Abstract
Changes in excitatory neuron and synapse structure have been recognized as a potential physical source of age-related cognitive decline. Despite the importance of inhibition to brain plasticity, little is known regarding aging-associated changes to inhibitory neurons. Here we test for age-related cellular and circuit changes to inhibitory neurons of mouse visual cortex. We find no substantial difference in inhibitory neuron number, inhibitory neuronal subtypes, or synapse numbers within the cerebral cortex of aged mice compared with younger adults. However, when comparing cortical interneuron morphological parameters, we find differences in complexity, suggesting that arbors are simplified in aged mice. In vivo two-photon microscopy has previously shown that in contrast to pyramidal neurons, inhibitory interneurons retain a capacity for dendritic remodeling in the adult. We find that this capacity diminishes with age and is accompanied by a shift in dynamics from balanced branch additions and retractions to progressive prevalence of retractions, culminating in a dendritic arbor that is both simpler and more stable. Recording of visually evoked potentials shows that aging-related interneuron dendritic arbor simplification and reduced dynamics go hand in hand with loss of induced stimulus-selective response potentiation (SRP), a paradigm for adult visual cortical plasticity. Chronic treatment with the antidepressant fluoxetine reversed deficits in interneuron structural dynamics and restored SRP in aged animals. Our results support a structural basis for age-related impairments in sensory perception, and suggest that declines in inhibitory neuron structural plasticity during aging contribute to reduced functional plasticity.
SIGNIFICANCE STATEMENT Structural alterations in neuronal morphology and synaptic connections have been proposed as a potential physical basis for age–related decline in cognitive function. Little is known regarding aging-associated changes to inhibitory neurons, despite the importance of inhibitory circuitry to adult cortical plasticity and the reorganization of cortical maps. Here we show that brain aging goes hand in hand with progressive structural simplification and reduced plasticity of inhibitory neurons, and a parallel decline in sensory map plasticity. Fluoxetine treatment can attenuate the concurrent age–related declines in interneuron structural and functional plasticity, suggesting it could provide an important therapeutic approach for mitigating sensory and cognitive deficits associated with aging.
Introduction
A large fraction of persons older than 85 years have significant cognitive impairment and dementia (Gallo and Lebowitz, 1999; Hebert et al., 2013; Wu et al., 2016). The decline in cognitive abilities begins much earlier (for review, see Zec, 1995), and although in many cases it is related to dementias such as Alzheimer's Disease, a decline in cognitive capacity has also been associated with normal aging (Petersen et al., 2018). Most common are declines in memory, performance of language, and executive function tasks (for review, see Hof and Morrison, 2004; Kirova et al., 2015), as well as sensory perception (Kalisch et al., 2009; Hickmott and Dinse, 2013; for review, see Andersen, 2012).
Despite common belief, loss of neurons due to cell death is quite limited during normal aging and unlikely to account for age-related functional impairments (for review, see Peters et al., 1998). Rather, it seems that structural alterations in neuronal morphology and synaptic connections are features most consistently correlated with brain age, and may be considered as the potential physical basis for the age-related decline in cognitive function (for review, see Morrison and Baxter, 2012). Multiple cellular changes have been reported in the aging brain. These include reductions in the size of dendritic and axonal arbors, in spine numbers, and in synaptic efficacy (Jacobs and Scheibel, 1993; Masliah et al., 1993; Kawaguchi et al., 1995; Anderson and Rutledge, 1996; Ou et al., 1997; Shankar et al., 1998; Grill and Riddle, 2002; Bloss et al., 2011; Mostany et al., 2013). Although a decrease in adult neurogenesis has also been noted, this is mainly relevant to the hippocampus (Kuhn et al., 1996).
Studies of structural alterations related to brain aging have mostly focused on the excitatory network. Recently, chronic two-photon in vivo imaging has revealed that in contrast to the stability of dendritic arbors of pyramidal neurons in the adult brain, inhibitory interneurons in cortical layer (L)2/3 retain the capacity to remodel in the adult on a day-to-day basis (Lee et al., 2006, 2008; Chen et al., 2011a; Chen and Nedivi, 2013). Furthermore, it is increasingly evident that inhibition plays a key role in cognitive abilities as well as experience-dependent plasticity (Dan and Poo, 2004; Fernandez and Garner, 2007). Inhibitory circuit development is crucial for both the onset and closure of developmental critical periods, and reducing inhibition pharmacologically can restore a juvenile state of plasticity to the adult neocortex (for review, see Hensch, 2004, 2005; Sale et al., 2010). In particular, the widely used antidepressant drug fluoxetine, which reduces intracortical inhibition, has been shown to promote interneuron branch remodeling in the adult mouse visual cortex (Chen et al., 2011b), and restore juvenile levels of ocular dominance plasticity (Maya Vetencourt et al., 2008). Despite the importance of the inhibitory circuitry to adult cortical plasticity and the reorganization of cortical maps (Jacobs and Donoghue, 1991; Jones, 1993), little is known regarding aging-associated changes to inhibitory neurons.
Here we show that morphological plasticity mechanisms that enable inhibitory cell remodeling in L2/3 of the adult mouse visual cortex are lost during the aging process, that the loss is gradual, and already begins in young adults. We characterize the time course of the decline in inhibitory circuit structural plasticity and show that it is congruent with loss of functional synaptic plasticity. Further, we show that use of the antidepressant fluoxetine has a beneficial effect, preventing declines in both structural and functional manifestations of inhibitory circuit aging, and that this effect is dependent on time of onset and length of treatment.
Materials and Methods
All animal work was approved by the Massachusetts Institute of Technology Committee on Animal Care and conforms to National Institutes of Health guidelines for the use and care of vertebrate animals.
Surgical procedures and fluoxetine administration.
To allow long-term visualization of in vivo neuronal morphology, cranial windows were implanted bilaterally over the visual cortices of adult male thy1-GFP-S mice as previously described (Lee et al., 2008) 21 d before they reached 3, 6, 9, 12, or 18 months of age, with the following modifications. Windows were 4 mm, and older animals usually needed additional dosing of anesthetics compared with previous studies in young adults. Animals were housed singly after window insertion. Sulfamethoxazole (SMX; 1 mg/ml) and trimethoprim (TMP; 0.2 mg/ml) were chronically administered in the drinking water through the final imaging session to maintain optical clarity of implanted windows. For animals subjected to fluoxetine treatment, fluoxetine-hydrochloride (160 mg/L) was chronically administered in the drinking water with SMX-TMP supplemented in the food pellet (6 mg and 1 mg, respectively, per pellet; Bio-Serv). SMX-TMZ dosage was approximately equivalent for both administration routes, based on measured consumption of 5–13 ml of water, and 1–3 pellets per day. Because animals were not water or food restricted, consumption per day varied for individual animals and between animals, within the same range of SMX-TMP regardless of administration route. Water usage at ∼5–13 ml per day is sufficient to maintain fluoxetine serum levels at ∼200 ng/ml, well above the recommended therapeutic range of 15–55 ng/ml for humans (according to fluoxetine drug label prescribing information).
Two-photon imaging.
Starting 3 weeks after cranial window surgery, mice were anesthetized with 1.25% avertin (7.5 ml/kg, i.p.). Anesthesia was monitored by breathing rate and foot pinch reflex, with additional doses administered as needed. In vivo two-photon imaging was performed using a custom-built microscope modified for in vivo imaging with a stereotaxic restraint affixed to a stage insert and custom acquisition software. The light source for two-photon excitation was a commercial Mai Tai HP Ti:Sapphire laser (Spectra-Physics) pumped by a 14 W solid-state laser delivering 100 fs pulses at a rate of 80 MHz with the power delivered to the objective ranging from ∼37–50 mW depending on imaging depth. Z-resolution was obtained with a piezo actuator positioning system (Piezosystem Jena) mounted to the objective. The excitation wavelength was set to 950 nm, with the excitation signal passing through a 20×/1.0 NA water-immersion objective (Plan-Apochromat, Zeiss) and collected after a barrier filter by a photomultiplier tube. Given the sparse GFP expression in the thy1-GFP-S line, typically only one cell was imaged per animal. Interneurons were identified based on their complex, local dendritic arborization, lack of apical dendrite, small diameter dendritic processes, and absence or sparseness of dendritic spines. While there has been some debate regarding an inflammatory response in the cranial window preparation affecting dendritic spine dynamics (Xu et al., 2007), we have previously shown that cranial window insertion and imaging by our protocol does not affect dendritic arbor or spine dynamics directly or through recruitment of an immune response (Lee et al., 2008).
Imaging data analysis.
Using MATLAB (MathWorks) and ImageJ (National Institutes of Health), 16-bit two-photon raw scanner data were converted into an 8-bit image z-stack. For each cell, 4-D (x, y, z, and t) stacks were manually traced in Neurolucida (MBF Biosciences) and analyzed blind to age or treatment. Every imaged cell was independently reconstructed at each time point and branch tip length measurements were obtained for all branches. The only branch tips (segments of dendrite from the last branch point to the branch terminal ending) included in the analysis were those that could be confidently identified across all imaging sessions in relation to local landmarks, did not extend beyond the imaging volume, and were not obscured by blood vessels. The branch length refers to the distance from the last branch point to the tip. For unbiased identification of dynamic branch tips, a Fano factor (FF) value was calculated based on branch tip length measurements obtained from Neurolucida across all imaging sessions. The FF is defined as the variance in the measured length of an individual branch tip across all imaging sessions, divided by the mean length of that branch tip . The FF has been previously shown to represent the best nonbiased indicator of branch tip dynamics compared with other methods of analysis and robustly identifies small branch tip changes while accounting for measurement variability during imaging or manual reconstructions (Lee et al., 2008). The reason that in practice the FF best represents the population of dynamic branch tips likely hinges on the fact that a majority of remodeling occurs on relatively short branch tips. Normalizing by the average branch tip length helps identify shorter branch tips that are dynamic and minimizes the effects of measurement variability/error in longer branch tips. Branch tips with FF > 1.09, representing the 1.5 × interquartile range above the upper quartile of the sample population, were confirmed as dynamic by visual examination. A mean FF > 0.35 across all monitored branch tips of individual L2/3 interneurons was previously determined to represent a dynamic cell (Lee et al., 2008). To determine the percentage of extending and retracting branches, the FF was calculated for each branch tip for each two consecutive imaging sessions. A branch tip changing sufficiently between two imaging sessions for FF > 1.09 was considered a dynamic event. Thus, each branch tip could produce several dynamic events. Each dynamic event was scored as a retraction or as an extension. The percentage of extensions versus retractions was then calculated across the entire population of cells for each age.
Immunohistochemistry.
Cellular immunohistochemistry was performed as in (Chattopadhyaya et al., 2004), with modifications as follows: Animals were perfused transcardially with PBS containing heparin (1:1000; Sigma-Aldrich), followed by 4% paraformaldehyde (PFA). Brains were then removed and postfixed overnight in 4% PFA. Fifty micrometer sections containing the visual cortex were cut using a vibratome (Leica VT1000S) and blocked for 2 h in solution containing 10% normal goat serum (NGS) or 10% fetal calf serum (FCS) and 1% Triton X-100. Sections were incubated 24–60 h at 4°C in a solution containing PBS, 5% NGS or FCS, and 0.1% Triton X-100 and either guinea pig α-GABA (1:2000; Millipore, AB175) or rabbit α-GABA (1:2000; Sigma-Aldrich, A2052) with one of the following primary antibodies: rabbit α-parvalbumin (PV; 1:5000; Swant, PV28), rabbit α-vasoactive intestinal peptide (1:2000; Immunostar, 20077), goat α-calretinin (1:1000; Swant, CG1), or rat α-somatostatin (1:200; Millipore, MAB354). Sections were washed three times in PBS then incubated with a secondary antibody solution containing PBS, 5% NGS or FCS, and 0.1% Triton X-100 and the appropriate AlexaFluor 555- and AlexaFluor 647-conjugated secondary antibody (1:400). Sections were then washed three times in PBS, where 4,6-diamino-2-phenylindole (DAPI; Sigma-Aldrich) was included in the first wash, then mounted for visualization in Fluoromount-G (Southern Biotech). Images were collected on an upright epi-fluorescent scope (Nikon) using a 10× objective/N.A. 0.3 (Nikon). For each section, images of the same area and focal plane of the visual cortex were taken using separate filters to identify DAPI (blue), GFP (green), AlexaFluor 555 (red), and AlexaFluor 647 (far red) labeled cells. ImageJ (NIH) was used for identification and quantification of double- or triple-labeled cells within each layer. Quantifications were done by layer as defined by DAPI staining. Because layer size is not uniform across sections, a rectangle was defined encompassing the maximum area for each layer within a section and all the GABA-positive cells were counted, whether costained for a second marker or not.
For synapse density analysis, immunohistochemistry was performed as described above except sections were cut at 40 μm (for excitatory synapses) or 50 μm (for inhibitory synapses) and primary incubation was performed overnight. Primary antibodies were guinea pig α-VGLUT1 (1:1000; Synaptic Systems), rat α-GFP (Nacalai Tesque), rabbit α-GABA (Sigma-Aldrich), and rabbit α-VGAT (1:1000; Synaptic Systems).
Images were obtained with a z-spacing of 0.5 μm from layer 2/3 of the visual cortex with a FluoView IX70 laser-scanning confocal microscope (Olympus), using a 60× UPlanApo water-immersion objective (Olympus) and the FluoView software (Olympus). Excitatory and inhibitory cells were identified based on morphology and immunohistochemistry. Neurons were defined as excitatory if they possessed a triangular cell body, one clear apical dendrite, and spiny dendrites. GABA-expressing neurons were considered inhibitory.
Images were analyzed using ImageJ (National Institute of Health). Synapses were counted if they consisted of puncta (multiple contiguous pixels brighter than background) that touched, or overlapped with, clearly visible dendrites extending from the cell of interest. To separate actual synapses from background noise, only points that were visible in at least two subsequent focal planes were included in the analysis. Dendrites were traced in Neurolucida (MBF Bioscience) software, and their lengths obtained from Neurolucida Explorer (MBF Bioscience).
Visually evoked potential recordings.
For electrode implantation, thy1-GFP-S mice were anesthetized with an intraperitoneal injection of 50 mg/kg ketamine and 10 mg/kg xylazine, and a local anesthetic of 1% lidocaine hydrochloride (1 mg/ml) was injected over the scalp. For purposes of head fixation, a post was fixed to the skull just anterior to bregma using cyanoacrylate and dental cement. Two small (<0.5 mm) burr holes were made in the skull overlying the binocular visual cortex (3 mm lateral of lambda) and tungsten microelectrodes (FHC) were inserted 450 μm below the cortical surface along the dorsal-ventral stereotaxic axis, positioning the electrode tip at the depth yielding the maximal field potential, which has been shown by current-source density analysis to be sensitive to changes in current sinks in layer 2/3 and 4 (Cooke et al., 2015). Reference electrodes were placed bilaterally in prefrontal cortex. Electrodes were secured in place using cyanoacrylate and the entire exposure was covered with dental cement. Mice were monitored postoperatively for signs of infection or discomfort and were allowed at least 24 h recovery before habituation to the restraint apparatus.
Visually evoked potential (VEP) recordings were conducted in awake mice. Mice were habituated to the restraint apparatus before the first recording session, and were alert during recording. Visual stimuli were presented to left and right eyes randomly. A total of 100–400 stimuli were presented per condition. VEP amplitude was quantified by measuring trough to peak response amplitude, as described previously (Frenkel and Bear, 2004; Frenkel et al., 2006).
Visual stimuli consisted of full-field sine wave gratings of varying contrast (0–100%) generated by a VSG2/2 card (Cambridge Research System) and presented on a computer monitor suitably linearized by γ-correction. VEPs were elicited by horizontal, vertical, or oblique (45° or 135°) bars. The display was positioned 20 cm in front of the mouse and centered on the midline, thereby occupying 92° × 66° of the visual field. Mean luminance, determined by a photodiode placed in front of the computer screen, was 27 cd · m−2.
Experimental design and statistical analysis.
Only male mice were used for this study. The cells used for the 3 month time points in Figures 1D, 2, and 4 were previously described by Chen et al. (2011b) and Lee et al. (2008), and some of the 3 month time points in Figure 3C were previously published by Chen et al. (2011b). The 3, 6, and 9 month data points used in Figure 4B were also used in Figure 1D. Cells from all ages, including previously analyzed cells, were independently analyzed blind to age and treatment by two people. For Figures 3, 5, and 6, each n corresponds to one animal (Fig. 3A: 3 months, n = 3–5; 24 months, n = 3–6; B: GABA-positive cells, n = 3; thy1-GFP-nonpyramidal cells, n = 6–8; C–E: n = 7–9 for all conditions; 5: 3 months, n = 5; 6 months, n = 10; 9 months, n = 5; 6A: treated, n = 6; untreated, n = 10. 6B: 9 months untreated, n = 5; 9 months with 3 month treatment, n = 11; 9 months with 6 month treatment, n = 5). For Figures 1, 2, and 4, each n corresponds to one cell, with only one cell imaged per animal (Fig. 1: 3 months, n = 16; 6 months, n = 14; 9 months, n = 10; 12 months, n = 8; 18 months, n = 11; 2: 3 months, n = 51; 12 months, n = 38). Figure. 2 made use of a larger dataset, because morphological analysis only requires a single time point per cell, and could include cells imaged only once, and the first session of control cells for experiments with other imaging time lines (Fig. 4: n = same as for 1, and in addition n = 8 for 6 months with 3 month treatment, n = 8 for 9 months with 3 month treatment, n = 6 for 9 months with 6 months treatment).
Statistical significance was primarily evaluated using Student's t test (unpaired for across-animal comparison, paired for within-animal comparison) and MANOVAs (repeated-measures for within-animal comparisons) with posttests for multiple comparisons. Differences in variances were evaluated using the Brown–Forsythe test. Fractions were compared using Fisher's exact test. Distributions were compared using the Kolmogorov–Smirnov test. See Results and figure legends for specifics on statistical tests for each experiment. For Figures 1 and 4, the large within group variance in cell dynamics per animal at some ages, precluded use of tests that assume equal variance. We therefore used a bootstrap approach to test if the mean of two conditions was different: For each bootstrap repeat, we resampled the data in each condition with replacement and computed the difference of the means of the resampled datasets. We performed 106 repeats. If the majority of repeats showed a difference >0, we report as p values the fraction <0; if the majority of repeats showed a difference <0 the p value is the fraction >0. Statistical analysis was performed with MATLAB (MathWorks) and GraphPad Prism v7 for Mac.
Results
Interneuron dendritic arbor remodeling progressively decreases with age
In vivo imaging studies using thy1-GFP-S transgenic mice have revealed that dendrites of inhibitory interneurons in the adult visual cortex remodel on a day-to-day basis (Lee et al., 2006). Remodeling interneurons in these mice represent all known interneuron subtypes residing within a “dynamic zone” corresponding to a superficial strip of extragranular L2/3 (Lee et al., 2008), a cortical lamina known to retain a capacity for sensory map plasticity beyond development into adulthood (LeVay et al., 1980; Daw et al., 1992; Hirsch and Gilbert, 1993). The structural rearrangement of dynamic zone interneurons is a general feature of sensory cortices. It is shared across modalities and between primary and higher sensory areas (Chen et al., 2011a), thereby providing a common mechanism for experience-dependent functional plasticity in the adult neocortex. To test whether inhibitory interneuron structural plasticity mechanisms are retained during normal brain aging, we monitored dendritic arbor dynamics of superficial L2/3 interneurons in the visual cortex of adult thy1-GFP-S transgenic mice (Lee et al., 2006). Three weeks before reaching 3, 6, 9, 12, or 18 months of age, mice were surgically implanted with bilateral cranial windows over the visual cortices. Following 3 weeks of recovery, L2/3 interneurons in the dynamic zone (65–150 μm below the pial surface) were identified and a two-photon imaging volume encompassing the cell and its entire dendritic arbor was acquired. The same cells were imaged again 7 and 14 d later (Fig. 1A). Branch tips were reconstructed and monitored for length changes, which were then quantified using a FF analysis (Fig. 1B–D).
Our previous results (Lee et al., 2008) showed that at 3 months of age (young adults) every neuron within the dynamic zone shows a FF >0.35, indicating at least 1 dynamic branch tip (Fig. 1D). When we compared previously collected 3 month data points (Lee et al., 2008; Chen et al., 2011b) to cells from older mice, we noticed a clear difference in variance between the groups (F(4,54) = 2.859, p = 0.032, Brown–Forsythe test, n = 16 for 3 months, 14 for 6 months, 10 for 9 months, 8 for 12 months, 11 for 18 months). This change in variance was paralleled by a reduction in mean FF with age. When comparing the 3 month group to 6-month-old mice, an age considered mature but not old, we saw a sharp drop in dynamics measured as a statistically significant reduction in mean FF (p = 0.0012, by bootstrap; see Materials and Methods) using Holm–Sidak's test to correct for multiple comparisons, n = 16 for 3 and 14 for 6 months), with less than half the cells retaining a FF >0.35, the threshold for a dynamic cell. By 9 and 12 months of age the number of dynamic cells had further dropped, to the point where all cells had a FF ∼<0.35, so that by 9 months, the highest value was 0.353. At 18 months, the average FF was significantly lower than at 6 months (p = 0.0229, by bootstrap using Holm–Sidak's test to correct for multiple comparisons, n = 14 for 6 and 11 for 18 months). These findings show a precipitous drop in the dynamic index of cortical inhibitory neurons with age, sooner than might have been anticipated based on other measures of aging. By the time mice reach an age considered mature, rather than old, remodeling of their inhibitory arbors is already significantly compromised.
Age-related simplification of interneuron dendritic arbors
We next asked whether interneuron arbors showed age-related changes in morphology. When comparing the number of dendrites in 3 month versus 18 month or older animals, we found no difference in the number of primary dendrites between these populations (Fig. 2A; t(87) = 1.215, p = 0.23, unpaired two-tailed Student's t test, n = 51 for 3 months, 38 for ≥ 18 months). We also did not see a difference in the distribution of dendritic lengths over this time period (Fig. 2B; D = 0.2843, p = 0.059, Kolmogorov–Smirnov test, n = 51 for 3, 38 for ≥ 18 months). However, in the 3–18 month period we did see a shift in the distribution of the number of dendritic nodes (branch points; Fig. 2C; D = 0.3122, p = 0.029, Kolmogorov–Smirnov test, n = 51 for 3 months, 38 for ≥18 months), suggesting an age-dependent reduction in arbor complexity. Our previous work has shown that in young adult mice interneuron dendritic dynamics are comprised of balanced additions and retractions (Chen et al., 2011a). The simplified arbors on interneurons in older mice suggested that at some point past 3 months and before 9 months, when most cells are stable (Fig. 1D), retractions should predominate. We therefore examined all dynamic branch tips at 3 and 6 months and classified them as extensions or retractions. At 3 months of age, the number of branch extensions and retractions was slightly shifted toward extensions (43.1% of all events were retractions). However, at 6 months branch retractions had grown more prevalent (61.2% of all events were retractions), such that retractions constituted a significantly larger fraction of the total number of events at 6 months than at 3 months (p = 0.0258, two-sided Fisher's exact test, 41 extensions and 31 retractions at 3 months, 33 extensions and 52 retractions at 6 months). The age-dependent reduction in branch dynamics, together with a shift toward increased branch retractions, explain our finding of simplified and stable interneuron arbors in older mice.
Aging does not affect interneuron cell or synapse numbers or subtype representation
To confirm that the specific age-related characteristics we find, of a simplified and stable arbor in aged mice, is general to all interneuron types and is not due to a selective retention of a particular interneuron subtype, we compared the prevalence of the four major interneuron subtypes in all layers of visual cortex at 3 and at 24 months. Immunohistochemical costaining for GABA and PV, vasointestinal peptide (VIP), somatostatin (SOM), or calretinin (CR) showed that the percentage of these different interneuron subtypes within the neocortical GABAergic population was consistent with previous results in young adult mice (Xu et al., 2006; Gonchar et al., 2007; Lee et al., 2008), and remained largely unchanged throughout a 2 year life span (Fig. 3A; p = 0.044 for SOM in layer 1, p = 0.69 for SOM in layer 5, p = 0.84 for PV in layer 6, and p > 0.99 for all other subtypes across layers, multiple two-tailed unpaired t tests with Holm–Sidak's correction for multiple comparisons, n = 3–5 mice for 3 months and 3–6 mice for 24 months). Further, at 24 months of age there was no difference in the representation of the different interneuron subtypes within the thy1-GFP labeled population compared with unlabeled interneurons in the same brains (Fig. 3B; p = 0.95 for PV in upper layer 2/3, p = 0.98 for CR in layer 1, and p > 0.99 for all other subtypes across layers, multiple two-tailed unpaired t tests with Holm–Sidak's correction for multiple comparisons, n = 3 mice for the GABA-positive cells and 6–8 for the thy1-GFP-nonpyramidal cells). We also examined whether the number of excitatory or inhibitory synapses (VGLUT-positive or VGAT-positive, respectively) onto L2/3 dendrites or L5 pyramidal apical tufts changed between 3 and 18 months. Although such colocalization studies based on immunohistochemistry are not definitive due to their limited resolution and specificity, we found no evidence of changes in synapse densities onto the cells we examined (Fig. 3C–E; C: t(14) = 0.4117, p = 0.69, unpaired two-tailed Student's t test, n = 8; D: t(14) = 0.02314, p = 0.99, unpaired two-tailed Student's t test, n = 7–9; E: t(12) = 1.473, p = 0.17, unpaired two-tailed Student's t test, n = 7).
Fluoxetine attenuates the age-related decline in interneuron complexity
Because the antidepressant drug fluoxetine has been shown to promote interneuron branch remodeling in the visual cortex of young adult mice (Chen et al., 2011a), we asked whether it could prevent the reduced dynamics of interneuron dendritic arbors that occurs with aging. We treated animals with fluoxetine in their drinking water for 3 months starting at either 3 or 6 months of age and then compared their dynamic index to age-matched untreated controls (Fig. 4A). When mice were treated from 3 to 6 months, no effect was seen on the mean FF (Fig. 4B; p = 0.13, by bootstrap, n = 14 for untreated, 8 for treated). A few of the animals at 9 months showed an improvement in FF to levels clearly above the dynamic threshold of 0.35, resulting in a statistically significant increase in the mean FF average at this age (Fig. 4B; p = 0.043, by bootstrap using Holm–Sidak's test to correct for multiple comparisons, n = 10 for untreated, 8 for treated). Given the indication that some improvement might be possible, we next tried extending the fluoxetine treatment to 6 months, starting at 3 months of age (Fig. 4A). Here, we found that 67% of the cells from treated animals showed dynamic arbors at 9 months of age (compared with 25% with only 3 months of treatment), again resulting in an increase in the average FF compared with untreated age-matched controls (Fig. 4B; p = 0.043, by bootstrap, using Holm–Sidak's test to correct for multiple comparisons, n = 10 for untreated, 6 for treated). Although in the case of 9-month-old animals treated with fluoxetine for 3 months the statistical significance was driven by two animals with a FF >0.35, one of which may be considered an outlier (2 SD above the mean), it is important to note that the untreated control group at 9 months does not have a single animal with a FF >0.35 (the highest FF = 0.3531). The 9 month animals with 6 months of treatment has 4 of 6 animals >0.35, and none of the values are even 1 SD above the mean. The difference between 6 months of fluoxetine and no treatment is therefore robust and not driven by outliers. These data are consistent with 3 months of treatment having a variable effect on arbor plasticity (and no effect on functional plasticity, see next section), in contrast to the more penetrant effects of the 6 month treatment on both structural and functional plasticity. These results suggest that fluoxetine treatment can somewhat attenuate the age-related decline in interneuron dendrite remodeling, in particular if treatment is started early and maintained throughout the lifespan.
Fluoxetine attenuates the age-related decline in visual cortex functional plasticity
To investigate whether there is a functional correlate to the decline in interneuron structural plasticity in mouse visual cortex, we recorded VEPs and induced stimulus-selective response potentiation (SRP). This NMDA-dependent form of plasticity bears many of the hallmarks of long-term potentiation (LTP) but is induced through visual experience (Frenkel et al., 2006; Cooke and Bear, 2010). VEPs were recorded 450 μm below the pia of binocular visual cortex as previously described (Frenkel and Bear, 2004) and SRP was induced through repeated presentations of sinusoidal gratings of a specific orientation. We found robust SRP in 3-month-old wild-type mice (Fig. 5A). However, levels of SRP decreased in 6-month-old mice and were virtually non-existent in 9-month-old animals (Fig. 5A; 3 months: t(4) = −5.145, p = 0.0068, paired t test, n = 5; 6 months: t(9) = −5.564, p = 0.00035, paired t test, n = 10; 9 months: t(4) = −0.462, p = 0.675, paired t test, n = 5). To confirm that the decrease in SRP was not a result of defective vision in 9-month-olds, we assessed visual acuity and contrast sensitivity in 3-, 6-, and 9-month-old mice and found no difference between the three groups tested (Fig. 5B: F(2,26) = 0.623, p = 0.542; repeated-measures MANOVA). These findings show there is a similar time course for the loss of functional plasticity in visual cortex as for the loss of interneuron structural plasticity.
In addition to promoting interneuron branch remodeling in the visual cortex of young adult mice (Chen et al., 2011a), chronic treatment with fluoxetine has also been show to restore juvenile levels of ocular dominance plasticity in adult mice (Maya Vetencourt et al., 2008). To test whether fluoxetine could also reverse SRP deficits in aged mice, we assessed SRP after fluoxetine treatment equivalent to that used for the structural plasticity studies (Fig. 6A). We found that 3 months of fluoxetine treatment did not improve SRP in 6- or 9-month-old mice (Fig. 6B: t(14) = −1.184, p = 0.256, unpaired two-tailed Student's t test, n = 6 for treated, n = 10 for untreated; C: t(14) = 0.992, p = 0.339, unpaired two-tailed Student's t test, n = 11 for treated, n = 5 for untreated). However, 6 months of fluoxetine treatment significantly increased SRP in 9-month-old animals (Fig. 6C; t(8) = 1.996, p = 0.0405, unpaired two-tailed Student's t test, n = 5 for treated, n = 5 for untreated). This suggests that chronic fluoxetine treatment maintains plasticity in neuronal circuits during aging, if initiated early.
Discussion
Here we examine the effects of aging on morphological and functional plasticity of inhibitory interneurons in mouse primary visual cortex (V1). We find that interneurons undergo two progressive changes with age; they become less dynamic and their dynamics shift from balanced branch additions and retractions to the prevalence of retractions over additions. Starting at 3 months of age, interneuron arbors undergo structural simplification and reduced plasticity, so that by 9 months their structure is less complex and more stable. These morphological changes run in parallel to an age-related decline in SRPs as measured by VEPs, and both can be partially prevented by long-term fluoxetine treatment.
To what extent age-related cognitive decline derives from changes to brain structure, and whether such changes are a normal part of brain aging unrelated to degenerative processes, has been hotly debated for many years. Although some longitudinal fMRI studies in human subjects suggest there is a reduction in volume of some brain areas during aging (for review, see Fjell et al., 2014a,b), the low resolution of such studies precludes an understanding of the underlying cellular basis. Loss of GABAergic neurons in aging cat striate cortex has been reported (Hua et al., 2008). Yet, overall, the evidence for neuronal loss during normal aging has not been strong (Morrison and Hof, 1997; for review, see Peters et al., 1998; Freeman et al., 2008), and we see no evidence of GABAergic cell loss in our studies (Fig. 3A,B). In contrast, a growing body of literature points to an age-related deterioration in neuronal connectivity (Morrison and Baxter, 2012; Dickstein et al., 2013; for review, see Hof and Morrison, 2004; Dickstein et al., 2007).
A major problem with obtaining data regarding age-related changes to neuronal numbers or structure, is that with classic anatomical methods small scale changes are difficult to detect against the enormous variance in the size and shape of individual cortical neurons. This is particularly true for inhibitory neurons, that as a group show immense morphological diversity (Kawaguchi and Kubota, 1997, 1998; Markram et al., 2004). Indeed, using conventional immunohistochemistry (Fig. 3C) we do not see gross age-related changes in inhibitory synaptic staining. The age-related reduction in interneuron complexity we describe is discernible thanks to our ability to chronically track the same cells, visualizing their gradual simplification. Chronic in vivo two-photon imaging of dendritic spines in neocortex of up to 1-year-old mice also suggests some age-dependent spine loss (Zuo et al., 2005; Spires-Jones et al., 2007; Mostany et al., 2013), and alterations in the size and stability of spines and boutons (Bloss et al., 2011; Mostany et al., 2013). The scale of net morphological change thus far detected by in vivo imaging in mice is overall small, and its analog in humans would be unlikely to explain large volumetric changes reported in some human fMRI studies.
Given that most branch tip dynamics are on the scale of tens of micrometers, at most, one might ask whether such changes in branch tip length represent sufficient synapse gain and loss to alter neuron and circuit function? Or in the case of brain aging, is the gradual loss of synaptic contacts due to branch tip retractions, and the eventual loss of branch dynamics, functionally meaningful? Post hoc electron microscopy (EM) reconstructions of interneuron dendrites imaged in vivo, show that new branch tips harbor synaptic contacts at a density similar to stable dendrites (Chen et al., 2011b), implying that branch tip extensions and retractions reflect synapse gain and loss, respectively. EM estimates of synaptic density on L2/3 interneuron dendrites is ∼1 synapse per μm (Kubota et al., 2007). Approximately 80% of the synapses on distal interneuron dendrites represent excitatory inputs (Kubota et al., 2007) from a large number of local pyramidal neurons that each contributes 3–7 synapses (Markram et al., 2004). The estimate is that as few as 10 presynaptic, temporally correlated excitatory inputs are sufficient to trigger an action potential in inhibitory cells (Buhl et al., 1997). For bitufted interneurons, a train of action potentials from even a single synaptic contact can produce an action potential (Kaiser et al., 2004). Thus, a branch tip length change of ∼30 μm (or ∼30 synapses) per cell could potentially alter the connectivity of 5–10 excitatory presynaptic partners, each with significant ability to initiate interneuron firing.
Further, structural remodeling in the adult brain has been proposed as a means to increase spatial access to potential synaptic contacts (Chklovskii et al., 2004). The geometry and space occupied by a dendritic spine in mouse cortex renders it capable of making ∼4 potential synaptic contacts (Stepanyants et al., 2002). An entire dendritic branch tip likely contains an even higher number of possible synaptic connections considering its size in relation to a dendritic spine. Increasing an interneuron's potential for sampling distinct circuits would be most readily achieved by modest length changes in several branch tips spatially distributed throughout the dendritic field compared with a substantial length change of a single branch tip. The cell type and laminar-specificity of dendritic structural dynamics (Chen et al., 2011a,b) argue that partner sampling that occurs during synaptic remodeling is circuit specific. Thus, it may not be the net number of synapses lost during aging that matters to a given neuron, but rather its reduced connectivity with specific neuronal partners within the local circuit, and the inability to trade or shuffle remaining partners.
Perhaps due to the fact that the overwhelming majority of synapses in the neocortex are excitatory, prior ex vivo postmortem as well as newer imaging studies have mostly focused on dendritic spines on excitatory pyramidal neurons. Yet, inhibitory neurons play an important role in modulating excitatory neuron function and plasticity within the cortical circuit (for review, see Singer, 1996; McBain and Fisahn, 2001; Dan and Poo, 2004; Foeller and Feldman, 2004) and have also been implicated in sensory map plasticity in the adult neocortex (Jacobs and Donoghue, 1991; Jones, 1993). Further, age-related deterioration of visual function in nonhuman primates, manifested as increased responsiveness to all orientations and directions as well as an increase in spontaneous activity, is consistent with a decline in intracortical inhibition (Schmolesky et al., 2000).
We have previously shown interneuron dendritic remodeling in superficial L2/3 of primary visual cortex (Lee et al., 2008), which is experience-dependent (Chen et al., 2011b), and seems to be a general feature of the adult cortical microcircuit across different sensory modalities and primary versus higher-order sensory regions (Chen et al., 2011a). Here we show that inhibitory neuron remodeling declines with age with a time course that mirrors the decline in visual cortex plasticity. These findings are in line with functional data suggesting that deficits in inhibitory circuit function contribute to age-related impairments in visual perception (Schmolesky et al., 2000; Leventhal et al., 2003) and perhaps more broadly to sensory, motor, and cognitive tasks that rely on the neocortex.
Given that interneuron remodeling in superficial L2/3 is not restricted to specific interneuron subtypes (Lee et al., 2008) and imaged neurons are sampled randomly, the finding that at 3 months every imaged interneuron is dynamic (FF > 0.35), and at 9 months none are dynamic (FF < 0.35), suggest that the general phenomenon of reduced dynamics with age, is unlikely to be subtype specific. It could be that some subtypes are affected earlier than others. However, a much larger sample size would be required to discern such an effect.
During development, the functional maturation of inhibitory circuits is thought to trigger the onset and closure of the critical period for OD plasticity in V1 (Jiang et al., 2005; Levelt and Hübener, 2012; for review, see Hensch, 2004, 2005;). Increasing cortical inhibition, for example with the GABA agonist benzodiazepam, can induce a precocious critical period (Fagiolini and Hensch, 2000). Conversely, treatments that limit inhibition in the adult, such as injection of the GABA(A) antagonist bicuculline (Duffy et al., 1976; Sillito et al., 1981), or treatment with the antidepressant fluoxetine (Sale et al., 2007; Maya Vetencourt et al., 2008; Harauzov et al., 2010), can restore juvenile plasticity. Fluoxetine treatment during normal vision enhances structural dynamics of cortical interneurons, and during MD provides a permissive environment that facilitates OD plasticity (Chen et al., 2011b). Here we show that fluoxetine can also ameliorate the age-related decline in structural and functional plasticity of visual cortex neurons. Longer treatment time improved outcome, which could be because of the more prolonged treatment duration, or alternatively, to the earlier intervention. Although there was a measurable and significant improvement in the fluoxetine-treated older animal groups, it is interesting to note that not all animals showed improvement, suggesting that therapeutic intervention may be an avenue for reducing age-related neuronal deficits, but potentially not in all individuals. Clinical trials testing treatment with fluoxetine for patients with mild cognitive impairment, show improvement in the Mini-Mental Status Examination not seen in the placebo group (Mowla et al., 2007). This preliminary study needs to be confirmed in larger double-blind placebo-controlled cohorts. It would be interesting to note in this case too, if there is interpatient variability in response, as well as an efficacy dependence on treatment duration or time of intervention.
The traditional form of OD plasticity induced by MD in adult animals seems to be lost with aging (Lehmann and Löwel, 2008). However, some forms of experience-dependent plasticity persist in adults (Sawtell et al., 2003; Frenkel et al., 2006), and much less is known regarding how these are affected by inhibition and modulatory systems across the lifespan. SRP is particularly robust in juvenile mice (Schecter et al., 2017), but can still be induced reliably in adults. Induction of SRP appears to share many features with the phenomenon of LTP (Cooke and Bear, 2010), and the expression of this plasticity is highly dependent upon intact cortical inhibition (Kaplan et al., 2016). Thus, it was of considerable interest to understand how this type of plasticity changes at ages when GABAergic neurons become structurally stable. Although we cannot ascribe a causal role of age-related changes in interneurons to the dramatic decline in SRP, it is particularly interesting that chronic treatment with fluoxetine, which partially restored malleability to interneurons in old age, also restored juvenile levels of SRP. Our finding that fluoxetine treatment in aging mice can attenuate the concurrent age-related declines in interneuron structural and visual cortex functional plasticity suggests it could provide an important therapeutic approach toward mitigation of sensory and cognitive deficits associated with aging, provided it is initiated before severe network deterioration.
Footnotes
This work was supported by National Eye Institute Grants RO1EY01756 (E.N.), RO1EY025437 (E.N.), and RO1EY023037 (M.B.), a Julie Martin Mid-Career Award in Aging Research to E.N. from the AFAR and Ellison Medical Foundation, a Machiah Foundation fellowship to R.E., and K99-NS076364 from NINDS to J.S. We thank Jae Won Cha and Peter So for two-photon microscopy support, and members of the Nedivi laboratory for critical reading of the paper.
The authors declare no competing financial interests.
- Correspondence should be addressed to Elly Nedivi, Massachusetts Institute of Technology, 43 Vassar Street, Cambridge, MA 02139. nedivi{at}mit.edu