Deciphering the causal relationship between blood pressure and regional white matter integrity: A two‐sample Mendelian randomization study

Elevated arterial blood pressure (BP) is a common risk factor for cerebrovascular and cardiovascular diseases, but no causal relationship has been established between BP and cerebral white matter (WM) integrity. In this study, we performed a two‐sample Mendelian randomization (MR) analysis with individual‐level data by defining two nonoverlapping sets of European ancestry individuals (genetics–exposure set: N = 203,111; mean age = 56.71 years, genetics–outcome set: N = 16,156; mean age = 54.61 years) from UK Biobank to evaluate the causal effects of BP on regional WM integrity, measured by fractional anisotropy of diffusion tensor imaging. Two BP traits: systolic and diastolic blood pressure were used as exposures. Genetic variant was carefully selected as instrumental variable (IV) under the MR analysis assumptions. We existing large‐scale genome‐wide association study summary data for validation. The main method used was a generalized version of inverse‐variance weight method while other MR methods were also applied for consistent findings. Two additional MR analyses were performed to exclude the possibility of reverse causality. We found significantly negative causal effects (FDR‐adjusted p < .05; every 10 mmHg increase in BP leads to a decrease in FA value by .4% ~ 2%) of BP traits on a union set of 17 WM tracts, including brain regions related to cognitive function and memory. Our study extended the previous findings of association to causation for regional WM integrity, providing insights into the pathological processes of elevated BP that might chronically alter the brain microstructure in different regions.


| INTRODUC TI ON
Elevated blood pressure (BP) which will progressively develop into hypertension is among the major modifiable risk factors for cerebrovascular and cardiovascular disease (CVD) (Chobanian et al., 2003;Fuchs & Whelton, 2020). Studies have shown that high BP is associated with significant brain structure alterations, particularly in multiple frontal, striatal, and temporal regions (Gons et al., 2010).
The cerebral white matter (WM) is considered critical for connecting local and distance brain regions for efficient cognitive functioning (Madden et al., 2009). Previous studies on both young and old adults have shown that strong associations between BP and WM are particularly compelling, often by measuring the fractional anisotropy (FA) of water diffusion through diffusion tensor imaging (DTI) (Maillard et al., 2012;Salat et al., 2012). Despite their strong associations, the causal directions between BP and WM integrity have not been fully established (Williamson et al., 2018). It is commonly assumed that high BP impacts WM decline (Allen et al., 2016;Taylor-Bateman et al., 2022), but there is also the possibility that impaired WM connectivity may affect central regulations of BP. Recent studies have documented the causal effect of high BP on white matter hyperintensities (WMHs) (Georgakis et al., 2020;Sargurupremraj et al., 2020) as well as WM measures that average over the whole brain (Taylor-Bateman et al., 2022). However, they tend to ignore the heterogeneous effects on different brain regions with specialized functions. Establishing the causality link between BP and regional WM is critical for identifying effective preventative approaches to reduce the clinical impact of hypertension on specific parts of the brain.
Imaging genetics is an emerging field that performs integrative analyses of imaging and omics data to gain more insights into the mechanism of disease. Large-scale consortiums, such as UK Biobank (UKB) (Sudlow et al., 2015) and Enhancing Neuro Imaging Genetics through Meta Analysis (ENIGMA) (Thompson et al., 2014), have been conducted to collect comprehensive genetic and neuroimaging data from a very large population. In this study, we used genetic, regional neuroimaging features as well as BP from the UKB cohort to investigate the causal relationship between BP and regional WM integrity by performing a two-sample Mendelian randomization (MR) analysis.
Traditional observational studies only assess associations between exposure and outcome variables. Causal interpretation of observed associations must account for both confounding and reverse causality as well as other biases (Tilaki, 2012). Making use of nature's random assortment of genetic makeup, the MR methods treat genetic variant as instrumental variable (IV) to minimize the effect of confounding and evaluate the causal effect of a modifiable exposure on outcome in observational studies (Lawlor et al., 2008). The selection of genetic variants as valid IVs that satisfy the IV assumptions is central to the MR analysis. The genetic determinants of BP as well as brain WM microstructure are increasingly characterized in the literature (Evangelou et  was carefully selected as instrumental variable (IV) under the MR analysis assumptions. We existing large-scale genome-wide association study summary data for validation. The main method used was a generalized version of inverse-variance weight method while other MR methods were also applied for consistent findings. Two additional MR analyses were performed to exclude the possibility of reverse causality.
We found significantly negative causal effects (FDR-adjusted p < .05; every 10 mmHg increase in BP leads to a decrease in FA value by .4% ~ 2%) of BP traits on a union set of 17 WM tracts, including brain regions related to cognitive function and memory.
Our study extended the previous findings of association to causation for regional WM integrity, providing insights into the pathological processes of elevated BP that might chronically alter the brain microstructure in different regions.

K E Y W O R D S
blood pressure, diffusion tensor imaging, genome-wide association study, Mendelian randomization, white matter

Significance
Causality between blood pressure (BP) and cerebral white matter (WM) integrity has not been fully established. BP has possibly heterogeneous effects on different brain regions with specialized functions. A generalized inversevariance weighted method and other MR methods were used to evaluate the causal effects of BP on regional WM integrity measured by the fractional anisotropy of diffusion tensor imaging. Elevated BP was found to lead to the decline in WM integrity in 17 different brain regions, related to cognitive function and memory. et al., 2021), which made it possible to identify better instruments for MR study to investigate their causal relationships.
In this study, we hypothesized that increase in BP caused decline of WM integrity in different brain regions. We considered two BP traits: systolic blood pressure (SBP) and diastolic blood pressure (DBP), and performed a two-sample MR analysis to evaluate their causal effects on 39 WM tracts measured by FA that described WM microstructures in different brain regions. We strictly followed the MR analysis assumptions to select the genetic variants as valid IVs and implemented a generalized version of inverse-variance weights (gen-IVW) MR method using data from the UKB cohort. Additionally, we also used existing large-scale meta-analyzed genome-wide association study (GWAS) summary statistics of BP to select IVs and perform MR analysis to validate our finding. Our MR results showed consistently negative causal effects of BP traits on a union set of 17 WM tracts. Among them, elevation in both SBP and DBP had the strongest adverse impacts on bilateral posterior thalamic radiation (PTR) and subfornical organ (SFO) regions, related to cognitive function and memory Walsh et al., 2011). We further performed two additional MR analyses by treating FA as the exposure and ruled out the possibility of reverse causality. These findings extended previous findings (Georgakis et al., 2020;Sargurupremraj et al., 2020;Taylor-Bateman et al., 2022) to provide insights into the pathological processes related to elevated BP that chronically alter the WM microstructure in different brain regions, which in turn affects the cognitive functioning.

| UK Biobank cohort and variables
The UKB is a large prospective study that recruited ~500,000 participants between 40 and 69 years at baseline from 2006 to 2010 in 22 assessment centers throughout the United Kingdom and collected comprehensive genetic and phenotypic details (Sudlow et al., 2015).
It has continued to collect more data such as imaging-derived phenotypes starting from 2014 and repeat assessment data between 2012 and 2013 (Conroy et al., 2019). Since the repeat assessment data were drawn from ~20,000 participants only with a low overall response rate (~21%) (Biobank, 2013(Biobank, , 2014, our analyses were restricted to family-unrelated individuals of European ancestry (i.e., primarily British, Irish) and utilized data from all sites and the initial (2006)(2007)(2008)(2009)(2010) and imaging (2014+) assessment visits of the UKB. Figure 1 shows the number of subjects included at each step of the analysis. The UKB data are split into genetic/outcome and genetic/exposure data and each carefully processed with inclusion and exclusion criteria (see details in subsequent subsections). The twosample MR analysis in UKB cohort (1) was performed on two nonoverlapping sets: the set for exploring genetic-exposure association (denoted GEset) that includes genotype and BP data but no regional WM FA data; and the set for exploring the genetic-outcome association (denoted GOset) that includes genotype, BP, and regional WM FA data. We also use large-scale meta-analyzed GWAS summary statistics of BP to select IVs and combine with the GOset to perform another two-sample MR analysis (2) to validate our findings, details of this cohort can be found in Section 2.2.

| Blood pressure traits
As the exposures of interest, we analyzed the following two BP traits in nonpregnant individuals with complete data collected from the initial (2006-2010) assessment visit. For each trait, two sets of measurements were taken within a 1-min interval and automated reading from Omron device (0-255) as described in the UKB protocol (https://bioba nk.ctsu.ox.ac.uk/cryst al/cryst al/docs/Blood press ure.pdf). We then further calculated the mean of two measurements for each BP trait: 1. Systolic blood pressure (SBP). The SBP was calculated as the mean of two non-null BP measurements using phenotype codes 4080 (Systolic blood pressure, automated reading) in the UKB.
2. Diastolic blood pressure (DBP). The DBP was calculated as the mean of two non-null BP measurements using phenotype codes 4079 (Diastolic blood pressure, automated reading) in the UKB.
We excluded individuals with discordance between self-reported sex and genotype-inferred sex from genomic data. Furthermore, we excluded individuals who took antihypertensive treatments at baseline using phenotype code 20003 (Treatment/medication code) because their observed BP did not reflect the genetically predicted BP (Malik et al., 2021). We also performed the same MR analysis when we included individuals having taken antihypertensive treatments as a sensitivity analysis (Table S8).

| White matter integrity data
As our primary outcome, we analyzed regional WM integrity measured by FA. FA measure is a scalar value ranging from 0 to 1, which describes the degree of anisotropy of a diffusion process (Timpe et al., 2011). A greater FA value indicates that the corresponding localized WM fiber bundles are more intact reflected by higher probability of water diffusion along the longitudinal axis of WM bundles (Timpe et al., 2011). The UKB consists of multi-modal braining imaging data covering structural, functional, and diffusion imaging based on acquisition protocol described elsewhere (Alfaro-Almagro et al., 2018).
Particularly, they averaged the skeletalized images across a set of 48 standard-space tract masks according to the ENIGMA protocol for the DTI FA images to dispose of the FA images into a standard-space white matter skeleton using a tract-based spatial statistics (TBSS) analysis (Smith et al., 2006). In this study, we concentrated on a total of 39 FA tracts for multiple brain WM regions in individuals with complete data recruited in UKB at the imaging (2014+) assessment visit (see the full list of names of 39 regional WM FA tracts and their abbreviations in Table S1). We excluded individuals having extreme values of total volume of WMH using phenotype code 25781 (Total volume of white matter hyperintensities (from T1 and T2_FLAIR images) to minimize the bias ( Figure S2, see more details in the Supplementary Methods) (Jones et al., 1999)).

| Potential confounders
The potential confounders used in our two-sample MR study as recommended by previous study  include: sex, age, body mass index (BMI), alcohol consumption and smoking status, fruit and vegetable consumption, and sedentary lifestyle (see details on their UKB phenotype codes in the Supplementary Methods). We F I G U R E 1 Flow chart showing the number of subjects included at each step of the analysis. On the left, the UKB data are split into genetic/outcome and genetic/exposure data and each carefully processed with inclusion and exclusion criteria. After QC step, we performed two-sample MR analysis (1) on two nonoverlapping sets: the set for exploring genetic-exposure association (denoted GEset) that includes genotype and BP data but no regional WM FA data; and the set for exploring the genetic-outcome association (denoted GOset) that includes genotype, BP, and regional WM FA data. We also used existing large-scale meta-analyzed GWAS summary statistics of BP to select IVs and combine with the GOset to perform another two-sample MR analysis (2) to validate our finding. Shown in the conceptual diagram of twosample MR analysis, the genetic variants serve as the instrumental variables (IVs), U represents the confounder(s), BP trait is the exposure of interest and FA is the outcome in our study. The solid lines with arrows indicated the causal relationship in directed acyclic graph. Under this two-sample MR model, we aimed to assess the hypothesized causal effects of BP on regional WM FA (arrow highlighted in red). The dashed line marked with a "cross" indicates that the valid IV must be satisfied with the IV assumptions (ii) IVs should be independent of the confounders and (iii) IVs should be independent of the outcome given the exposure. restricted our two-sample MR analysis to individuals with no missing values in the aforementioned set of confounders (i.e., complete case analysis) and we performed two-sample t-test for each confounder separately comparing the FA values of those with versus without missing confounder values to make sure missing completely at random assumption is not violated for our complete case analysis (Table S2).

| Existing large-scale blood pressure GWAS summary data
To strengthen the IV selection step and validate our causal finding, we also collected the existing large-scale GWAS summary data on BP from the meta-analysis of over 750,000 participants of European ancestry recruited from a total of 78 different cohorts (mainly from UKB and International Consortium of Blood Pressure (ICBP))   (Lyon et al., 2021). We used this existing meta-analyzed GWAS summary data to select IVs and conduct MR analysis as a validation of our causal findings using UKB alone ( Figure 1, right).

| Mendelian randomization analyses
In this study, we performed a two-sample MR analysis to evaluate potential causal effects of SBP and DBP on FA of 39 WM tracts ( Figure 1). We considered genetic variants as IVs, BP as the exposure, and FA as the outcome. In the GEset, we restricted to individuals with both genotype and BP data but no regional WM FA data to investigate the genetic-exposure associations. In the GOset, we restricted to individuals with both genotype, BP and regional WM FA data to investigate the genetic-outcome associations. There were no shared individuals between these two sets of samples. The two-sample MR analysis consists of two main analytical steps: (1) Selecting valid IVs that satisfied the IV assumptions; and (2) evaluating the causal effects using the selected IVs. For step (1), we also used the existing meta-analyzed GWAS summary statistics to select IVs to validate the findings. In addition, we performed two more MR analyses by switching the exposure and the outcome to rule out the reverse causality. We describe these steps in more details in the following subsections.

| Instrumental variable selection and assumption checking
There are three standard assumptions for IV selection in MR analysis (Angrist et al., 1996): (i) The IV is associated with the exposure; (ii) the IV is independent of the confounding factors; and (iii) the IV is independent of the outcome given the exposure (i.e., the IV does not exert horizontal pleiotropy).
We selected valid IVs based on the above assumptions. First (i), we performed GWAS analysis of BP traits under an additive genetic model adjusting for sex, age, BMI, genotyping chip type (i.e., UKBL/UKB chip) and top 10 principal components (PCs) of population admixture using PLINK (version 1.9, http://www.cog-genom ics.org/plink/ 1.9/) (Chang et al., 2015). We adopted a conservative cutoff (e.g., GWAS p-value <1 × 10 −6 ) to select potential IVs and performed linkage disequilibrium (LD) clumping with r 2 threshold of .5 within window of 1 Mb to remove redundant highly correlated variants using PLINK (version 1.9, http://www.cog-genom ics.org/plink/ 1.9/) (Chang et al., 2015). This choice of r 2 threshold and window size follows from the recent studies (Chen et al., 2021;Hou et al., 2022;Karlsson Linnér et al., 2019) to balance between being too stringent (too small r 2 , too wide window size, very few IVs selected) and too conservative (too large r 2 , too narrow window size, too many IVs selected). We also performed a sensitivity analysis by using the other thresholds and results remain largely consistent, reflecting that our parameter setting can select fewer but more specific and sufficiently strong IVs to achieve a good detection power (Table S3). Next (ii), we eliminated IVs associated with potential confounders by performing association test between each IV and each potential confounder (FDR-adjusted p-value >.05). Lastly (iii), we performed conditional independence tests to exclude IVs with evidence of horizontal pleiotropy (FDR-adjusted p-value >.1). In addition to satisfying the standard IV assumptions, we also followed more recent guidance for IV selection from Burgess et al. (2019) to select IVs that more biologically relevant to the exposure by performing the follow-

| Generalized version of inverse-variance weighted method and other MR methods to evaluate causality
We first performed an association analysis of FA in each WM tract with BP traits and restricted our causal analysis only to FA in regional WM with significantly negative association with BP (Leritz et al., 2010;Suzuki et al., 2017;van Dijk et al., 2004). We then used the selected IVs to perform the two-sample MR analysis using gen-IVW method developed by Burgess et al. (2016). The gen-IVW method generalized the inverse-variance weighted method in MR analysis by taking correlations among IVs into account . If a negative causal effect of BP on any WM tract was detected, we further performed MR analyses using other robust MR methods such as outlier-robust method (e.g., Mendelian randomization pleiotropy residual sum and outlier (MR-PRESSO) test (Verbanck et al., 2018)) and consensus method (e.g., weighted-median approach ) for consistent findings (see details in Supplementary Material). A leave-one-out approach (LOOA) (Corbin et al., 2016) (i.e., leave one IV out at a time and rerun gen-IVW) was also performed as a sensitivity analysis (see details in Supplementary Material).

| Two additional two-sample MR analyses for reverse causality
A paradoxical relationship between low BP and reduced volumes of WM has been reported in some previous studies (Foster-Dingley et al., 2015;Td et al., 2003), suggesting that brain WM integrity might also cause changes in BP. To exclude the possibility of reverse causality, we preformed two additional two-sample MR analyses (i.e., MR reverse I and II) treating FA as the exposure and BP as the outcome. Due to the high correlation among human WM tracts treated as multiple exposures (Telford et al., 2017), we first estimated a general factor of FA (gFA) using factor analysis implemented in an R package "pysch" (version 1.7.8) (Revelle & Revelle, 2017) (see Figure S1). Then, we performed GWAS analysis on gFA adjusting for age, sex, BMI, genotyping chip type (i.e., UKBL/UKB chip), and top 10 PCs of population admixture by using PLINK (version 1.9, http:// www.cog-genom ics.org/plink/ 1.9/) (Chang et al., 2015).
In MR reverse I, following the scheme of bidirectional twosample MR analysis (Zheng et al., 2017), we selected IVs according to IV assumptions (i)-(iii) with the switched exposure and outcome.
In MR reverse II, we selected IVs that had significant association with both BP and gFA. Our MR reverse II was beyond the scope of bidirectional two-sample MR analysis, adding stronger evidence of whether the reverse causality existed.
The MR-PRESSO test was implemented by using the R package "MR-PRESSO" (version 0.1.0) (Verbanck et al., 2018). For both association and MR analyses, we controlled false discovery rate (FDR) at 5% to adjust for multiplicity using the Benjamini-Hochberg procedure (Benjamini & Hochberg, 1995).

| RE SULTS
The GEset includes a total of 203,111 individuals to investigate the genetic-exposure associations, while the GOset includes 16,156 individuals to investigate the genetic-outcome associations. Baseline characteristics of individuals from the two sample sets are summa- years. There are a few self-reported diseases (Table S4) that may or may not confound our findings ((non-)insulin-dependent diabetes mellitus, hypertensive heart disease and chronic ischemic heart disease, etc.), we decide to include these individuals to ensure we have sufficient sample size for GWAS and MR analysis. But at the same time, we also perform sensitivity analysis to ensure our results are consistent while including versus excluding these subjects.
Based on the GWAS results from UKB cohort, we identified 8685 and 10,281 genetic variants associated with SBP and DBP (p-value <1 × 10 −6 ). We further obtained 458 and 527 weakly correlated genetic variants by applying LD clumping for SBP and DBP, respectively. We then checked IV assumptions (ii) and (iii) to refine the pool of candidate IVs (see the number of candidate IVs passing each step for each FA measure in Table S5). A majority of these variants were mapped to BP-related genes (e.g., CACNB2, MECOM, ADRB1, and ARHGAP42) that have been reported in previous studies (Ehret & Caulfield, 2013;Fjorder et al., 2019;Levy et al., 2009;Tikhonoff et al., 2008) (full gene annotations summarized in Table S6). Likewise, IVs selected from the existing meta-analyzed GWAS summary data were also annotated to similar BP-related genes, which validated our IV selection step.
From the association analysis, we identified 37 and 36 WM tracts significantly negatively associated with SBP and DBP, respectively (Table S7; FDR-adjusted p < .05) and our MR analysis were restricted to only these tracts. The MR analyses further identified significantly negative causal effects of BP on 21 and 18 WM tracts for SBP and DBP, respectively (Table S7; FDR-adjusted p < .05). Figure 2 displays their standardized causal effect estimates (standardized coefficient ranging from −5.83 to −2.18; every 10 mmHg increase in BP leads to a decrease in FA value by .4% ~ 2%, see more details in Table S7) from gen-IVW using IVs selected from UKB alone or existing metaanalyzed GWAS summary data. Higher BP caused a reduction in FA values of multiple WM tracts and such negative causal effects were consistent for both SBP and DBP, using IVs selected from UKB GWAS or existing meta-analyzed GWAS. We then applied alternative MR methods including MR-PRESSO (Verbanck et al., 2018) and weighted-median approach  and identified 13 and 14 WM tracts (p-value <.05, Figure 3a) with consistently negative causal effects in all methods for SBP and DBP (standardized coefficient ranging from −5.05 to −2.58; every 10 mmHg increase in BP leads to a decrease in FA value by .4% ~ 2%, see more details in Table S8), respectively. The LOOA also showed consistent causal effects on these WM tracts (see Supplementary File). To account for the effects of antihypertensive treatments, we also conducted a sensitivity analysis by including individuals who took antihypertensive treatments and detected similar results (see Table S8). Figure 3b shows the location of a union set of 17 WM tracts adversely impacted by SBP or DBP in the brain (Rorden & Brett, 2000). TA B L E 1 Baseline characteristics of UK Biobank (UKB)'s individuals from the two sample sets used for two-sample Mendelian randomization (MR) analysis.
We also performed two additional two-sample MR analyses by switching exposure and outcome. In MR reverse I where gFA was treated as the exposure and BP as the outcome, we selected 27 valid IVs for SBP and DBP, respectively (Table S10). Using these IVs, we did not observe significant causal effects of gFA on any BP (FDRadjusted p < .05). In MR reverse II, none of variant was associated with both gFA and BP (Table S10). The results of these two additional MR analyses ruled out the reverse causality of gFA and BP, posing strong supports to our hypothesized causal pathway of BP on regional WM integrity.

| DISCUSS ION
In this study, we integrated neuroimaging, genetics, and BP to conduct a two-sample MR analysis for evaluating the causal effects of BP on regional WM integrity measured by FA. We primarily conducted GWAS analysis on BP traits (i.e., SBP and DBP) and checked MR analysis assumptions using data from UKB cohort to select valid IVs for our two-sample MR analysis. The rigorous IV selection were further validated by existing large-scale GWAS summary data on BP . Most selected IVs were located in genes related to BP, functioning in intervention of heart rate via basal and agonist-stimulated receptor response (Johnson et al., 2011) and modulation of vascular resistance through calcium signals Simonyte et al., 2018). Using these IVs, we applied the gen-IVW and other MR methods, and observed negative causal effects of BP traits on a union set of 17 WM tracts, including brain regions related to cognitive function, motor ability and memory, revealing that brain has differential vulnerability of WM deterioration to BP. These findings were consistent with the previous study F I G U R E 2 Forest plot to display the standardized causal effects of BP on WM FA tracts (21 and 18 with FDR-adjusted p < .05 for SBP (left) and DBP (right), respectively) using gen-IVW method and IVs selected from UKB GWAS alone (blue solid dots) or existing metaanalyzed GWAS summary (orange solid dots). The pooled effect sizes were obtained by a fixed effect model and represented by solid diamonds with blue and orange color, respectively. The 95% confidence interval (CI) was shown as bands. Y-axis refers to regional WM FA tracts, including bilateral posterior thalamic radiation ( by considering the higher BP leads to WM abnormalities (Suzuki et al., 2017;Taylor-Bateman et al., 2022). Moreover, we identified specific WM tracts highly sensitive to changes in particular BP trait (e.g., bilateral EX more sensitive to SBP), which are worth of further explorations. The possible reverse causality was also carefully evaluated and eliminated to disentangle the causality link between BP and WM.
We found that 10 mmHg increases in BP can lead to a decrease in the FA values by .4% ~ 2%, suggesting that the vascular health influences WM structure. The WM regions of SFO and PTR have been identified as being causally affected by high BP because the fibers within these regions are related to memory and cognitive functions with a relatively high risk of brain atrophy. The SBP-specific causal effect on bilateral EX was consistent with the findings in the previous studies (Acosta et al., 2022;Taschler et al., 2022). The DBPspecific causal effect on bilateral ALIC, RLIC-R, and PCR-L has also been reported in the previous literature (Hyett et al., 2018;Takeuchi & Kawashima, 2022), which is supportive of causal relationships between increasing DBP and the development of psychomotor dysfunctions. The present findings demonstrate the regional-specific neural implications of elevated BP on brain structure, which could possibly aid in the prevention of white matter loss in the future.
Genetic variants selected as IVs in the study mainly resided in genes CACNB2 and ARHGAP42 and were shared in both BP traits.
Their encoded proteins have been identified to be essential to intracellular calcium homeostasis by modulating calcium channel activity (Foster-Dingley et al., 2015) and limiting contractility of vascular smooth muscle cells to exert its protective function , respectively. Mutations occurred within these two genes will damage calcium buffering capacities and malfunction calcium channel activities, resulting in neuronal autophagy and neurodegenerative diseases such as Alzheimer's disease and Parkinson's disease (Zündorf & Reiser, 2011). Most of these variants were located in intronic regions of the genes, and further studies need to be conducted to understand the biology behind these genes in regulating BP which in turn alters white matter microstructure in the brain, by integrating with various other omics data such as epigenomic profiles and incorporating expression quantitative trait locus (QTL) database.
This study has several strengths. The two-sample MR study is a powerful method to make causal inferences of modifiable exposure on the outcome in observational studies by using genetic variants as IVs. We applied it to two large cohorts (primarily UKB cohort, and then large existing meta-analyzed GWAS cohort as validation) with high-quality data, providing a greater statistical power to better understand the causal relationship between BP with regional WM FA diminishing confounding and reverse causality. We also restricted our study population to European ancestry to minimize the confounding raising from population stratification. Additionally, we excluded individuals taking antihypertensive treatments and having extreme values of WMH to eliminate the neuroprotective effects of antihypertensive treatments and bias effects of total volume of WMH on brain structures and functions, leading to unbiased causal effects of genetically determined BP across brain regions. We conducted various sensitivity analyses and used several robust MR methods to test the validity and consistency of our causal findings. For example, MR-PRESSO (Verbanck et al., 2018) was adopted to evaluate pleiotropy bias, weighted-median approach  to assess invalid instrument bias of IVs and LOOA (Corbin et al., 2016) to test the sensitivity of single IV, the results of which were consistent with our primary gen-IVW results, indicating the reliability of our causal findings.
However, this study has some limitations with potential extensions. First, although we avoided the confounding effects of antihypertensive treatments by removing the participants taking this treatment, further study is encouraged to account for the effects of antihypertensive medication. Also, the UKB cohort has a healthy F I G U R E 3 (a) The number of WM tracts identified for SBP and DBP at each step of analysis and a Venn diagram showing the union set of 17 regional WM FA tracts adversely impacted by SBP or DBP.
(b) Location of the union set of 17 WM tracts in the brain. The color represents -In(p-value) generated from a meta-analysis of gen-IVW results from two BP traits and IVs selected from two GWAS results using adaptive-weighted Fisher's method.
volunteer selection bias (Fry et al., 2017), it is critical to validate the generalizability of our results using populations with a wider range of socioeconomic profiles, instead of restricted to the European ancestry. Second, in this study, we investigated the causal effect of BP on multiple correlated outcomes (i.e., FA tracts in different brain regions). The multivariate MR analysis enables assessing multiple correlated exposures simultaneously (Burgess & Thompson, 2015;Sanderson et al., 2019) through including genetic variants from each risk factor into the same model .  (Bowden et al., 2015).
High BP is a common risk factor for the development of CVD with progressive hypertension, which greatly damages the WM integrity in the brain. Our results demonstrated strong evidences for the causal role of genetically determined BP in WM degradation of different parts of brain. These findings are critical to our understanding of the relationship between CVD risk factors and regional brain structural change and can be used for prevention of WM loss in the future.

D ECL A R ATI O N O F TR A N S PA R EN C Y
The authors, reviewers and editors affirm that in accordance to the policies set by the Journal of Neuroscience Research, this manuscript presents an accurate and transparent account of the study being reported and that all critical details describing the methods and results are present.

CO N FLI C T O F I NTE R E S T S TATE M E NT
The authors report no conflict of interest.

PEER R E V I E W
The peer review history for this article is available at https:// www.webof scien ce.com/api/gatew ay/wos/peer-revie w/10.1002/ jnr.25205.

DATA AVA I L A B I L I T Y S TAT E M E N T
The raw genetic and phenotypic data used in the current study are available from the UK Biobank (UKB), which can be accessed via https://www.ukbio bank.ac.uk/.  Data S1.Transparent Science Questionnaire for Authors Table S1. A list of 39 regional white matter (WM) integrity measured by fractional antitropy (FA).  determined at each step and numbers of significant causal effects detected implemented with our statistical method by separately using different LD clumping settings. Table S4. Self-report neurological characteristics of UK Biobank's participants from two independent association sample sets for the two-sample Mendelian randomization (MR) analysis. determined at each step of our statistical method by separately using UK Biobank (UKB) GWAS and existing meta-analyzed GWAS.