Main

Ageing is accompanied by a loss of proteostasis, which involves the maintenance of a balanced and functional proteome1,4. All aspects of proteostasis are disrupted with ageing, including the balance of protein synthesis with protein degradation, protein transport and protein folding1,4. The loss of proteostasis in the brain contributes to age-associated vulnerability to reduced cognitive and motor abilities and neurodegenerative diseases2. Indeed, experimentally compromising broad proteostasis pathways can induce dementia-like phenotypes2,5,6. Understanding the dynamics of proteostasis in a neuron-specific manner may define mechanisms or individual proteins that could be exploited for therapeutic purposes. However, despite the emergence and application of several tools to study cellular proteomes3,7,8,9,10,11,12,13,14,15,16, such research has been hindered by a lack of robust models to examine protein dynamics in a cell-specific manner in mammals. Here we develop in vivo models that enable robust tagging of nascent proteomes with non-canonical biorthogonal amino acids in a cell-specific manner through the expression of mutant aminoacyl-tRNA synthetases (aaRSs). We leverage these models to study key features of neuronal proteostasis dynamics with ageing to provide detailed insights into the decline of neuronal proteostasis with age. We also describe a microglia-mediated mechanism that maintains neuronal proteostasis.

Proteome labelling by BONCAT models

Expanding on our previous in vitro studies3, we generated two new bioorthogonal non-canonical amino acid tagging (BONCAT) knock-in mouse lines with cassettes that express the mutant aaRSs flox-stop-flox-eGFP-p2a-PheRS(T413G) and flox-stop-flox-eGFP-p2a-TyrRS(Y43G) (hereafter, termed PheRS* and TyrRS*, respectively) (Fig. 1a). In vitro studies using these constructs confirmed that proteome tagging depended on the incorporation of the non-canonical amino acids during protein synthesis (Extended Data Fig. 1a). We first compared the protein tagging efficacy of our models to each other and to that of the current standard BONCAT knock-in mouse line based on the expression of a mutant methionine aaRS (hereafter, termed MetRS*)7,8. Each of the BONCAT lines was crossed to a Camk2a-cre driver, and the resulting offspring (Camk2a-cre+/−;BONCAT*+/–*) were uniformly treated with their respective azido-modified amino acid (AzAA) to evaluate nascent protein labelling in CAMK2A+ neurons (Fig. 1a,b). Examination by in-gel fluorescence revealed that the Camk2a-cre;PheRS* model showed a high fluorescence signal over its respective background control, whereas the Camk2a-cre;TyrRS* and Camk2a-cre;MetRS* models did not show an appreciable difference relative to their respective background controls (Extended Data Fig. 1b). These results were supported by in situ tissue staining for the azide-modified proteins (Fig. 1c). Labelled proteins in brain sections from Camk2a-cre;PheRS* mice colocalized to GFP+ neurons with the expected spatial distribution of CAMK2A+ neurons (Fig. 1c and Extended Data Fig. 1c–e). Last, we evaluated protein labelling by performing liquid chromatography with mass spectrometry (LC–MS) on BONCAT-labelled proteins enriched by bead-based pull-down (Extended Data Fig. 1f–h). Principal component analysis (PCA) clearly separated the different models (Fig. 1d). We detected 3,787 proteins in Camk2a-cre;PheRS* mice, 2,320 proteins in Camk2a-cre;MetRS* mice and 4 proteins in Camk2a-cre;TyrRS* mice (Fig. 1e and Supplementary Table 1a). In particular, significant P values (Fig. 1f) and a high fold change in labelled signals relative to the background (Fig. 1g and Extended Data Fig. 1i,j) were observed in the Camk2a-cre;PheRS* model. The robust labelling achieved in the Camk2a-cre;PheRS* model (Extended Data Fig. 1k,l) did not induce HSP90 expression (Extended Data Fig. 1m) or microgliosis (Extended Data Fig. 1n). This result suggests that azide-modified residues do not induce proteostatic stress or a local immune response.

Fig. 1: Evaluation of nascent protein labelling of CAMK2A+ neurons by BONCAT mouse lines.

a, Schematics of BONCAT knock-in mice and methodology. The Camk2a-cre;PheRS* (Cam;PheRS*) and Camk2a-cre;TyrRS* (Cam;TyrRS*) lines were developed in this study. The Camk2a-cre;MetRS* (Cam;MetRS*) line is from a previous study7. NCAA, non-canonical amino acid. b, Timeline of NCAA administration. c, Images of BONCAT-labelled proteins (Click-555) in brain sections from the indicated Camk2a-cre+/−;BONCAT+/− mice. White outlines denote tissue border. Cb, cerebellum; Ctr, cortex; Ob, olfactory bulb. d, PCA based on the abundance of BONCAT-labelled proteins from the indicated Camk2a-cre+/−;BONCAT+/− mice. e, Venn diagram comparing the number of proteins identified in each Camk2a-cre+/−;BONCAT+/− mouse line. f, Heatmap comparing P values of proteins identified in the indicated Camk2a-cre+/−;BONCAT+/− mouse line. Proteins not identified in a particular line were assigned a –log10[P] value of zero. g, Top, volcano plots showing the enrichment of proteins identified in the indicated Camk2a-cre+/−;BONCAT+/− mouse lines relative to background controls (WT). Bottom, bar chart of the log2[fold change (FC)] in the signal of labelled protein relative to the background. h, Volcano plot for Camk2a-cre;PheRS* mice as in g, with dots colour-coded by cell-type enrichment. CellMarker and Panglao DB databases were used for analyses. Inset, barchart of the number of proteins per enrichment category (colour-coding consistent with key in top left of h). i, GO cellular component analysis of BONCAT-labelled proteins in Camk2a-cre;PheRS* mice. j, Images of BONCAT-labelled proteins in the motor cortex, striatum and hippocampus of a Camk2a-cre;PheRS* mouse. k, Venn diagram comparing the number of different proteins in the motor cortex, hippocampus and striatum of Camk2a-cre;PheRS* mice. l, PCA based on the abundance of BONCAT-labelled proteins from the motor cortex, striatum and hippocampus of Camk2a-cre;PheRS* mice. m, Heatmap comparing the z scored abundance of BONCAT-labelled proteins from the motor cortex, striatum and hippocampus of Camk2a-cre;PheRS* mice. Protein clusters enclosed by a white dotted line are regional marker proteins. n, Heatmap comparing pathway fold enrichment of the top ten GO biological processes for the motor cortex, striatum and hippocampus based on the regional marker proteins in m. n = 4–5 BONCAT mice and 4 respective background control mice in d–h. n = 4 mice per experimental group in k and l. P values in e,g,h and k were derived from two-tailed, two-sample Student’s t-tests. Scale bars, 20 µm (c,j (left and middle)) or 50 µm (j, right).

The performance of the Camk2a-cre;TyrRS* line was unexpected based on our previous observations3. To test whether the efficacy of protein labelling by each BONCAT line is related to the tissue examined, we performed similar experiments as described above in CMV-cre;BONCAT mice to induce ubiquitous cellular labelling in all tissues17. Depending on the tissue examined, different BONCAT lines had varying strengths in labelling tissue proteins (Extended Data Fig. 2a and Supplementary Table 1b). Several factors potentially contribute to the differences in labelling among tissues, including varying cell-type-specific expression of cognate tRNA molecules and cell states (Supplementary Text). These data demonstrate the strengths and potential utility of all three BONCAT lines in different tissue contexts.

The CAMK2A+ neuronal proteome

Given the robust labelling of neuronal proteins in Camk2a-cre;PheRS* mice, we further characterized this model. Many proteins (606) identified were annotated as neuronal (LY6H, SACS and SCNA, among others) with few (65) annotated as specific to other cell types (Fig. 1h and Supplementary Table 1c). As expected, many proteins labelled were marker genes of glutamatergic neurons (Extended Data Fig. 2b–e), a finding that was supported by in situ staining (Extended Data Fig. 2f,g). All major neuronal anatomical features were represented by hundreds of proteins (Fig. 1i and Supplementary Table 1d).

To assess regional proteomes in Camk2a-cre;PheRS* mice, we dissected the motor cortex, striatum and hippocampus, regions that exhibited robust labelling (Fig. 1j), and performed LC–MS on the enriched labelled proteins. A total of 3,054 proteins were commonly identified among all regions, but each region had 276–338 uniquely identified proteins (Fig. 1k and Supplementary Table 1e). PCA of the regional CAMK2A+ neuronal-labelled proteomes separated all three regions (Fig. 1l), which was reflected by hierarchical clustering and heatmap analyses (Fig. 1m and Supplementary Table 1f). We validated three regionally enriched proteins by immunostaining in situ (Extended Data Fig. 2h). Gene ontology (GO) biological process enrichment showed that each cluster or region was relatively unique in their pathway representation, which highlighted their regional specialization (Fig. 1n and Supplementary Table 1g).

Age-reduced neuronal protein degradation

Given the fundamental role of protein turnover in neurodegenerative diseases, particularly in long-lived, nonmitotic neurons18, we sought to determine how neuronal protein degradation changes with age. To rapidly deliver the BONCAT machinery to aged mice, we developed an adeno-associated virus (AAV) expression vector encoding a Camk2a-driven PheRS* (Extended Data Fig. 3a). Mice transduced with this construct showed significantly higher labelling than background controls (Extended Data Fig. 3b). On the basis of several measures, labelling in the AAV model was comparable with that of the analogous knock-in model (Extended Data Fig. 3c–h and Supplementary Table 1h). The AAV construct also enabled protein labelling in aged, 21-month-old mice, which facilitated comparisons of the ‘aged’ and ‘young’ neuronal-labelled proteomes (Extended Data Fig. 3i–k).

To study how protein degradation changes with age, young (4-month-old), middle-aged (12-month-old) and aged (24-month-old) mice were transduced with AAV:Camk2a-PheRS* by retroorbital injection with a pulse-chase AzF administration scheme (Fig. 2a). Mice were euthanized at 4 time points within the 2-week chase period, and brain regions were dissected immediately after brain extraction (Fig. 2a). In-gel fluorescence (Fig. 2b and Extended Data Fig. 4a) and in situ tissue staining (Fig. 2c) showed a dilution of tagged-protein fluorescence signals that progressed through the chase period, a result indicative of protein degradation. To quantify degradation rates for individual neuronal proteins across brain regions and ages, we enriched for tagged neuronal proteins, multiplexed enriched peptide fractions by tandem mass tag (TMT) labelling and analysed the plexes by LC–MS (Fig. 2a). We obtained degradation trajectories of the per cent protein remaining over time for every protein identified for each region and each age (Fig. 2d, Extended Data Fig. 4b,c and Supplementary Table 2a). The average degradation trajectories for all proteins among regions differed (Fig. 2d and Extended Data Fig. 4d). Moreover, the average degradation trajectories of regions in aged mice relative to their respective regions in young and middle-aged mice were broader or had lower slopes (Fig. 2d). This finding indicates that protein degradation slows with ageing and emerges after middle age, a result that was quantitatively supported (Extended Data Fig. 4d).

Fig. 2: Neuronal protein degradation slows with ageing and is regionally heterogeneous.

a, Schematic of the approach used to study protein degradation by BONCAT. n = 4 BONCAT mice per time point (TP1–TP4) for each age and n = 2 mice per age for background controls. b, In-gel fluorescence images (top and middle) and quantification (bottom) of BONCAT-labelled proteins in whole brain lysates derived from Camk2a-cre;PheRS* mice at the indicated time points in the chase period. AU, arbitrary units. c, Images of BONCAT-labelled proteins in the cortex of brain tissue sections from Camk2a-cre;PheRS* mice at the indicated time points in the chase period. Scale bars, 20 µm (left column) or 10 µm (right column). d. Trajectories of the per cent of BONCAT-labelled protein remaining through the chase period. Each thin line represents one protein derived from averaging four biological replicates. The single bold line represents the average of all proteins. Proteins were filtered to only exclude proteins with a 5% increase between any two time points. e, Plot of the estimated protein half-life in days for the indicated brain regions and ages (A, aged; MA, middle-aged; Y, young). Each dot represents one protein. For each individual brain region, only proteins commonly identified between all ages of that region are plotted. f, Plot of log2[FC] of estimated protein half-life values between the indicated brain regions and ages. Each dot represents one protein and is the same as those displayed in f. g, Bar plot of the number of proteins with an age-increased half-life that are also risk genes for the indicated brain disorders. Proteins highlighted are the top five most half-life-increased risk genes of the indicated diseases with age in the sensory cortex. h, Scatter plots comparing the log2[FC] of the estimated protein half-lives (young to aged) between proteins commonly detected between the indicated brain regions. Proteins with an absolute value difference of >1 were considered regionally vulnerable. P values were determined by two-sided Person’s correlation tests. i, Bar plot of the number of regionally vulnerable proteins among the indicated brain regions. j, Bar plot of the number of neurodegenerative risk genes in the identified regionally vulnerable proteins for the indicated brain regions. P values in e and f were determined by paired two-tailed t-tests between young and aged proteins. ***P < 0.0001 risk genes quantified in g and j were derived from the H-MAGMA study28 and considered only if the originally reported P value was <0.05. AD, Alzheimer’s disease; ALS, amyotrophic lateral sclerosis; CAD, coronary artery disease; DDA, data-dependent mode; MDD, major depressive disorder; MS, multiple sclerosis; PD, Parkinson’s disease; SCZ, schizophrenia. Mice in a were created in BioRender. Guldner, I. (2025) https://BioRender.com/9itwqmf.

We estimated protein half-life by using established modelling techniques19,20, which correlated well with direct interpolation of the half-life values from the trajectories (Extended Data Fig. 4e). The estimated half-life values further showed relatively stable average half-lives from young to middle age (except for the hypothalamus), with an average increase of 2.27–5.04 days from young to aged mice (Fig. 2e and Supplementary Table 2b). The average fold change in half-life among all regions was 1.2 (around 20% increase, mostly attributed to the hypothalamus) from young to middle age, but approximately 2.03 (100% increase or doubling) from young to aged (Fig. 2f and Supplementary Table 2b). The observation of reduced protein degradation with age is consistent with previous reports21,22,23, as was the lack of correlation between protein abundance and half-life24,25 (Extended Data Fig. 4f). Of the cortical proteins, those with the greatest fold change increase (top 10%) from young to aged were enriched for proteins of the synapse (PPP2R1A, DTNA and DNM2), cell junctions (GRM3, RTN3 and GJA1) and mitochondria (Extended Data Fig. 4g). These neuronal features have been observed to be compromised in ageing and dementias26,27. Notably, several hundred of the proteins that exhibited an aged-increased half-life are encoded by neurodevelopmental or neurodegenerative risk genes identified in a study that used Hi-C-coupled multimarker analysis of genomic annotation (H-MAGMA)28 (Fig. 2g). Some of the proteins with the most age-increased half-life are encoded by neurodegenerative risk genes such as CPLX1, DCLK1, FERMT2 and YWHAQ (Fig. 2g). These proteins localize to cell junctions and the actin cytoskeleton and are involved in cell–cell junction organization and signalling, which implies that reduced protein degradation has repercussions for both the host cells and their signalling partners. Age-reduced degradation only slightly correlated with certain protein features (Extended Data Fig. 4h and Supplementary Table 2c).

Next, we compared half-life fold changes of proteins shared across regions. We speculated that proteins with differing half-life changes with age could contribute to regional vulnerability or resilience to ageing and diseases. No two regions were perfectly correlated (Fig. 2h and Extended Data Fig. 4i). The visual cortex and hypothalamus displayed the fewest changes in half-life fold change (n = 3 proteins, |log2[FCAged/Young Visual cortex] – log2[FCAged/Young Hypothalamus]| > 1), with more changes present when comparing the sensory cortex to either the hippocampus (n = 39 proteins) or hypothalamus (n = 12 proteins) (Fig. 2h,i). In support of the hypothesis that these proteins could confer vulnerability or resilience, several of these proteins are encoded by known neurodegenerative risk genes (Fig. 2j). Some of the proteins identified, such as LRPPRC or RAPGEF2, have been experimentally demonstrated to contribute to Alzheimer’s disease progression, and others have strong disease correlations. However, the contribution of many other proteins, including DLG2, NLGN3, STXBP1 and TMED10, to ageing and neurodegeneration remains to be elucidated.

Coordinated degradation of proteins

As half-life values can oversimplify nuances of complex kinetic degradation trajectories such as those shown in Fig. 2d, we performed analyses on the degradation trajectories to extract additional information (Fig. 3). First, we clustered the degradation trajectories on a per region basis (Fig. 3a and Extended Data Fig. 5a). As an example, the ‘young’ sensory cortex had five clusters with visually distinguishable average profiles (Fig. 3b) and slope values (Fig. 3c). The top five most enriched GO biological process terms of each cluster in the sensory cortex were mostly unique to each cluster and were largely represented by one or two broad biological processes (Fig. 3d). This result supported the biological meaningfulness of the clustering, which was generally recapitulated among other brain regions (Extended Data Fig. 5b). These data suggest that proteins in similar pathways have coordinated degradation rates, an observation complementary to previous findings that individual proteins in multiprotein complexes share similar half-lives24,29.

Fig. 3: The coordinated degradation of proteins by biological function is differentially compromised with ageing.

a, Trajectories of the five clusters identified by clustering all of the protein-degradation trajectories from the sensory cortex of young mice. Red lines represent proteins closer to the cluster average. b, Plot of the average degradation trajectory for each cluster visualized in a from the sensory cortex of young mice. c, Bar plot comparing the slopes of the kinetic degradation trajectories of each protein in each cluster (C1–C5) for the sensory cortex of young mice. Each dot represents the slope of the kinetic degradation trajectory of one protein. d, Heatmap of the top five most significant GO biological processes identified for each cluster in the sensory cortex of young mice. Heatmap colours represent the fold enrichment for each pathway. e, Overlap of degradation trajectories of the six clusters identified in the sensory cortex of young (4-month-old) and aged (24-month-old) mice, with lines colour-coded by age. The average ΔIntegral, calculated by averaging the difference of the integral values of each protein in aged and young mice in the cluster, is provided on each plot. f, Bar plot comparing the integral values of proteins from young and aged mice in each cluster of the sensory cortex. Each dot represents the integral value for one protein in the indicated cluster. g, Bar plot comparing the integral values of proteins from young and aged mice on a per-region basis. n = 730, 380, 507 and 386 integral values for the sensory cortex, visual cortex, hippocampus and hypothalamus, respectively. Error bars represent the mean with s.d. h, Heatmap of the top five most enriched GO biological processes identified for each cluster in each brain region examined. Regions and respective clusters are indicated at the top of the heatmap. Heatmap colours represent the fold enrichment for each pathway. The annotation at the top of the heatmap represents the ΔIntegral of the indicated region or cluster. The data in c and e–g were determined by an ordinary one-way analysis of variance (ANOVA) with significant comparisons identified by Tukey tests. ****P *< 0.0001.

Selective vulnerability to degradation

We next compared how clusters change with age in the sensory cortex (Fig. 3e). For the sensory cortex and all other brain regions examined, profiles from aged mice had a larger integral value (area under the curve) than profiles from young mice for each cluster (Fig. 3f,g, Extended Data Fig. 5a and Supplementary Table 2d), which signified reduced degradation rates in aged mice. Only the visual cortex and hypothalamus showed increased average integral values from young to middle-aged mice (Extended Data Fig. 5c), which signified earlier degradation deficits in these regions.

We calculated the average difference of the integral values for proteins from aged and young mice of each cluster to obtain a delta integral (ΔIntegral) score for each cluster (Fig. 3e), a measure of the magnitude of protein degradation reduction between clusters. In the sensory cortex, cluster 1, which was enriched for synapse transmission functions (Fig. 3d), had one of the largest ΔIntegral scores (1.55) (Fig. 3e). By contrast, cluster 5, which was enriched for metabolic processes (Fig. 3d), had the smallest ΔIntegral score (0.77) (Fig. 3e). The ΔIntegral scores suggest that certain biological processes, such as synaptic transmission, are more vulnerable to the consequences of age-related degradation than other processes, such as metabolism, in the sensory cortex.

We extended this integral score and pathway analysis to all clusters of all regions (Fig. 3h). Although most clusters, regardless of region, had similar integral scores, a few clusters had substantially higher or lower ΔIntegral scores (Fig. 3h, ΔIntegral annotation on the heatmap), which suggests that there is more prominent vulnerability or resilience, respectively, with ageing.

The aged neuronal aggregome

There are many potential causes for the age-related reduction in protein degradation, including the formation of protein aggregates4. Neuronal protein aggregation increases with age in mice, and aggregates have been detected (using Proteostat30) in human brains from old individuals (Fig. 4a–c). Moreover, protein aggregates are commonly associated in age-related brain diseases18. Therefore, we followed up on the connection of aggregation and age-reduced protein degradation. Combining protein aggregate isolation techniques31 with neuronal BONCAT labelling enabled us to define the neuronal aggregome, a catalogue of neuronal proteins that contribute to protein aggregates in the aged brain (Fig. 4d). LC–MS analyses enabled us to identify 1,726 neuronal proteins present in aggregates in the aged brain (Fig. 4e, Extended Data Fig. 6a–c and Supplementary Table 3a), 392 of which have been identified in aggregates of human brains from old individuals32 (Extended Data Fig. 6d). We selected RTN3 and SRSF3 for orthogonal validation, as these have been previously identified in aggregate omics data32,33 and been implicated to have a role in dementias34,35,36. Through immunofluorescence staining of brain tissue, we confirmed that RTN3 and SRSF3 formed ubiquitin-tagged and p62-tagged aggregate-like puncta in aged mice but not in young mice, particularly in the hippocampus (Fig. 4f–h). Some proteins identified are well known to aggregate in neurodegenerative diseases, including TDP43, FUS and NSF, whereas most were previously not reported to aggregate (Supplementary Table 3a). In further support of the likely relevance of these aggregating proteins in contributing to ageing and diseases, 1,195 (69%) of the aggregating neuronal proteins are encoded by risk genes as defined by the H-MAGMA study28 (Fig. 4i–k and Supplementary Table 3b). Several protein features implicated in aggregation propensity were altered between aggregating and non-aggregating proteins in the sensory cortex (Extended Data Fig. 6e and Supplementary Table 3c). GO cellular component analysis revealed that aggregating neuronal proteins could be ascribed to several neuronal compartments. However, synapse-related terms were recurrently represented (Fig. 4l and Supplementary Table 3d) and represented a range of synaptic anatomy and function (Extended Data Fig. 6f). GO biological process analysis showed that several cellular functions were enriched, with protein localization recurrently represented (Extended Data Fig. 6g and Supplementary Table 3e).

Fig. 4: Aggregating neuronal proteins in aged brains have links to age-related degradation deficits, synaptic dysregulation and proteinopathies.

a, Images of young (4-month-old) and aged (24-month-old) mouse cortex sections stained for neurons (NeuN, red) and protein aggregates (Proteostat, green). b, Quantification comparing aggregate number (left) and area (right) between mouse cortices from young and aged mice. n = 3 mice per age group. Error bars represent the mean with s.d. c, Image of a human brain tissue section from an old individual stained for protein aggregates (Proteostat, green). d, Experimental approach used to determine the identity of neuronal proteins in aggregates from brains of aged (22-month-old) mice. n = 11 BONCAT-labelled mice and n = 11 background controls. DIA, data-independent acquisition. e, Volcano plot showing the enrichment of BONCAT-labelled neuronal proteins in protein aggregates from aged mice relative to the background. f, Images of hippocampus sections from young (4-month-old) and aged (24-month-old) mice (top) and other indicated brain regions from aged mice (bottom) for RTN3 aggregates (red). g, Images of hippocampus sections from young (4-month-old) and aged (24-month-old) mice for SRSF3 aggregates (red). h, Images of hippocampus sections from aged (24-month-old) mice visualizing the colocalization of RTN3 aggregates (red, left) and SRSF3 aggregates (red, right) with the protein aggregate tag ubiquitin (green, top) or p62 (green, bottom). i, Volcano plot showing the enrichment of BONCAT-labelled neuronal protein aggregates from aged mice relative to the background. Proteins are colour-coded based on the disease or disorder for which they have been identified as risk genes according to H-MAGMA. j, Bar plot of the number of aggregating neuronal proteins in aged brains that are also H-MAGMA risk genes of the indicated brain diseases and disorders. k, Donut plots showing the percentage of all aggregating neuronal proteins in aged brains that are risk genes of the indicated diseases. l, GO cellular component analysis on all aggregating neuronal proteins in aged brains. False discovery rate (FDR) values were derived from one-sided Benjamini–Hochberg tests. Bold terms are synapse-related. m, Donut plots showing the percentage of all proteins with an age-increased half-life in the indicated brain regions that were also identified in protein aggregates in aged mice. n, Upset plot showing the overlap of aggregating neuronal proteins with age-increased half-lives among the brain regions (colours are as for m). o, Density plots comparing protein half-life values of proteins identified in aggregates from aged mice compared with proteins not identified as aggregated in the indicated brain regions (colours are as for m). P values were determined by two-tailed Wilcoxon tests. P values in b,e and i were determined by two-tailed, two-sample Student’s t-tests. Risk genes used i–k were derived from the H-MAGMA study28 and considered only if the reported P value was <0.05 as determined by a two-tailed, two-sample Students t-test. Scale bars, 5 µm (h, top right three images), 10 µm (a (bottom row), f (top right), g,h (left and bottom right three images)), 20 µm (a (top row), c,f (bottom row)), or 250 µm (f, top left). NDev, neurodevelopment; NDeg, neurodegeneration.

Most proteins (1,352, or around 78%) identified were present in 2 other label-free datasets of aggregates in the aged brain (Extended Data Fig. 6h). Notably, no change in the total mass of insoluble proteins (Extended Data Fig. 6i) or Proteostat signal (Extended Data Fig. 6j) was observed between BONCAT-labelled brains and non-BONCAT labelled brains. These findings indicate that BONCAT labelling does not artificially induce aggregation.

Aggregation connects to degradation

We next queried whether the aggregation of proteins could explain their slower degradation that accompanies ageing. Overall, 46.8–54.6% of proteins with age-reduced degradation were also found in neuronal aggregates (Fig. 4m). In detail, 54 proteins, including 17 synaptic proteins (VCP, HSPA8 and EEF2, among others) displayed both reduced degradation with age and aggregation with age among all brain regions examined (Fig. 4n and Supplementary Table 3f), and a range of proteins (7–50) displayed both reduced degradation with age and aggregation with age among 2–3 regions (Fig. 4n and Supplementary Table 3f). This finding indicates that many proteins are prone to both reduced degradation and aggregation in a non-regional dependent manner. The distribution of protein half-lives of aggregating proteins in most brain regions in aged mice was slightly less than that of non-aggregating proteins in the same respective region (Fig. 4o), a result consistent with the observation that shorter-lived proteins are more prone to aggregate37. Collectively, these data suggest that protein aggregation could be a contributor to the reduced protein degradation observed with age.

Microglia accumulate neuronal proteins

Microglia maintain neuronal homeostasis by detecting, engulfing and processing neuron-derived proteins[38](https://www.nature.com/articles/s41586-025-09987-9#ref-CR38 “Cserép, C., Pósfai, B. & Dénes, Á Shaping neuronal fate: functional heteroge