Calorie Restriction Reduces the Influence of Glucoregulatory Dysfunction on Regional Brain Volume in Aged Rhesus Monkeys

Forty-four rhesus monkeys (Macaca mulatta) between 19 and 31 years of age were included in this study. Seventeen animals were fed ad libitum for approximately 8 h/day, whereas the remaining 27 subjects were fed 30% fewer calories relative to their own baseline intake. Length of CR diet was 12–17 years and initiated in middle age. Demographic data are available in Table 1. These monkeys were the remaining subjects of a longitudinal CR project begun at the Wisconsin National Primate Research Center in 1989. Details of the CR manipulation, housing, and husbandry have been described previously (13,27).

Scan session and image acquisition parameters have been detailed elsewhere (28). Briefly, animals were anesthetized while images were acquired using a General Electric 3.0 T scanner (GE Medical Systems, Milwaukee, WI). T1-weighted volumetric scans were used to determine regional volumes. Diffusion tensor imaging was acquired to quantitatively estimate coherence of white matter tracts by computing fractional anisotropy, axial diffusivity, and radial diffusivity, whereas microstructural density of brain parenchyma was determined using mean diffusivity (29). T1-weighted scan parameters were: repetition time = 8.772 ms; echo time = 1.876 ms; inversion time = 600 ms; flip angle = 10°; number of excitations = 2; matrix = 256 × 224; and field of view = 160 mm. A total of 124 coronal slices were acquired with a thickness of 0.7 mm and no gap, resulting in 0.625 × 0.625 × 0.7 mm voxels. For diffusion tensor imaging, a diffusion-weighted echo-planar imaging sequence with 12 gradients was used with the following parameters: repetition time/echo time = 10,000/77.2 ms; b = 816 s/mm²; number of excitations = 6; field of view = 140 mm; in-plane matrix = 256 × 256; and slice thickness = 2.5 mm, no gap, and 35 slices. Voxels were resampled to 0.5 × 0.5 × 0.5 mm for T1-weighted and diffusion tensor imaging sequences. T2-weighted scans were acquired for global volume estimation and are described elsewhere (28).

All images were inspected to locate scan artifacts that could adversely affect the voxel-wise analyses. Three diffusion tensor-imaging scans and one volumetric scan had artifact abnormalities (motion, respiratory, phase, other) or gross misregistration with the template brain that rendered data unusable. Six additional diffusion tensor-imaging scans (controls = 2, CR = 4) had one or more 1- to 2-mm abnormalities located in the white matter of the dorsal convexity that precluded analysis.

Preprocessing of magnetic resonance images for use in voxel-wise analyses has been described previously (28,30). Briefly, T1-weighted images were segmented and normalized to 112RM-SL atlas space (28) with Diffeomorphic Anatomical Registration Through Exponentiated Lie algebra (31). Volumetric segments were smoothed using a 4-mm full width at half maximum Gaussian kernel. Fractional anisotropy, mean diffusivity, axial diffusivity, and radial diffusivity maps were computed using DTIFIT in FSL (www.fmrib.ox.ac.uk/fsl/). Diffusion tensor-imaging measurements were aligned to the 112RM-SL atlas space in SPM and smoothed with a 4-mm smoothing kernel.

We have previously described the frequently sampled glucose tolerance test procedure that provides data on basal and acute glucose and insulin signaling dynamics (13). Derived indices relevant to this report are described in Table 1. Glucoregulation measures were collected within 6 months of the magnetic resonance imaging (MRI) scan. Insulin sensitivity, or S_I, was calculated using the Minimal Method Model (32). This measure provides an accurate quantification of S_I, as well as other aspects of glucose kinetics, and is far more practical than other methods like the glucose-clamp technique. Plasma glucose was measured using the glucose oxidase method (Yellow Springs Instruments, Yellow Springs, OH). Insulin levels were determined using double antibody radioimmunoassay (Millipore, Billerica, MA). The homeostatic assessment of insulin resistance (HOMA-IR) was calculated using basal glucose and basal insulin (33). HOMA-IR was calculated in order to see if basal insulin resistance explained variance beyond its converse construct, which would be insulin sensitivity measured by S_I. Homocysteine was collected using methods previously described (30). Glucoregulatory data near the time of the MRI scan were not available for one CR and one control monkey.

Animals were classified by an expert (R.J.C.) as having normal, prediabetic/at-risk, or type 2 diabetes-like profiles using established criteria (15).

In order to independently assess (23) if glucoregulatory dysfunction might influence motor learning via S_I predicted variation in brain, an expert (A.A.W.) drew a mask on the 112RM-SL atlas to isolate the bilateral hippocampus using methods previously described (34). Mean GM volume was extracted from the ROI for all normalized monkey brains and was used in conjunction with S_I to predict performance on a motor task described next.

It was of interest to see if the relationship between insulin signaling and hippocampal volume predicted changes in motor task learning and performance. To this end, our cohort has previously been tested on the mMAP (25), which required subjects to retrieve an appetitive stimulus from a flat platform, straight rod, or curved rod (24). An automated system recorded the number of seconds necessary to reach from the home cage to the first area of the affixed apparatus (reaction time), from the first to second area (coarse motor movement), and from the second area to a small receptacle that held the stimulus (fine motor movement). We used fine motor performance data from the most difficult task (i.e., curved rod), because CR animals acquire and perform this task more quickly than controls (25). Animals did not differ on simpler tasks. To test our hypothesis regarding insulin signaling and hippocampal volume, these performance measures were correlated with hippocampal volume adjusted or not adjusted by S_I. The sample size for this analysis was 26 animals (C = 7; CR = 19).

To investigate regional brain associations of S_I, multiple regression voxel-wise analyses were conducted in SPM8 (12). For the purposes of this report, this type of regression technique produces t-statistic, color-coded result maps that are the product of a regression model performed at every voxel in the brain for a given modality. Contiguous groups of voxels that attain statistical significance, called clusters, will thus overlap with and implicate different brain regions. In this study, volumetric or diffusion tensor-imaging scans were entered as the dependent variable. The independent variable was S_I. Covariates included age at scan, sex, dietary condition, and the S_I term when testing an interaction (see below). Analyses of volume additionally covaried a global index of either gray or white matter (28). Our primary analysis of interest was testing a S_I × dietary condition interaction term to see if CR monkeys showed more volume or microstructure per S_I unit change beyond the association seen with S_I alone, suggesting a further beneficial effect. The voxel and cluster level thresholds were set at P < 0.005 (uncorrected) and P < 0.05 (corrected). Type 1 error was minimized by first using an omnibus F-contrast (P < 0.05, uncorrected) for S_I, dietary condition, and S_I × dietary condition to mask subsequent contrasts, followed by Monte Carlo simulations to estimate cluster sizes that would occur due to chance (25,30). Clusters required 280 contiguous voxels to reach significance at P < 0.05 (corrected). Reported whole-brain cluster coordinates correspond to the Saleem-Logothetis atlas (35) and are displayed on the 112RM-SL template image (28). Standard rhesus monkey atlases were used to identify fibers (36) and subcortical structures (37).

Tests were conducted using SPSS 18.0 (SPSS Inc., Chicago, IL) at an α of 0.05. Logarithmic transformations were used to adjust nonnormally distributed indices. ANOVA tested group differences for demographic and biological variables. Multiple regression was used to determine which basal and frequently sampled glucose tolerance test variables significantly explained error variance in the S_I association and interaction voxel-wise analyses. The first regression block included HOMA-IR to account for basal insulin regulation. The second block contained a priori variables of interest directly related to S_I: basal glucose and insulin, compensatory pancreatic sensitivity represented by Disposition Index (13), and glycosylated hemoglobin levels. The third block contained the vascular risk factor homocysteine.

Brain-behavior mediational models tested whether or not S_I-predicted variation in brain was correlated with motor learning and performance. To avoid circular analysis (23), an independent anatomical ROI approach was used to derive mean image signal (e.g., mean GM volume) within the bilateral hippocampus. S_I was then linearly regressed onto the hippocampal signal. The Pearson’s statistic was then used to correlate the predicted change in hippocampal volume due S_I with the mean number of seconds it took for a monkey to complete the fine motor portion of the curved rod mMAP task during initial acquisition (i.e., the learning phase) and when the animal reached proficiency (25).

See Table 1 for descriptive data and statistics. The mean age, sex composition, and total brain volume of the two dietary conditions did not differ. CR monkeys showed expected benefits in insulin signaling and related indices (22).

Six controls and zero CR monkeys were classified as having preclinical (n = 4) or diabetes-like glucoregulatory dysfunction (n = 2). Diagnosis was not a significant covariate in regression analyses.

To examine the association of S_I on regional GM, S_I was regressed onto GM volume voxels across all subjects. As indicated in Fig. 1A–C (yellow-orange areas) and Table 2, higher S_I predicted more GM in motor and somatosensory cortices. Fig. 1G depicts this relationship by illustrating the correlation between S_I and the voxel with the highest (peak) t-statistic located in left primary motor cortex. The association of S_I and GM was comparable for both control (R² = 0.516; P < 0.001) and CR (R² = 0.324; P < 0.001) monkeys.

Measurements directly or indirectly related to glucoregulation that are described in Table 1 may elucidate potential mechanisms underlying the relationship between S_I and regional GM. Therefore, follow-up multiple linear regression was conducted. The dependent variable was the predicted volume change in left motor cortex shown in Fig. 1G. Different regression models using stepwise, forward, or enter methods produced similar results (data not shown). The final adjusted regression model predicted 20.7% of the variance across both groups [F(6,35) = 4.170; P < 0.01]. Although the influence of S_I on GM was not significantly mediated by HOMA-IR or basal glucose, increased basal insulin was modestly related to less GM (R² = 0.133; P < 0.05). Higher levels of homocysteine (R² = 0.169; P < 0.05) and glycosylated hemoglobin (R² = 0.148; P < 0.05) were also associated with less frontal GM volume.

Given the salubrious effects of CR on vascular health and metabolic indices related to improved glucose regulation (22), we tested for a similar beneficial effect on the brain in CR monkeys. Thus, this contrast examined if CR monkeys had more GM volume per unit increase in S_I compared with controls beyond the association with S_I alone, suggesting a protective effect. As shown in Table 2 and depicted in Fig. 1A (purple areas), voxel clusters were present in bilateral anterior hippocampus and both inferior and middle temporal gyri. The interaction is represented in Fig. 1H using the peak t-statistic voxel. CR monkeys showed more GM as S_I increased (R² = 0.159; P < 0.05). For controls, however, higher S_I unexpectedly corresponded to less GM (R² = 0.449; P < 0.01). To validate this result, a mean index of S_I was calculated using data from all glucose tolerance tests since the start of the project in 1989, which was up to 14 assessments depending on the length of time an animal spent in the study. This index or area under the curve estimates resulted in clusters and graphs similar to those produced using the S_I value nearest to the MRI scan date (data not shown). Additional significant brain areas found in the interaction analysis were caudal perirhinal, entorhinal, and parahippocampal cortices, insula, amygdala, temporal pole, anterior cingulate cortex, and orbital and medial prefrontal cortices (Fig. 1A–F).

Multiple linear regression was simultaneously performed on each dietary condition group to detect mediators that might explain the S_I × dietary condition interaction effect. The dependent variable was the predicted change in hippocampal GM depicted in Fig. 1H. The same regression model was used as in the S_I association analysis (see research design and methods). Table 3 indicates the result. Fig. 2A–C shows that control monkeys had strong associations between HOMA-IR, basal insulin, or basal glucose and GM related to the interaction. Fig. 2D–F shows nonsignificant relationships for the same variables in CR monkeys.

We finally wished to test if predicted changes in hippocampal GM related to S_I were associated with mMAP fine motor performance during the curved rod task (24). An independent analysis was conducted using S_I measured at the time closest to scan in 7 controls and 19 CR monkeys that had successfully learned the task. As reported elsewhere (25), this task was chosen because CR monkeys performed it significantly more quickly during acquisition (CR: 4.84 s; control: 6.26 s) and after gaining proficiency (CR: 3.08 s; control: 3.73 s). Unadjusted hippocampal volume was not related to motor learning (R² = 0.034; not significant) or proficient (R² = 0.01; not significant) mMAP performance. By contrast, when first adjusting hippocampal volume by S_I, this brain measure significantly explained 12% of the variance for mMAP performance during the acquisition phase (P = 0.042) and 24.1% (P = 0.005) after monkeys became proficient at the task (25).

No clusters exceeded the minimum number of voxels required for type 1 error correction in white matter volume or diffusion tensor imaging modalities.

We hypothesized that higher insulin sensitivity, indexed by S_I, would predict more GM or denser microstructure in brain regions that are influenced by insulin signaling and impacted by neurodegenerative and neurovascular disorders. Across CR and control subjects, higher S_I was associated with more GM volume in somatosensory and motor cortices. These areas have low insulin receptor density relative to medial temporal lobe and prefrontal cortex in rodents (38). Interestingly, S_I interacted with dietary condition, for which CR monkeys with higher S_I had significantly more GM in hippocampus, insula, prefrontal cortex, and other regions with a high density of insulin receptors. Higher S_I among controls was unexpectedly related to less GM in these regions. Models suggest that prediabetes or type 2 diabetes in controls did not influence this result. Similar relationships between impaired glucoregulation and brain volume have been seen in AD patients with no history of type 2 diabetes (3). As such, mild to moderate insulin signaling dysregulation may negatively affect key brain areas.

Although the relationship between S_I and GM among controls in the interaction was unexpected, it is not likely due to assay error or a sudden change in glucoregulatory dynamics. A similar interaction result was found when using a mean S_I index derived from every glucose tolerance test conducted since 1989. More importantly, an estimate of basal insulin resistance, HOMA-IR, corresponded to less hippocampal GM in controls, a result that has been observed in humans (21). No relationship was seen between S_I and white matter volume or tissue microstructure, which may reflect the role of insulin-like growth factor signaling rather than insulin in mediating oligodendrocyte development and growth factor-mediated preservation (39). Although tissue microstructure is negatively impacted by type 2 diabetes (40), it may be due to vascular pathology or other consequences instead of antecedent decreases in insulin sensitivity.

Our results suggest that there is wide variation in how insulin signaling may affect energy metabolism and other functions in brain. The central action of insulin on reducing oxidative damage or maintaining synaptic plasticity appears to be area-specific due to receptor density and binding dynamics (2,17,38). For example, intracerebrovascular treatment of rats with low doses of streptozotocin, which is normally toxic to pancreatic insulin-secreting cells and creates a diabetes-like state, reduced downstream phosphatidylinositol-3 kinase activity primarily in hippocampus but to a much lesser extent in frontal cortex without affecting peripheral glucose regulation (41). Intracerebrovascular administration of insulin also affected adenosine 5′-triphosphate storage in hippocampus but not parietotemporal cortex (4).

By extension, the relatively moderate relationship between higher S_I and more GM in motor and somatosensory areas may be due to indirect mechanisms attributed to higher insulin concentrations. Chronic peripheral hyperinsulinemia is typically characterized by hyperglycemia and breakdown of the epithelial vasculature. Levels of glycosylated hemoglobin and homocysteine, a biomarker for vascular health, were significant mediators of the association between S_I and GM. In brain, it is conceivable that a microvascular insult combined with other neuropathologies might synergistically reduce perfusion and glucose transport into parenchyma, leading to damage (42). For example, streptozotocin alone in transgenic APP/PS1 mice produced more advanced glycation end products in microvasculature and worse spatial performance; similar pathophysiological effects were also seen in human cerebromicrovascular cells exposed to streptozotocin and amyloid β (9).

The interaction of S_I and GM between dietary groups revealed several important findings related to insulin signaling and brain in CR versus control monkeys. Ameliorative effects may be due to the direct action of insulin, although CR may act through related mechanisms such as less atherogenic dyslipidemia, lower expression of proinflammatory cytokines, and fewer prothrombotic factors due to reduced insulin resistance (43). S_I represents the capacity for insulin to facilitate glucose uptake and inhibit hepatic production. Lower S_I would correspond to a higher insulin secretion peak followed by prolonged hyperinsulinemia. In this vein, Burns and coworkers (3) found that a higher insulin area under the curve during a glucose tolerance test predicted more hippocampal GM in early AD without a history of type 2 diabetes, but not in nondemented participants. This compensatory response may reflect glucoregulatory and metabolic dysfunction that can occur in early AD (2,6). Similarly, mediational models suggest that control animals with high basal insulin, but relatively lower basal glucose levels, would benefit from reduced insulin sensitivity in areas that rely on insulin signaling for optimal neural function. Previous studies have shown that infusion of insulin or antagonism via streptozotocin affects adenosine 5′-triphosphate storage in hippocampus (4), dopamine metabolism in striatum (44), and 2-deoxy-d-[¹⁴C]glucose utilization in rat medial temporal lobe, striatum, and frontal cortex (18). All of these brain regions were implicated in the current study. Although this compensatory mechanism for rhesus monkey controls may benefit these areas, hyperinsulinemia and lower insulin sensitivity is harmful to skeletal muscle (45) and may also cause vascular damage (42).

CR monkeys showed the expected relationship between higher S_I and more GM in the S_I × dietary condition interaction. This positive correlation was similar to CR results for the S_I association analysis. We have previously established that CR in rhesus monkeys reduced basal glucose and insulin, lowered insulin resistance, and increased lean muscle intracellular receptor substrate-1 expression (14,46). Such effects for glucoregulatory indices were also found in the current sample. Either intermittent fasting or CR also lowers serum glucose and insulin, as well as protects hippocampal neurons from excitotoxic injury induced by kainite (47).

Finally, hippocampal GM volume adjusted by S_I was significantly correlated with learning and memory performance for the mMAP task, whereas there was no relationship with unadjusted volume. This analysis was independent from voxel-wise analyses to minimize circularity concerns (23). Both approaches are complementary and can elucidate potential relationships among glucoregulatory dysfunction, brain, and behavior. As S_I increases during CR, there might be less hippocampal atrophy over time that could positively influence task learning and performance. A comparable increase in insulin sensitivity among controls, by contrast, may be detrimental to hippocampus and could negatively affect mMAP performance.

There are several limitations that should be noted. It is not yet clear if improved peripheral insulin signaling in primates reflects similar processes in the brain. It must also be established if the relationship between insulin sensitivity and GM primarily reflects glucose uptake dynamics or other functions of insulin. For controls, the discrepancy between HOMA-IR and S_I findings for GM warrant caution in interpretation, although several measures of insulin sensitivity produced the same result. Given that CR monkeys have less age-related morbidity and mortality (15), there may also be a survivor bias that influenced the voxel-wise results. This bias may make current results more conservative, however, given that surviving controls are likely more resilient to age-related pathophysiology. Regarding mediational models, glucoregulatory variables are sometimes multicollinear and may affect the interpretation of coefficients but not the overall model. Although basal insulin and HOMA-IR were very highly correlated (data not shown), this relationship is sensible both physiologically and statistically. Finally, the existing data are cross-sectional, and causation cannot be inferred.

In summary, increased S_I among CR monkeys was associated with more GM in parietofrontal cortices, hippocampus, and other regions that vary in insulin receptor density and signaling. Among controls, higher S_I showed a positive relationship with GM volume in parietofrontal cortices with low insulin receptor density, but predicted less GM in structures and areas that have high receptor density. CR or CR mimetics may benefit some specific brain regions and aspects of task learning and performance. Nevertheless, additional studies are needed to validate and clarify the association between glucoregulatory dysfunction and GM volume.

This study was supported in part by the National Institutes of Health grants RR-000167, AG-011915, AG-000213, GM-007507, MH-085051, and MH-062015. This study was also supported with resources and facilities at the William S. Middleton Memorial Veterans Hospital. This research was conducted in part at a facility constructed with support from Research Facilities Improvement Program grants RR-15459-01 and RR-020141-01.

No potential conflicts of interest relevant to this article were reported.

A.A.W. researched the data, analyzed the data, and wrote the manuscript. B.B.B., E.K.K., A.S.F., A.L.A., A.S., D.B.A., R.A., and M.-L.V. offered expertise and reviewed and edited the manuscript. J.W.K. researched the glucoregulation data in addition to offering expertise and editing the manuscript. R.J.C., R.H.W., and S.C.J. contributed resources and reviewed and edited the manuscript. S.C.J. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

The authors thank R. Fisher of the University of Wisconsin-Madison and the Waisman Center for Brain Imaging at the University of Wisconsin-Madison for assistance.