Observation of reduced homeostatic metabolic activity and/or coupling in white matter aging

Abstract

BACKGROUND AND PURPOSE:

Transvascular water exchange plays a key role in the functional integrity of the blood brain barrier (BBB). In white matter (WM), a variety of imaging modalities have demonstrated age-related changes in structure and metabolism, but the extent to which water exchange is altered remains unclear. Here, we investigated the cumulative effects of healthy aging on WM capillary water exchange.

METHODS:

38 healthy adults (aged 36–80 years) were studied using 7T dynamic contrast enhanced MRI. Blood volume fraction (v_b) and capillary water efflux rate constant (k_po) were determined by fitting changes in the ¹H₂O longitudinal relaxation rate constant (R₁) during contrast agent bolus passage to a two-compartment exchange model. WM volume was determined by morphometric analysis of structural images.

RESULTS:

R₁ values and WM volume showed similar trajectories of age-related decline. Among all subjects, v_b and k_po averaged 1.7 (± 0.5) mL/100 g of tissue and 2.1 (± 1.1) s⁻¹, respectively. While v_b showed minimal changes over the 40-year age span of participants, k_po declined 0.06 s⁻¹ (ca. 3%) per year (r= −0.66; P< .0005), from near 4 s⁻¹ at age 30 to ca. 2 s⁻¹ at age 70. The association remained significant after controlling for WM volume.

CONCLUSIONS:

Previous studies have shown that k_po tracks Na⁺,K⁺-ATPase activity-dependent water exchange at the BBB and likely reflects neuro-glio-vascular unit (NGVU) coupled metabolic activity. The age-related decline in k_po observed here is consistent with compromised NGVU metabolism in older individuals and the dysregulated cellular bioenergetics that accompany normal brain aging.

INTRODUCTION

Increasing evidence suggests microvascular changes accompany normal brain aging. In gray matter, histological studies have revealed a variety of age-related changes in capillary ultrastructure, including thickening of the basal lamina and loss of endothelial cells. Decreased branching, increased tortuosity and reduced microvascular density are also commonly observed.^1,2 In contrast, age-related changes in white matter (WM) microvessels have been inconsistently reported. While several groups have noted an increase in basal lamina thickness and endothelial thinning in older brains,³ others have observed only subtle differences.^4,5 Stereological measurements of capillary diameter, density, and other microarchitectural features are not robust and the effects of normal aging on the functional integrity of WM microvessels remain unclear.

The ¹H₂O (water proton) longitudinal relaxation rate constant (R₁ ≡ 1/T₁) has been shown to be a sensitive in vivo marker of tissue characteristics throughout the central nervous system and represents a potentially important probe of capillary integrity in the brain. However, R₁ values reflect relaxation characteristics of all water protons contributing to the MR signal. Thus, while microvascular features that affect longitudinal relaxation are likely to be reflected in tissue R₁ values, so too are neuronal and gliotic changes that alter myelin lipids, proteins and other extravascular macromolecules.⁶ Moreover, the water fraction of the microvascular space is very small (typically 1–3%),^7,8 reducing sensitivity to that component under normal MRI acquisition conditions. As a result, static R₁ values provide limited insight into microvascular changes that accompany normal brain aging. In contrast, dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) yields R₁ values before, during and after passage of an intravenous bolus of paramagnetic contrast agent (CA). Pharmacokinetic modeling of the R₁ time-course to a two-compartment exchange model incorporating the effect of vascular CA concentration, [CA_b], on longitudinal relaxation then yields estimates of the blood plasma and (with hematocrit assumption) blood volume fractions (v_p and v_b, respectively), and the steady-state capillary water molecule efflux rate constant (k_po).^9,10 Because water exchange between plasma and red blood cells is very fast in whole blood,¹¹ k_po can be equated to k_bo, the blood water extravasation rate constant. Since water extravasation most likely reflects mechanisms other than paracellular diffusion,¹² k_po may be a particularly important probe of the subtle microvascular abnormalities expected to accompany healthy brain aging.

The purpose of this study is to quantify capillary water exchange in the WM of healthy individuals across a wide age range. We hypothesize that the bioenergetics of processes that drive water exchange within the tightly coupled cells of the neuro-glio-vascular unit (NGVU) are compromised during normal aging, resulting in reduced k_po in older individuals. Since CA detection sensitivity increases with magnetic field strength,¹³ we performed our measurements at 7T. Together with the increased signal to noise ratio (S/N) afforded by high field,¹⁴ we expected the approach to improve the precision of R₁ measurements and the pharmacokinetic parameters derived from them.

METHODS

Participants

Thirty-eight healthy volunteers (17 male/21 female) aged 60.7 (± 12.8) yrs., with no functional impairment were recruited from the Oregon Health & Science University (OHSU) community. Vascular risk factors, including diabetes, prior transient ischemic attack or stroke were absent in all individuals. Volunteers who had undergone cancer treatment in the previous 12 months, had a history of brain surgery or head trauma with more than momentary loss of consciousness were not enrolled. In an effort to include participants more representative of the general population, subjects with hypertension or mild cardiovascular heart disease well-controlled by oral medications were included. Since hypertension has been shown to be associated with small vessel disease,^15,16 WM signal hyperintensities were visually assessed on T₂-weighted images in all subjects. When present, these were seen as small, focal lesions scattered throughout the WM (≤ grade 1 on the Fazekas rating scale).¹⁷ The study was approved by the OHSU Institutional Review Board and all participants signed written informed consent.

MRI Acquisition and Analysis

Images were acquired using a whole-body 7T instrument (Siemens, Erlangen, Germany) with 8-channel (13 subjects) or 24-channel phased-array RF head coils. Anatomic imaging consisted of a T₁-weighted 3D inversion recovery (IR) MPRAGE (inversion time, TI: 1050 ms; repetition time/echo time, TR/TE: 2300/2.8 ms; nominal flip angle, FA: 6°; 0.8–1 mm isotropic resolution) and a 2D proton density/T₂-weighted turbo spin echo scan (echo times, TEs: 11 and 87 ms; TR 10000 ms; slice thickness: 2 mm with 2 mm gap; (0.44 mm)² in-plane resolution.

Single-slice DCE image sets positioned superior to the lateral ventricles in the centrum semiovale were obtained using a fast IR-gradient echo (turboFLASH) technique.¹⁸ Each image set consisted of a series of 5.5–10 mm thick axial images (TR/TE/FA: 285 ms/1.25 ms/6°; matrix: 64×64; field of view, FOV: (192 mm)² or TR/TE/FA: 285 ms/1.15 ms/6°; matrix: 128×96; FOV: 256×192 mm²) collected at 8 TIs after a non-selective inversion pulse. TIs ranged from 153–2035 ms with 264 ms spacing. A bolus of ≤ 0.05 mmol/kg gadoteridol (ProHance; Bracco Diagnostics, Monroe Township, NJ) was delivered into an antecubital vein by power injector (2 mL/s) followed by 20 mL saline flush. This dose provided good image contrast while reducing potential saturation of the blood signal due to high [CA]. To allow magnetization to reach a steady state prior to CA arrival, the 6–8 s injection began after collection of the sixth image set. Pharmacokinetic temporal resolution was 2.3 s. The S/N was estimated from long TI images as the ratio of the mean signal intensity on the sixth image and the standard deviation of pixel values in a difference image formed by subtraction of the eighth and sixth images.¹⁹ Correction of the ratio for the increased standard deviation of a difference image yielded a S/N of 92 (± 17).

R₁(t) maps were calculated by voxelwise fitting the full Bloch equation incorporating all RF pulses and delays to the signal magnitude at each TI. Inversion recovery was modeled with a two-parameter single exponential, using a gradient expansion algorithm and Levenberg-Marquardt optimization implemented in Python (v 2.7.12). Initial R₁ (R₁₀) estimates were calculated by voxelwise fitting the simple IR equation to the average magnitude of the last five image frames of the variable inversion time data prior to CA injection.²⁰

WM masks were prepared by fitting bimodal Gaussian distributions to R₁₀ histogram data.¹³ All masks were eroded to reduce partial volume contamination from cortical gray matter and manually edited to ensure at least a 1 voxel gap between WM and voxels containing gray matter or mixed tissue types. Application of masks to R₁(t) maps yielded R₁ time-courses in a WM region of interest [ROI; 18.9 (± 4.8) cm³]. The venous output function (R_1b(t)) was determined from a voxel centered within the sagittal sinus. After temporal alignment of tissue and blood signals to correct for delay in venous signal change onset, v_b and k_po were determined by fitting the R₁ time-course to the two-compartment Shutter-Speed Paradigm (SSP) model. The mathematical formulation of the SSP is shown in Equation [1].^9,10 Here, R₁(0), R_1b(0) and R₁(t), R_1b(t) are the measured R₁ values of tissue and blood ¹H₂O, respectively, prior to and at each time, t, after CA injection, R_1e is the R₁ of extravascular tissue ¹H₂O corrected for blood contributions, r₁ is the gadoteridol relaxivity (3.3 s⁻¹ mM⁻¹ at 7T),²¹ h is the hematocrit (ca. 0.45), f_w is the tissue volume fraction accessible to mobile aqueous solutes (ca. 0.8), and [CA_b] is the blood CA concentration.²² Curve fitting errors were minimized by Levenberg-Marquardt least squares minimization²³ and goodness of fit assessed by l₂-norm.

$${\text{R}}_1\left(\text{t}\right)=\frac{1}{2}\left\{ \left[{\text{R}}_{1\text{b}}\left(\text{t}\right)+{\text{ R}}_{1\text{e}}+{\text{ k}}_{\text{po}}+\frac{{\text{k}}_{\text{po}}{\text{p}}_{\mathrm{b}}}{1-{\text{p}}_{\mathrm{b}}}\right]-{\left[ {\left({\text{R}}_{1\text{e}}-{\text{R}}_{1\text{b}}\left(\text{t}\right)-{\text{ k}}_{\text{po}}+\frac{{\text{k}}_{\text{po}}{\text{p}}_{\mathrm{b}}}{1-{\text{p}}_{\mathrm{b}}}\right)}^2+\frac{4{\text{k}}_{\text{po}}^2{\text{p}}_{\mathrm{b}}}{1-{\text{p}}_{\mathrm{b}}}\right]}^{\frac{1}{2}}\right\}$$ [1]

where

$${\text{p}}_{\mathrm{b}}={\text{v}}_{\mathrm{b}}/{\text{f}}_{\mathrm{w}},{\text{R}}_{1\text{e}}=\frac{{\text{R}}_1(0)-{\text{R}}_{1\text{b}}(0){\text{p}}_{\mathrm{b}}}{1-{\text{p}}_{\mathrm{b}}},\text{and}{\text{R}}_{1\text{b}}(\text{t})={\text{R}}_{1\text{b}}(0)+{\text{r}}_1(1-\text{h})[{\text{CA}}_{\text{b}}](\text{t})$$

Monte Carlo simulations were used to investigate the S/N influence on fitted parameters. Random noise was added to signal intensities in each DCE image set. Noise was estimated by sampling a uniformly distributed random number generator with zero mean and standard deviation equal to the signal intensity divided by S/N. The S/N of simulated datasets ranged from 50–70. R₁(t) values were estimated a total of 1000 times and parametric estimates obtained from fittings to Eqn. [1], as described above.

Total WM volume was determined from anatomical MPRAGE images after skull removal and three class segmentation using tools from the FMRIB Software Library (http://www.fmrib.ox.ac.uk/fsl). Visual assessment of extracted images showed that, in general, use of bias field and neck cleanup options produced optimal brain extraction while removing non-brain regions. Variation in WM volume due to head size was corrected by normalization to total intracranial volume (ICV).

Statistical Analyses

Statistical analyses were performed in Stata (v. 12, Statacorp, College Station, TX) and JMP (v. 12, SAS Institute, Cary, NC). Distributional normality was assessed using the Kolmogorov-Smirnov test. Skewness, when present, was removed by common data transformations. Univariate associations were analyzed by least squares linear regression. Adequacy of linear models was examined using residual plots. Homogeneity of residual variance was assessed using graphical (residual v fitted plots) and non-graphical (Breusch-Pagan test) methods.²⁴ Influence of data points with extreme values was assessed using difference-in-fits and Cook’s distance measures.²⁵ Strength of association was measured by Pearson correlation coefficient (r). Numerical optimization (Matlab v. 9.4, Mathworks, Natick, MA) of piecewise linear regression modeling was used to estimate regression coefficients and change point location in R₁ and WM volume by age plots.²⁶ Models were iteratively fitted with change point values constrained between 50 and 70 years of age and the model with lowest residual error selected to summarize results. P values were calculated at the 0.05 significance level in two-tailed tests.

RESULTS

The age-dependences of WM ROI-averaged ¹H₂O R₁ values and volume fraction were first examined. R₁ values remained near 0.925 s⁻¹ during mid-life but decreased by 0.017 s⁻¹ per decade (95% CI; −0.0035, −0.031) after age 57 (Fig. 1a). While subjects scanned with the 8-channel coil were significantly older than those scanned with the 24-channel coil (69 ± 6 vs. 57 ± 13 years, respectively, P= .0004), the precision of R₁ values was unaffected. Likewise, WM volume fraction differed little among subjects younger than 54 years of age, but declined by 0.96% per decade (95% CI: −1.6, −0.34) among older subjects (Fig. 1b). Figure 1c shows the WM R₁ vs. volume fraction scatter plot. After controlling for age, an R₁ decline of 0.012 s⁻¹ was found to accompany a 1% loss in tissue volume (P =0.003; r= 0.57) or, since water fraction ≡ (1- tissue volume fraction), a 1% increase in water content.

Figure 1.

Linear regression plots with age of ROI-averaged WM (a) ¹H₂O R₁ values, and (b) WM volume, normalized to total intracranial volume (ICV). Both R₁ and volume fraction are independent of age in subjects younger than ca. 60 years, but decline significantly thereafter. Arrows indicate the change points determined by numerical optimization. Outlying, influential R₁ values in two subjects are excluded. (c) Regression plot of R₁ values vs. WM volume fraction. Inset shows fit residuals.

Figure 2 shows voxelwise R₁ histograms and maps prior to (bottom panel) and at two times after CA injection in a 68 year-old subject. The expected rightward shift to increased median R₁ values is observed approximately 30 s after injection as CA arrives in the tissue (middle), decreasing to near baseline levels after most has washed out (top panel). A representative plot of ROI-averaged R₁ vs. blood R_1b values is shown in Figure 3. The associated DCE time-courses are also shown (inset). As expected, a hyperbolic dependence of tissue R₁ on blood R_1b values is observed. Consistent with the negligible CA extravasation during the first post-injection minute in intact brain tissue, the peak enhancement in R₁ values was 0.04 (± 0.02) s⁻¹, roughly 1% of the maximum R_1b increase (ca. 7 s⁻¹ at 1 min). The solid curve in the R₁ vs. R_1b plot represents the best fitting of Eqn. [1] to the data.

Figure 2.

(a) Pixel-wise WM ¹H₂O R₁ values in a 68 year-old male. The horizontal axis shows R₁ values across the imaged slice, after skull removal and masking of gray matter and CSF regions. The vertical axis (removed for clarity) is proportional to the number of voxels and the scale is constant across the three plots. The time after CA injection is given at left. The median value of the R₁ distribution increases ca. 30 seconds (s) after injection as [CA] increases, and then decreases at longer times as CA washes out. The corresponding R₁ maps are shown at right, with identical grayscale applied to each.

Figure 3.

A representative WM ROI (yellow, upper left) R₁ vs. R_1b plot from a 69 year-old male. The solid curve represents the best fitting of the data to Eqn. [1]. Residual l₂-norm is 0.016 (0.020 ± 0.013, all datasets). The inset shows the blood (filled markers) and tissue (open) R₁ time-courses. A temporal correction of 2–3 sec (1 frame) has been applied to the blood data to account for delay in CA bolus arrival in the sagittal sinus (red voxel, upper left).

Fittings of tissue R₁ vs. blood R_1b plots from all subjects yielded mean (± SD) estimates of v_b [ 1.7 (± 0.5) mL/100 g tissue] and k_po [2.1 (± 1.1) s⁻¹]. It should be noted that standard deviations are dominated by between-subject variation. Monte Carlo simulations showed the uncertainty in individual estimates is much less (SD = 0.04 and 0.10 in v_b and k_po, respectively). The mean v_b is in good agreement with the 1.5 mL/100 g tissue in young volunteers based on T₁-weighted EPI measurements,²⁷ and the 1 mL/100 g measured by quantitative PET.²⁸ k_po values have not been widely reported, particularly in older individuals. In healthy 20–45 year old subjects, Rooney and colleagues reported WM k_po values of 3.6 (± 0.77) s⁻¹ and 3.2 (± 0.56) s⁻¹.^10,12 For the seven subjects younger than 45 years in the present study, k_po averaged 3.2 (± 1.2) s⁻¹, in excellent agreement with these findings.

The age-dependencies of ROI-averaged v_b and k_po values for all subjects are shown in Figure 4. v_b declined an average of 0.02% per year, but the change is not significant (inset). Although the magnitude may be somewhat obscured by between-subject variance, this reduction is substantially less than the 0.3% per year reported using DSC-MRI²⁹ and the 0.5% per year found in healthy 22–82 year-old subjects by ¹⁵O-PET⁸. In contrast, k_po was strongly associated with age (r= −0.66), declining by 0.059 (± 0.011) s⁻¹ per year (P< .0005). The association remained significant after controlling for WM volume (P< .0005). There was no significant effect of sex on any parametric estimate.

Figure 4.

Linear regression plots of v_b (inset) and k_po by age. The error bars (not all of which are visible on this scale) reflect standard deviation of ROI mean estimates for individual subjects based on Monte Carlo simulations. The solid lines represent the best linear fittings to the data.

DISCUSSION

DCE-MRI combined with pharmacokinetic modeling has been shown to be an extremely sensitive probe of vascular changes in many pathological states.³⁰ Nevertheless, the relatively low vascularity of WM and subtlety of disturbances that characterize normal aging makes accurate determination of age-related changes in WM microvasculature challenging. In the present study, we have taken advantage of the increased S/N, reduced ¹H₂O R₁₀ values and improved CA detection sensitivity afforded by high magnetic field to increase the precision of R₁ measurements.^13,14,31 In addition, pharmacokinetic estimates were obtained by fitting the R₁ time-course to a two-compartment SSP model that explicitly accounts for the effects of equilibrium water exchange on the MR signal. We^{10, 32–34} and others^7,35 have shown that this approach can yield an extremely sensitive probe of BBB pathology, particularly in the early stage of disease when disruption is likely to be subtle. The appropriateness of the approach in the context of normal WM aging is confirmed by the observed non-linearity of R₁ vs. R_1b plots (characteristic behavior in human tissues^10,34,36 and the signature feature of the departure of a compartmental water exchange system from the fast exchange limit^9,37), the fittings of Eqn. [1] to the data (Fig. 3) and the resultant precision of parametric estimates.

The main finding of our study is that normal WM aging is accompanied by decreased k_po. This result has important implications. The molecular pathways by which water exchanges across plasma membranes are not well understood. In the brain, passive processes, i.e., simple diffusion through the membrane, aqueous channels (aquaporins) or paracellular leakage, generally contribute little to the overall extravasation flux.¹² Rather, most water is transported through capillary endothelial cell membranes, driven by processes coupled with metabolically active ion transport via membrane-bound proteins.^38–41 In mammalian cells, the P-type Na⁺,K⁺-ATPase (NKA) pump creates the electrochemical ion gradients that drive transmembrane ion exchange and water cotransport.⁴² The finding that k_io, the cellular equivalent of k_po, correlates with H⁺-ATPase activity in yeast led us to postulate that k_io reflects metabolically active water transport.⁴³ Subsequent ex vivo DCE studies (tabulated and described in more detail in Li, et al.³⁷) provide evidence that k_io is responsive to a wide variety of metabolic alterations (NKA activity, substrate/inhibitor concentrations, ATPase gene dosage) and likely to be altered by pathology. In some cases, results have been validated in ultra-low field measurements not subject to DCE uncertainties. Importantly, these findings have recently been extended to the human brain where, in combination with ³¹P MR spectroscopy, k_po was shown to track ATP production/consumption in vivo, declining in hypometabolic regions of the brain.¹²

Taken together, these results suggest a possible physiological interpretation of k_po. The defining feature of cerebral capillaries is the tight junctions between endothelial cells that provide a physical barrier to most blood solutes. However, endothelial cells are tightly coupled spatially and metabolically to pericytes, astroglial cells, and neurons and function within the greater context of the multicellular NGVU.⁴⁴ Rooney and coworkers proposed that as neuronal activity varies, the demand for substrate from the blood space to cells behind the BBB changes and endothelial cell activity is altered.¹² In response, ion and water cycling within the NGVU, which fluctuates constantly to maintain ionic gradients, is affected. Thus, by reporting NKA activity of water cycling processes within the NGVU, the authors hypothesized that k_po provides a quantitative metric of on-going brain metabolism. Our finding of a continuous decline in k_po through mid-life and into later decades is consistent with this interpretation, and mirrors the progressive reduction in cerebral metabolism revealed in many cross-sectional and longitudinal studies.^45,46

Despite the simplicity of the linear relationship suggested by Fig. 4, the mechanism of age-related k_po decline is likely to be complex. Metabolic dysregulation is multifactorial with processes coupled in a highly complex manner.⁴⁷ In WM, altered glucose metabolism⁴⁵ and impaired oxygen delivery⁴⁸ have been shown to contribute to the observed decline. Moreover, mitochondria undergo profound age-related changes resulting in increased generation of reactive oxygen species, decreased Ca²⁺ uptake and abnormal organelle networks critical to regulating energy metabolism.^49,50 As homeostatic mechanisms become increasingly disrupted, the NGVU is likely to become more vulnerable to these and other cellular events, compromising metabolic activity and ion transport processes coupled to water cycling.

Although accumulating evidence supports interpretation of k_po as a marker of NGVU metabolism, additional validation is needed. We anticipate the increased frequency with which SSP models are used to analyze DCE results will provide opportunities to examine this more closely. Of note, Dickie et al. recently used DCE-MRI and SSP modeling to measure P_wS [the capillary water permeability coefficient (P_w)•capillary surface area (S) product¹²] as a function of occludin (a transmembrane protein enriched at endothelial tight junctions) content in the rat brain.³⁵ Although their terminology differs slightly from ours, they find P_wS (= v_b•k_po) increases with decreasing occludin content, despite the absence of significant change in lectin immunohistochemical staining. Assuming lectin provides a valid metric of S, though, this result implies that P_w (and thus k_po) increases with decreasing occludin; and, since cerebral tissue occludin content is inversely related to age,⁵¹ with increasing age- the opposite of Fig. 4. Since occludin is essential to tight junction structural integrity, the authors’ implicit implication is that BBB water extravasation is driven by passive, diffusion-mediated processes via paracellular pathways and the importance of these processes increases with decreasing occludin content. Importantly, though, occludin is also an NADH oxidase, and has been recently shown to affect NGVU mitochondrial metabolic activity.⁵² Thus, it is possible that occludin stimulates the active transendothelial water efflux pathway, the paracellular pathway playing essentially an insignificant role. Reporting occludin’s effect on the individual v_b and k_po estimates could provide unique insight into the relative contributions of the two pathways and interpretation of k_po as a marker of NGVU metabolism.

A second finding of our study is the similarity in the age trajectories of both R₁ and WM volume, with both remaining essentially constant during mid-life and decreasing sharply thereafter. Progressive loss of WM volume in healthy elderly, particularly after age 60, has been noted previously^53–56 and likely reflects atrophic processes that accelerate in later years. In a recent report, WM R₁ values determined across a wider age range were found to follow a non-linear trajectory similar to that observed in the present study.⁵⁷ Ex vivo studies have shown that in the absence of CA, the dominant (ca. 90%) source of R₁ variation is myelin content.⁵⁸ Together, these results confirm that age-dependent changes in WM tissue are likely due to a combination of atrophic processes (and the requisite increase in water content) and alterations in the molecular composition of the tissue itself. The similarity in age trajectories between WM volume and R₁ values observed here suggests a temporal concordance between the two in mid- to later life.

Our study is not without limitations. First, the number of subjects is small. While the sample size is sufficient to reveal a significant age-related effect on k_po, the low blood volume of WM coupled with between-subject variability could be obscuring a significant effect on v_b. Sample size also limited modeling of age-related changes to linear models and higher order models were not examined. Although a linear model provides a good fit to aging effects on k_po, R₁ changes appear to accelerate in later years. Age-related changes in WM R₁ have been reported to follow a second order polynomial in individuals 10–80 years of age, increasing until about 45 years of age and decreasing thereafter⁵⁷. The age range of participants in the current study is a subset of this distribution and heavily weighted towards older adults. While this likely limits the accuracy of our change point analysis, the trajectory of WM R₁ changes observed in the present study is nevertheless consistent with that observed across the lifespan. Second, imaging data were acquired with either an 8- or 24-channel coil. While there is no difference in sex, subjects scanned with the 8-channel coil are significantly older. Simulations show a slight increase in v_b standard errors in 8-channel datasets that could have contributed to our failure to find an age-dependency. However, the maximum change in R_1b is not significantly different between acquisitions. Moreover, the precision of R₁ and k_po estimates does not vary (likely the result of the more or less constant voxel volumes of the acquisitions) and there are no coil-related discontinuities or meaningful effects on k_po vs. age plots. Thus, it is unlikely that coil differences introduce important systematic error or compromise the main conclusions of our study. Finally, it is important to note that while high magnetic field offers potential advantages for DCE MRI, increased field strength may also be a source of error. Power deposition rises significantly with field strength and often requires compromises in sequence parameters (with an accompanying reduction in sensitivity) to comply with specific absorption rate (SAR) limitations.⁵⁹ In addition, spatial variation in the B₁ field due to dielectric inhomogeneities can reduce image quality and decrease the accuracy of parametric estimates. While the IR sequence used in R₁ mapping mitigates these effects somewhat, B₁ field inhomogeneities may nevertheless contribute to our difficulty in obtaining reliable voxelwise estimates. Therefore, analyses were done at the ROI level. This approach greatly improved stability of the estimates. However, the ROIs upon which our conclusions are based incorporate at least three major association tracts (cingulum, inferior fronto-orbital and superior longitudinal fasiculii) and, given the 2D acquisition, reflect only a relatively small fraction of the total WM volume. Moreover, the slice thickness in some of our DCE studies is relatively large and ROIs could include gray matter contributions from underlying structures. Three dimensional FLASH acquisitions optimized for water exchange effects have been designed with sufficient S/N and contrast-to-noise to produce DCE MRI images with high spatial and temporal resolution.⁶⁰ Their use could reduce partial volume contamination and allow assessment of the regional specificity of pharmacokinetic estimates.

In conclusion, our results demonstrate that equilibrium water exchange at the BBB declines in an age-dependent manner in normal WM. The decline is apparent in midlife, continues into later decades and persists in the presence of atrophic changes. Additional studies are underway to confirm k_po as a measure of active water cycling and a potential biomarker of NGVU metabolic activity during normal and pathological brain aging.