. Heterogeneity of SNPs was estimated using the Cochran Q statistic. If there is heterogeneity, random-effects IVW models are applied; otherwise, the fixed-effect IVW model is applied. To estimate the influence of outlying or pleiotropic genetic variants, we conducted a leave-one-out analysis in which we reestimated the effect by sequentially dropping one SNP at a time.
The odds ratios (ORs) with their 95% confidence intervals (CIs) per one standard deviation log-transformed genetically predicted increase in insulin resistance were used to represent the association between insulin resistance and outcomes (small vessel stroke and AD). We additionally plotted the association of each genetic variant with insulin resistance against its effect on the outcomes. The Bonferroni-adjusted significance for small vessel stroke and AD was calculated as p < 0.025 (0.05/2 = 0.025) to ensure the validity of our conclusions. In the present analyses, two-sided p values < 0.05 were taken to indicate statistical significance for a potential, yet to be confirmed, causal association, and a two-sided p value < 0.025 was taken to indicate statistical significance for a causal association. All analyses were conducted using R 3.5.3 (R Development Core Team).

Causal association of insulin resistance with small vessel stroke
The IVW MR analyses showed positive correlations between insu-  Figures S1 and S2).

Causal association of insulin resistance with AD
The IVW method showed a positive causal association between insulin resistance, assessed by fasting insulin, and the risk of AD (OR 1.13, 95% CI 1.04-1.23, p = 0.004; Figure 3) and a potential causal association of higher insulin resistance, assessed by gold standard measures, with AD (OR 1.02, 95% CI 1.00-1.03; p = 0.03; Figure 3).
Associations between each instrumental variable with insulin resistance and risk of AD are shown in Figure 4c,d.
However, no significant association was found for the risk of AD using simple median, weighted median, and weighted MBE methods The result of leave-one-out sensitivity analyses indicated that the association between insulin resistance and AD was not affected by any individual SNP (data shown in Figures S3 and S4).

DISCUSS ION
In the present study, genetically predicted insulin resistance, either based on fasting insulin or gold standard measures (including euglycemic-hyperinsulinemic clamp and insulin suppression test), showed potential causal associations with increased risk of small vessel stroke and AD. Findings on the causal association between insulin resistance and small vessel stroke were more robust in sensitivity analyses using different instruments and statistical models.
A previous study showed that the frequency of vascular risk factors differed among subtypes of ischemic stroke [28]. Insulin resistance is widely considered a core feature of metabolic disorders but is not limited to type 2 diabetes mellitus, and it was found to be an independent risk factor for small vessel stroke and a predictor of its severity in Korean individuals [12]. Previous studies have used different indices of insulin resistance, such as homeostasis model assessment-estimated insulin resistance (HOMA-IR) index and insulin resistance score, to demonstrate that insulin resistance is associated with occurrence of lacune, although the results of the different indices are not identical [11,12]. The HOMA-IR and triglycerideglucose index, calculated as proxies of insulin resistance, were also indicated to be risk factors for increased CSVD burden [13,14].
However, previous MR studies on the relationship between insulin resistance and small vessel stroke, where insulin resistance was assessed by a proxy measure (fasting insulin), did not yield significant results [29,30], and the GWAS summary data on small vessel stroke TA B L E 2 Genetic association of insulin resistance related genetic variants with small vessel stroke and Alzheimer's disease  A previous observational study indicated that insulin resistance was associated with cognitive impairment in elderly patients with primary hypertension [31]. Radiographic studies taking advantage of positron emission tomography found that higher insulin resistance was associated with reginal cortical hypometabolism in the frontal,  [32,33]. In a post-mortem study, insulin resistance was found to be associated with β-amyloid plaques [34]. However, the evidence of a genetic effect of insulin resistance on AD was conflicting in previous MR studies [16,17]. The MR analysis performed by Walter et al. [17] suggested that insulin sensitivity assessed by subscores formed from a subset of type 2 diabetes mellitus-associated SNPs affects the risk of AD. Nevertheless, the SNPs classified as related to insulin sensitivity were not valid instrumental variables, which may lead to violation of assumptions. Ostergaard et al. [16]performed an MR study and found no association between insulin resistance and AD. We used data from a much larger and more recent consortium and assessed insulin resistance not only through fasting insulin but also through gold standard measures, thus providing a more reliable causal effect of insulin resistance on the risk of AD.
There are several potential mechanisms involved in insulin resistance and CSVD and cognitive impairment. First, futile response of adipocytes to the actions of insulin is observed in patients with insulin resistance, which can lead to lipolysis and dyslipidemia and increase the risk of atherosclerosis [35]. Second, subclinical inflammation and oxidative stress are common in patients with insulin resistance, which can result in endothelium impairment and increased blood-brain barrier permeability, eventually contributing to initial onset and subsequent progression of CSVD [13,14]. Third, insulin was demonstrated to increase cerebral perfusion, which can be impaired owing to insulin resistance and subsequently lead to neuronal dysfunction [36]. Fourth, accumulation of β-amyloid and hyperphosphorylation of tau protein are core features of AD [37]. It was reported that insulin can accelerate βamyloid clearance from the brain and prevent extracellular deposition as well as fibril and plaque formation in normal conditions [38]. Finally, insulin and insulin resistance are suggested to implicate in the aggregation of tau protein. Dysfunction of insulin can lead to tau hyperphosphorylation of specific amino acids such as Ser and Thr [39].
A major strength of the present study is the MR design. In our MR analysis we used genetic variants to assess the causal effect of exposure (insulin resistance) for disease (small vessel stroke and AD) based on multiple insulin resistance-related SNPs and effects of SNP  [21]. Therefore, caution is needed in interpreting the causal relationships between insulin resistance assessed by gold standard measures and small vessel stroke and AD. In addition, our MR analysis was based on individuals of European ancestry, which may limit the generalizability of our findings to other ethnicities.
In conclusion, the present results provide genetic support for a potential causal effect of insulin resistance on small vessel stroke and AD. It is widely known that both type 1 and type 2 diabetes mellitus share long-term microvascular injury/dysfunction, but previous studies have focused more on the microvasculature of eyes and kidneys.
The results of the present study suggest that insulin resistance might contribute to cerebral small vessel injury. Further investigations are needed to verify the association and the mechanism involved.

ACK N OWLED G M ENTS
We thank the MAGIC and GENESIS consortia for providing insulin resistance-related SNPs. Data on small vessel stroke were provided by the MEGASTROKE (PubMed ID: 29531354) investigators. Data on AD were provided by the PGC-ALZ consortium (PubMed ID: 30617256).

CO N FLI C T O F I NTE R E S T
All authors declare that they have no competing interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data of the present study are publicly available and may also be available from the corresponding authors upon request.