HIV infection and cerebral small vessel disease are independently associated with brain atrophy and cognitive impairment

Abstract

Objective:

The objective of this study is to investigate whether cerebral small vessel disease (CSVD) is more common in virologically suppressed HIV-positive participants compared with HIV-negative controls and examine the potential synergistic effects of HIV and CSVD on brain structure and cognition.

Design:

Cross-sectional analysis of 119 treated, virologically suppressed HIV-positive and 55 HIV-negative participants. Forty-six HIV-positive and 30 HIV-negative participants had follow-up 2 years later. All participants underwent MRI and neuropsychological testing.

Methods:

Volume of white matter hyperintensities (WMH) was used as a surrogate measure of CSVD severity. Tensor-based morphometry and cortical modeling estimated brain volumes and cortical thickness, respectively. Rasch measurement theory was applied to neuropsychological test scores to estimate overall cognition. Linear models compared WMH loads, brain volumes, and cognition between groups; evaluated the association of WMH loads with brain volumes and cognition; and tested the interaction between HIV and WMH loads on brain volumes and cognition. Mixed-effects models compared the change in WMH loads between groups.

Results:

WMH loads and change in WMH loads were similar between the groups. HIV-positive participants had poorer cognition, thinner cortex and reduced subcortical volumes compared with HIV-negative controls. Higher WMH loads were associated with reduced cortical thickness and subcortical volumes and worse cognition, regardless of HIV serostatus. No significant interactions were observed between HIV and WMH loads with regards to brain volumes or cognition.

Conclusion:

These findings suggest that the contributions of HIV and CSVD on brain atrophy and cognitive impairment are independent but additive processes. This argues that optimizing vascular health may mitigate brain injury and cognitive decline, especially in treated, virologically suppressed HIV-positive individuals.

Introduction

Despite the introduction of suppressive combination antiretroviral therapy (cART), people living with HIV continue to experience HIV-associated neurocognitive disorders (HAND) at rates similar to the pre-cART era [1]. Although impairment is now less severe, HAND remains clinically relevant, affecting performance in everyday activities and reducing quality of life [1].

Longitudinal data have provided a clearer understanding of the causes of this impairment [2–4]. The cognitive deficits reflect, in part, brain injury that occurred soon after initial infection, before cART was started [2–4]. However, there are likely other contributing factors. Cerebral small vessel disease (CSVD) could be a key contributor to HAND [5–8]. CSVD is a term used to describe a group of pathological processes that affect the small vessels in the brain and is thought to be among the main causes of vascular cognitive impairment in the general population [9,10]. The consequences of CSVD on the brain are heterogeneous, including ischemic and hemorrhagic injury [10]. The sequela of the ischemic manifestations are visible as white matter lesions on brain imaging, and are easily detected through conventional T₂-weighted MRI as they appear hyperintense [10]. These hyperintense lesions are commonly referred to as white matter hyperintensities (WMH). The total volume of WMH is typically used as a proxy measure for overall CSVD severity; prior work has shown that WMH volume is strongly associated with cardiovascular risk factors, especially hypertension, diabetes, hyperlipidemia and smoking [10,11].

There has been indirect evidence that people living with HIV may have increased risk of CSVD [5,12,13]. However, whether CSVD is more common in HIV-positive compared with HIV-negative individuals remains unclear. A recent study reported that a group of HIV-positive individuals had a greater burden of WMH compared with controls [5], whereas several other studies did not find any association between HIV and WMH lesion loads [6–8,14,15]. Furthermore, the combined effect of HIV infection and CSVD might also exacerbate cognitive impairment. However, the hypothesis concerning synergistic effects of HIV infection and CSVD on brain structure and cognition has not been definitively tested. One study reported that the presence of WMH in HIV-positive participants was associated with decreased frontal lobe volume [16], but this study did not have a HIV-negative comparison group making it difficult to disentangle effects related to HIV and CSVD, and did not measure cognition.

In this study, we investigated a large sample of virologically suppressed HIV-positive participants and demographically similar controls. A subset of these participants had follow-up visits 2 years later. The total volume of WMH because of vascular causes, as seen on T₂-weighted MRI, was used as a surrogate marker of CSVD severity [9]. We characterized brain volumetrics by applying multiple advanced neuroimaging processing methods, including tensor-based morphometry (TBM) and cortical modeling, and assessed cognitive function using a battery of neuropsychological tests. We sought to: identify whether the HIV-positive participants had greater WMH lesion loads compared with HIV-negative controls; estimate the strength of the relationship between WMH lesion loads and standard clinical markers of HIV infection severity in the HIV-positive group; estimate the extent to which HIV and WMH lesion loads contribute to changes in brain volumetrics and cognitive function; and contribute evidence concerning the longitudinal evolution of WMH lesion loads and the extent to which the HIV-positive cohort shows a greater rate of lesion load change than HIV-negative controls.

Methods

Participants

Washington University in St. Louis (WUSTL) Institutional Review Board approved the study. Written informed consent was obtained from all participants. HIV-positive participants were acquired from ongoing studies conducted by the infectious disease clinic and the AIDS clinical trial unit at WUSTL from 2 February 2012 to 18 April 2017. Demographically similar HIV-negative controls were recruited from the St. Louis community using leaflets and from a research participant registry at WUSTL. Participants were excluded if they had a history of confounding neurological disorders, current or past opportunistic central nervous system (CNS) infection, traumatic brain injury (loss of consciousness >30 min), major psychiatric disorders, or were actively abusing substances or had a dependence diagnosis according to Diagnostic and Statistics Manual of Mental Disorders fourth edition criteria. The present study included HIV-positive and HIV-negative participants more than 40 years old and who had MRI (T₁-weighted and T₂-weighted scans) and neuropsychological testing. All HIV-positive participants were on stable cART for at least 6 months and had an undetectable viral load (<50 copies/ml). A total of 120 HIV-positive and 55 HIV-negative participants were included. From this cohort, 46 HIV-positive and 30 HIV-negative participants had follow-up examination approximately 2 years later (HIV-positive: 2.1 ± 0.2 years; HIV-negative: 1.8 ± 0.3 years).

Clinical variables commonly associated with CSVD, including blood pressure, BMI, waist circumference, and smoking history, were collected from all participants. Participants were considered to have hypertension if they had a resting SBP at least 140 mmHg or diastolic blood pressure at least 90 mmHg. Smoking status was defined as a current smoker or not from self-reported questionnaire.

Neuropsychological assessment

Participants underwent a comprehensive neuropsychological assessment that included 10 standard tests recommended to assess HAND, including [17]: Grooved Pegboard (dominant and nondominant), Hopkins Verbal Learning Test - Revised immediate and delayed recall, Trail Making Test Part A and B, Digit Symbol Substitution Task, Letter-Number Sequencing, Letter Fluency (FAS) and Action (verb naming) Fluency.

Rasch Measurement Theory was applied to the neuropsychological test scores, yielding an estimate of overall cognitive ability. This approach uses item response theory to determine the extent to which a set of items (e.g. neuropsychological tests) and responses to those items reflect a single latent construct (i.e. cognitive ability) [18,19]. Rasch analysis arranges items and participant responses on the same scale (logits), such that items that most participants correctly answer are considered easy items and participants who incorrectly respond are considered to have less cognitive ability. Likewise, items that few participants pass are harder items, and participants who pass them have more cognitive ability [18,19]. This approach yields a single estimate of each person’s cognitive ability that can be treated as a continuous measure (i.e. higher score corresponds with more cognitive ability). This approach does not require normative data and accommodates missing data, allowing all available neuropsychological test scores to be used [19].

MRI acquisition

All participants underwent neuroimaging using the same 3T Siemens Tim TRIO whole-body magnetic resonance scanner at WUSTL. The scanning protocol included T₁-weighted three-dimensional magnetization-prepared rapid acquisition gradient echo sequence [repetition time (T_R)/echo time (T_E)/inversion time (T_I) voxel = 2400/3.16/1000 ms; voxel = 1.0 mm³] and T₂-weighted Fast Spin Echo sequence [T_R/T_E = 3200/460 ms; voxel = 1.0 mm³].

MRI processing

Cross-sectional data were processed using a previously described standard pipeline [20]. Preprocessing included intensity inhomogeneity removal [21] and brain masking [22]. Images were initially linearly registered to the Montreal Neurological Institute ICBM152 template using a nine-parameter affine transform, followed by nonlinear registration [23]. For longitudinal data, a participant-specific template was created using an unbiased template creation approach to ensure good registration to the ICBM152 space and to maintain intra-participant consistency across visits [3]. All data were carefully inspected for unacceptable processing outcomes. All data passed visual quality control.

White matter hyperintensity segmentation

WMH were segmented using T₁ and T₂-weighted scans with a previously validated technique [24]. This approach has been shown to produce valid WMH segmentations that strongly correlate with manual segmentations in this data set and others [24]. In brief, this technique uses a set of features that best inform a random forest classifier of the likelihood that a voxel is a WMH. The feature set includes T₁ and T₂ voxel intensities as well as an intensity and spatial WMH probability map were created by averaging manually segmented WMH maps [24]. The quality of all segmentations were visually assessed. One participant was removed because of a poor segmentation outcome. The burden of WMH was defined as the volume (cm³) of all segmented WMH voxels in ICBM152 space and is thus normalized for head size.

Tensor-based morphometry

TBM provides a voxel-wise estimate of brain structure volume relative to the ICBM152 template. Structural volumes were calculated by taking the log Jacobian determinant of the deformation field from the nonlinear transform [25].

Cortical modeling

Cortical modeling provides a quantitative measure of cortical thickness. Cortical thickness estimates were extracted with Fast Accurate Cortical Extraction by deforming polygonal meshes to fit the gray-white matter and pial surface boundaries [26]. Thickness estimates were mapped to the ICBM152 average cortical template using an iterative feature-based registration algorithm [26].

Statistical analysis

Linear models were used to determine the factors that were significantly associated with WMH lesion loads in all participants. Significant factors were identified using a stepwise model selection approach, where factors that had P less than 0.1 were included in the final model. All models consistently included HIV serostatus, age, sex and race. Within the HIV-positive group, separate linear models tested whether standard clinical markers of HIV disease, including nadir and current CD4⁺, duration of HIV infection, duration on cART, and type of antiretroviral medication (e.g. protease inhibitor), were associated with the burden of WMH, controlling for age, sex and race.

Linear mixed-effects models were used to examine longitudinal changes in WMH lesion loads. This model included HIV serostatus, time (years from initial visit), age at initial visit, sex and HIV serostatus by time interaction as fixed effects, along with participant-specific random intercepts.

Finally, linear models were used to determine the variables that were associated with cognitive function, as indexed by the Rasch score; voxel-wise linear models were used to assess the factors related to regional brain volumes and vertex-wise linear models were used for cortical thickness in all participants. Optimal models were identified using the stepwise procedure as described above. All models included HIV serostatus, age and sex as covariates. Whole-brain statistical maps were corrected for multiple comparisons using a standard false discovery rate with a false-positive rate of 5%. WMH lesion loads were log-transformed to normalize the distribution for all analyses. Statistical significance was set at P less than 0.05 (two-sided) for all models.

Results

Participants

Table 1 summarizes demographic and clinical characteristics of all participants. Although the HIV-positive and HIV-negative groups were comparable with respect to age, education and race, the HIV-positive group had significantly more male individuals than HIV-negative controls (P = 0.001). Factors commonly associated with CVSD were similar between the groups, except for smoking: the HIV-positive group had more current smokers (P = 0.05).

Table 1.

Demographic and clinical characteristics of study participants.

	HIV-positive (n = 119)a	HIV-negative (n = 55)	P valueb
Age [years, mean (SD)]	55.8 (7.9)	56.2 (11.7)	0.83
Sex [n, % (male)]	96 (81)	28 (51)	0.001
Race [n, % (African American)]	67 (56)	32 (58)	0.78
Education [years, mean (SD)]	13.2 (2.8)	13.8 (2.2)	0.14
Duration of HIV infection [years, mean (SD)]	16.4 (8.2)	NA
Current CD4+ [cells/μl, median (IQR)]	580 (397–767)	NA
Nadir CD4+ [cells/μl, median (IQR)]c	121 (21–272)	NA
CPE score [median (IQR)]	7.0 (7–9)	NA
cART duration [years, median (IQR)]d	14.3 (7.3–18.0)	NA
Nucleoside reverse transcriptase inhibitor [n, (%)]	86 (72)	NA
Nonnucleoside reverse transcriptase inhibitor [n, (%)]	7 (6)	NA
Protease inhibitor [n, (%)]	62 (52)	NA
Integrase inhibitor [n, (%)]	22 (18)	NA
Fusion inhibitor [n, (%)]	2 (2)	NA
Blood pressure - systolic [mmHg, mean (SD)]	123.1 (10.5)	125.4 (10.8)	0.20
Blood pressure - diastolic [mmHg, mean (SD)]	79.4 (8.0)	82.0 (9.2)	0.11
Hypertension [n, % (hypertensive)]	23 (19)	7 (13)	0.39
BMI [mean (SD)]	27.2 (6.1)	28.5 (5.9)	0.18
Waist circumference [inches, mean (SD)]	39.7 (6.6)	40.0 (5.8)	0.78
Currently smoking [n, % (smokes)]	62 (52)	18 (35)	0.05
Rasch score [mean (SD); higher is better]	−0.34 (1.16)	0.26 (1.31)	0.008
WMH lesion loads [cm3, median (IQR)]	1.4 (0.8–3.0)	1.4 (0.9, 2.2)	0.94

cART, combination antiretroviral therapy; CPE, central nervous system penetration effectiveness; IQR, interquartile range; NA, not applicable; NC, not collected; WMH, white matter hyperintensities. Bold indicates statistically significant values.

^aOne participant was removed because of poor WMH segmentations.

^bComparisons were made using Wilcoxon signed-rank test for continuous variables and chi-square test for categorical variables.

^cTwenty-five HIV-positive participants were missing nadir CD4⁺.

^dThirty-four HIV-positive participants were missing cART duration.

Factors associated with white matter hyperintensity lesion load

The WMH burden was similar between the two groups (Fig. 1). The optimal linear model revealed that older age (P = 0.001) and hypertension (P = 0.002) were significantly associated with greater WMH burden, smoking had a trend level effect (P = 0.07), whereas HIV serostatus was not significantly associated with WMH (Fig. 3a). Moreover, no significant interactions existed amongst these factors, HIV serostatus and WMH lesion load. Within the HIV-positive group, standard clinical markers of HIV disease were not associated with WMH lesion load.

Fig. 1. White matter hyperintensity lesion load in HIV-positive and HIV-negative participants.

Lesion loads were not log-transformed in the plot to facilitate interpretability.

Fig. 3. Standardized weights of factors that were associated with (a) white matter hyperintensity lesion loads and (b) overall neuropsychological test performance, as summarized by the Rasch score.

Error bars indicate 95% confidence intervals.

Longitudinal change in white matter hyperintensity lesion load

Statistically significant increases in WMH lesion load over an approximately 2-year interval were seen for both groups, after correcting for HIV serostatus, age at initial visit and sex (Fig. 2). Observed changes in WMH burden were similar for the HIV-positive and HIV-negative groups. No significant interactions were observed between HIV serostatus and the time between scans or age with regards to WMH burden. In addition, preexisting hypertension (i.e. hypertension at baseline visit) did not predict subsequent WMH volumes.

Fig. 2. White matter hyperintensity burden over approximately 2 years for HIV-positive (blue) and HIV-negative participants (red).

The solid line represents the best linear fit for each group. The grey band indicates the 95% confidence interval on the linear fitted model.

Neuropsychological test performance

The HIV-positive group had significantly worse overall cognitive ability compared with the HIV-negative controls (P = 0.005; Table 1). The optimal linear model revealed that age (P = 0.0008), education (P = 0.0005), HIV serostatus (P = 0.004) and WMH burden (P = 0.02) were significantly associated with cognitive ability (Fig. 3b). Specifically, we observed that fewer years of education and older age were the strongest predictors of worse cognition, followed by HIV serostatus and greater burden of WMH. Other variables, including hypertension, smoking, BMI and waist circumference, were not associated with cognition.

Brain volume and cortical thickness

Comparing brain measures between the groups revealed thinner cortex and smaller subcortical volumes in the HIV-positive group compared with the HIV-negative group (Fig. 4). The cortex was thinner in bilateral primary sensory and motor cortex, superior temporal gyrus, anterior and middle cingulate cortex and frontal lobe (Fig. 4a). TBM revealed significantly smaller subcortical volumes in the thalamus, putamen, globus pallidus, brainstem and midbrain (Fig. 4b). These results remained significant even after controlling for WMH load, age and sex.

Fig. 4.

(a) Cortical thickness reductions in HIV-positive participants compared with HIV-negative controls. (b) Brain volume reductions in HIV-positive participants compared with HIV-negative controls as revealed with TBM. (c) Cortical thickness reductions associated with greater WMH burden for all participants. (d) Brain volume reductions associated with greater burden of WMH for all participants as revealed with TBM. TBM, tensor-based morphometry; WMH, white matter hyperintensities.

Increased WMH lesion loads were significantly associated with reduced cortical thickness and smaller subcortical volumes in all participants, an effect that remained significant even after controlling for HIV serostatus, age and sex. Cortical thickness reductions associated with WMH burden were found in bilateral frontal lobe, anterior cingulate cortex and temporal lobe (Fig. 4c), whereas smaller subcortical volumes were observed in the thalamus, putamen, globus pallidus and brainstem (Fig. 4d). No interactions were observed between HIV serostatus and WMH lesion load with regards to brain volumes or cortical thickness.

Discussion

Although several studies have investigated possible causes of HAND in the cART era [3–5], the potential synergistic impact of HIV infection and CVSD on the brain structure and cognition in people living with HIV is not well characterized. In the present study, we observed that there was no interaction between HIV and WMH lesion loads on brain volumes or cognitive function. Instead, it appears that HIV infection and CSVD are independent processes that additively contribute to brain atrophy and cognitive impairment.

Using the total volume of WMH of presumed vascular origin as a surrogate marker of CSVD severity, we observed that the higher WMH lesion loads in all participants, independent of HIV serostatus, were associated with reduced subcortical volumes, thinner cortical thickness and poorer cognitive function, which corresponds with the current literature [5,6,8,9,16]. The brain regions associated with WMH lesion loads were similar to those seen in previous work that reported volume reductions throughout the basal ganglia [9] and cortical thickness reductions in the frontal lobe [16,27] in HIV-negative individuals with CSVD, similar in age to our sample. Integrating these results with our previous work in similar cohorts of HIV-positive individuals [2–4], a potential unifying explanation regarding the development of cognitive impairment in HIV-positive individuals in the cART era could be formulated: neurobiological changes occur soon after initial HIV infection and worsen in the absence of cART, resulting in cognitive deficits [2–4], whereas subsequent CSVD may independently contribute to brain injury leading to additional cognitive problems. The combination of these independent processes, likely amongst other factors, has a cumulative detrimental effect on brain structure and cognitive function resulting in the mild cognitive impairment common in older people with longstanding, well controlled HIV infection [19].

Although HIV-related brain injury may not be completely reversible with cART [4], further injury because of the ischaemic consequences of CSVD could be mitigated by addressing modifiable CSVD risk factors through lifestyle changes or medication. We observed that smoking and hypertension were associated with greater lesion loads, and greater lesion loads were associated with poorer cognitive function. This could mean that interventions focused on reducing smoking and controlling blood pressure could lead to a reduction in WMH lesion loads, which, in turn, could lead to better brain health and cognition, especially in treated HIV-positive individuals with good viral suppression. Reduction of CSVD risk has been shown to prevent further decline in cognition in the general population [28]. Future longitudinal interventional studies are warranted to assess the effectiveness of vascular health optimization on cognition in people living with HIV.

The WMH lesion loads were similar between the groups, suggesting that treated, virologically suppressed HIV-positive individuals do not have more severe CSVD.

Although this result lends support to previous findings [6–8,15,29], it differs from a recent study that reported greater lesion loads in an HIV-positive group compared with controls [5]. Discrepancies between these results may reflect cohort differences. The prior study only included HIV-positive men, which could be a source of bias [30], reducing the generalizability [5]. The longitudinal data here reinforced our cross-sectional finding that HIV was not associated with WMH lesion loads, showing that the HIV-positive group did not have more rapid worsening of WMH. However, our longitudinal data may be susceptible to selection bias as only a limited number of participants returned for follow-up visits. Additional longitudinal studies are warranted to verify whether HIV is associated with greater rate of change in WMH lesion load. No associations were observed between clinical measures of HIV disease and WMH, arguing against a direct causal link between HIV and CSVD. Furthermore, no associations were found with cART duration or specific antiretroviral regimens. This is in contrast to prior work reporting that prolonged exposure to cART, specifically protease inhibitors, increased the risk of CSVD [12,31]. However, given the limited number of treatment regimens prescribed, this sample of HIV-positive participants was not appropriate for demonstrating a potential association between cART and CSVD.

Despite uncertain evidence as to whether people living with HIV have increased risk of CSVD, multiple hypotheses have been introduced to explain such a possibility, including: HIV infection and antiretroviral medications alter the pathogenic process underlying CSVD [5,12]; and HIV serostatus may be an indicator of a sub-population that has altered prevalence of CSVD risk factors unrelated to the virus itself (e.g. smoking rates) [12,32]. Our findings indicate that HIV and cART do not affect the pathogenic process underlying CSVD in a clinically meaningful manner. Instead, we provide evidence that the presence of CSVD in people living with HIV is most likely because of traditional vascular factors, such as ageing and hypertension [6–8,14,29,33]. Our results offer some support for HIV serostatus serving as an indicator of a sub-population that has altered CSVD risk factors, as the HIV-positive group had more current smokers than HIV-negative group. However, this difference did not translate into greater WMH lesion loads in our HIV-positive group, which could be explained, in part, by the small, trend-level contribution that smoking had on the WMH lesion loads in our study. This may mean that the perceived relationship between HIV and CSVD in the current literature is driven by increased CSVD risk factors that are unrelated to the virus itself but, for complex reasons [32], are more common in HIV-positive populations [12].

This study has limitations. First, measures of other CSVD risk factors, such as diabetes and cholesterol, were not acquired. Indeed, these CSVD-associated risk factors could be more frequent in people living with HIV because of metabolic effects of HIV infection and cART [34]. Second, although WMH segmentation using the T₁ and T₂-weighted data has been shown to provide valid estimates of lesion loads, FLAIR images would allow a more complete estimate of the full extent of the white matter lesions [24]. However, it remains unknown whether the white matter lesions not captured in the T₁ and T₂-weighted data are clinically significant [24]. Third, the study included participants greater than 40 years old, limiting the generalizability to similarly aged individuals. Additional studies focusing across a greater age spectrum (particularly ≥55 years) are required as possible synergistic effects between HIV and ageing on CSVD may occur [29]. Finally, the HIV-positive group had more men than the HIV-negative group. Although sex was included in all models, there are substantial sex differences in the prevalence of cardiovascular disease [35]. This could be a confounding factor and warrants further investigation.

In conclusion, we provided insight into the impact that HIV and CSVD has on brain structure and cognition in people living with HIV. We showed that although CSVD is not more common in HIV-positive participants compared with HIV-negative controls, cognitive impairment in the cART era may reflect a combination of injury caused by early HIV infection and subsequent CSVD, likely amongst other factors. These results suggest that optimizing vascular health in people living with HIV who are on stable cART and virologically suppressed may be a useful route to improve brain health and protect against decline.