One hundred and eleven community-dwelling older adults were initially recruited for the experiment (64 women, age range 56–86). All participants provided informed consent under a protocol approved by the Institutional Review Board of the University of Kentucky. Participants were recruited from an existing longitudinal cohort at the Sanders-Brown Center on Aging (SBCoA) and the Lexington, KY community. Twenty-eight participants were also co-enrolled in the MarkVCID consortium study (51). Participants completed the Montreal Cognitive Assessment (MoCA) (42) within 6 months of their scan date. Participant MoCA scores ranged from 18 to 30. A total of 72 participants had MoCA scores within the cognitively normal range (26–30) while a total of 39 participants had scores within the mild cognitive impairment (MCI) range (18–25).

Exclusion criteria were significant head injury (defined as loss of consciousness for more than 5 min), stroke, neurological disorders (e.g., epilepsy, Alzheimer's disease) or major psychiatric disorders (e.g., schizophrenia, active clinical depression), claustrophobia, pacemakers, the presence of metal fragments or implants that are incompatible with MRI, or significant brain abnormalities detected during imaging. A neuroradiologist (FDR) evaluated the T1W and FLAIR images for evidence of stroke or other clinically relevant abnormalities. This resulted in exclusion of two participants due to evidence of previous stroke (one participant) and hydrocephalus (one participant), not known at the time of enrollment. Three additional participants were excluded due to MoCA scores suggestive of dementia (MoCA <18), an exclusionary criterion for this study. Finally, one participant was excluded on the basis of being a statistical outlier in WMH volume. Detailed characteristics of the final group of 105 participants involved in data analyses are shown in Table 1.

Participants were scanned in a Siemens 3T Prisma scanner (software version E11C), using a 64-channel head coil, at the University of Kentucky's Magnetic Resonance Imaging and Spectroscopy Center (MRISC). Prior to scanning, all participants were screened to ensure magnetic safety for scanning. The following scans were acquired: (1) a 3D multi-echo, T1-weighted magnetization prepared rapid gradient echo (T1) scan, (2) a 3D fluid-attenuated inversion recovery (FLAIR) scan, (3) a 3D, multi-echo gradient-recalled echo scan used for quantitative susceptibility mapping (QSM). Several other sequences were collected during the scanning session related to other scientific questions and are not discussed further here.

The T1 sequence covered the entire brain [1 mm isotropic voxels, 256 × 256 × 176 mm acquisition matrix, parallel imaging (GRAPPA) acceleration = 2, repetition time (TR) = 2,530 ms, inversion time = 1,100 ms, flip angle (FA) = 7°, scan duration = 5.88 min] and had four echoes [first echo time (TE1) = 1.69 ms, echo spacing (ΔTE = 1.86 ms)]. The 3D FLAIR sequence covered the entire brain (1 mm isotropic voxels, 256 × 256 × 176 acquisition matrix, TR = 5,000 ms, TE = 388 ms, inversion time = 1,800 ms, scan duration = 6.45 min). A high-resolution, flow compensated, multi-echo, 3D spoiled GRE sequence with eight echoes (TR/TE1/ΔTE/FA = 24ms/2.98ms/2.53ms/15°) was acquired and used to create QSM images as described elsewhere (52). The entire brain was covered [acquisition matrix = 224 × 224 × 144, parallel imaging (GRAPPA) acceleration = 2, 1.2 mm isotropic voxels and scan duration = 6.18 min].

We used a previously validated, visual rating method for quantification of region-specific ePVS burden developed via collaboration between multiple consortia and intended to standardize ePVS assessment across the field of cSVD research (47). The method involves manually counting ePVS on T1 images, with additional reference to T2 FLAIR images. Counts were performed in each hemisphere in a single, axial slice of T1 images, within four regions of interest that are known to have the greatest burden of ePVS (11, 14, 27, 33, 47, 53): the centrum semiovale, 1 cm above the lateral ventricles (Figure 1A); the basal ganglia, in the plane of the columns of the fornix (Figure 1B); the midbrain, at the level of the cerebral peduncles; the hippocampus, at the level of the midbrain.

Previous work by multiple consortia such as STandards for ReportIng Vascular changes on nEuroimaging (STRIVE) and Uniform Neuro-Imaging of Virchow-Robin Spaces Enlargement (UNIVRSE) have demonstrated the reliability of counting ePVS on T1, with very high correlation to counts on T2 images (5, 27, 48) as well as a high correlation between single-slice and multi-slice counts (47, 48, 54–57). A total ePVS score was created for each participant by combining counts across the four ROIs. All counts were conducted by the lead author (TJL) blinded to participant demographics and under the supervision of an experienced neuroradiologist (FDR), who clarified unclear imaging.

In accordance with STRIVE and UNIVRSE consensus guidelines (5, 8, 48, 58), ePVS were identified using T1, FLAIR and susceptibility weighted images. ePVS were identified as hypointense and less than 3 mm in diameter to differentiate them from lacunes, which tend to be larger (5, 59, 60). ePVS were further differentiated from lacunes based on their lack of hyperintensity on FLAIR (5, 48). ePVS were differentiated from cerebral microbleeds (CMBs) by their absence of prominent associated blooming artifact on QSM. We used QSM for differentiation of ePVS and CMBs due to evidence that QSM images outperform traditional single-echo susceptibility weighted images in this regard (61–66). Intra-rater reliability for ePVS was assessed on a subset of 20 randomly selected participants, using intra-class correlation (ICC).

Whole brain white matter hyperintensity (WMH) volumes were computed using the ADNI pipeline, specifically the UCD WMH segmentation toolkit (Version 1.3), which employs a validated 4-tissue segmentation method (67). Briefly, participants' T1 image [the four echoes averaged into a root mean square (RMS) image] were first registered to their FLAIR image using FLIRT from FMRIB Software Library version 6.0.1 (68). The FLAIR image was then skull stripped, corrected for inhomogeneities using a previously published local histogram normalization (69), and then non-linearly aligned to a standard atlas (67). WMHs were estimated in standard space using Bayesian probability based on histogram fitting and prior probability maps. Voxels labeled as WMHs based on these maps exceeded 3.5 SDs above the mean WM signal intensity. WMH volumes were calculated in participants' native FLAIR space after back-transformation and reported in cubic millimeters.

Statistical analyses were performed using SPSS (IBM, Chicago, IL, USA, version 28), with results considered statistically significant at P < 0.05. Two main linear regression models were performed. The predictor variable in each model was total ePVS count. The dependent variable in the first model was MoCA score and the dependent variable in the second model was whole brain WMH volume. In the case of significant omnibus results, post-hoc comparisons were conducted to assess relationships between ePVS in specific ROIs and MoCA scores or WMH volume. Age, sex, years of education and estimated total intracranial volume were included as covariates in all models. Estimated total intracranial volume (eTIV) was computed using FreeSurfer as described elsewhere (70).

Three additional analyses were run to control for potential confounders in the relationship between ePVS and MoCA scores. In the first follow-up model, we added other cSVD neuroimaging measures as additional covariates (lacune counts and cerebral microbleed counts). In a second follow-up model, we added available self-reported cSVD risk factors as covariates [body mass index (BMI), hypertension, and diabetes]. Finally, due to a correlation between ePVS counts and whole brain WMH volume, post-hoc regression models were conducted to determine if the relationships between ePVS in our ROIs and MoCA scores remained significant after controlling for whole brain WMH volume.

All predictors and dependent variables were tested for the assumption of normality using the Shapiro-Wilk test. Collinearity between predictors in all models was explored using the variance inflation factor (VIF), with a value of 5 implemented as a threshold value (71). For the generation of scatterplot figures, predictor and dependent variable scores were z-scored within our participant sample to aid identification of potential outliers.

High intra-rater reliability was achieved for ePVS (ICC = 0.9) on a subset of 20 randomly selected participants. MoCA scores were skewed (W statistic = 0.914; P ≤ 0.001) and therefore log-transformed. The distribution of whole brain WMH volume was also skewed, as is typical (W statistic = 0.542; P ≤ 0.001), and thus log-transformed. Variance inflation factor for all predictors was <2 and tolerance was >0.5 in all analyses. Error residuals from all ePVS analyses were normally distributed indicating that the assumption of normality was met.

Results indicated that total ePVS counts were negatively associated with MoCA scores (N = 105, β = −0.352, P ≤ 0.001, SE = 0.099, 95% CI = −0.549 to −0.156, VIF = 1.239) after controlling for age, gender, eTIV, and years of education (Figure 2A). Post-hoc comparisons were conducted to assess relationships between ePVS in specific ROIs and MoCA scores. Results indicated that number of ePVS in the centrum semiovale was negatively associated with MoCA scores (N = 105, β = −0.351, P ≤ 0.001, SE = 0.097, 95% CI = −0.543 to −0.160, VIF = 1.179) after controlling for age, gender, eTIV, and years of education (Figure 2B). ePVS in the basal ganglia was marginally associated with MoCA scores (P = 0.07), while hippocampus (P = 0.561), and midbrain (P = 0.447) were not associated with MoCA score.

Next, we assessed the impact of controlling for additional covariates on our observed relationships between ePVS counts and MoCA scores. First, lacunes and cerebral microbleeds, known cSVD neuroimaging markers, were added as continuous covariates. The relationship between total ePVS and MoCA (P = 0.002) as well as the relationship between centrum semiovale ePVS and MoCA (P ≤ 0.001) remained significant. ePVS in the basal ganglia (P = 0.205), hippocampus (P = 0.579), and midbrain (P = 0.629) were still not associated with MoCA score.

Second, we investigated participant-reported risk factors of cSVD as covariates in the ePVS-MoCA model. BMI was treated as a continuous variable and hypertension and diabetes were treated as dichotomous variables. Once again, the relationship between total ePVS and MoCA (P = 0.002), as well as the relationship between centrum semiovale ePVS and MoCA (P = 0.001), remained significant. ePVS in the basal ganglia (P = 0.145), hippocampus (P = 0.629), and midbrain (P = 0.542) were not associated with MoCA score.

Neither the addition of other neuroimaging markers of cSVD as covariates in our models, nor the addition of cSVD risk factors, affected the significance of our results. Therefore, we summarize statistical values (Table 2) and present figures (Figure 2) from the original linear regression model which includes age, gender, years of education, and eTIV as covariates.

Linear regression models were used to evaluate the relationship between ePVS and WMH volume. Results indicated that total ePVS count was positively associated with whole brain WMH volume (N = 104, β = 0.245, P = 0.010, SE = 0.093, 95% CI = 0.060–0.430, VIF =1.239) after controlling for age, sex, eTIV, and years of education (Figure 3A). Post-hoc comparisons indicated that basal ganglia ePVS were positively associated with whole brain WMH volume (N = 104, β = 0.324, P ≤ 0.001, SE = 0.089, 95% CI = 0.148–0.500, VIF = 1.191) (Figure 3B). In addition, centrum semiovale ePVS were positively associated with whole-brain WMH volume (N = 104, β = 0.206, P = 0.027, SE = 0.092, 95% CI = 0.023–0.388, VIF = 1.179) (Figure 3C). ePVS in the midbrain (P = 0.117) and hippocampus (P = 0.860) were not significantly related to WMH volume. Linear regression statistical values for the relationships between WMH volume and ePVS by ROI are reported in Table 2.

Consistent with our hypotheses, our results thus far indicate a relationship between ePVS and both MoCA and WMH volume. In order to determine if our hypothesized relationship between ePVS and MoCA is independent of WMH volume, a final set of regression models were conducted between ePVS and MoCA scores after controlling for whole brain WMH volume. Standard covariates of age, gender, eTIV, and education were also included. The relationship between total ePVS and MoCA (N = 104, β = −0.358, P ≤ 0.001, SE = 0.104, 95% CI = −0.564 to −0.153, VIF = 1.326) as well as the relationship between centrum semiovale ePVS and MoCA (N = 104, β = −0.354, P ≤ 0.001, SE = 0.100, 95% CI = −0.552 to −0.155, VIF = 1.240) remained significant after controlling for total WMH volume. ePVS in the basal ganglia (P = 0.094), hippocampus (P = 0.572), and midbrain (P = 0.386) were still not associated with MoCA scores. Statistical values for these relationships across all ROIs are reported in Table 3.

This study has limitations that highlight the need for additional follow-up studies. First, our cross-sectional study cannot determine if ePVS predict cognitive dysfunction and white matter damage. Future research using longitudinal imaging and clinical data collection is needed to identify if ePVS are baseline predictors of longitudinal cognitive decline and WMH change. While longitudinal studies are the gold-standard for investigating the predictive capacity of ePVS, results from cross-sectional studies such as the present one may prove useful in optimizing the selection of relevant variables for use in future longitudinal ePVS studies. It should also be noted that our cohort included primarily highly educated, White participants. Our findings will need to be replicated in more diverse cohorts. Future studies should also explore the associations between ePVS and MoCA scores in a more clinically heterogeneous set of participants, including those with dementia.

Finally, future longitudinal studies should consider mechanistic contributions to ePVS development. For example, longitudinal studies considering the potential contributions of biofluid markers of AD, atrophy, and inflammatory processes to subsequent ePVS development could prove informative (12, 14, 17, 18, 86, 91–93). In particular, the inflammatory processes of arterial stiffening and blood-brain-barrier breakdown are both established markers of cSVD and should be explored as predictors of ePVS development or as modifiers of ePVS effects on global cognition (10, 17, 24, 94–96). A better understanding of the potential mechanisms of perivascular space enlargement could point to early intervention targets intended to slow or prevent cognitive dysfunction.

Our results indicate that ePVS burden in older adults is negatively associated with performance on the MoCA, a standardized clinical measure of global cognitive function. Further, ePVS burden was positively associated with WMH volume, an established marker of cSVD. Our results are consistent with a view that ePVS are clinically significant and motivate future longitudinal studies exploring the accuracy of ePVS in predicting subsequent cognitive decline and cSVD progression.