Cardiovascular health, infection burden and their interactive association with brain volumetric and white matter integrity outcomes in the UK Biobank

Abstract

Background.

Cardiovascular health is associated with brain magnetic resonance imaging (MRI) markers of pathology and infections may modulate this association.

Methods.

Using data from 38,803 adults (aged 40–70 years) and followed-up for 5–15 years, we tested associations of prevalent total (47.5%) and hospital-treated infection burden (9.7%) with brain structural and diffusion-weighted MRI (i.e., sMRI and dMRI, respectively) common in dementia phenome. Poor white matter tissue integrity was operationalized with lower global and tract-specific fractional anisotropy (FA) and higher mean diffusivity (MD). Volumetric sMRI outcomes included total, gray matter (GM), white matter (WM), frontal bilateral GM, white matter hyperintensity (WMH), and selected based on previous associations with dementia. Cardiovascular health was measured with Life’s Essential 8 score (LE8) converted to tertiles. Multiple linear regression models were used, adjusting for intracranial volumes (ICV) for subcortical structures, and for demographic, socio-economic, and the Alzheimer’s Disease polygenic risk score for all outcomes, among potential confounders.

Results.

In fully adjusted models, hospital-treated infections were inversely related to GM (β±SE: −1042±379, p=0.006) and directly related to WMH as percent of ICV (Log_e transformed) (β±SE:+0.026±0.007, p<0.001). Both total and hospital-treated infections were associated with poor WMI, while the latter was inversely related to FA within the lowest LE8 tertile (β±SE:−0.0011±0.0003, p<0.001, P_LE8×IB<0.05), a pattern detected for GM, Right Frontal GM, left accumbens and left hippocampus volumes. Within the uppermost LE8 tertile, total infection burden was linked to smaller right amygdala while being associated with larger left frontal GM and right putamen volumes, in the overall sample. Within that uppermost tertile of LE8, caudate volumes were also positively associated with hospital-treated infections.

Conclusions.

Hospital-treated infections had more consistent deleterious effects on volumetric and white matter integrity brain neuroimaging outcomes compared with total infectious burden, particularly in poorer cardiovascular health groups. Further studies are needed in comparable populations, including longitudinal studies with multiple repeats on neuroimaging markers.

INTRODUCTION

Dementia is characterized by a marked loss in cognitive abilities resulting in dependence across activities of daily living in formerly healthy adults (Huang et al., 2014). In the developed world, Alzheimer’s Disease (AD), the most common sub-type of dementia, is also the leading cause of old age disablement (Helmer et al., 2006) carrying the greatest health care burden ($236 billion in long-term care and hospice care cost in the US in 2016 ascribed to dementia)(Alzheimer’s, 2016; Honjo et al., 2009). Given the lack of effective treatment, prevention of both AD and all-cause dementia is crucial, resulting in the need to further uncover modifiable risk factors. Late-onset AD is known to have a strong genetic heritability, with the leading genetic risk factor being ε4 allele of APOE. The 2020 Lancet commissions concluded that approximately 40% of dementia population attributable risk is explained by factors such as early-life education, mid-life hearing loss, traumatic brain injury, hypertension, alcohol use and obesity, as well as later life smoking, depression, social isolation, physical inactivity, air pollution and diabetes (Livingston et al., 2020). As such, identification of novel mid-life risk factors is crucial for planning cost-effective interventions.

Whether infectious agents, both CNS and systemic, are involved in AD has gained the attention of researchers over the past few years (Honjo et al., 2009). For over a century, infections have been shown to trigger central nervous system (CNS) disorders. Some of these infections, including herpes simplex virus type 1 (HSV-1), Helicobacter pylori (H. pylori), Chlamydophila pneumoniae, and Borrelia burgdorferi have been shown to reduce neurocognitive performance over time in humans and experimental animal models, and were additionally implicated in AD etiology among humans (Beydoun et al., 2018; Beydoun et al., 2021; Doulberis et al., 2018; Letenneur et al., 2008; Mawanda and Wallace, 2013; Mawanda et al., 2016). These pathogens could induce AD directly through CNS infection or indirectly via systemic inflammation affecting the brain. An alternative hypothesis was that infectious pathogens can trigger autoimmunity responses that could target the brain, causing neuro-inflammatory processes leading to AD (Mawanda and Wallace, 2013). To date, no specific pathogen has been conclusively related to late-onset AD among humans, and a polymicrobial hypothesis has been proposed instead (Mawanda and Wallace, 2013).

Moreover, the American Heart Association (AHA) measure of optimal cardiovascular health (CVH), known as Life’s simple 7 (LS7) was recently linked with reduced risk of all-cause dementia (Malik et al., 2021) and was causally related to early brain MRI markers of dementia (Acosta et al., 2022). An updated version of LS7, known as Life’s Essential 8 (LE8) included sleep health as a component, with modification to the original 7 components, namely obesity, hypertension, dyslipidemia, diabetes, smoking, diet and physical activity. Both LS7 and LE8 followed an algorithm whereby a higher total score reflected better CVH.

A few biological components included in LE8, such as diabetes and hypertension, might threaten blood-brain barrier integrity, thereby allowing neurotoxin and possibly infectious agents’ entry into the brain (Chow and Gu, 2015). Similarly, some bacterial infections too have been associated with damage to the blood-brain barrier (Chow and Gu, 2015). Moreover, a compromised blood-brain barrier has been linked to increased dementia risk (Janelidze et al., 2017). Further, certain infectious agents have been associated with early cardiovascular disease (Mendy et al., 2013). Brain magnetic resonance imaging (MRI) offers an opportunity to measure in vivo brain structural pathologies and vascular brain injuries that often accompany age-related cognitive decline and are considered part of the dementia phenome. Recent evidence points to an association between various infections and brain structural and diffusion MRI markers of the dementia phenome (Albaret et al., 2020; Andreou et al., 2021; Andreou et al., 2022; Boska et al., 2014; Chougar et al., 2020; Duggan et al., 2022; Horacek et al., 2012; Kalpana et al., 2011; Kandemirli et al., 2020; Kumar et al., 2017; O’Connor et al., 2018; Park et al., 2021; Peters van Ton et al., 2022; Reichardt et al., 2022; Sanford et al., 2018; Sugden et al., 2016; Walker et al., 2018; Zilli et al., 2021). This suggests a potential for interaction between infection burden and cardiovascular health in determining early brain MRI markers of dementia.

Whether hospital-treated and total infection burdens are involved in the dementia phenome, particularly in association with brain MRI markers of dementia is still largely unknown. Moreover, whether such associations differ across cardiovascular health groups, as measured by LE8, remains uncertain. This study examines the association of total and hospital-treated infection burden with several dementia phenome-related brain MRI volumetric and white matter integrity measures among non-demented older adults, across strata of cardiovascular health as measured by the LE8, using a large sample from the UK Biobank.

MATERIALS AND METHODS

Database

The UK Biobank is a prospective study of about 500,000 adults initially aged 37–73 years old residing in the UK who were recruited between 2006 and 2010 (Mutz et al., 2021; UK Biobank, 2007). Study rationale, design and protocol measures are detailed elsewhere (Bycroft et al., 2018; UK Biobank, 2007). Briefly, participants were examined in centers located within England, Scotland and Wales. The exam included self-administered, and interviewer administered questionnaires, clinical measures, performance measures, bio-imaging, and obtaining bio-specimens. (Ho et al., 2022; UK Biobank, 2007). The study was approved by the North West Multi-Centre Research Ethics Committee. Participants provided written informed consent for data collection, data analysis, and record linkage, provided that data were de-identified (Foster et al., 2018; Mutz et al., 2021; UK Biobank, 2007). The present study is part of the UK Biobank application #77963, was approved by the National Institutes of Health Institutional Review Board.

Study Sample

Of the initial 502,389 UK Biobank participants, we excluded participants who had incomplete key sMRI and dMRI outcomes, including intracranial volume (ICV), subcortical structural volumes such as hippocampal volumes, Mean FA, Mean MD, total brain volume (TBV), GM, and WM volumes, and WMH phenotype (as percentage of ICV, Log_e transformed). All volumes were measured in mm³. Moreover, we also excluded participants who had no available data on LE8 total and/or sub-scores and/or no available AD polygenic risk score (PRS) scores. We also excluded individuals with prevalent dementia at baseline (N=3, supplementary method 1). No additional exclusions were made based on exposure availability. All other socio-economic and socio-demographic covariates were imputed using chain equations, with 5 imputations and 10 iterations, given that their missingness rate was <10% each. Thus, the final analytic sample consisted of 38,803 participants aged 40–70 years at the baseline assessment visit (47% men), (Figure 1).

FIGURE 1. Participant Flowchart.

Abbreviation: AD=Alzheimer’s Disease; LE8=Life’s Essential 8; dMRI=Diffusion-weighted Magnetic Resonance Imaging; PRS=Polygenic Risk Score; sMRI=Structural Magnetic Resonance Imaging; UK=United Kingdom.

Brain MRI acquisition and processing

Brain MRI was obtained on a sub-sample of ~45k participants in 3 MRI centers located throughout the study area (de Groot et al., 2013; Navale et al., 2022). Supplementary Method 1 provides more details regarding IDP measures and processing (de Groot et al., 2013). All brain MRI data were acquired on three 3T Siemens Skyra scanners, according to a freely available protocol (http://www.fmrib.ox.ac.uk/ukbiobank/protocol/V4_23092014.pdf), documentation (http://biobank.ctsu.ox.ac.uk/crystal/docs/brain_mri.pdf), and publication. (Alfaro-Almagro et al., 2018; Cox et al., 2019). Scans from the top of the head to the neck were conducted using a 256-cm superior– inferior field of view (de Groot et al., 2013; Navale et al., 2022). The global tissue volumes, and white matter tract-averaged water molecular diffusivity indices were processed by the UK Biobank team and made available to approved researchers as imaging-derived phenotypes (IDPs); the full details of the image processing and QC pipeline are available elsewhere. (Alfaro-Almagro et al., 2018; Cox et al., 2019)

Volumetric outcomes were labelled sMRI outcomes, whereas FA/MD phenotypes were labelled dMRI outcomes. The selected imaging phenotypes were a priori shown to be associated with worse cognitive ability and decline, namely for sMRI those were total white and grey matter volumes (WM, GM, respectively), bilateral Frontal GM volume (e.g. (Tank et al., 2021)), sub-cortical volumes, including hippocampal volume (See Supplementary Table 1 for details) and Log_e transformed white matter hyperintensity (WMH), expressed as % of ICV. For dMRI, average FA and MD across white matter tracts listed in supplementary Table 1 were considered as key outcomes of interest. ICV was also used as a potential confounder in the case of subcortical volume outcomes. Tract-specific FA and MD were also considered among secondary outcomes of interest.

Infection burden

Linked hospital admission records through the hospital electronic system (HES) were used to identify a primary or secondary diagnosis of infection, as were other sources (e.g., primary care, self-report, death certificate). ICD-10 codes utilized to identify central nervous system (CNS) infections includedA17,A80-A81, A85-A89, B00.3-B00.4, B01.0-B01.1, B02.0-B02.2, B05.0-B05.1, B06.0, B2.61-B26.2, G00-G01, G02.0, G03, G04.2, G05.0-G05.1,(e.g., meningitis, viral encephalitis); those for gastrointestinal infections were in the following ranges: A00-A05, A08 (e.g., salmonella, shigellosis). Moreover, liver infections, including hepatitis A, covered ICD-10 codes of B15-B19, whereas respiratory infections (e.g., pneumonia, laryngitis) covered A15-A16, A36-A38, J00-J06, J09-J18 and J20-J22 ICD-10 codes. We used other ICD-10 codes for sepsis, namely A40-A41 (e.g., streptococcal sepsis), for skin infections: A46, B00-B09, L00-L05, L08 (e.g., cellulitis, measles), for urogenital infections: N30.0, N39.0, N41.0–41.1,N71-N72 (e.g., cystitis, prostatitis), and for other infections, including bone infections and mastitis, the ICD-10 code list was: A18-A19, A31-A32, A39, A42-A44, A48-A49, B25-B27, B30, B33-B34, B95-B98, H62.0-H62.1, H67.0-H67.1, M00, M01.0-M01.5, N61 (Andersson et al., 2016; Ronaldson et al., 2022), (see supplementary Tables 5 and 6 for details). The count of infection types was used to reflect infection burden (IB), as long as the date of occurrence was prior the start date of baseline assessment. The source was further specified as “all sources” (IB_total) vs. “hospital-treated” (IB_hosp) and additional analysis was conducted with the “hospital-treated infection burden” as the main exposure of interest. For most analyses, Tertiles of IB_total and IB_hosp were used for ease of interpretation.

Life’s Essential 8

In 2010, the AHA widened its scope of interest by prioritizing wellness over illness, by defining a new measure of cardiovascular health (CVH) aiming at individual and population-level health promotion.(Hayman and Martyn-Nemeth, 2022; Lloyd-Jones et al., 2010) CVH was initially operationalized with 7 potentially modifiable biological and lifestyle factors, that, when at optimal levels would result in greater cardiovascular disease (CVD)–free survival, longevity, and better quality of life. This measure of CVH was labelled “Life’s Simple 7” (LS7), with its 7 components including indicators of diet quality, greater physical activity, reduced cigarette smoking, lower body mass index, total cholesterol, fasting blood glucose, and optimal blood pressure levels. Using clinical thresholds that were accepted for both children and adults, each metric was categorized as poor (0), intermediate (1), or ideal (2). The overall summary score of LS7 could then range from 0 (all metrics at poor levels) to 14 (all 7 metrics at ideal levels)(Hayman and Martyn-Nemeth, 2022; Lloyd-Jones et al., 2010). Since the 2010 AHA statement was published, CVH was re-evaluated and an AHA Presidential Advisory proposed an updated and enhanced version of CVH, reflecting advances made over a decade’s worth of research, while LS7’s methodological limitation were remedied.(Hayman and Martyn-Nemeth, 2022; Lloyd-Jones et al., 2022). This new measure was labeled “Life’s Essential 8” (LE8), retaining all 7 components of LS7 with major modifications to definitions and scales (described below). Sleep health due to its now known influence on promoting CVH across the life span was included in LE8 (Hayman and Martyn-Nemeth, 2022; Lloyd-Jones et al., 2022). The detailed algorithm is provided in supplementary Table 4 and the dietary component is described in supplementary Tables 2 and 3.

Covariates

Socio-demographic variables included as potential confounders in all of our analyses included age, sex, race/ethnicity (recoded as: White, Black, South Asian and Others; and in part of the analysis as Non-White vs. White) and household size. Three separate measures of socio-economic status were utilized, namely educational attainment, household income and Townsend deprivation index (TDI). Baseline assessment used a touch-screen questionnaire to elicit information on educational attainment among others, which was regrouped on at least one previous study (Chadeau-Hyam et al., 2020), as: 0=Low, combining None, “CSEs/Equivalent”, “NVQ/HND/HNC/Equivalent” and “Other professional qual”; 1=Intermediate, combining “O Levels/GCSEs/Equivalent” and ”A/AS Levels Equivalent; 2=Higher level or “College/University.” Total household income before tax was categorized into 1= as less than £18,000, 2= £18,000–£29,999, 3= £30,000–£51,999, 4= £52,000–£100,000, and 5= as greater than £100,000. TDI scores were generated based on national census data reflecting residential postcode-level car ownership, household overcrowding, owner occupation, and unemployment. TDI score reflects higher socioeconomic deprivation with higher TDI scores (Townsend P, 1987). Time elapsed between baseline assessment visit and the neuroimaging visit was measured in days. AD PRS was also included among potential confounders and is detailed in supplementary methods 2. The intracranial volume (ICV) was included as a covariate when subcortical sMRI volumes were the main outcomes of interest. ICV was accounted for in the WMH measure.

Statistical methods

Analyses were completed using Stata 17.0 (StataCorp, College Station, TX). Missing data (for most covariates at less than 5%) was imputed using multivariate imputation by chained equations (Lee and Carlin, 2010) for socio-demographic and socio-economic covariates (5 imputations with 10 iterations). Univariate descriptive statistics comparing means and proportions of variables of interest were examined across LE8 tertiles using ordinary least square (OLS) linear and multinomial logit models with LE8 tertile was the only predictor. In the main analysis, we used conducted OLS multiple linear regression models to conduct sets of analyses to: examine the association of the imaging markers with infection burden exposures, overall and as stratified by tertiles of LE8, as well as the joint association of infection burden and LE8 tertiles with the imaging markers.

In all models, we adjusted for sample selectivity due to missing exposure and outcome data, relative to the initially recruited sample, using a two-stage Heckman selection strategy (Heckman, 1979). Initially, we predicted an indicator of selection bias with socio-demographic factors, namely, age, race/ethnicity and sex using probit regression, which yielded an inverse mills ratio (IMR) – a function of probability of being selected given those socio-demographic factors. Thus, OLS multiple linear regression models that testing key exposure-outcome relationships were adjusted for age, sex, race/ethnicity, household size, household income, education, TDI, AD PRS, time elapsed from baseline to neuroimaging visit (days), LE8 total score and the IMR in the total sample. In the stratified models, LE8 total score was excluded from the model. Interaction by LE8 tertile was formally tested in the unstratified sample by including a 2-way interaction term between LE8 tertile and the main exposure (IB tertile). ICV was only included as a covariates for models with preselected subcortical sMRI measures. For dMRI analyses, FA and MD global means were the main outcomes of interest. While total infection burden was the primary exposure of interest, “Hospital-treated infection burden” was another important exposure. Type I error was set at 0.05.

Further secondary analyses were also conducted to examine the association between IB tertile with tract-specific FA and MD (48 tracts for each of FA and MD), in models adjusted only for main socio-demographic variables that were complete in this sample, using unimputed data, namely age, sex, race/ethnicity, household size, AD PRS, time elapsed from baseline to neuroimaging visit (days) and the IMR. Tract-specific FA and MD were standardized z-score transformed, while the main exposure (IB_total) was expressed as a tertile entered as an ordinal variable (range: 1–3), which yielded an effect size estimate that was comparable to Cohen’s D. In this part of the analysis, type I error was corrected for multiple testing using Bonferroni correction (P_corr<0.05), while false discovery rates (q-values) were also estimated. Statistically significant findings based on P_corr<0.05 were plotted on a standard Montreal Neurological Institute (MNI) brain image using FSLeyes software (URL: https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FSLeyes), contrasting effect sizes for FA and MD tract-specific outcomes. Heat maps were also produced for tract-specific FA and MD summarizing all results for the OLS multiple linear regression models, focusing on the effect sizes and directions of the infection burden tertile, using R software version 4.2.2 (https://www.r-project.org/). Finally, we carried out an additional analysis whereby a series of age and sex-adjusted logistic regression models were conducted examining the odds of hospital-treated vs. non-hospital treated infection for infections that were common to both grouped sources. Odds ratios of having a hospital source for each infection UKB code (yes vs. no) are visualized as a volcano plot along with their associated p-values, using R software version 4.2.2 (https://www.r-project.org/). Prevalence of each type of infection within each of the two main sources was also estimated.

RESULTS

Table 1 displays the distribution of sample characteristics across LE8 tertiles. Overall, our sample consisted of individuals with mean age 55.5y, 53% of whom were female, 96.8% self-reported as White, and 49.5% had higher educational attainment, while 12% had incomes < £18,000. Mean LE8 total score in the overall sample was 530, with lifestyle and biological sub-scores around 263 and 268, respectively. Individuals in the uppermost tertile of LE8 were generally younger, with higher proportion female, and higher socio-economic status in terms of education, income and the TDI. Furthermore, ICV and TBV were both significantly larger for the lowest LE8 tertile compared with the highest, as was the total WM volume. In contrast, GM volumes (total, frontal left and frontal right), showed a positive association with LE8 tertile. With the possible exceptions of amygdala, caudate, and right putamen, there was a similar trend whereby the second and third tertiles of LE8 were characterized by significantly larger subcortical volumes compared with the first tertile of LE8. Nevertheless, a dose-response relationship was only uncovered for the accumbens (L and R) and the thalamus (L and R). Such dose-response relationship was also found for FA and MD, whereby LE8 tertile was positively and inversely related with these two WMI measures, respectively. While the total infection burden did not differ across LE8 tertiles, notably there were fewer hospital-treated infections in the higher LE8 tertile.

Table 1.

Study sample characteristics by Life’s Essential 8 (LE8) tertile: UK Biobank 2006–2021

		LE8 tertile
	Overall	T1	T2	T3	PLE8tert T2 vs. T1	PLE8tert T3 vs. T1
		120–485	486–566	567–800
	N=38,803	N=12,190	N=12,806	N=13,807
Socio-demographic
Baseline age, y	55.50±0.04	56.31±0.06	55.98±0.06	54.28±0.07	0.001	<0.001
Sex, % female	53.0	44.5	50.3	63.1	<0.001	<0.001
Race/ethnicity
White	96.8	96.6	96.8	97.1	__	__
Black	0.6	0.8	0.7	0.4	0.57	<0.001
South Asian	1.0	1.0	1.0	0.9	0.93	0.56
Other	1.5	1.5	1.4	1.6	0.40	0.64
Household size	2.55±0.01	2.45±0.01	2.54±0.01	2.64±0.01	<0.001	<0.001
Socio-economic status
Education
Low	15.6	18.8	15.6	12.8	<0.001	<0.001
Intermediate	34.9	38.1	34.6	32.4	<0.001	<0.001
High	49.5	43.2	49.8	54.9	__	__
Income
Less than £18,000	11.7	13.8	11.0	10.3	<0.001	<0.001
£18,000–£29,999	22.2	22.9	22.7	21.1	0.29	0.42
£30,000–£51,999	30.2	31.3	29.8	29.6	__	__
£52,000–£100,000	28.2	25.6	28.9	29.7	<0.001	<0.001
greater than £100,000	7.8	6.3	7.6	9.2	<0.001	<0.001
Townsend Deprivation Index	−1.88±0.01	−1.65±0.03	−1.96±0.02	−2.02±0.0	<0.001	<0.001
Life’s Essential 8 total score	530.5±0.5	424.5±0.5	527.6±0.2	626.8±0.4	<0.001	<0.001
Lifestyle sub-score	263.1±0.3	214.1±0.5	263.8±0.4	305.7±0.4	<0.001	<0.001
Biological sub-score	267.9±0.3	210.3±0.5	264.2±0.4	322.1±0.4	<0.001	<0.001
sMRI outcomes, mm3
ICV	1,548,596±781	1,554,823±1,384	1,553,167±1,390	1,538,859±1,280	0.39	<0.001
Total brain volume	1,159,394±565	1,160,201±1,016	1,162,163±996	1,156,113±930	0.16	0.003
Total GM	614,429±283	611,740±510	615,599±497	615,716±466	<0.001	<0.001
Total WM	544,966±312	548,461±561	546,564±549	540,397±512	0.015	<0.001
Frontal GM, Left Brain	75,878±41	75,605±73	75,964±70	76,039±68	<0.001	<0.001
Frontal GM, Right Brain	75,540±40	75,260±73	75,637±70	75,698±67	<0.001	<0.001
White matter hyperintensity, WMH, % ICV, Loge transformed	−1.629±0.005	−1.457±0.009	−1.601±0.009	−1.806±0.008	<0.001	<0.001
Subcortical volumes
Accumbens, Left	491.4±0.6	483.7±1.0	491.9±1.1	497.9±1.0	<0.001	<0.001
Accumbens, Right	385.4±0.6	377.5±1.0	383.6±1.0	394.1±0.9	<0.001	<0.001
Amygdala, Left	1,262.2±1.3	1,269.1±2.3	1,265.3±2.2	1,253.1±2.1	0.22	<0.001
Amygdala, Right	1,226.6±1.4	1,267.9±2.5	1,231.2±2.4	1,218.8±2.3	0.84	0.001
Caudate, Left	3,376.4±2.1	3,367.8±3.9	3,383.7±3.8	3,377±3.5	0.003	0.077
Caudate, Right	3,559.5±2.3	3,552.8±4.1	3,566.5±4.0	3,558.8±3.7	0.015	0.28
Hippocampus, Left	3,771.4±2.5	3,758.8±4.4	3,768.4±4.3	3,785.3±4.1	0.12	<0.001
Hippocampus, Right	3,886.7±2.5	3,872.6±4.5	3,894.7±4.4	3,892.7±4.2	0.001	0.001
Pallidum, Left	1,753.9±1.3	1,748.1±2.2	1,759.2±2.2	1,754.1±2.1	<0.001	0.050
Pallidum, Right	1,797.9±1.3	1,787±2.3	1,802.9±2.2	1,802.3±2.1	<0.001	<0.001
Putamen, Left	4,756.9±3.1	4,744±5.7	4,763.2±5.5	4,762.2±5.0	0.014	0.017
Putamen, Right	4,813.7±3.0	4,808±5.6	4,820.6±5.3	4,811.8±4.9	0.12	0.69
Thalamus, Left	7,740.0±3.9	7,707±7.0	7,746.9±6.8	7,762.2±6.5	<0.001	<0.001
Thalamus, Right	7,548.6±3.8	7516.8±6.8	7,558.2±6.6	7,567.9±6.3	<0.001	<0.001
dMRI outcomes
Mean FA	0.5612±0.0001	0.5590±0.0002	0.5611±0.0002	0.5633±0.0002	<0.001	<0.001
Mean MD	0.0008±1.64e-07	0.00080±3.03e-07	0.000794±2.88e-07	0.000791±2.60e-07	0.001	<0.001
Genetic risk, follow-up time
AD PRS	0.039±0.005	0.063±0.009	0.0428±0.009	0.0130±0.008	0.099	<0.001
T1: -3.42;-0.46	32.9	32.0	32.8	34.0	0.34	<0.001
T2 : -0.46; +0.31	33.3	33.5	33.0	33.4	0.78	0.086
T3 : +0.31;+4.98	33.7	34.5	34.3	32.6	__	__
Follow-up time, days	3,289.9±3.2	3,249.3±5.8	3,289.9±5.6	3,325.9±5.3	<0.001	<0.001
Infection burden, IB
Total	1.177±0.009	1.157±0.015	1.191±0.015	1.180±0.014	0.12	0.27
T1: 0	52.5	53.3	52.0	52.3	__	__
T2 : 1	18.0	17.8	18.5	17.6	0.070	0.17
T3 : 2–15	29.5	28.9	29.4	30.0	0.77	0.040
Hospital treated	0.232±0.005	0.2775±0.0089	0.2214±0.0077	0.2041±0.0071	<0.001	<0.001
T1: 0	90.2	88.5	90.6	91.4	__	__
T2 : n/a	0.0	0.0	0.0	0.0	…	…
T3 : 1–15	9.8	11.5	9.4	8.6	<0.001	<0.001

Abbreviations: AD=Alzheimer’s Disease; dMRI=Diffusion Magnetic Resonance Imaging; FA=Fractional Anisotropy; GM=Gray Matter; IB=Infection Burden; ICV=Intracranial Volume; LE8=Life’s Essential 8; MD=Mean Diffusivity; PRS=Polygenic Risk Score; sMRI=Structural Magnetic Resonance Imaging; T₁=First tertile; T2=Second tertile; T₃=Third tertile; UKB=UK Biobank; WM=White Matter; WMH=White Matter Hyperintensity.

Table 2 displays results from a series of multiple OLS linear regression models examining the association of total and hospital-treated infection tertile with sMRI volumetric outcomes, overall and across LE8 tertiles. Our results indicated that prevalent total infection burden tertile was linked to a smaller left accumbens in the lowest LE8 tertile (β=−3.5±1.12, p<0.010, P_{LE8tert×IBtert}=0.002 for T₂ vs. T₁). In the uppermost LE8 tertile, total IB tertile was linked to a smaller right amygdala (β=−6.9±2.5, p<0.010) with only a marginally significant interaction between LE8 tertile and IB tertile (P=0.057). In contrast and in the overall sample, the higher total infection burden tertile was associated with a larger frontal GM in the left brain (β=+78±39, p=0.045), a pattern also observed for the right putamen (β=5.91±2.7, p=0.028).

Table 2.

Infection burden (total and hospital-treated) and sMRI volumetric outcomes, overall and stratified by Life’s Essential 8 (LE8) tertiles: UK Biobank 2006–2021^a

	Overall		By LE8 tertile
	IB tertile, per unit increase	PIB	T1	T2	T3	PLE8tert×IBtert T2 vs. T1	PLE8tert×IBtert T3 vs. T1
			120–485	486–566	567–800
	N=38,803		N=12,190	N=12,806	N=13,807
Total Infection burden
Total brain volume	+270±506	0.59	+945±915	+1,024±886	−969±836	0.98	0.076
Total GM	+176±257	0.49	+439±464	+451±499	−383±422	0.92	0.12
Total WM	+93±285	0.75	+506±516	+573±449	−581±469	0.89	0.079
Frontal GM, Left Brain	+78±39	0.045	+97±71	+144±68	+8±65	0.69	0.26
Frontal GM, Right Brain	+71±39	0.068	+81±70	+118±67	+25±65	0.79	0.42
White matter hyperintensity, WMH, % ICV, Loge transformed	+0.009±0.005	0.082	+0.016±0.009	+0.003±0.009	+0.008±0.008	0.25	0.46
Subcortical volumes
Accumbens, Left	−0.928±0.620	0.14	−3.55±1.12 **	+1.22±1.08	−0.592±1.028	0.002	0.067
Accumbens, Right	−0.465±0.571	0.42	−1.33±1.02	+0.02±0.99	−0.051±0.956	0.38	0.49
Amygdala, Left	+0.246±1.340	0.85	+1.26±2.43	−0.61±2.34	+0.34±2.20	0.50	0.81
Amygdala, Right	−3.250±1.508	0.031	+0.52±2.72	−2.77±2.64	−6.99±2.49**	0.39	0.057
Caudate, Left	+3.30±2.04	0.11	−1.31±3.70	+5.22±3.61	+5.80±3.33	0.22	0.15
Caudate, Right	+2.87±2.17	0.19	−2.61±3.95	+4.58± 3.81	+6.29 ±3.56	0.20	0.087
Hippocampus, Left	+3.18±2.54	0.21	+8.84±4.55	+2.93±4.45	−1.10±4.22	0.34	0.11
Hippocampus, Right	+4.07±2.58	0.11	+5.57±4.59	+8.04±4.53	−0.47±4.29	0.73	0.35
Pallidum, Left	+2.33±1.27	0.067	+4.55±2.28 *	+2.21±2.24	+0.71±2.09	0.49	0.24
Pallidum, Right	+0.70±1.26	0.58	+0.45±2.28	+0.98±2.21	+0.89±2.09	0.87	0.90
Putamen, Left	+3.91±2.77	0.16	+2.94±5.10	+4.38±4.86	+4.88±4.48	0.90	0.80
Putamen, Right	+5.91±2.69	0.028	+4.95±4.96	+9.96±4.71*	+3.40±4.35	0.44	0.85
Thalamus, Left	+1.93±3.18	0.54	+4.68±5.71	+0.46±5.52	+0.84±5.27	0.62	0.57
Thalamus, Right	+1.62±2.99	0.59	+2.43±5.36	+0.17±5.23	+2.10±4.98	0.75	0.92
Hospital-treated infection burden
Total brain volume	−1,157±748	0.12	−2,402±1,253	+1,073± 1,326	−2,001± 1,313	0.048	0.76
Total GM	−1,041±379	0.006	−1,811±636 **	+243± 673	−1,460±663 *	0.019	0.63
Total WM	−116±421	0.74	−591±707	+830±746	−540±737	0.16	0.91
Frontal GM, Left Brain	−87±58	0.13	−177±97	+114 ±101	−188 ±103	0.036	0.96
Frontal GM, Right Brain	−98±58	0.087	−198±96 *	+108±101	−196± 102	0.026	0.97
White matter hyperintensity, WMH, % ICV, Loge transformed	+0.026±0.007	<0.001	+0.043±0.012 ***	+0.022±0.013	+0.014±0.013	0.24	0.12
Subcortical volumes
Accumbens, Left	−2.040±0.915	0.026	−4.583 ±1.535 **	−0.261±1.619	−1.107±1.615	0.044	0.10
Accumbens, Right	−2.134±0.843	0.011	−3.283±1.400 *	−1.553±1.488	−1.459±1.501	0.40	0.39
Amygdala, Left	−1.325±1.978	0.50	−4.065±3.333	3.680±3.507	−3.253±3.462	0.10	0.83
Amygdala, Right	−2.086±2.226	0.35	+0.723±3.717	−1.422±3.957	−6.043±3.917	0.74	0.21
Caudate, Left	+1.871±3.015	0.54	−7.499 ±5.062	+0.733±5.408	+13.947±5.230 **	0.23	0.002
Caudate, Right	+6.023±3.207	0.060	−3.880 ±5.409	+7.289±5.700	+16.411±5.588 **	0.13	0.007
Hippocampus, Left	−5.649±3.749	0.13	−12.616±6.231 *	+2.629±6.658	−5.769±6.628	0.078	0.38
Hippocampus, Right	−3.416±3.803	0.37	−0.549 ± 6.278	−3.687±6.779	−5.463±6.739	0.85	0.78
Pallidum, Left	−2.440±1.875	0.19	−2.414 ± 3.126	−3.951±3.358	−0.624±3.281	0.80	0.66
Pallidum, Right	−3.004±1.866	0.11	−5.925±3.114	−2.031±3.315	−0.314±3.289	0.37	0.21
Putamen, Left	−0.787±4.091	0.85	−3.497± 6.987	−2.336±7.278	+4.364±7.035	0.87	0.35
Putamen, Right	−0.547± 3.973	0.89	−6.980± 6.796	6.936±7.047	−0.293±6.840	0.14	0.40
Thalamus, Left	−3.343±4.675	0.48	−14.548±7.809	11.397±8.254	−5.366 ± 8.276	0.021	0.44
Thalamus, Right	−6.138±4.416	0.17	−12.013±7.339	2.919±7.821	−8.099±7.827	0.16	0.74

Abbreviations: AD=Alzheimer’s Disease; dMRI=Diffusion Magnetic Resonance Imaging; FA=Fractional Anisotropy; GM=Gray Matter; IB=Infection Burden; ICV=Intracranial Volume; LE8=Life’s Essential 8; MD=Mean Diffusivity; PRS=Polygenic Risk Score; sMRI=Structural Magnetic Resonance Imaging; T₁=First tertile; T2=Second tertile; T₃=Third tertile; UKB=UK Biobank; WM=White Matter; WMH=White Matter Hyperintensity.

^*P<0.05;

^**P<0.010;

^***P<0.001

^aModels are adjusted for age, sex, race/ethnicity, household size, household income, education, TDI, AD PRS, time elapsed from baseline to neuroimaging visit (days), LE8 total score (for unstratified models) and the IMR. ICV was adjusted for in models with subcortical volume outcomes. To test heterogeneity of the infection burden effect on each outcome, a 2-way interaction was added between LE8 and IB tertiles in the unstratified model.

As also shown in Table 2, when prevalent hospital-treated infection burden was considered as the exposure, individuals with at least one infection of this type were characterized by smaller total GM volume overall (β=−1,041±379, p=0.006) and in the lowest and uppermost LE8 tertiles (T₁: β=−1811±636, p<0.010; T₃=−1460±663, p<0.05, P_{LE8tert×IBtert}=0.019 for T₂ vs. T₁) and larger WMH volume as percent of ICV in the overall sample (β=0.026±0.007, p<0.001) and within the lowest LE8 tertile (β=0.043±0.012, p<0.05; p>0.05 for _{LE8tert×IBtert} for all tertile contrasts), compared to individuals not treated for an infection at a hospital prior to baseline assessment. Hospital-treated infection was also linked with a smaller right frontal GM volume within the lower LE8 tertile (β=−198±1.0, p<0.05, P_{LE8tert×IBtert}=0.026 for T₂ vs. T₁). Despite no association overall and only a trend towards heterogeneity across LE8 tertiles (P_{LE8tert×IBtert}=0.078 for T₂ vs. T₁), hospital-treated infections were inversely related to the left hippocampal volume within the lowest LE8 tertile (β=−12.62±6.23, p<0.05) LE8 tertile heterogeneity was also noted for sub-cortical structures in relation to hospital-treated infections. Most notably, an inverse relationship between hospital-treated infections and left accumbens volume was found overall (β=−2.04±0.92, p=0.026), but was mostly detected in the lowest LE8 tertile (β=−4.58±1.53, p<0.010, P_{LE8tert×IBtert}=0.044 for T₂ vs. T₁), consistently with the total infection burden. Another notable finding is a positive association of hospital-treated infection with caudate volumes (L/R) in the uppermost LE8 tertile (P<0.010), with significant heterogeneity between T₃ and T₁ (P<0.010).

Table 3 similarly shows results of multiple OLS linear regression models, focusing on WMI outcomes in relation to infection burden exposures, overall and across LE8 tertiles. Total and hospital-treated infections were both associated with poorer WMI, in terms of lower global mean FA and higher global mean MD. Nevertheless, upon stratification according to LE8 tertiles, notable patterns emerged. Total infection burden was associated with lower FA across all 3 tertiles with no detectable heterogeneity. In contrast, the association between total infection burden and MD was markedly attenuated with LE8 stratification. More importantly, hospital-treated infections were associated with reduced mean FA within the lowest two LE8 tertiles, with a dose-response relationship in the association (T₁: −0.0011±0.0003, p<0.001 ; T₂: −0.00073±0.00029, p<0.05 ; T₃: −0.00034± 0.00027, p>0.05) and significant heterogeneity detected between T₃ and T₁ (P=0.032). A similar pattern was observed for the positive association of hospital-treated infection with MD, although no significant heterogeneity was detected.

Table 3.

Infection burden (total and hospital-treated) and dMRI white matter integrity outcomes, overall and by Life’s Essential 8 (LE8) tertile: UK Biobank 2006–2021^a

	Overall		By LE8 tertile			PLE8tert×IBtert T2 vs. T1	PLE8tert×IBtert T3 vs. T1
	IB tertile, per unit	PIB	T1	T2	T3
	N=38,803		N=12,190	N=12,806	N=13,807
Total infection burden
Mean FA	−0.000751±0.000108	<0.001	−0.00082±0.00020 ***	−0.00077±0.00019 ***	−0.00068± 0.00017 ***	0.79	0.59
Mean MD	5.06e-07±1.65e-07	0.002	4.97e-07±3.10e-07	5.39e-07±2.92e-07	4.55e-07± 2.61e-07	0.38	0.56
Hospital treated infection burden
Mean FA	−0.000744±0.000159	<0.001	−0.00114±0.000270 ***	−0.00073±0 .000290 *	−0.00034± 0.000271	0.25	0.032
Mean MD	8.18e-07±2.44e-07	0.001	1.12e-06±4.24e-07 **	8.85e-07±4.37e-07 *	4.03e-07± 4.10e-07	0.62	0.17

Abbreviations: AD=Alzheimer’s Disease; dMRI=Diffusion Magnetic Resonance Imaging; FA=Fractional Anisotropy; GM=Gray Matter; IB=Infection Burden; ICV=Intracranial Volume; LE8=Life’s Essential 8; MD=Mean Diffusivity; PRS=Polygenic Risk Score; sMRI=Structural Magnetic Resonance Imaging; T₁=First tertile; T2=Second tertile; T₃=Third tertile; UKB=UK Biobank; WM=White Matter; WMH=White Matter Hyperintensity.

^*P<0.05;

^**P<0.010;

^***P<0.001

^aModels are adjusted for age, sex, race/ethnicity, household size, household income, education, TDI, AD PRS, time elapsed from baseline to neuroimaging visit (days), LE8 total score (for unstratified models) and the IMR. To test heterogeneity of the infection burden effect on each outcome, a 2-way interaction was added between LE8 and IB tertiles in the unstratified model.

Tract-specific standardized z-scores of FA and MD were also examined in relation to total infection burden tertiles, with each tertile approximating 1 SD increase in infection burden. While effect sizes generally indicated a weak association (|b|<0.10), certain regions were more sensitive to the total infection burden, including the posterior and anterior limbs of the internal capsule (ALIC and PLIC, respectively) and superior cerebellar peduncle or SCP (|b|≥0.04 up to 0.06). These findings are shown in Figure 2 plotted on standard brain images for tracts that survived multiple testing (p_corr<0.05 with |b|≥0.03) and all tracts with p<0.05 (heat maps). Detailed results are presented in Supplementary Datasheets 1 (for heatmap) and 2 (for brain images).

FIGURE 2. Tract-specific FA and MD in relation to total infection burden tertiles.

Abbreviations: FA=Fractional Anisotropy; MD=Mean Diffusivity. See Supplementary Table 1 for tract abbreviations.

Note: Based on a series of linear regression models adjusted for age, sex, race (Non-White vs. White), AD PRS, time elapsed from baseline to neuroimaging visit (days), household size and the inverse mills ratio. Tract-specific FA and MD are standardized z-scores, while the total infection burden tertile approximates 1 SD in the exposure. Effect sizes are plotted on a heat map highlighting strength of the association (D), while standard MNI brain images display effect sizes on the brain regions/tracts only when P_corr<0.05. Light blue (or yellow) color is for effect sizes in absolute values ≥0.04 and dark blue (or red) are for effect sizes in absolute value ≥0.03 but <0.04. Those are bolded rows in supplementary Datasheet 2 [baydounm/UKB4-supplementary-data: Supplementary data for UKB manuscript (github.com)] (B) is for fractional anisotropy after correcting for multiple testing: ALIC (L/R), CP (R), CH (L), EC (L/R), FCST (L), GCC, ML (R), MCP, PC, PCR (L/R), RPIC (L/R), SCC, SCP (L/R), SCR (L/R), UNC (L) (C) for mean diffusivity after correcting for multiple testing: ALIC (L/R), CP (R), CCG (L/R), GCC, ML (R), PLIC (L/R) (A) displays the entire JHU FA skeleton with different colors with the significant tracts (red/yellow and light blue/dark blue) based on (B) and (C) being annotated with appropriate labels where possible.

Among the most prevalent infections for both sources, there was a significantly higher prevalence of urinary tract infections and other genitourinary disorders among hospital-treated infections as compared to non-hospital treated infections (19% vs. 9%, age and sex-adjusted OR=2.48, p<0.001), as was the case for cellulitis and “other bacterial infections” with prevalence rates of 6.2% and 4.8%, respectively in the hospital-treated group (vs. 3.9% and 0% in the non-hospital treated group) (Figure 3 and supplementary Datasheet 3). In contrast, infections that were found to be more prevalent in the non-hospital treated group included those often prevented by vaccination, such as measles, mumps, pertussis, and most notably chicken pox (varicella). All supplementary datasheets, detailed code and related result datasets will be provided as a GitHub repository [baydounm/UKB4-supplementary-data: Supplementary data for UKB manuscript (github.com)].

FIGURE 3. Volcano plot of odds of hospital-treated infections vs. non-hospital-treated infections by type of infection.

Abbreviations: Ln=Log_e; OR=Odds Ratio. Note: Based on a series of age and sex-adjusted logistic regression models, with main predictor being the type of prevalent infection (1=yes, 0=no) and the outcome being hospital-treated vs. non-hospital-treated infection (1=yes, 0-no). The y-axis is the predictor’s associated p-value on a -Log10 scale and the X-axis is the Log_e(odds ratio) from the age and sex-adjusted logistic models for each predictor. A twofold increase or decreased odds of the infection being hospital-treated is marked by red and blue colors, respectively. An estimate with a p-value<0.05 is marked by the UKB field number (See UKB showcase URL: https://biobank.ndph.ox.ac.uk/showcase/), while a p-value<0.00065 (76 estimates) is marked with a dashed line. All infections occurred prior to baseline assessment.

DISCUSSION

Summary of findings

This is to our knowledge the first study to systematically examine whether total and hospital-treated infection burdens interact with overall cardiovascular health in their associations with brain MRI markers of the AD phenome in a large sample of middle-aged and older adults. Using UK Biobank data from 38,803 adults aged 40–70 years at baseline assessment and followed up to 2021 with neuroimaging data available 5 to 15 years after baseline (mean: approximately 9 years), we tested associations of total and hospital-treated infection burden with brain MRI measures reflecting in part the AD phenome, while stratifying by cardiovascular health. Among key findings, hospital-treated infections were inversely related to GM (β±SE: −1042±379, p=0.006) and directly related to WMH as percent of ICV (Log_e transformed) (β±SE:+0.026±0.007, p<0.001). Both exposures were associated with poor WMI, with hospital-treated infections inversely related to FA within the lowest LE8 tertile (β±SE:−0.0011±0.0003, p<0.001, P_LE8×IB<0.05), a pattern detected for GM and Right Frontal GM, left accumbens, and left hippocampus. Caudate volumes were positively associated with hospital-treated infections in the uppermost LE8 tertile. Total infectious burden was linked to smaller right amygdala within the highest LE8 tertile, whereas total infectious burden tertile was associated with larger left frontal GM and right putamen. Together, these findings suggest that infectious diseases may affect brain structure in regions that have been associated with Alzheimer’s disease and further that cardiovascular risk might moderate these associations.

Previous studies

Prior research has indicated that infections of various types, whether viral, bacterial, or parasitic may be involved in the etiology of dementia, including AD and vascular dementia (Aiello et al., 2006; Barnes et al., 2015; Gajewski et al., 2014; Miklossy, 2011; Readhead et al., 2018). Previous findings also have implicated some infectious diseases in structural brain changes. Using data on 40 Hepatitis C virus (HCV) seropositive cases and 31 matched controls, along with diffusion tensor tractography (DTT) metrics, a study indicated that HCV positive individuals demonstrated structural brain abnormalities and neurocognitive dysfunction coupled with changes in cell component and extracellular space in the WM regions of brain in asymptomatic HCV infection (Kumar et al., 2017). Similarly, among apolipoprotein E4 (APOE4) carriers but not among non APOE4 carries, herpes simplex virus was associated with decreased hippocampal volume and decreased FA and increased MD in the cingulum (Linard et al., 2021). In a study, based on a meta-analysis of 19 cross-sectional studies, standardized mean differences related to HIV serostatus were estimated at −0.65 (P = .002) for TBV, −0.28 for GM (P = .008), −0.24 (P = .076) for WM, and 0.56 (P = .001) for CSF volume (O’Connor et al., 2018). This stronger association with GM compared with WM is similar to our present study findings. Among 1,689 older adults participating in the Atherosclerosis Risk In Communities (ARIC) prospective cohort study, (mean age at MRI 76±5), 72% were hospitalized, 14% had a major infection, and 4% had a critical illness during the surveillance period (Walker et al., 2018). The study concluded that while all-cause hospitalization was primarily associated with lower white matter integrity using measures such as FA and MD, critical illness and major infection were associated with smaller brain volume, particularly within regions implicated in AD, including a smaller AD signature region (−1.28 cm(3), 95% CI=−2.21 to −0.35) (Walker et al., 2018). This is also in line with our current results with more comparable exposures and outcomes, although our study also found that total and hospital-treated infections were additionally associated with poorer WMI. Herpesviruses, including cytomegalovirus, along with SARS-Cov2 were among pathogens to receive additional recent attention in relation to the AD phenome, although the results were largely inconsistent (Andreou et al., 2021; Andreou et al., 2022; Chougar et al., 2020; Duggan et al., 2022; Kandemirli et al., 2020; Zilli et al., 2021). Among bacterial pathogens, H. pylori was at the forefront as a potential causal agent in AD along with periodontal pathogens such as Porphyromonas gingivalis, with different pathogens potentially interacting with each other in their associations with AD risk (Albaret et al., 2020; Beydoun et al., 2018; Beydoun et al., 2021). Later evidence suggested that in a sample of 822 men who underwent 3-Tesla brain MRI and had cross-sectional data on H. pylori infection using histological assessment, H. pylori infection was associated with overall (p = 0.022), parietal (p=0.008), and occipital (p=0.050) brain cortical thinning, even after adjustment for age, education, alcohol, smoking, and intracranial volume (Park et al., 2021). Using 3-dimentional topographical analysis, the study showed that H. pylori- infected men exhibited cortical thinning in several other smaller areas (false discovery rate corrected, Q < 0.050) (Park et al., 2021), even after further adjustment for an inflammatory marker (C-reactive protein) and several metabolic characteristics (obesity, dyslipidemia, fasting glucose, and blood pressure) (Park et al., 2021).While we are not aware any previous studies investigating interactions between cardiovascular risk, infection, and brain structure, a previous study examined associations between cardiovascular risk, an index of infection burden, and cognitive function. Although not investigating dementia or brain structure or function directly, the study found that in young and middle-aged adults, an index of infection burden interacted with an index of cardiovascular risk to predict worse cognitive function (Hedges et al., 2019), a finding broadly consistent with the results of the present study.

Biological mechanisms

Even though no single infection has been linked exclusively to the causation of AD, a number of infectious pathogens have been detected in brain of AD patients. This may have several interpretations. First, various neurotropic infectious agents may have a causal or promoting role in AD, depending on exposure circumstances coupled with host factors, including genetic predisposition (Mawanda and Wallace, 2013). Second, viral or bacterial pathogens may be found in AD brain because patients are more susceptible to infections. Third, infection could trigger neuroinflammation resulting in increased AD pathology (i.e., Aβ) (Komaroff, 2020). Fourth, a more consequential hypothesis is that those particular infections or groups of infections are directly playing a causal role in AD pathogenesis(Mawanda and Wallace, 2013; Readhead et al., 2018). This causal role in AD pathogenesis was recently explained by a hypothetical model whereby systemic and bacterial amyloids and other pathogen-associated molecular patterns (PAMPs) are capable of accelerating AD brain pathology (Ganz et al., 2022). The model suggests that Aβ’s triggering of neuroinflammation could render the brain more visible to the systemic milieu (Ganz et al., 2022). Nevertheless, Aβ may be insufficient by itself, but neurodegeneration then could be accelerated in the presence of PAMPs, which become more neurotoxic as a result of this enhanced visibility (Ganz et al., 2022). This then would trigger neurodegeneration (Ganz et al., 2022). Different individual infectious pathogens have unique or shared adverse effects on brain or neurocognitive function. In this regard, interactions between H. pylori seropositivity and concentration of the folate-cycle factor 5-methyltetrahydrofolate are associated with worsened cognitive function (Berrett et al., 2018). While it is unclear whether this interaction is associated with an increased risk of Alzheimer’s disease, it does suggest one possible mechanism by which an infectious disease could affect neurocognitive function. In addition, infectious pathogens can change the microbiome, and emerging evidence suggests that the gut microbiome can be associated with cognitive function (Davidson et al., 2018). Some infectious diseases can result in biochemical changes that could affect risk for neurodegeneration, such as the association between H. pylori and oxidative stress (Kao et al., 2006).

LE8 is a composite score of several lifestyle and biological factors that reflect optimal cardiovascular health with higher score. Nevertheless, its individual components as well as its sub-scales termed “biological LE8” and “lifestyle LE8” have been linked to both the AD phenome and increased susceptibility for infections. Thus, it was expected that the association between infections and AD-related brain markers would be modified by the LE8 score, with the expectation that the infection burden and low LE8 scores would act synergistically to increase AD risk, as mediated through brain MRI markers of AD. Such a finding would have the implication of cardiovascular health potentially reducing the impact of previous infections on AD risk. In addition, cardiovascular disease may damage the blood-brain barrier, potentially increasing entry of infectious pathogens into the brain (Lyon, 2017).

Strengths and limitations

The present study has several strengths. First, it is one of few and the largest study to date examining the association between hospital-treated infections and various brain MRI markers of the dementia phenome, reflecting both volumetric and white matter integrity outcomes. It is the also the first to test those associations across measures of CVH, with adequate power to conduct such stratified analysis. Second, the UK Biobank included a wide array of variables that could be controlled for allowing for unbiased estimates of exposure-outcome associations, by adjusting for potential confounders. Nevertheless, our study findings should be interpreted with caution in light of several limitations. First, although infection burden was measured prior to potential incidence of dementia, the dementia phenome was measured using sMRI and dMRI outcome variables that were measured only once ~5–15 years after baseline assessment (mean~9 years). Thus, the study has a cross-sectional design with a lag between exposure and outcome which was controlled for in the analysis. A repeat is being conducted but only on a small sub-set of participants, which will preclude a longitudinal analysis into the association between infection burden and change in brain volumes or white matter integrity with age. In addition, infection burden status (yes vs. no) as well as titers of infections are relatively stable over time, making our findings valid when it comes to exposure measurement error and time dependence. Second, residual confounding is still possible, even though we have included a large number of independent potential confounders in our models. Third, despite partial adjustment using a 2-stage Heckman selection model, selection bias cannot be ruled out. Finally, several findings labelled as statistically significant (e.g., IB_total vs. left frontal gray matter volume) may be due to chance, given the multiplicity of brain sMRI outcomes.

Conclusions

In sum, hospital-treated infections had more consistent deleterious effects on volumetric and white matter integrity brain neuroimaging outcomes compared with the total infection burden, particularly in poorer cardiovascular health population groups. Further studies are needed to further investigate associations between infectious diseases, brain structure, and dementia with attention to how cardiovascular risk might modify these associations, including longitudinal studies with multiple repeats on neuroimaging markers in comparable samples.

ABBREVIATIONS

Aβ

Amyloid beta

AD

Alzheimer’s Disease

AHA

American Heart Association

AMICO

Accelerated Microstructure Imaging via Convex Optimization

APOE

Apolipoprotein E gene

CAPI

computer-assisted personal interview

CNS

Central Nervous System

CSF

Cerebrospinal Fluid

CVD

cardiovascular disease

CVH

Cardiovascular Health

dMRI

Diffusion magnetic resonance imaging

DTI

Diffusion Tensor Imaging

DWI

Diffusion weighted imaging

FA

Fractional Anisotropy

FIRST

FMRIB’s Integrated Registration and Segmentation Tool

FLAIR

Fluid-attenuated inversion recovery

FoV

Field of View

GM

Grey Matter

GWAS

Genome-wide association studies

HDS

Healthy Diet Score

HES

Hospital Electronic System

HSV-1

herpes simplex virus type 1

IB

Infection burden

IB_total

Total infection burden

IB_hosp

Hospital-treated infection burden

ICD-10

International Classification of Diseases, 10^th revision

ICV

Intracranial Volume

IDP

Imaging Derived Phenotype

LE8

Life’s Essential 8

LS7

Life’s Simple 7

MD

Mean Diffusivity

MNI152

Montreal Neurological Institute template with 152 structural images

MPRAGE

Magnetization Prepared - RApid Gradient Echo

MRI

Magnetic Resonance Imaging

NODDI

neurite orientation dispersion and density imaging

PAMPs

pathogen-associated molecular patterns

PRS

Polygenic Risk Score

QC

Quality Control

sMRI

Structural magnetic resonance imaging

T1

T1 weighted images

T2

T2 weighted images

TBSS

Tract-Based Spatial Statistics

TDI

Townsend Deprivation Index

TR

Trace

UK

United Kingdom

WM

White Matter

WMH

White Matter Hyperintensity

WMI

White Matter Integrity

Data availability statement

These data are subject to the following licenses and restrictions: UK Biobank is a large-scale biomedical database and research resource containing in-depth genetic and health information from about 500,000 United Kingdom participants. The data are augmented regularly with additional records and are globally accessible to approved researchers undertaking vital research into the most common and life-threatening diseases. Requests to access these datasets should be directed to https://www.ukbiobank.ac.uk/.