Research Article

A brainwide atlas of synapses across the mouse life span

See allHide authors and affiliations

Science  17 Jul 2020:
Vol. 369, Issue 6501, pp. 270-275
DOI: 10.1126/science.aba3163

Brain synapses through the life span

Excitatory synapses connect neurons in the brain to build the circuits that enable behavior. Cizeron et al. surveyed synapses in the mouse brain from birth to old age and present the data as a community resource, the Mouse Lifespan Synaptome Atlas (see the Perspective by Micheva et al.). Molecular and morphological features defined 37 subtypes of synapses. Although synapse density generally increased in early development and declined in old age, the details differed in different brain areas.

Science this issue p. 270; see also p. 253

Abstract

Synapses connect neurons together to form the circuits of the brain, and their molecular composition controls innate and learned behavior. We analyzed the molecular and morphological diversity of 5 billion excitatory synapses at single-synapse resolution across the mouse brain from birth to old age. A continuum of changes alters synapse composition in all brain regions across the life span. Expansion in synapse diversity produces differentiation of brain regions until early adulthood, and compositional changes cause dedifferentiation in old age. The spatiotemporal synaptome architecture of the brain potentially accounts for life-span transitions in intellectual ability, memory, and susceptibility to behavioral disorders.

Excitatory synapses are the main class of brain synapse, and their postsynaptic proteins regulate both innate and learned behaviors (16). Mutations in these proteins cause >130 brain diseases (7), including disorders that characteristically arise in childhood, in adolescence, and in young or elderly adults.

Using synaptome mapping (8), we mapped excitatory synapse diversity and spatiotemporal synaptome architecture in >100 brain regions from birth until 18 months of age in mice (fig. S1). Synapses were labeled using fluorescent tags on endogenous PSD95 (PSD95-eGFP) and SAP102 (SAP102-mKO2) (8), two postsynaptic scaffold proteins that assemble multiprotein signaling complexes (912) necessary for synaptic plasticity and innate and learned behaviors (2, 3, 57, 12). Disrupting the normal expression of these scaffold proteins or their associated proteins results in human neurodevelopmental and psychiatric disorders including autism, schizophrenia, and intellectual disability (6, 7, 9, 12).

Our findings reveal a spatiotemporal program of synapse diversity across the brain, which we call the life-span synaptome architecture (LSA). The LSA shows how synapse diversity is generated as brain regions become dissimilar and how the synaptome architecture changes through development to adulthood and old age. The LSA provides a framework for understanding stereotypical life-span trajectories of behavioral changes and psychological functions (1315) and why gene mutations characteristically result in synaptic pathology in certain brain areas and at certain ages. The Mouse Lifespan Synaptome Atlas and interactive visualization and analysis tools (16) provide a community resource for investigation of synapse function across all brain regions and the 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).

Fig. 1 Life-span trajectories of synapse parameters.

(A) PSD95-eGFP (green) and SAP102-mKO2 (magenta) expression acquired at low [20× (i and ii)] and high [100× (iii)] magnification in the whole brain (i), hippocampus (ii), and molecular layer of the dentate gyrus (iii) at 10 ages across the mouse postnatal life span. Scale bars, 4 mm (i), 500 μm (ii), 3.5 μm (iii). D, day; W, week; M, month. (B) Life-span trajectories of synapse density, intensity [normalized to the mean intensity, arbitrary units (AU)], and size in the whole brain. Shown are PSD95-eGFP (green) and SAP102-mKO2 (magenta). Points represent individual mice, with beta-spline-smoothed curve of mean values and SEM shown. (C) Differences (Cohen’s d) in synapse parameters between 3M and 18M in brain subregions (numbered; see table S1). *P < 0.05, Bayesian test with Benjamini-Hochberg correction. CB, cerebellum; CTXsp, cortical subplate; HPF, hippocampal formation; HY, hypothalamus; MB, midbrain; MY, medulla; OLF, olfactory areas; P, pons; PAL, pallidum; STR, striatum; TH, thalamus.

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 (1922). 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 (2426).

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.

Fig. 2 Life-span trajectories of synapse types, subtypes, and diversity.

(A) Stacked bar plot of percentage of synapse type density (type 1, PSD95 only; type 2, SAP102 only; type 3, colocalized PSD95+SAP102) in the whole brain across the life span. (B) Percentage of synapse subtype density in the whole brain across the life span. Key: synapse subtypes (1 to 37). (C) Percentage of synapse subtype density in hippocampus and cerebellum across the life span. (D) Life-span trajectories of synapse type density in 12 regions and 109 subregions (rows; see table S1). Density in each subregion was normalized (0 to 1) to its maximal density across the life span (columns). Twelve brain regions are shown (abbreviations as in Fig. 1C). (E) Life-span trajectories of three representative synapse subtypes (2, 16, and 31) in each of 109 subregions (rows; see table S1). Density in each subregion was normalized (0 to 1) to its maximal density across the life span (columns). (F) Life-span trajectories of synapse diversity (Shannon entropy) for whole brain (top) and main regions from the cerebrum (middle) and brainstem and cerebellum (bottom). Beta-spline-smoothed curve of mean and SEM are shown. (G) Unsupervised synaptome maps showing the spatial patterning of synapse diversity (Shannon entropy) per area (pixel size 21.5 × 21.5 μm) in representative parasagittal sections [for all ages, see fig. S15 and (8)].

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).

Fig. 3 Life-span synaptome architecture.

(A) Matrix of similarities between pairs of subregions (rows and columns) at 1W, 3M, and 18M (for all ages, see fig. S16). Small white boxes indicate the subregions that belong to the same main brain region (see color code, left and top) and larger white boxes indicate main clusters: cerebrum, brainstem, and cerebellum. Note the reduction in similarity from 1W to 3M and the increase to 18M. Iso, isocortex; other abbreviations as in Fig. 1C. (B) Similarity ratio comparing the relative similarity of the synaptome in each main brain region with that of every other region (8). See the materials and methods for details. There were significant differences in the ratio between 3M and other ages: **P < 0.01, ***P < 0.001, two-way ANOVA with post hoc multiple-comparisons test. (C) Whole-brain hypersimilarity matrix showing the similarity between pairs of subregions at all ages. White boxes indicate the three main clusters corresponding to LSA-I, LSA-II, and LSA-III. Yellow boxes indicate the increased similarity of the old brain with the young brain (see higher-magnification image in fig. S18). (D) Hippocampus hypersimilarity matrix showing the similarity of pairs of hippocampal subregions at all ages. White boxes indicate the three main clusters corresponding to LSA-I, LSA-II, and LSA-III. Yellow boxes indicate the increased similarity of the old brain with the young brain. (E) Average small worldness across the life span. Scatter plots indicate the average small worldness per mouse brain section at different ages. There were significant differences in small worldness between 3M and other ages: *P < 0.05, **P < 0.01, ***P < 0.001, two-way ANOVA with post hoc multiple-comparisons test.

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.

Fig. 4 Life-span changes in hippocampus architecture and electrophysiological properties.

(A) Schematics of the hippocampus showing radial and tangential gradients in CA1sr subfield. Graphs show gradients of normalized synapse intensity (AU) of PSD95 and SAP102 at 1W, 3M, and 18M. CA1, cornu ammonis 1; CA2, cornu ammonis 2; CA3, cornu ammonis 3; DG, dentate gyrus; gr, granular layer; mo, molecular layer; po, polymorphic cell layer; slm, stratum lacunosum-moleculare; slu, stratum lucidum; so, stratum oriens; sp, stratum pyramidale; sr, stratum radiatum. (B) Summed response (EPSP amplitude) to three patterns (gamma, theta, and theta burst) of 20 action potentials of the 11 × 11 matrix of hippocampus synapses at three ages. Histograms show changes (summed Euclidean distance, ED) between 1W and 3M (purple) and 3M and 18M (yellow). (C) Schematic of the flow of information (arrows) in the trisynaptic hippocampal circuit connecting the DG molecular layer (DGmo), CA3 stratum radiatum (CA3sr), and CA1 stratum radiatum (CA1sr) and the life-span trajectory of synapse subtype density (normalized) in each region. EC, entorhinal cortex.

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).

Discussion

The dynamic temporal trajectories of excitatory synapse number, protein composition, morphology, and type and subtype diversity in >100 brain areas reveal a life-span synaptome architecture for the mouse brain. The continuum of changes in the synaptome architecture is divided into three epochs that broadly correspond to childhood and adolescence, early adulthood, and late adulthood. Synapse diversity expands between birth and early adulthood, driving the differentiation of brain regions, before changes in synapse composition progressively dedifferentiate brain regions in old age. These changes alter brain network and hippocampal physiological properties, and are potentially relevant to the trajectory of cognitive functions described in life-span studies of human behavior (3134) and changes in the behavioral repertoire of animals across the life span (1315).

The LSA reveals how factors that modify the expression of synaptic proteins (including genetic mutations, toxic proteins, inflammation, and drugs) can target particular synapses and brain regions at different ages and lead to behavioral changes. Expanding our approach to other synaptic proteins labeling greater synapse diversity, examining the synaptome of neuron types and dendritic morphology, and linking these brainwide synapse resolution data to transcriptional mechanisms that control brain gene expression across the life span (35) should enhance the effort to uncover the mechanisms and impacts of brain development, aging, and disease. Our highly scalable synaptomic methods and the Mouse Lifespan Synaptome Atlas described here provide new tools for addressing these issues.

Supplementary Materials

science.sciencemag.org/content/369/6501/270/suppl/DC1

Materials and Methods

Figs. S1 to S23

Table S1

References (3843)

MDAR Reproducibility Checklist

References and Notes

Acknowledgments: We thank C. McLaughlin and K. Elsegood for mouse colony and laboratory management; D. Kerrigan and D. Fricker for genotyping; N. G. Skene for statistical advice; S. Munni, O. Kealy, and H. Taczynski for image calibration; D. Maizels for artwork; and C. Davey for editing. Funding: This work was supported by the Wellcome Trust (Technology Development grant no. 202932), the European Research Council (ERC) under the European Union’s Horizon 2020 Research and Innovation Programme (695568 SYNNOVATE), and the Simons Foundation Autism Research Initiative (529085). Author contributions: M.C. designed the animal cohort; collected, prepared, and imaged brain samples; delineated brain regions and subregions; calibrated synapse detection; analyzed synapse parameters and hippocampal gradients; and interpreted the data. Z.Q. developed the methodology and optimized the life-span SYNMAP pipeline; analyzed data of the whole life-span mouse cohort; segmented images and quantified and classified puncta; performed unsupervised and supervised mapping of synapse parameters, types, and subtypes; and analyzed diversity and network topology. E.F. performed statistical analysis and computational modeling of synaptome physiology. B.K. constructed the Synaptome Explorer and the Synaptome Homology Viewer. R.G. constructed the website. N.H.K. provided advice and supervision. S.G.N.G. conceived the project, performed analyses, supervised the project, and wrote the paper. Competing interests: The authors declare no competing interests. Data and materials availability: All data are available at the Mouse Synaptome Atlas (16) and Edinburgh DataShare (36), and code is available at Zenodo (37). Requests for materials should be addressed to S.G.N.G.

Stay Connected to Science

Navigate This Article