Altered linear coupling between stimulus-evoked blood flow and oxygen metabolism in the aging human brain

Abstract

Neural-vascular coupling (NVC) is the process by which oxygen and nutrients are delivered to metabolically active neurons by blood vessels. Murine models of NVC disruption have revealed its critical role in healthy neural function. We hypothesized that, in humans, aging exerts detrimental effects upon the integrity of the neural-glial-vascular system that underlies NVC. To test this hypothesis, calibrated functional magnetic resonance imaging (cfMRI) was used to characterize age-related changes in cerebral blood flow (CBF) and oxygen metabolism during visual cortex stimulation. Thirty-three younger and 27 older participants underwent cfMRI scanning during both an attention-controlled visual stimulation task and a hypercapnia paradigm used to calibrate the blood-oxygen-level-dependent signal. Measurement of stimulus-evoked blood flow and oxygen metabolism permitted calculation of the NVC ratio to assess the integrity of neural-vascular communication. Consistent with our hypothesis, we observed monotonic NVC ratio increases with increasing visual stimulation frequency in younger adults but not in older adults. Age-related changes in stimulus-evoked cerebrovascular and neurometabolic signal could not fully explain this disruption; increases in stimulus-evoked neurometabolic activity elicited corresponding increases in stimulus-evoked CBF in younger but not in older adults. These results implicate age-related, demand-dependent failures of the neural-glial-vascular structures that comprise the NVC system.

Introduction

Neural activity within a given region of the brain is constantly shifting in response to exogenous stimuli (Ogawa et al. 1992), cognitive processes (Le Bihan et al. 1993), and intra- and internetwork connection signaling (Biswal et al. 1995). Lacking local reserves of energy (e.g. glycogen storage), neurons depend upon astrocytes to relay their activity-related metabolic needs by signaling blood vessels to dilate in response to increases in neurometabolic activity (Attwell et al. 2010). Research using murine models has elucidated fundamental molecular and cellular mechanisms that comprise this neural-vascular coupling (NVC) process. At excitatory synapses, astrocytic endfeet uptake glutamate (Gottfried et al. 2002) and potassium ions (Jukkola et al. 2013) following action potentials. Astrocytes then signal local vasculature to dilate through deposition of epoxyeicosatrienoic acid and prostaglandins, thus increasing local blood flow and oxygen (O₂) delivery (i.e. functional hyperemia; Jukkola et al. 2013). Dual-echo calibrated functional magnetic resonance imaging (cfMRI Hyder, 2004; Hoge, 2012) permits specific measurement of each component of a three-tier neural-glial-vascular model. Neurometabolic activity can be functionally indexed by the rate of oxygen metabolism by neurons (tier I), calculated using a scaled product of measured blood flow and blood-oxygen-level-dependent (BOLD) signal. The function of neural-vascular mediators, such as astrocytes (Metea and Newman 2006; Takano et al. 2006; Attwell et al. 2010; Tian et al. 2010; Petzold and Murthy 2011; Lee et al. 2012; MacVicar and Newman 2015; Cauli and Hamel 2018; Gu et al. 2018) as well as others such as endothelial cells (Chen et al. 2014; Hillman 2014) and pericytes (Krueger and Bechmann 2010; Dalkara et al. 2011), can be indexed by the proportion of blood flow increase per unit change in neural oxygen metabolism (tier II). Vascular activity (cerebral arterioles and capillaries, tier III) can be functionally indexed by direct measurement of blood flow.

Healthy aging affects NVC (for review, see Abdelkarim et al. 2019; Tsvetanov et al. 2021). However, precisely which constituent processes of NVC are altered in healthy aging remains unknown. We hypothesize that, as neurometabolic demand increases, adverse effects on neural function aggregate due to the inability of metabolite supply to keep pace with metabolic demand (the metabolic insufficiency hypothesis; Shetty et al. 2011; Turner 2021; see also Zhao et al. 2021). Mitochondrial production of adenosine triphosphate (ATP) decreases with age (Harper et al. 2004), while concentrations of reactive oxygen species increase (Leutner et al. 2001), damaging cells and disrupting calcium homeostasis (Chandran et al. 2019). This age-related oxidative stress also disrupts functional hyperemia by limiting nitric oxide signaling and damaging vascular endothelial cells (Podlutsky et al. 2009; Trott et al. 2011), leading to stiffening of arteries and arterioles (Kohn et al. 2015; Zimmerman et al. 2021). Oxidative stress may also limit the ability of astrocytes to produce functional hyperemia in healthy aging (cf. Jiang and Cadenas 2014; Abdelkarim et al. 2019). In this case, age-related impairment of NVC would also be expected to disrupt coordination of the fundamental supply–demand relationship between neurometabolic activity and subsequent delivery of nutrients to neurons. Previous work by Hutchison, Lu, et al. (2013) and Hutchison, Shokri-Kojori, et al. (2013) used cfMRI to explore age-reductions in BOLD signal in terms of the underlying neurophysiologic quantities of relative cerebral metabolic rate of oxygen (CMRO₂, tier I), the NVC ratio (tier II), and relative cerebral blood flow (CBF, tier III) in response to visual stimulation. Their observation of age-equivalent blood flow despite age-related increases in oxygen metabolism, yielding a significantly greater NVC ratio in younger than in older adults, supports the hypothesis of age-related disruption of the glial intermediaries of the neural-glial-vascular system.

Effects of metabolic demand on neural systems complicate interpretations of results from studies of healthy human aging which invoke assumptions related to the NVC system. NVC is affected by different aspects of experimental paradigms in functional magnetic resonance imaging (fMRI) research, including the nature of a presented stimulus (e.g. duration, intensity; Liang et al. 2013), the parameters of the task that participants perform (e.g. task difficulty, speed/accuracy demands; Moradi et al. 2012), or alteration of vascular tone from exogenous neuropharmacological agents (e.g. caffeine; Perthen et al. 2008; Griffeth et al. 2011). The strength of NVC also varies spatially as a function of cortical region (Chiarelli et al. 2007; Ances et al. 2009). In the present study, building on prior work (Hutchison, Lu, et al. 2013; Hutchison, Shokri-Kojori, et al. 2013), we modulated demand on the NVC system in occipital lobe by parametrically varying the frequency of a visual stimulus presented to older and younger participants during measurement of BOLD and its neurophysiologic constituents using calibrated fMRI scanning.

Materials and methods

Participants

A total of 60 participants were enrolled in this study at 2 sites. Thirty-three younger (M_age = 22.5, SD_age = 3.3, 14 female) and 27 older (M_age = 60.5, SD_age = 6.8, 17 female) participants were recruited from the greater Dallas-Fort Worth Metroplex and the Baltimore-Washington metropolitan area through posted flyers, advertisements placed online and through local radio stations, and word-of-mouth referrals. All participants provided informed written consent prior to scanning, and all were compensated financially for their participation. Procedures were jointly approved by the Institutional Review Boards of the University of Texas at Dallas, the University of Texas Southwestern Medical Center, and Johns Hopkins University.

Participants were screened for eligibility prior to study enrollment. General exclusion criteria were the presence of any intracranial or intraocular metal, or implanted devices (e.g. pacemaker, neurostimulation system), or if they were pregnant or claustrophobic. To limit additional risk or discomfort during hypercapnia, exclusion criteria also included a history of chronic obstructive pulmonary disease, interstitial lung disease, asthma, sarcoidosis, tuberculosis, asbestosis, silicosis, emphysema, systemic lupus erythematosus, pulmonary hypertension, or airway obstruction. To control for potential vascular confounds, exclusion criteria also included a history of transient ischemic attack, cerebrovascular accident (stroke), arteriovenous malformation, cardiac disease, angina, myocardial infarction (heart attack), congestive heart failure, coronary artery bypass graft, heart arrhythmia, migraine headaches (depending on frequency and the incidence of aura), or smoking (defined as smoking >1 cigarette a day, >3 times a week).

Experimental paradigm

Protocol

Inclusion and exclusion criteria were confirmed on the day of scanning. To account for potential effects on perfusion measurements due to caffeine or nicotine ingestion (Tsvetanov et al. 2021), each participant completed a questionnaire that ascertained their last consumption of caffeine, alcohol, nicotine, and any prescription or nonprescription drugs as well as any exercise and/or health supplements within 24 h prior to scanning. Once the participant entered the scanner room, a mouthpiece and a nose clip were positioned for the breathing challenge (see “Hypercapnia task”). The participant was then scanned for approximately 12 min (comprising the dual-echo cfMRI scan as well as phase-contrast (PC) under both normocapnic and hypercapnic conditions) and was withdrawn from the scanner bore afterwards. The hypercapnia apparatus was then removed, and the participant once again entered the scanner and completed 3 runs of a visual task (see “Visual task”), lasting approximately 20 min. For the remainder of the time the participant spent in the scanner, structural scans were collected (approximately 25 min; see “Structural MRI scanning parameters”). Total scan time was approximately 1 h.

Hypercapnia task

To permit calculation of CMRO₂, a hypercapnic breathing challenge was employed in which the participant breathed both room air and an iso-oxic CO₂ solution (medical-grade gas composed of 74% N₂, 21% O₂, and 5% CO₂). A mouthpiece and nose clip were positioned to ensure breathing could only occur by mouth (see Fig. 1A). During this part of scanning, the participant breathed room air through the mouthpiece for approximately 4 min, followed by the CO₂ solution for approximately 6 min. A researcher inside the scanner room (at sufficient distance to not affect field homogeneity) monitored the participant’s vital signs and switched the breathing valve between room-air and CO₂ once 4 min had passed. After scanning, the researcher switched the breathing valve back to room air, and the participant was removed from the scanner.

Fig. 1.

Depictions of (A) the hypercapnia apparatus and of (B) the paradigm and flickering checkerboard stimulus used in the visual task.

Visual task

The visual task was block-designed to maximize signal detection power (Birn et al. 2002) and to accommodate the effective repetition time (TR) of the dual-echo sequence (see “cfMRI scanning parameters”). The task was presented in 3 sequential runs, each containing 6 task blocks and 6 rest blocks. Task blocks were alternated with rest blocks, each 32 s long (both task blocks and rest blocks), and each run began with a rest block and ended with a task block (see Fig. 1B).

During rest blocks, a single fixation cross was presented on a black background. During task blocks, a single fixation cross was presented at the center of a radial yellow and blue flickering checkerboard that maximized contrast. The frequency of the flicker was parametrically modulated between 2, 4, and 8 Hz (Cliff et al. 2013) to vary the frequency of visual cortex stimulation. Flicker frequency presentation was pseudorandomized with the constraints that each condition did not follow linear, quadratic, or cubic patterns, and every condition occurred an equal number of times (i.e. twice) per run. To ensure the consistency of the participant’s focus of attention, the fixation cross presented at center screen changed in luminance every 2–6 s (mean = 4 s) throughout the task (through both task and rest blocks). Participants were instructed to indicate via thumb button press each time they saw the luminance of the cross change. Each run lasted 6 min and 24 s (48 dynamics), and all 3 runs lasted 19 min and 12 s.

MRI data acquisition

Imaging data were collected at 2 sites: (i) the Advanced Imaging Research Center at the University of Texas Southwestern Medical Center and (ii) the Department of Radiology at the Johns Hopkins University School of Medicine. Data at both sites were collected on identical-model Philips Achieva 3 Tesla magnetic resonance imaging (MRI) scanners (Philips Medical Systems, Best, the Netherlands). Both scanners were equipped with 32-channel SENSE head coils. Prior to data collection, an equivalence study was performed using the same participant (1 of the authors, H.L.) to ensure that signal measured in both scanning environments did not show site effects.

Structural MRI scanning parameters

A T1-weighted magnetization-prepared rapid gradient-echo (MPRAGE) was used to acquire structural MRI data in 160 sagittal slices (no gap), each containing a 256 × 256 matrix of 1 mm³ isovoxels (flip angle = 12°, echo time = 3.7 ms, TR = 8.1 ms).

cfMRI scanning parameters

A pseudo-continuous arterial spin labeling (pCASL) sequence (Alsop et al. 2015; Aslan et al. 2010) without background or vascular suppression was used to acquire dual-echo fMRI data with echo times at TE₁ = 13 ms and TE₂ = 30 ms, and TR = 4,000 ms (flip angle = 90°, multislice echo-planar-imaging acquisition, 22 transverse slices parallel to the AC-PC plane, voxel size 3.44 × 3.44 × 6 mm³, labeling duration = 1,400 ms, postlabeling delay = 1,450 ms). Data from Echo 1 were used to obtain CBF-weighted images, and data from Echo 2 were used to obtain BOLD images. Acquisition geometry for functional volumes was situated to ensure whole-brain coverage, with emphasis placed on acquiring both occipital (visual) and precentral (motor) cortices.

During the hypercapnia session, in order to account for variations in pCASL labeling efficiency during CO₂ inhalation (Aslan et al. 2010), PC quantitative flow MRI scans were performed before and after the cfMRI sequence. Global flow values were used to correct for individual differences in pCASL labeling efficiency using procedures described previously (Hutchison, Lu, et al. 2013; Hutchison, Shokri-Kojori, et al. 2013).

Structural MRI data processing

Freesurfer was used for cortical parcellation and reconstruction of T1-weighted images using the “recon-all” function (Fischl 2012). Individual regions of interest (ROIs) were generated in participants’ native space by segmentation of the reconstructed T1-weighted images using labels from the Desikan-Killiany atlas (Desikan et al. 2006). The structural ROI we selected to represent occipital (visual) cortex comprised cuneus, lingual gyrus, lateral occipital gyrus, and pericalcarine cortex (that surrounding the calcarine fissure) and was generated using Freesurfer’s “mri_label2vol” function.

cfMRI data preprocessing

Figure 2 depicts a diagram of the preprocessing workflow for cfMRI data. Raw images were converted to AFNI’s HEAD/BRIK format, as AFNI was used for all preprocessing and first-level data analysis. Images were checked for spikes using AFNI’s “3dDespike” function (values >4 standard deviations (SDs) above the mean value), and any detected spikes were attenuated to between 2.5 and 4 SDs above the mean value based on their location in the distribution of values in the volume. BOLD images (i.e. Echo 2 images) from each run were motion-corrected using AFNI’s “3dvolreg” function by a 6-degree-of-freedom rigid-body transformation (3 translation and 3 rotation parameters) to the fifth functional volume within the time-series of the run. Due to the inherently greater gray-white contrast of BOLD data relative to pCASL (i.e. Echo 1) data, and because data from both echoes were gathered with identical geometry, the transformation matrix output by AFNI’s “3dvolreg” function for BOLD volumes was applied to pCASL volumes. Volumes from subsequent runs of the task were coregistered to volumes from the first run of the task using AFNI’s “3dAllineate” function.

Fig. 2.

Preprocessing pipeline for BOLD and relative CBF data.

Relative CBF maps were generated from pCASL label and control images using the surround-subtraction method (Liu and Wong 2005; Lu et al. 2006). Surround-subtraction minimizes BOLD influence on relative CBF data and upsamples those data (Echo 1) to an effective TR of 4 s, matching the TR of the BOLD (Echo 2) data. BOLD volumes were averaged with neighboring BOLD volumes in time to remove the contaminating effects of the pCASL labeling pulse prior to acquisition of label images to attenuate high-frequency noise in the BOLD time-series data (Liu and Wong 2005; Hutchison, Lu, et al. 2013; Hutchison, Shokri-Kojori, et al. 2013; Hubbard et al. 2017; Sivakolundu et al. 2020; West et al. 2021). Functional data were then coregistered to a participant’s structural T1 image in native space using AFNI’s “align_epi_anat.py” script, which exhibited superior intermodal coregistration performance compared to AFNI’s “3dAllineate” function and resampled into structural-resolution voxels (1 mm³). Data were then spatially smoothed by convolving an 8-mm FWHM Gaussian kernel with BOLD and relative CBF volumes, high-pass filtered to remove low-frequency artifacts (e.g. scanner drift), and extracranial voxels were removed.

cfMRI data analysis

To calculate percent signal change (PSC) estimates for each modality (BOLD and CBF; Davis et al. 1998; Hoge et al. 1999a, 1999b), a reduced general linear model (GLM) was run with a single boxcar regressor using AFNI’s “3dDeconvolve” function to identify voxels where the visual task reliably elicited signal. The resulting parameter map was intersected with the occipital ROI generated from Freesurfer, which served as the structural mask. Task data were functionally masked using an approach that allowed an equal number of voxels to be analyzed for each participant and for each condition. t-score maps output from the single-regressor GLM were converted to rank-scores within the structural ROI, and a conjunction image was created using the minimum t-score of BOLD and relative CBF data (Friston et al. 2005; Hutchison, Lu, et al. 2013; Hutchison, Shokri-Kojori, et al. 2013; see also Schäfer et al. 2012). The top 5,000 voxels of this conjunction image served as the ROI for the initial by-condition analysis, which was then expanded to concentric ROIs of 10,000, 20,000, 30,000, and 40,000 voxels. Three ROIs were chosen for analysis: the focal ROI (the top 5,000 voxels), the proximal ROI (the top 10,000 voxels but not including the focal ROI), and the distal ROI (the top 20,000 voxels but not including the proximal or focal ROIs). Finally, a second GLM used 3 individual boxcar regressors to model PSC in the resulting ROI for each modality (BOLD and relative CBF) and each condition (2, 4, and 8 Hz). The resulting PSC estimates for BOLD and relative CBF from the second GLM were then used in subsequent analyses to calculate both relative CMRO₂ and the NVC ratio. It is important to note that these measures all reflect changes from baseline rather than absolute values.

Calculation of M, CMRO₂, and NVC ratio

Figure 3 depicts the schematic for the calculation of M, CMRO₂, and NVC ratio. The deoxyhemoglobin dilution model of BOLD signal change (Davis et al. 1998; Hoge et al. 1999a, 1999b) allows calculation of CMRO₂ change from BOLD signal change and CBF change using the relationship in Equation (1):

(1)

Fig. 3.

(A) Schematic for the calculation of M, CMRO₂, and NVC ratio. Relative CBF and BOLD measured during hypercapnia are used to calculate M (the theoretical ceiling of BOLD signal change), which can be combined with relative CBF and BOLD measured during the visual task to calculate relative CMRO₂ (and NVC ratio). (B) Each tier of the neural-glial-vascular model can be represented by one measure: CMRO₂ (tier I) represents neurometabolic activity, CBF (tier III) represents vascular responses to changes in that activity, and the NVC ratio (tier II) represents the intermediaries that link blood flow and metabolism.

In Equation (1), M is the calibration constant and can be conceptualized either as a baseline deoxyhemoglobin level (Griffeth and Buxton 2011), or as the theoretical ceiling of BOLD signal change (Bulte et al. 2012), α is a constant relating CBF to cerebral blood volume (Grubb Jr. et al. 1974; Chen and Pike 2009), and β is a constant based on the strength of the magnetic field (Lu and van Zijl 2005). This equation, in addition to the measured quantities of BOLD PSC (ΔBOLD/BOLD₀) and CBF ratio (CBF/CBF₀) and constants α (=0.23) and β (=1.3), contains 2 unknowns. The first of these unknowns and the ultimate goal of the calculation, CMRO₂ ratio (CMRO₂/CMRO_2|0; i.e. CMRO₂ PSC/100 + 1), can only be determined after the second unknown, M, has been estimated via the hypercapnia calibration experiment. Because CO₂ is a vasodilator to which cerebral vasculature is particularly sensitive (Ainslie and Burgess, 2008), imaging participants while breathing a 5% CO₂ solution allows for calculation of M relative to BOLD signal change and CBF change under conditions of negligible CMRO₂ change (but see Zappe et al. 2008; Xu et al. 2011; Thesen et al. 2012). ROI-level PSC estimates for both modalities (BOLD and CBF) during hypercapnia were used to calculate the M value for each participant’s visual cortex using a modified form of Equation (1):

(2)

After regional M values were calculated from hypercapnic calibration data, CMRO₂ change was calculated from BOLD PSC and CBF changes evoked during performance of the visual task using another modified form of Equation (1) that has been solved for CMRO₂:

(3)

The NVC ratio, relating stimulus-evoked CBF to stimulus-evoked CMRO₂, can be calculated by dividing CBF PSC by CMRO₂ PSC. Thus, the NVC ratio represents the relative amount of stimulus-evoked CBF change in response to a given unit change in CMRO₂. However, the quotients of CBF PSC and CMRO₂ PSC have inordinately high variance, acting similarly to a Cauchy random variable (Cassela and Berger 2002), making them, by definition, highly inefficient estimators of the NVC ratio. We instead chose an approach used previously (Kim et al. 1999; Hoge et al. 1999a, 1999b; Wey et al. 2011) to quantify the NVC ratio with much lower variance for a given group and condition using the parameter estimates from a GLM that predicts CMRO₂ from CBF. Using this method, the NVC ratio is the reciprocal of the regression parameter estimate produced by this GLM fit by ordinary least squares. Using a first-order approximation (Wolter 2007), the variance of the reciprocal of the parameter estimate from the GLM predicting CMRO₂ from CBF is given by Equation (4):

(4)

Quality assurance

Eleven participants were excluded from subsequent analyses due to data quality. One younger participant was excluded for noncompliance. One older participant was later determined to have been diagnosed with sleep apnea (and thus may have had altered sensitivity to CO₂). One younger participant was excluded for having an outlier M-value (>10 SDs from both the mean of the group and the entire sample). Three participants (2 young and 1 old) were excluded for having negative baseline CBF values, which led to large negative or incalculable CMRO₂ values. Three older participants were excluded for excessive head motion (in each case, rotation about the x-axis was >3°). One younger participant was excluded for aliasing artifacts on the MPRAGE structural image. Finally, 1 older participant was excluded due to failures of cortical parcellation and reconstruction during Freesurfer processing.

Results

Behavioral performance

Younger adults and older adults showed comparable performance on the in-scanner fixation task, confirming visual fixation and attention across groups. Formal tests of an age-group difference in reaction time (RT) using an independent-samples t-test revealed that younger adults (M_Y = 728.20 ms, SEM_Y = 72.39 ms) and older adults (M_O = 741.03 ms, SEM_O = 85.03 ms) did not differ significantly in RT on the visual attention task, t(58) = 0.152 (P = 0.880). Similar tests for group differences in accuracy rates on the task revealed that younger adults (M_Y = 83.08%) and older adults (M_O = 82.30%) did not differ significantly in accuracy, t(58) = 0.261 (P = 0.794).

Age differences in cortical volume

Prior to analysis of functional neurophysiologic data, we assessed whether there were significant differences in the volume of occipital cortex between age-groups. Group means of voxel counts within occipital lobe as defined by cortical parcellation in Freesurfer were numerically similar, and a formal test of significance (independent-samples t-test) revealed that younger (M_Y = 87,819 mm³, SEM_Y = 2339 mm³) and older adults (M_O = 86,521 mm³, SEM_O = 2088 mm³) did not differ significantly in the volume of occipital cortex, t(58) = 0.403 (P = 0.689).

Effects of age on BOLD, blood flow, oxygen metabolism, and NVC

In the focal region of visual cortex most stimulated by the flickering checkerboard, represented by the 5,000 mm³ with highest signal, we observed greater BOLD signal in younger relative to older adults across stimulus frequency conditions, F(1, 47) = 11.450 (P = 0.001; see Fig. 4). Conversely, we observed greater CMRO₂ in older compared to younger adults across stimulus frequency conditions, F(1, 47) = 7.018 (P = 0.011). There was no significant difference in relative CBF between younger and older adults, F(1, 47) = 0.781 (P = 0.381). These results replicate previous findings from our lab (Hutchison, Lu, et al. 2013; Hutchison, Shokri-Kojori, et al. 2013).

Fig. 4.

Main effects of age-group on (A) BOLD, (B) relative CBF, (C) relative CMRO₂, and (D) NVC ratio, main effects of flicker frequency, and age-group × flicker frequency interaction effects on (E) BOLD, (F) relative CBF, and (G) relative CMRO₂. Both (A) BOLD and (D) NVC ratio were significantly higher in younger than in older adults, whereas relative (C) CMRO₂ was significantly higher in older than in younger adults, and no group difference was observed in (B) relative CBF. The effects of flicker frequency on both (E) BOLD and (F) relative CBF were significant, and the quadratic main effect of flicker frequency was significant on (G) relative CMRO₂. The only significant age-group × flicker frequency interaction effect was observed in NVC ratio (see Fig. 5B, top row).

Calculation of CMRO₂ depends critically upon the calibration constant M (see Materials and methods), the theoretical maximum change in BOLD signal. M is known to vary not only across participants but also across cortical regions and even from 1 scanning session to another (Leontiev et al. 2007; Bright et al. 2019). Age-related differences in M would have important consequences for the assessment of age-related differences in CMRO₂ (and thus also in NVC ratio). Inspection of Table 1 suggests that differences in M are minimal between younger (M_Y = 9.492, SD_Y = 2.606) and older adults (M_O = 10.722, SD_O = 2.946). To assess whether these differences in M could influence the estimation of age-group differences in CMRO₂, we formally tested for age-group differences in M using an independent-samples t-test. There was no significant age-group difference in M, t(47) = 1.477 (P = 0.146).

Table 1.

Group means and SDs of M and quantities used in the M calculation.

Quantity	Mean, younger group	SD, younger group	Mean, older group	SD, older group
BOLD PSC (hypercapnia)	9.012	2.545	10.273	2.707
CBF PSC (hypercapnia)	25.134	15.098	26.821	11.706
M (%)	9.492	2.606	10.722	2.946

We observed that the NVC ratio was higher in younger than in older adults across all conditions (see Fig. 4). An independent samples t-test revealed that the NVC ratio was significantly higher in younger than in older adults, t(47) = 4.027 (P = 2.044 × 10⁻⁴).

When examining the neurophysiologic measures in the focal region of visual cortex, no age-group × flicker frequency interaction effects were significant for BOLD (d = 0.018), relative CBF (d = 0.145), or relative CMRO₂ (d = 0.252). Effectively, the nature of the change in measured signal with flicker frequency did not depend on age-group for these quantities. The mixed-effects linear model did reveal a significant age-group × flicker frequency interaction effect on the NVC ratio in the focal ROI (see Table 2 and Fig. 5; see also Supplemental Information).

Table 2.

Age-group × flicker frequency interaction effect on NVC ratio by mask size, including results of individual t-contrasts between 2 Hz versus 4 Hz and 2 Hz versus 8 Hz conditions.

Region of occipital lobe	F-score (df = 2, 86)	P-value	t-score, 2 Hz versus 4 Hz (df = 86)	P-value	t-score, 2 Hz versus 8 Hz (df = 86)	P-value
Focal (innermost 5,000 mm3)	9.325	2.160 × 10  −4	2.295	2.415 × 10  −2	4.317	4.216 × 10  −5
Proximal (surrounding 5,000 mm3)	3.781	2.668 × 10  −2	1.029	3.065 × 10  −3	2.712	8.071 × 10  −3
Distal (outermost 10,000 mm3)	1.526	2.232 × 10−1	0.980	3.298 × 10−1	1.746	8.441 × 10−1

Note. Bolded P-values represent significant results.

Fig. 5.

A) ROIs from a representative participant. Colors indicate the size of the maximum extent of the ROI: Focal (red; 5,000 mm³), proximal (yellow; 10,000 mm³), and distal (green; 20,000 mm³). Also depicted for reference are more outlying ROIs in blue (30,000 mm³) and purple (40,000 mm³). B) The first 2 columns depict PSC values of CMRO₂ by CBF for each frequency in younger (left) and older (center) adults, and the right column depicts the NVC ratio for each group at each frequency, calculated as the reciprocals of the by-condition parameter estimates from the left two columns. Each row depicts results for each of the focal (top), proximal (middle), and distal (bottom) regions corresponding to representative ROIs above in red, yellow, and green, respectively. Results for younger adults are represented in blue, and those for older adults are represented in red. Lighter and darker shades represent lower and higher frequency conditions, respectively. Error bars in the right column represent the extent to which individual observations vary from the least-squares regression line in the left and center columns for younger and older adults, respectively, and were calculated using Equation (4) (see Materials and methods). Greater variability was observed in younger adults.

Analysis of focal, proximal, and distal regional signal differences

To assess the spatial distribution of flicker frequency effects on NVC, and whether those effects varied with age, we compared NVC ratios from the focal ROI to NVC ratios from nonoverlapping concentric ROIs. Discrete volumes of visual cortex that shared a geometric center with the Focal ROI were chosen to be included for analysis. These volumes were labeled as the proximal ROI (the top 10,000 mm³ excluding the focal ROI) and the distal ROI (the top 20,000 mm³ excluding the proximal and focal ROIs). Significant interaction effects were not observed in more outlying ROIs (i.e. 30,000 and 40,000 mm³), although significant age main-effects remained. Representative masks are depicted in Fig. 5A.

Greater NVC ratio in younger relative to older adults across stimulus frequency conditions was observed at all volumes. This main effect of age-group was significant at all mask sizes. Additionally, the NVC ratio in the 4 and 8 Hz conditions was higher than in the 2 Hz condition in younger adults in all mask sizes, but the NVC ratio did not appear to change with condition in older adults at any mask size. Importantly, a formal test of significance revealed that the age-group × flicker frequency interaction effect was significant in the focal (Cohen’s d = 1.67) and proximal (d = 1.11) ROIs but not in the distal ROI (see Table 2 and Fig. 5).

The parameters α and β are fundamental in the calculation of CMRO₂ using the deoxyhemoglobin dilution model and thus in calculation of the NVC ratio. α represents the nonlinear coupling between CBF and CBV (Grubb Jr. et al. 1974; Chen and Pike 2009), whereas β represents features of the anatomy of microvasculature (Bright et al. 2019) and depends on the standing magnetic field strength. Although work investigating age-group differences in these parameters remains to be done, it is supposable that such differences may develop with age. To investigate the potential influences of age-group differences in these parameters upon our results, we conducted sensitivity analyses, wherein we recalculated M and CMRO₂ values while varying both α and β between extremes of values reported previously in the calibrated fMRI literature (Grubb Jr. et al. 1974; Pasley et al. 2007; Chen and Pike 2009; Mohtasib et al. 2012). Results of these analyses indicated that our result of a significant age-group × flicker frequency interaction effect was robust to group-equivalent and group-independent variation in α and β (see Supplemental Information for more detail).

A substantial gap exists in calibrated fMRI studies of aging in that no prior studies have empirically tested for age-group differences in α and β. Mohtasib et al. (2012) did simulate the effects of such differences, positing that, across the lifespan, α would slowly increase (at a rate of half a percent per year), while β would decrease at the same rate. However, it is important to note that this rate was extrapolated from a single result showing age-reductions in gray-matter perfusion (Parkes et al. 2004). As discussed above, the relationships between α and β to BOLD and CMRO₂ are complex, both mathematically (acting in nonlinear fashion) and conceptually (representing very specific physical or physiologic relationships), and as such, while both quantities probably vary with age, more work is required to determine the precise fashion in which α and β change over the lifespan.

Discussion

Using a dual-echo calibrated fMRI approach, we separated the BOLD signal into its neurophysiologic constituents of CBF and CMRO₂ to examine the NVC ratio, and we assessed the separate and combined effects of age and modulations in demand on NVC in the visual system. Results from this approach yielded 3 key findings. First, in response to flickering checkerboard stimuli, we observed reduced BOLD response with age across primary visual cortex, together with age-group equivalent blood flow responses and age-increases in oxygen metabolism, resulting in overall lower NVC ratios in older than in younger adults. These results replicate earlier results from our laboratories (Hutchison, Lu, et al. 2013; Hutchison, Shokri-Kojori, et al. 2013) and support a growing body of evidence in animal (e.g. Grosche et al. 2013; Jessen et al. 2017; Tarantini et al. 2017, 2021; Yabluchanskiy et al. 2021) and human studies (Howarth et al. 2021; Hutchison, Lu, et al. 2013; Hutchison, Shokri-Kojori, et al. 2013; Tsvetanov et al. 2021), showing that NVC is disrupted in aging. Second, analysis of BOLD and its physiologic constituents (i.e. changes in blood flow and oxygen metabolism) at multiple frequencies indicated that NVC was sensitive to the modulation of visual system demand in younger adults but was insensitive to demand in older adults. Previous work has shown that the intact NVC system of healthy younger adults is responsive to variations in stimulus characteristics (e.g. intensity, duration; Leontiev et al. 2007; Lin et al. 2009, 2010; Moradi et al. 2012; Liang et al. 2013; Moradi and Buxton 2013; Buxton et al. 2014; Mark et al. 2015). Results observed here demonstrate that the NVC system is also responsive to variations in stimulus frequency in younger but not in older adults. Third, while the NVC ratio was greater in younger than older adults throughout the primary visual cortex, the sensitivity of NVC in younger adults to stimulus characteristics was limited to the focal regions of visual cortex. These results are congruent with murine and primate studies demonstrating the deleterious effects of aging upon the systems that subserve NVC (Peters and Sethares 2004; Tarantini et al. 2017).

The combination of our blood-flow change measurements and BOLD signal, together with the hypercapnia calibration experiment, permitted us to calculate both the rate of oxygen metabolism and the NVC ratio within brain parenchyma and to assess age-group differences in these quantities. In the focal ROI, we observed age-equivalent stimulus-evoked blood flow together with age-related increases in stimulus-evoked oxygen metabolism. These patterns explain both the higher BOLD signal and the higher NVC ratio (i.e. the change in blood flow per unit change in oxygen metabolism) observed in younger relative to older adults. The greater oxygen metabolism we observed in older adults would not be explained by observations of age-increases in baseline oxygen metabolism in prior work (Lu et al. 2011), as such an elevated baseline would yield the opposite result (i.e. greater oxygen metabolism in younger than in older adults). Rather, our result of age-decreases in BOLD signal reflects age-differential increases in stimulus-evoked oxygen metabolism together with age-equivalence in stimulus-evoked blood flow (cf. Ross et al. 1997). Further, physiologic studies present compelling evidence that the metabolic rate of oxygen in brain tissue actually increases in the presence of restricted O₂ availability (Harik et al. 1995; Hochachka et al. 1994; Richards et al. 2007; Smith et al. 2013; Xu et al. 2012). These observations in the focal ROI replicate prior research from our laboratories (Hutchison, Lu, et al. 2013; Hutchison, Shokri-Kojori, et al. 2013) in a larger sample using a similar task.

The calibrated fMRI method permitted us to assess the neural and vascular phenomena underlying age-related changes in BOLD signal. Change in BOLD signal over time, known as the hemodynamic response function (HRF), has been used extensively in the study of both healthy (e.g. Rypma and D'Esposito 2000; Rypma et al. 2001; Cabeza 2002; Reuter-Lorenz 2002; Aizenstein et al. 2004; Hedden and Gabrieli 2004; Rypma et al. 2006; Fabiani et al. 2014; Thomas et al. 2014) and pathological aging of the brain (e.g. Machulda et al. 2003; Hao et al. 2005; Rombouts et al. 2005; Cole et al. 2006; Goekoop et al. 2004; Hämäläinen et al. 2007). The HRF features a reliable time-course (over a range of 15–30 s following stimulus) and regular shape in healthy, young adults, though regional and interindividual variability has been observed (Aguirre et al. 1998; Miezin et al. 2000; Handwerker et al. 2004). Study of age-group differences in shape parameters of this function have shown significant age-related increases in variability (D'Esposito et al. 1999, 2003; Handwerker et al. 2007; Tsvetanov et al. 2015, 2019; Abdelkarim et al. 2019; Turner, Fischer, et al. 2019; West et al. 2019). The age-related reductions in NVC we observed in this study provide a plausible mechanism for this increased variability. Disruption in the delivery of oxygen (and other metabolites) in the context of the increased neurometabolic activity that we observed in older adults would be expected to alter processes underlying the HRF. For instance, shape parameters reflecting the timing of hemodynamic activity (e.g. time-to-peak and time-to-trough; Turner, Hubbard et al. 2019) which have been observed to increase with age (West et al. 2019) could reflect restrictions in the speed with which neural assemblies generate action potentials required for timely signal transmission, as has been observed in murine models of aging (e.g. Insel et al. 2012). Importantly, the age-reductions we observed in the NVC ratio do not reflect reduced neurometabolic activity in older adults. On the contrary, the reduced NVC ratio in older adults actually reflects disproportionate increases in stimulus-evoked neurometabolic activity (i.e. demand) which cannot be matched sufficiently by the blood flow response (i.e. supply). Such a desynchronization between blood supply and neurometabolic demand would lead to disruption of the homeostatic mechanisms necessary to maintain ionic gradients at resting potential (Abdelkarim et al. 2019), in turn limiting the frequency with which neural assemblies transmit signals reflected in age-related HRF shape changes.

Parametric modulation of the checkerboard flicker frequency (2, 4, and 8 Hz) revealed that the linear BOLD increases observed here, and also observed in many prior studies (e.g. Kwong et al. 1992; Thomas and Menon 1998; Zhu et al. 1998; Ozus et al. 2001; Hagenbeek et al. 2002; Pastor et al. 2003; Singh et al. 2003; Parkes et al. 2004; Cliff et al. 2013), result from complicated relationships between the physiologic components underlying the BOLD signal (i.e. stimulus-evoked blood flow and oxygen metabolism) and stimulus frequency. While stimulus-evoked blood flow increases between 2 and 8 Hz, stimulus-evoked oxygen metabolism assumes a quadratic pattern, increasing from 2 to 4 Hz, but decreasing from 4 to 8 Hz. Previous studies investigating oxygen metabolism changes with flicker frequency in visual cortex have also identified such a CMRO₂ peak at 4 Hz (Zhu et al. 1998; Vafaee et al. 1999; Lin et al. 2010; see also Jessen et al. 2017). A study by Lin et al. (2010) also observed peak oxygen extraction fraction and ATP concentration at 8 Hz, suggesting a shift in the source metabolites utilized for ATP production at lower versus higher frequencies (i.e. aerobic metabolism at lower frequencies and anaerobic glycolysis at higher frequencies). More work is certainly needed to understand the complexities underlying the apparently straightforward monotonic increases in BOLD signal with increasing frequency and possibly cognitive demand (e.g. Rypma et al. 1999; Cappell et al. 2010; see also Csipo et al. 2021).

The introduction of stimulus frequency into a study of the effects of aging allowed us to test for interactions between the effects of stimulus frequency and age-group. Within the focal ROI, no such interaction effects were observed for BOLD signal, stimulus-evoked blood flow, or stimulus-evoked oxygen metabolism, but a significant interaction was observed for the NVC ratio. Interestingly, this interaction effect was attenuated in more distal ROIs (outside the metabolic focal point). In other words, the frequency dependence of stimulus-induced change in blood flow per unit change in oxygen metabolism was different for younger and older adults. In younger adults, the NVC ratio actually increased with increasing frequency; similar changes in NVC ratio with varying stimulus properties have been reported previously for both visual (e.g. Moradi et al. 2012; Liang et al. 2013) and cognitive (Csipo et al. 2021) stimuli. By contrast, for older adults, the NVC ratio remained relatively unchanged with increasing frequency.

NVC increases with increasing frequency could serve to regulate oxygen delivery mechanisms (Hyder et al. 1998; Hudetz 1999; Lin et al. 2010; Celaya-Alcala et al. 2021) as metabolic demand increases. Effective preservation of oxygen delivery to tissue is a critical homeostatic process because very little oxygen is available to neurons and glia that require it from sources other than erythrocytes flowing through the vasculature (Hyder et al. 1998; Buxton 2010). Successful maintenance of tissue oxygen availability could require additional vascular support from the immediately surrounding areas of cortex in younger adults. Such a vascular support network that extends beyond immediate regions of active tissue has been observed in other studies (e.g. West et al. 2021). Age-related disruptions to the NVC system would result in mismatches between not only stimulus-evoked blood flow and stimulus-evoked oxygen metabolism but also in mismatches between oxygen extraction, which could result in failures to maintain oxygen delivery due to diminished vascular support in cortical areas immediately surrounding the focal regions of neurometabolic activity (cf. Hubbard et al. 2021). Future studies that include the collection of hyperoxia data will be necessary to more directly examine the role of tissue oxygen delivery in NVC differences between age-groups.

In contrast to our finding that NVC sensitivity to stimulus frequency in younger adults is specific to the center of neurometabolic activity, we observed overall age-differences in the NVC ratio which persisted throughout visual cortex. Based on murine models of astrocyte function in NVC (Haydon and Carmignoto 2006; Carmignoto and Gómez-Gonzalo 2010; Petzold and Murthy 2011; Otsu et al. 2015), we hypothesize that the spatially persistent NVC age main-effects might reflect brain-wide structural and functional changes to astrocytes (Figley and Stroman, 2011; Tarantini et al. 2017; Howarth et al. 2017) that pervade the aging brain. Such changes include astrocyte hypertrophy (Rodríguez-Arellano et al. 2016), loss of endfoot domain specificity (Bors et al. 2018), and toxic reactivity (e.g. increased inflammatory signaling; see Abdelkarim et al. 2019). Reductions in cerebrovascular reactivity (CVR) do not account for this result (Taneja et al. 2020). Indeed, studies of CVR changes with age have found minimal age-differences in occipital cortex (Peng et al. 2018). More work is certainly needed to explicitly test this hypothesis. Work underway in our laboratories is examining the specific nature of individual NVC components and how they contribute to the phenomena we observed here.

We observed age-related increases in stimulus-evoked oxygen metabolism, similar to previous studies (e.g. Lu et al. 2011; Hutchison, Lu, et al. 2013; Hutchison, Shokri-Kojori, et al. 2013; Jessen et al. 2017). The mechanisms underlying this phenomenon are not yet known. They may be related, however, to known decreases in ATP availability with age (for review, see Harper et al. 2004). Experimental work examining the relationship between ATP availability and oxygen metabolism remains to be done (but see Salin et al. 2015; Andersen et al. 2019). However, one possibility is that the neurometabolic system responds to this ATP deficit through increased aerobic metabolism (i.e. increased CMRO₂) to generate higher levels of ATP in mitochondria (Camandola and Mattson 2017). Despite these age-related increases in oxygen metabolism, however, ATP deficits persist. One mechanistic rationale for such mitochondrial inefficiency (i.e. producing less ATP per molecule of oxygen) might be to reduce the production of inflammatory reactive oxygen species byproducts (Brand 2000; Mookerjee et al. 2010). A second possibility involves the levels of oxygen available to neural tissue. Xu et al. (2012) found that manipulation of the fraction of inspired O₂ has a dose-dependent effect on brain oxygen metabolism. Their results showed that oxygen metabolism increases significantly during hypoxia but actually decreases during hyperoxia. Thus, a dearth of oxygen that may accumulate following supply–demand mismatches by the NVC system may produce conditions wherein oxygen metabolism in older adults is actually higher. Irrespective of the mechanism, our results show that age-related increases in oxygen metabolism remain unmatched by blood flow responses, implicating glial intermediaries as points of failure within the aged neural-glial-vascular system.

Conclusion

In this study, we assessed age-differences in the effects of stimulus frequency on the NVC system in visual cortex. We observed age-equivalent stimulus-evoked blood flow (tier III of the neural-glial-vascular model) in the presence of greater stimulus-evoked oxygen metabolism (tier I of the neural-glial-vascular model) in older compared to younger adults. Measurement of these processes, via measurement of BOLD signal, permitted us to calculate the NVC ratio (i.e. the change in blood flow per unit change in oxygen metabolism). This NVC ratio (tier II of the neural-glial-vascular model) was reliably greater in younger than in older adults regardless of stimulus frequency or the volume of cortex over which it was measured. Moreover, blood flow increased disproportionately with stimulus-evoked metabolic increases (specifically within the center of neurometabolic activity) in younger but not in older adults. The age-related disruption of NVC that we observed rendered the older system unresponsive to increases in stimulus frequency. These results implicate the glial component of the neural-glial-vascular signaling system as a significant contributor to age-related decline in brain function.

Funding

This work was supported by National Institute of Health grants (R01AG047972 to B.R. and H.L., R01AG029523 to B.R., R01NS106711 to H.L., R01NS106702 to H.L., F32MH114525 to N.A.H.; P20GM130461[6026] to N.A.H.), the Brain and Behavior Research Foundation and Vital Projects Investigator Fund, Inc. (N.A.H.), and the Nebraska Biomedical Research Development Funds (N.A.H.).

Data availability

The data that support the findings of this study are available from the corresponding author (BR) upon reasonable request.

Conflict of interest statement

None declared.