A brainwide atlas of synapses across the mouse life span

Life-span synaptome mapping pipeline and data resource

Parasagittal brain sections from cohorts of PSD95-eGFP; SAP102-mKO2 male mice were collected at 10 postnatal ages: 1 day (1D), 1 week (1W), 2 weeks (2W), 3 weeks (3W), 1 month (1M), 2 months (2M), 3 months (3M), 6 months (6M), 12 months (12M), and 18 months (18M) (Fig. 1A and figs. S2 and S3). Whole-brain sections were imaged at single-synapse resolution (17) on a spinning disc confocal microscope (pixel resolution 84 nm and optical resolution ~260 nm), and the density, intensity, size, and shape parameters of individual puncta were acquired using computer vision methods described previously (8). Synapses were classified into three types: type 1 express PSD95 only, type 2 express SAP102 only, and type 3 express both PSD95 and SAP102 (8). Thirty-seven subtypes were defined on the basis of molecular and morphological features (8). Supervised synaptome maps were generated by registering the data to the Allen Reference Atlas (18). Data were delineated into 109 anatomical subregions within 12 overarching regions, comprising isocortex, olfactory areas, hippocampal formation, cortical subplate, striatum, pallidum, thalamus, hypothalamus, midbrain, pons, medulla, and cerebellum (table S1).

• Download high-res image
• Open in new tab
• Download Powerpoint

All data and analysis tools are available in the Mouse Lifespan Synaptome Atlas (16). Synaptome Explorer enables in-depth exploration of raw and processed image data in single sections at single-synapse resolution, and the Synaptome Homology Viewer enables comparison of brain regions within and between mice of different ages.

The synaptome continuously changes across the life span

Raw images at low and high magnification revealed that each synaptic protein has a distinct spatiotemporal pattern and that the synaptome changes with age (Fig. 1A and figs. S2 and S3). To quantify the spatiotemporal differences in the synaptome, the life-span trajectories of PSD95 and SAP102 puncta density, intensity, and size were plotted as graphs and heatmaps revealing characteristic patterns for the whole brain, 12 regions, and 109 subregions (Fig. 1B, figs. S4 and S5, and table S1). Each parameter continuously changed across the life span. Synapse density rapidly increased during the first month in all brain areas and then fluctuated before declining in old age (Fig. 1, A to C, and figs. S3 to S5) (whereas the adult brain size remained unchanged; fig. S6), consistent with previous studies of synapse number quantified using electron microscopy in the rat brain (19–22). Each brain area underwent a specific program of synapse development, maturation, and aging. For example, the density of synapses peaked in the brainstem before cerebrum structures, potentially reflecting the requirement for the brainstem in early postnatal functions (fig. S7). The two synapse proteins showed different spatiotemporal trajectories (Fig. 1, B and C, and figs. S4 and S5), with SAP102 puncta density peaking before that of PSD95 in most brain areas (fig. S7), consistent with previous literature (23). Although, together, PSD95 and SAP102 label most excitatory synapses, additional markers would be required for an assessment of total excitatory synapse number in all brain regions and at all developmental stages.

Between 3M and 18M, most brain regions and subregions showed significantly (P < 0.05, Bayesian test with Benjamini-Hochberg correction) reduced synapse density (Fig. 1C, top panel, 70/109 subregions for PSD95; 78/109 for SAP102; fig. S8) and increased size (Fig. 1C, bottom panel, 56/109 subregions for PSD95, 80/109 for SAP102; fig. S8). Examination of the size distribution of the synapse populations showed a shift toward larger synapses with age (effect size >0.25 with P < 0.01, Kolmogorov-Smirnov test), consistent with previous electron microscopy studies in the aging macaque dorsolateral prefrontal cortex (24–26).

Life-span changes in the synaptome architecture can be divided into three broad phases. During the first phase (LSA-I), from birth to 1M, the numbers of puncta increased rapidly. The second phase (LSA-II) began as the rate of increase in puncta density slowed and was characterized by relative stability until 6M (adulthood). The third phase (LSA-III), late adult life, was characterized by a decline in puncta density and an increase in synapse size (Fig. 1C and fig. S8).

Synapse diversity across the life span

Each synapse type (Fig. 2, A and D, and figs. S9 and S10) and subtype (Fig. 2, B, C, and E, and figs. S11 and S12) had a specific trajectory in each brain region and subregion, reaching their peak values at different ages. Thus, the synapse composition of brain regions continued to change throughout the life span and was not restricted to LSA-I, when synapse density increased. Moreover, the presence of more than one peak at different ages (e.g., subtypes 17 and 18, P < 0.05, paired t test and Kolmogorov-Smirnov test, Cohen’s d > 1.2; figs. S10 to S12) suggests that shaping synapse composition (through processes such as transcriptional regulation, synapse pruning, and growth) is an ongoing process. In LSA-III, some subtypes were decreased (P < 0.05, Bayesian test with Benjamini-Hochberg correction), whereas others were increased (P < 0.05, Bayesian test with Benjamini-Hochberg correction), with differing specificity to brain regions (fig. S13). For example, subtypes 2, 27, and 34, which are large synapses, increased in many brain regions (Fig. 2E and figs. S12 and S13), whereas subtypes 12, and 14 to 16, which are small synapses, were lost in olfactory areas and thalamus in the old brain (Fig. 2E and figs. S12 and S13). Thus, subtypes of excitatory synapses are selectively gained or lost with aging, and different regions of the brain age in distinct ways.

• Download high-res image
• Open in new tab
• Download Powerpoint

LSA-I was initially dominated by a small subset of synapse types and subtypes, and these were overtaken by expanding populations of other types and subtypes (Fig. 2, A to C, and figs. S9 and S11). For example, type 2 and subtype 16 synapses dominated in the first postnatal week (Cohen’s d > 2 with P < 0.01, two-way ANOVA with post hoc multiple-comparisons test) but were reduced by 1M (Cohen’s d < –2, P < 0.001, two-way ANOVA with post hoc multiple-comparisons test). We next quantified synapse diversity and found that all regions and subregions showed a rapid initial increase in the first 3 postnatal weeks (Fig. 2F and fig. S14). Brain areas responsible for higher cognitive functions (isocortex, cortical subplate, hippocampus, striatum) continued to expand their excitatory synaptic diversity (as reflected by the markers PSD95 and SAP102) after LSA-I, reaching a peak at 2M, whereas brain areas serving basal neurophysiological functions (midbrain, pons, medulla) peaked at 3W to 1M during LSA-I (Fig. 2F and fig. S14). Synapse diversity plateaued from 3M on in most brain areas (Fig. 2F). Unsupervised synaptome maps of the mouse brain, which visualize the anatomical distribution of synapse diversity (Fig. 2G and fig. S15), clearly show the increase in diversity in LSA-I, with the emergence of layers in the isocortex and subregional differentiation in the hippocampus.

Synaptome architecture first specializes, then dedifferentiates

To reveal how changes in synapse composition might contribute to differences between brain areas, we plotted similarity matrices of brain subregions at each age (Fig. 3A and fig. S16) (matrices were nonrandom, P < 0.05, Cohen’s d > 2, permutation test). Similarity across all brain areas was highest in the first postnatal week and diminished until 3M (P < 0.001, two-way ANOVA with post hoc multiple-comparisons test) (Fig. 3, A and B, and fig. S16). As the brain aged beyond 3M, there was a progressive increase in the similarity between brain areas (P < 0.001, two-way ANOVA with post hoc multiple-comparisons test) (Fig. 3, A and B, and fig. S16). Individual brain regions also showed a reduction in similarity with the rest of the brain (i.e., differentiation) during the first 3 months (P < 0.001, two-way ANOVA with post hoc multiple-comparisons test) (fig. S17), and from 3M to 6M on all regions except the cerebellum and medulla showed an increase in similarity (i.e., dedifferentiation) (P < 0.01, two-way ANOVA with post hoc multiple-comparisons test) (fig. S17).

• Download high-res image
• Open in new tab
• Download Powerpoint

We investigated whether dedifferentiation in LSA-III represents a return to a synaptome resembling that of a young brain or to a distinct, elderly-specific synaptome architecture. Using a hypersimilarity matrix that compares all subregions at all ages, we found that the 18M brain, in contrast to the 3M brain, was more similar to the 2W brain (Cohen’s d = 1.5, P < 0.01, Bayesian test) (yellow boxes in Fig. 3C and figs. S18 and S19A). The hypersimilarity matrix also revealed three major blocks corresponding to the LSA phases (white boxes in Fig. 3C and fig. S18), with a transition between LSA-I and LSA-II at 3W, which corresponds to the behavioral transition from dependence on maternal care to independent living. The hypersimilarity matrix of hippocampal subregions showed a similar pattern (Cohen’s d = 1.7, P < 0.01, Bayesian test) (Fig. 3D and fig. S19B).

To identify the synapse types and subtypes contributing to the differentiation-dedifferentiation trajectory, we correlated the abundance of each synapse subtype with the similarity ratio using brain-wide data and regional data (fig. S20), revealing a role for all three synapse types and a subset (21/37) of synapse subtypes (r > 0.5 or r < –0.5, P < 0.05, Mantel test with Benjamini-Hochberg correction).

The functional connectivity between brain areas, measured using resting-state functional magnetic resonance imaging, correlates with the topology (small worldness) of the synaptome network (8). Small worldness increased from birth to 3M (P < 0.001, two-way ANOVA with post hoc multiple-comparisons test) and then declined to 18M (P < 0.001, two-way ANOVA with post hoc multiple-comparisons test) (Fig. 3E), suggesting that the differentiation-dedifferentiation trajectory influences the integrative property of brain circuits.

Life-span synaptome changes alter functional outputs

To explore how the age-dependent changes in synaptome architecture may cause changes in cognitive functions, we focused on the hippocampal formation, which is key for spatial navigation, learning, and memory (27). In the CA1 stratum radiatum of the adult mouse, there are orthogonal (radial and tangential) spatial gradients in PSD95 and SAP102 synaptic parameters that produce a local architecture of molecularly diverse synapses (8, 28). Quantification of these gradients at 1W, 3M, and 18M showed age-dependent changes for PSD95 intensity in both radial and tangential directions (Fig. 4A and fig. S21). Conversely, for SAP102 no radial gradient was observed, and the tangential gradient was established by 1W and thereafter remained unchanged (Fig. 4A and fig. S21). This shows that these two closely related synaptic proteins undergo distinct spatiotemporal changes within the dendrites of CA1 pyramidal cells, producing a changing two-dimensional synaptome map across the life span. Using a computational simulation approach that tests the response (excitatory postsynaptic potential, EPSP) of CA1 synaptome maps to patterns of neural activity (8), we found that gamma and theta-burst patterns produced differential responses between 1W to 3M and 3M to 18M (P < 0.05, paired t test, Kolmogorov-Smirnov test), in contrast to theta trains, which produced a stable response at all ages (Fig. 4B and fig. S22). This illustrates how life-span synaptome changes affect synaptic responses to distinct temporal patterns of neural activity.

• Download high-res image
• Open in new tab
• Download Powerpoint

Different subregions of the hippocampal formation contribute distinct cognitive functions, which together produce an integrated behavioral output (27, 29, 30). This integrated function is exemplified by the trisynaptic circuit, in which axons project from neurons in the dentate gyrus to CA3 neurons, which project to CA1 neurons (27). Our data showed that each subregion in the trisynaptic circuit undergoes a different life-span trajectory of synaptic subtype composition, indicating that the memory functions controlled by these hippocampal subregions are highly likely to change with age (Fig. 4C and fig. S23).