|Home | About | Journals | Submit | Contact Us | Français|
Age-associated memory impairment (AAMI) occurs in many mammalian species including humans. In contrast to Alzheimer’s disease (AD) where circuit disruption occurs through neuron death, AAMI is due to circuit and synapse disruption in the absence of significant neuron loss and thus may be more amenable to prevention or treatment. We have investigated the effects of aging on pyramidal neurons and synapse density in Layer III of area 46 in dorsolateral prefrontal cortex of young and aged, male and female Rhesus monkeys (Macaca mulatta) that were tested for cognitive status through the delayed non-matching-to-sample (DNMS) and delayed response (DR) tasks. Cognitive tests revealed an age-related decrement in both acquisition and performance on DNMS. Our morphometric analyses revealed both an age-related loss of spines (33%, p< 0.05) on pyramidal cells and decreased density of axo-spinous synapses (32%, p<0.01) in layer III of area 46. In addition, there was an age-related shift in the distribution of spine types reflecting a selective vulnerability of small, thin spines, thought to be particularly plastic and linked to learning. While both synapse density and the overall spine size average of an animal were predictive of number of trials required for acquisition of DNMS (i.e., learning the task), the strongest correlate of behavior was found to be the head volume of thin spines, with no correlation between behavior and mushroom spine size or density. No synaptic index correlated with memory performance once the task was learned.
The dorsolateral prefrontal cortex (dlPFC) of nonhuman primates is highly vulnerable to aging. This vulnerability is reflected in both cognitive decline (Gallagher and Rapp, 1997, Moss, Killiany and Herndon, 1999) and structural alterations in Brodman area 46, such as spine loss (Duan et al., 2003), loss of subcortical white matter (Peters and Sethares, 2002, Peters and Rosene, 2003), and synapse loss (Peters, Sethares and Moss, 1998, Peters, Sethares and Luebke, 2008). Unlike AD in humans where extensive neuron loss occurs in dlPFC (Morrison and Hof, 1997, Bussiere et al., 2003), these alterations occur in the absence of significant neuron loss in dlPFC of aged rhesus monkey (Peters et al., 1998). Nonetheless, certain alterations correlate with the degree of cognitive decline in individual monkeys (Peters, Sethares and Luebke, 2008, Makris et al., 2007).
Our recent analyses of the interactive effects of estrogen treatment and aging in area 46 of ovariectomized female rhesus monkeys have highlighted the selective vulnerability of a particular class of spines. We demonstrated that the thin spines with small spine heads on layer III pyramidal neurons are particularly vulnerable to aging, and this same class of spines/synapses is restored by estrogen treatment (Hao et al., 2007), as is cognitive performance (Hao et al., 2007, Rapp, Morrison and Roberts, 2003). In vivo imaging studies in rodent neocortex have demonstrated that small spines are NMDA receptor (NMDAR)-dominated, highly motile, and capable of expanding and stabilizing or retracting, whereas large spines are highly stable and AMPAR-dominated (Matsuzaki et al., 2001, Kasai et al., 2003, Zuo et al., 2005, Holtmaat et al., 2006, Bourne and Harris, 2007). These observations lead to the hypothesis that small spines are required for learning new information, whereas the large stable spines mediate stable circuits linked to memory (Kasai et al., 2003, Kasai et al., 2010, Bourne and Harris, 2007). We have hypothesized that this conceptual framework may be particularly relevant to area 46 of rhesus monkey (Hao et al., 2007). The dlPFC plays a dominant role in goal-directed behavior (Goldman-Rakic, 2002, Fuster, 2008) and mediates both learning “the rules of the game” as well as dynamically updating those rules as contingencies are altered (Miller, Freedman and Wallis, 2002). Such complex functions are likely to require dynamic shifts in synaptic inputs dependent on extensive availability, turnover, and potential stabilization of the small, plastic spines. We have hypothesized that the age-related loss of such spines results in the loss of plasticity essential for optimal function of dlPFC.
The current study was designed to test this hypothesis in gonad-intact young and aged rhesus monkeys with no hormone manipulations. By performing an unbiased classification of large numbers of spines based on previously published criteria (Sorra and Harris, 2000), the spines were separated into thin and mushroom types. Though both synapse number and overall spine size were predictive of one particular domain of cognitive performance, acquisition of DNMS, the correlation was driven solely by the subclass of thin spines.
All experiments were conducted in compliance with the National Institutes of Health Guidelines for the Care and Use of Experimental Animals approved by the Institutional Animal Care and Use Committee at the University of California-Davis and Mount Sinai School of Medicine.
Six young adult (age range, 9–14 years old; mean, 10.8±0.8 years old; 4 females, 2 males) and 9 aged (age range, 22–35 years old; mean, 30.2±1.5 years old, 7 females, 2 males) Rhesus monkeys (Macaca mulatta) were used in the behavioral phase of this study and electron microscopic analyses, and a subset of nine (4 aged, 5 young) were used in the morphometric analyses of filled neurons (see below). Behavioral testing extended over approximately 2 years.
The DR testing was equivalent to the approach used in our previous studies (Hao et al., 2007, Rapp, Morrison and Roberts, 2003). Subjects watched from behind a transparent screen while one of the wells of the apparatus was baited with a food reward, and both wells were then covered. During initial training, the screen was raised immediately, allowing monkeys to displace the cover and retrieve the reward if the correct location was chosen. Testing continued until animals met a criterion of 90% correct or better across 9 consecutive blocks of 10 trials. For the next phase of testing, a 1 s retention interval was imposed between baiting the well and the opportunity to respond. Delays were implemented by lowering an opaque screen such that the monkey could no longer see the reward wells. Training with a 1 s delay continued until the monkey reached criterion (≥90% correct over 90 trials), and the demands on memory were then made progressively more difficult by imposing successively longer retention intervals of 5, 10, 15, 30, and 60 s. Each delay was tested for a total of 90 trials (30 trials/daily session; intertrial interval (ITI) = 20 s).
DNMS testing was conducted as described in detail elsewhere (Rapp, Morrison and Roberts, 2003, Rapp and Amaral, 1991). Briefly, a sample object was placed over the baited central well of the test tray and after a response, the opaque barrier was lowered to impose a pre-designated retention interval. The same item was then presented simultaneously with a novel object that covered the food reward. Objects were drawn from a large enough pool such that the paired object was novel on every trial. During the acquisition phase, a 10 sec retention interval was used until the monkeys learned the non-matching rule to a criterion of 90% correct or better (across 100 trials, 20 trials/day, inter-trial interval = 30 sec throughout testing). Once the monkeys reached criterion, DNMS performance was measured by increasing the demand on recognition memory with successively longer delays of 15, 30, 60, and 120 sec (100 trials total at each delay, 20 trials/day) and 600 sec (50 trials total, 5 trials/day).
Rhesus monkeys were perfused transcardially in groups of 2 or 3 as they completed their behavioral assessments. If all behavioral data had been acquired but perfusion was not able to be scheduled immediately, monkeys continued mock training up until the day prior to perfusion in order to avoid testing-induced variability. All animals were deeply anesthetized with ketamine hydrochloride (25 mg/kg) and pentobarbital (20–35 mg/kg, i.v.), intubated, and mechanically ventilated. The chest was opened to expose the heart and 1.5 ml of 0.5% sodium nitrate was injected into the left ventricle to facilitate vasodilation. The descending aorta was clamped immediately following intubation, and the monkeys were perfused transcardially with cold 1% paraformaldehyde in 0.1 M phosphate buffer (PB; pH 7.2) for 2 min, followed by 4% paraformaldehyde in 0.1 M PB at 250 ml/min for 10 min, after which the flow rate was reduced to 100 ml/min for 50 min.
After perfusion, the brain was removed from the skull and dissected into standardized blocks, taking care to include the entire region surrounding the principal sulcus (Brodmann’s area 46) in the frontal block. The frontal block was post-fixed for 6 hours in 4% paraformaldehyde with 0.125% gluteraldehyde in PB and then cut serially on a Vibratome. Every 1.3 millimeter(mm) of tissue was divided into one 400 μm thick section, flanked by 200 μm sections on each side, with the remainder cut into 50 μm sections, with the same sequence repeated throughout the frontal block. Thereby, a series of 400 μm thick sections within area 46 were collected for intracellular injection of Lucifer Yellow (LY; Molecular Probes, Eugene, OR), while a systematic random set of 200 μm sections were embedded for electron microscopy (EM), and the 50 μm sections were stored at −20 degrees for light microscopic immunohistochemistry. Since the precise position of the first slab is random, this results in a systematic random series of at least 8 sections throughout area 46 available for cell loading. Multiple sections are used to obtain the total number of cells, such that the sampled neurons represent a random selection throughout area 46.
Nine out of the 15 monkeys had undergone optimal perfusion and could be used for cell filling. Intracellular injection of layer III pyramidal cells with LY and quantitative analysis was performed according to methods previously described (Hao et al., 2007). Printed datasets of dendritic tree maps were used to allow identification of dendritic segments chosen in a systematic-random fashion for spine analysis, as described in our previous studies (Hao et al., 2007, Hao et al., 2006). A clear acetate sheet, containing a series of scaled concentric rings 60 μms apart was placed over the working maps of neuronal reconstructions, with the cell soma at the center. Points of intersection between dendritic branches and the circles were used to identify those segments to be used for high magnification microscopic spine analysis. Consequently, although the somata of the sampled cells were always located in layer III, the dendritic segments chosen for analysis were located in layers I through III, with occasional basal dendrites extending into layer V. Confocal z-stacks of identified intersections were captured on a Zeiss LSM 410 (Oberkochen, Germany). Each designated segment (25–50 μms in length) was located in the microscope field, and confocal stacks of 100–300 digital images separated by a z-step of 0.1 um were captured using a Plan-Apochromat 100X/1.4 NA Zeiss oil-immersion objective (fig 1A). LY was excited with an Ar/Kr laser at 488 nm (attenuation set at 10). All confocal stacks included at least 1 μm above and below the identified dendritic segment. Settings for pinhole size, aperture gain, and offset were optimized initially and then held constant throughout the study to ensure that all images were digitized under the same illumination conditions at a resolution of 256 X 256 pixels. An average of 12 high-magnification z-stacks were captured per neuron (total 39 neurons). The confocal stacks were deconvolved with AutoDeblur (Media Cybernetics, Silver Spring, MD) (fig 1B) and imported to NeuronStudio for 3-dimensional (3D) spine density, and spine size analysis (Rodriguez et al., 2008) (fig 1C). All spine measurements (total 21,000 spines) were performed in 3D from the z-stacks. The density was calculated by dividing the total number of spines present by the dendritic length of the segment. The automated measures of spine volume, spine head diameter and spine length were then imported into Matlab for classification of spines into thin and mushroom types based on categories previously established from serial section EM reconstruction of dendritic segments (Sorra and Harris, 2000) (fig 1D). A spine was labeled thin if its head was below 0.6 μm in diameter and had a max length that was at least twice as big as the head diameter. A spine was classified as mushroom if its head diameter exceeded 0.6 μm. The remainder of spines were classified as “other.” This group presumably contained primarily stubby spines. However, because stubby spines are classified based the absence of a neck (Sorra and Harris, 2000) and because no data was available for this index, no further analysis was performed on this group of spines. All spines included in this study had a clear head, though in some instances the head swelling was more evident from an increase in fluorescence than an increase in diameter (see figs 6E and and8B8B inset). Thus it is unlikely that filopodia (long and thin protrusions without spine heads (Grutzendler, Kasthuri and Gan, 2002)) contributed to the spine density and size measurements.
100 μm thick sections chosen in a systematic random fashion and containing area 46 that had been postfixed with cold 4 % paraformaldehyde and 0.125% glutaraldehyde in PBS were embedded with Lowicryl Resin [Electron Microscopy Sciences(EMS), Hatfield, PA]. Consecutive ultrathin sections (90 nm thick) from layer III of area 46 were cut and mounted on formvar carbon-coated copper slot grids (EMS). Since the tissue sections were sampled in a systematic-random fashion, two consecutive ultrathin sections could be arbitrarily selected from the ultrathin sections and analyzed in an unbiased manner. The ultrathin sections were viewed on a JEOL 1200EX electron microscope (JEOL, Tokyo, Japan). To constitute physical disector pairs, the corresponding tissue areas from the two consecutive ultrathin sections were captured using the Advantage CCD camera (Advanced Microscopy Techniques Corporation, Danvers, MA).
The captured disector images were viewed using the IGL trace program (Harris Laboratory, Boston University, Boston). Axo-spinous and axo-dendritic synapses in area 46 were designated and counted with different markers by using the physical disector (Fig 2), as previously described (Sterio, 1984, de Groot and Bierman, 1986, Tigges, Herndon and Rosene, 1996, Adams et al., 2001). In short, only those axo-spinous or axo-dendritic synapse profiles contained in the reference image but not in the corresponding look-up image were counted and the synapse profiles that appeared in both images were not counted. In order to increase sampling efficiency, the reference image and look-up image were reversed. The size of the disector area used was 80 μm2 and the average section thickness was 90 nm. Therefore, the height of the disector was 180 nm because the synapses were obtained by counting both ways. The volume of each disector pair was the area of the counting area multiplied by the height of the disector, which was 14.4 μm3 in the present study. On average, 20 disector pairs from each monkey were analyzed, generating a total volume of 288 μm3. Approximately 100 synapses per monkey were counted. The criteria for inclusion as an axo-spinous synapse included the presence of synaptic vesicles in the presynaptic terminal and a distinct asymmetric postsynaptic density (PSD) separated by a clear synaptic cleft. The synapse density was calculated as the total number of counted synapses divided by the total volume of the disectors used.
A systematic random series of 50 μm sections was taken from the same prefrontal cortex used for electron microscopic analysis and provided for volume estimate. The interval between the sampled sections was 1.3 mm. The sampled sections were stained with the Nissl stain, and area 46 was outlined and measured using Stereoinvestigator software (MicroBrightField, Williston, VT) in each Nissl-stained PFC section. The volume of area 46 was estimated with the Cavalieri principle.
Differences in performance on DR and DNMS were assessed using a repeated measures ANOVA. Statistical analyses were performed using ANOVA to assess possible differences in the various morphometric parameters between groups. The values are shown as means ± SEM, calculated based on one aggregate (i.e. average) per animal. Observed power was calculated in all ANOVA to confirm that the sample size was sufficient to support the data. Pearson correlations were used to determine the relationships between morphological indices and acquisition and performance on DNMS and DR. The statistical significance level was set at p< 0.05.
The results of DR testing in this group of monkeys were atypical in that there was an overall lack of age effect, contrary to previous reports (Rapp and Amaral, 1989, Bachevalier et al., 1991). All monkeys, regardless of age or sex, learned the DR procedure with a 0 and 1 sec delay at comparable rates (two-way ANOVA, p>0.1), demonstrating that all groups were motivated to learn and were capable of high levels of accuracy. After DR acquisition, young adult and aged monkeys showed no significant differences in accuracy when successively longer retention intervals were imposed (Fig 3A); however, males (2 young, 2 aged) had significantly higher scores compared to females (4 young, 7 aged) (repeated measures ANOVA; age effect, p>0.7; sex effect, F(1, 13)=6.31, p=0.03, observed power=0.63).
The results of DNMS testing, on the other hand, were similar to previous studies (Rapp, Morrison and Roberts, 2003, Shamy et al., 2006), demonstrating that aged monkeys required significantly more trials than young animals to learn the DNMS test initially with a 10-sec delay (174±50 trials for young, 1030±113 trials for aged, age effect: F(1, 13)=25.16, p=0.0004, observed power=0.995). Once the monkeys reached the learning criteria (90% correct at delay of 10 sec), the memory demands of the DNMS task were increased by imposing successively longer delays, where aged monkeys performed significantly worse compared to the young (Fig. 3B; repeated measures ANOVA; main age effect; F(1, 13)=13.25, p=0.004, observed power=0.91). Aged monkeys displayed impaired recognition compared to young adult monkeys at 15, 30, 120, and 600 second delays (Fig. 2B; 15 s, F(1,13)=6.39, p=0.03; 30 s, F(1,13)=15.03, p=0.002; 120 s, F(1,13)=5.61, p=0.03; 600 s, F(1,13)=15.33, p=0.002).
Microinjection of individual cells for morphometric analysis was focused on layer III pyramidal cells because they constitute the main corticocortical circuit (as opposed to layer V cells, which are the main cortical output to multiple subcortical sites and therefore very heterogeneous with respect to their targets) (Jones, 1984). Furthermore, although layer II cells also provide corticocortical interactions, layer III neurons have been found to be the most vulnerable cell population in Alzheimer’s disease, pointing to their vulnerability to aging and relevance to cognitive decline (Morrison and Hof, 1997). The morphometric analyses of LY-filled neurons (Fig 1) revealed an age-related loss of spines in layer III pyramidal neurons (32.9% decrease from 1.52±0.13/μm in young to 1.02±0.07/μm in aged group, F(1,7)=9.97, p=0.02, observed power=0.77). Figure 4A and B show representative dendrites from the young and aged group respectively, indicating that this remarkable difference is readily visible by eye. The decrease was consistent across all branch orders of both apical and basal dendrites and at all distances examined. To confirm the relevance of spine density to synaptic density, quantitative electron microscopic (EM) analysis (fig 2) was performed on the same cohort of monkeys. Consistent with the light-level data, EM analysis demonstrated an age-related loss of synapses (31.6%, F(1,7)=13.29, p=0.003, observed power=0.92) in the aged group (0.27±0.02/μm3) as compared to the young monkeys (0.39±0.03/μm3) (Fig 5A). This change in synaptic density is mainly driven by the loss of axo-spinous synapses (36.6% decrease from 0.30±0.03/μm3 to 0.19±0.02/μm3, p<0.01). For the nine animals for which both types of analyses were performed, EM density was highly correlated with spine density assessed from LY-filled cells (Pearson correlation, r=0.89, p=0.001; Fig 5B), further supporting the premise that changes in dendritic spines are indicative of changes in synaptic indices. In contrast to the spine and synapse density, stereologic analyses demonstrated that there was no age-related alteration in the volume of area 46 (117.8±8.4 mm3 in young and 123.1±8.4 mm3 in aged group, p>0.05).
Changes in dendritic spine morphology were first assessed by looking at aggregate morphological measurements per individual animal. Aged animals had significantly bigger mean spine head diameters (17.5% increase from 385±5 nm to 453±7 nm, F(1,7)=72.31, p=0.6 × 10−6, observed power=1.00; Fig 6A) and mean spine head volumes (27.1% increase from 152±6 nm3 to 193±11 nm3, F(1,7)=11.97, p=0.01, observed power=0.84; Fig 6B). Although there was a trend toward shorter lengths in the aged group, this difference did not reach statistical significance (p>0.1). To examine in greater detail this difference in size, we plotted the cumulative distribution of the spine head diameters for each individual animal. As can be seen in figure 6D, the distributions from animals in the two groups are virtually completely non-overlapping, with each animal segregating in accordance to its age group.
In addition, the greatest separation in the cumulative frequency is readily observed visually among the smaller spines, primarily in the 0.3–0.4 μm ranges. This feature of the cumulative distribution led us to hypothesize that only a subset of spines is susceptible to aging-related loss. In order to test this, we adapted previously published criteria for spine classification (Sorra and Harris, 2000) and defined a thin spine as one with a head diameter less than 0.6 μm and a max length that is at least twice the head diameter; a spine was classified as mushroom if its head diameter was greater than 0.6 μm. The remainder of spines was left unclassified and labeled as “other” (see Materials and Methods). Based on this classification, the prevalence of thin spines was 75±1.6% in the young group and 61±0.9% in the aged group (data not shown), which is consistent with previous reports from serial section EM (Sorra and Harris, 2000). The mean spine density per unit dendritic length was then calculated for each subclass of spines. Figure 6E shows that the overall loss of spines seen in figure 4 is primarily due to a selective loss of thin spines (45.6% decrease from 1.14±0.09/μm to 0.62±0.05/μm, F(1,7)=23.83, p=0.002, observed power=0.99), with no statistically significant difference in mushroom or other spines (p>0.2).
Are the thin spines that remain with aging congruent with thin spines found in young monkeys? To investigate this question, average morphometric indices for each subtype of spine were calculated for each animal. Interestingly, we found that thin spines from aged animals have significantly larger mean head diameters (14.1% increase from 354±5 nm to 405±2 nm, F(1,7)=79.33, p=0.5 × 10−6, observed power=1.00, Fig 7A) and mean head volumes (26.2% increase from 129±4 nm3 to 163±7 nm3, F(1,7)=21.91, p=0.002, observed power=0.98, Fig 7D). In contrast, mushroom spines remain unchanged in terms of both mean head diameter and mean head volume (p>0.05). As can be seen in figure 7B and E, the main shift in size is primarily in the 50% smallest thin spines; this implies that not only are thin spines lost in aging, but that it is the smallest of the thin spines that are the most vulnerable. The cumulative distribution of the head diameter and volume of mushroom spines (fig 7C and F) clearly demonstrate the lack of an aging-induced change in this subclass of spines. The average length of spines was unaffected by age for both thin and mushroom spines (p>0.05, data not shown).
Synapse density in layer III of area 46 showed a significant inverse correlation (Pearson correlation, r=−0.58, p=0.02; fig 8A) with DNMS acquisition (i.e., number of trials to reach criterion), demonstrating that a high density of axo-spinous synapses in layer III of area 46 facilitated learning the DNMS task. For the subset of animals for which LY-filled cells were analyzed, the correlation between DNMS acquisition and spine density did not reach significance overall (Pearson correlation, r=−0.60, p=0.09; supplementary Table 1). However, when the correlation was run for the density of thin spines alone, a similar inverse correlation was found as with synaptic density (Pearson correlation, r=−0.69, p=0.04; fig 8B). In contrast, the density of mushroom spines was unrelated to cognitive performance (Pearson correlation, r=0.58, p=0.10; Supplementary Table 1).
Most importantly, the strongest correlations were found when morphometric indices of thin spines were run against DNMS trials to criterion. As can be seen in figure 8C, the spine head volume of thin spines was 97% correlated with learning (Pearson correlation, p=0.1 × 10−6). This correlation is clearly not driven merely by group differences, since significant correlations can be observed within the individual groups as well (Pearson correlation, r=0.89, p=0.04 in young, r=0.97, p=0.03 in aged). A similar correlation was seen for thin spine head diameter (Pearson correlation, r=0.86, p=0.003; Supplementary Table 1). In stark contrast, mushroom spine morphology was not predictive of DNMS acquisition neither in terms of head volume (Pearson correlation, r=−0.37, p=0.33, fig 8D) nor in terms of head diameter (Pearson correlation, r=−0.34, p=0.37; Supplementary Table 1).
Following learning of the DNMS task, there was no significant correlation between any synaptic measure and DNMS performance (Supplementary Table 1). Furthermore, DR acquisition and performance were not correlated with any synaptic index (see Supplementary materials). Taken together, this suggests that the learning component of DNMS is particularly tightly linked to the prevalence of very small axo-spinous synapses in layer III of area 46.
There are four important findings that we present here (Fig 9): Firstly, there is a substantial aging-induced loss of spines in layer III pyramidal neurons from monkey dlPFC area 46, and this loss was highly consistent between our two methods (33% spine loss in LY-filled cells and 32% synapse loss with EM). The magnitude of the synaptic depletion with aging is consistent with a previous report (Peters, Sethares and Luebke, 2008), which found a 30% loss of asymmetric synapses in layer II/III of area 46 of monkey. Secondly, the loss of spines is specific to spines previously referred to in the literature as “thin”. These small spines are highly motile and plastic and have been hypothesized to comprise a specialized subpopulation, distinguished by its capacity for expansion in relation to new learning (Kasai et al., 2003, Kasai et al., 2010, Bourne and Harris, 2007). Thirdly, the thin spines that remain in aging are larger than the thin spines found in young animals. Fourthly, the focal novel finding presented here is that the most important structural correlate with cognitive function is the average volume of thin spines (r=0.97, p=1 × 10−5), which is far more predictive of function than overall axo-spinous synapse density (r=−0.58, p=0.02), or the actual density of thin spines (r=−0.69, p=0.04). This directly supports the notion that spine size has emerged as a key morphologic index linked to functional attributes of the synapse as well as potential for synaptic plasticity in the context of learning (Kasai et al., 2003, Bourne and Harris, 2007, Hao et al., 2007).
This study has the limitation of being a snapshot of age-related changes in spine structure. Thus, although we have shown an indisputable change in the distribution of spine size and an overall increase in mean spine diameter and volume with aging, the events underlying these changes remain unknown.
The discrepancy in the size of thin spines could indicate a difference in turnover rates between young and aged animals. Live imaging has shown that spine turnover rates in rodents decline with age (Yang, Pan and Gan, 2009). If aged monkeys have a diminished capacity to generate new spines, the thin spine assembly at any one time-point would have a much older average age and therefore be on average bigger, a proposition that is supported quantitatively by two-photon time-lapse imaging in cultured hippocampal pyramidal cells (Yasumatsu et al., 2008).
It is also conceivable that the lack of change in mushroom spine density and size is a result of a loss of large spines with a concomitant increase in the size of thin spines. This would be consistent with the observed increase in mean size of thin spines.
“Matters of size” are at the forefront of current neurobiological thinking, as pointed out by the title of a 2004 review by Kopec and Manilow (Kopec and Malinow, 2006). In order to investigate the effects of aging on different subpopulations of spines, we developed a method of classification that both allows for high throughput, as well as results in entirely unbiased data. NeuronStudio was used for automated unbiased quantification of spine size and density (Rodriguez et al., 2008, Rodriguez et al., 2006). This enabled us to analyze over 20,000 spines, a sample size that would not be feasible for manual measurements. Spines were then classified based on parameters adapted from Sorra and Harris (2000): a spine was deemed thin if its head diameter was less than 0.6 μm and its total length was at least twice its head diameter; a spine that had a head above 0.6 μm in diameter was classified as mushroom. Using this classification, we found the percent thin spines to vary between 59% and 80% in our animals, with an average of 75±1.6% in the young group and 61±0.9% in the aged group. These data are in agreement with reports from the Harris laboratory, stating that about 65% of spines are thin in both the hippocampus and prefrontal cortex of rats (Bourne and Harris, 2007). In addition, in a serial section EM study, the mean volume of the head of thin spines in the rat adult CA1 was estimated to be 0.03 μm3 (Harris, Jensen and Tsao, 1992). Because the heads of small spines are roughly spherical, we can use the formula for the volume of a sphere to derive that the average diameter in that study was 0.386 μm. Remarkably, the average head diameter in our study was 0.385 μm in young and 0.452 μm in aged. (Note that 3D images were first deconvolved to compensate for the point-spread function (PSF).) In contrast, the average volume of thin spines in our young monkeys was 0.15 μm3, a five-fold difference from the EM data mentioned above. This is an unfortunate, though expected, consequence of the oval-shaped PSF in confocal microscopy (Coling and Kachar, 2001), giving rise to a Z-plane “smear” that cannot fully be resolved by deconvolution. Nonetheless, because all dendrites are imaged and processed identically, the relative difference between animals and groups should hold true with other imaging methods such as EM.
Surprisingly, DR acquisition and performance did not differ between the young and aged monkeys in this cohort. However, the lack of an effect was due to a few poor performing young monkeys rather than no age-induced decline in performance, since aged monkeys scored similarly to previous results (O’Donnell, Rapp and Hof, 1999). No correlation between DR acquisition/performance and synaptic measures was found in this cohort (Supplementary Table 1).
Conversely, both the number of trials required to reach criterion for acquisition and recognition following a delay in DNMS were impaired in the aged monkeys. However, the only behavioral measure that correlated with synaptic indices in layer III of area 46 was acquisition of the DNMS task (Supplementary Table 1). A growing body of work now suggests that the two phases of the DNMS task tap into two different memory systems, with DNMS acquisition mostly dependent on the dlPFC and subsequent DNMS performance more reliant on the medial temporal lobe. For example, learning (Zola et al., 2000)(Zola et al., 2000, Zola et al., 2000)(Zola and Squire, 2001) or relearning (Murray and Mishkin, 1998) the DNMS task is often intact in young monkeys with experimentally induced lesions in the medial temporal lobe (although see Beason-Held et al., 1999) even though DNMS performance is predictably impaired (for reviews, see Zola and Squire, 2001, Baxter and Murray, 2001). Furthermore, action potential firing rates of layer II/III neurons in area 46 of aged monkeys are significantly related to DNMS acquisition (Chang et al., 2005) and there is a correlation between DNMS acquisition and volume of the PFC (Shamy et al., 2006).
Furthermore, the degree of age-related impairment of DNMS acquisition far exceeds the degree of impairment in subsequent performance on the task. This behavioral divergence, along with the correlation between prefrontal cortical synaptic indices and acquisition but not subsequent performance, gives rise to an intriguing hypothesis: the prefrontal cortex and hippocampus age differently. This difference might simply be related to the rate of decline in the two brain areas, or might be triggered by a complex interplay between environment and biological age.
The data presented here begin to refine the vulnerable class of synapses in area 46 that contributes to age-related cognitive decline, and provide further support for our hypothesis that the small highly plastic class of spines is particularly important for optimal performance on cognitive tasks mediated by dlPFC. However, there is a class of spine that is not vulnerable to aging: the mushroom spine. Although the precise nature of the events underlying this stability remain unknown (see “Alternative interpretations” above), our data shows that neither density nor any morphological parameter analyzed varied significantly between the mushroom spines found in aged versus young monkeys. Recently, there has been an increased interest in the field of aging of evaluating the question of how the brain maintains expertise (such as playing piano or chess) throughout life in spite of its pronounced decline in the ability to learn new things (Kramer et al., 2004, Horton, Baker and Schorer, 2008). A variety of hypotheses have been proposed, such as preservation of innate superior abilities, development of compensatory skills, and selective maintenance through sustained practice (Horton, Baker and Schorer, 2008). To our knowledge, this report is the first to show that the number and mean size of mushroom spines are unchanged by aging. These large spines are known to be highly stable, contain strong synapses (Harris and Stevens, 1989, Takumi et al., 1999), and have been proposed to be the locus of long term memories (Kasai et al., 2003). We would therefore propose the following new hypothesis for aging-related maintenance of prefrontal-dependent expertise: selective synaptic preservation of “memory” spines.
The bad news may be that we can’t make an old dog learn new tricks. The good news however, is that we can age gracefully by emphasizing learning in our youth.
Supplementary Table 1. Correlations between behavior and synaptic indices. Correlations were run for all behavioral scores against all synaptic indices measured. Pearson’s correlations were obtained for all pairs except for DR acquisition (DTtc0 and DRtc1) were Spearman’s rank correlations were used because the behavioral data was non-parametric. The correlation between DRtc0 and mean mushroom spine volume was disregarded as it arose on the basis of multiple comparison statistics (26 correlations were run on DR acquisition; at a p value of 0.05, one correlation is expected to emerge as significant on the basis of chance alone; in addition, if DR acquisition is indeed correlated with mean mushroom spine size, this correlation should hold for DRtc0 as well as DRtc1). *denotes that these correlations were run with all 15 animals for which EM data had been obtained. DRtc0= delayed response trials to criterion at 0 delay, DRtc1= delayed response trials to criterion at 1 second delay, DRavrg= average performance across all DR delays used, DNMStc= delayed non-matching-to-sample trials to criterion; DNMSavrg= average performance across all DNMS delays used, r=Pearson’s or Spearman’s correlations coefficient.
This work was supported by NIA grants AG 006647 (JHM) and AG 010606 to (JHM and PRR), AG 016765 (JHM and PRR), NRSA training grant 1F30MH083402 (DD), and supported in part by the Intramural Research Program of the NIA. We thank Ljiljana Minwalla and Mary Roberts for expert technical assistance.