Protein accumulation in the aging brain: made locally or acquired from other cells?

01The label first identifies the producer

Proteostasis includes synthesis, folding, transport and clearance. The same abundance can accompany rapid synthesis and clearance or slow synthesis and clearance. Abundance, half-life and function provide complementary information. s2

Guldner and colleagues' 2026 tools express an engineered enzyme in selected neurons, allowing newly synthesized proteins to incorporate a chemical label that can subsequently be enriched. Material can retain this origin information after leaving its producer. Coverage mainly concerns Camk2a-labeled neurons and detectable proteins, rather than every neuron and protein. s1

Several controls support origin assignment: reporter mice did not show the corresponding microglial labeling; sorted transcriptomes enriched microglial markers; and tissue antibody imaging located neuronal SV2A and SV2B signals inside microglia. The last control weakens an explanation based entirely on neuronal debris introduced during sorting, while leaving the intact form, intracellular location and route of other proteins individually unvalidated. Independent purity experiments estimated 97.23% microglia and 2.77% border-associated macrophages among CD11b-positive cells. Cell proportions do not establish the source proportion of each protein signal. s1 s12

Labeling concerns proteins newly synthesized during the experiment. A protein species occurring in persistence, aggregate and microglial lists does not show that the same years-old damaged molecules traveled through all three stages.

02More microglial stock: greater acquisition or slower clearance?

The transfer experiment compared male mice aged 3–5 and 18–23 months. Young, aged and background groups contained 10, 8 and 6 pools, respectively, with about 900,000 cells per pool. Methods specify three mice per pool; the figure legend says three to four. The pool is the sample. Its constituent mice and detected proteins cannot all be counted as independent replicates. s1

Reapplying the authors' background and age thresholds gives 119 protein groups shared between ages and 1,212 unique to the aged list. Of the merged 1,335 groups, 1,027 are higher in aged pools and one is higher in young pools. BH adjustment within the released age-comparison family gives corresponding counts of 1,024 and one. This check preserves the main direction, but the family follows background selection and missing-value imputation; it does not guarantee unconditional error control for the full original pipeline. The authors' qvalue outputs and BH outputs are kept separate. s4

These counts describe detectable protein groups and relative signals in pools with a fixed cell count. They cannot be converted into the number of synapses removed from an aged brain. Fixed cell counts still allow differences in cell size, state and composition; protein-specific neuronal labeling input was not measured alongside every recipient pool. Imputed low signals are not measured zero expression.

For recipient stock M, input u and clearance constant k, a minimal model is dM/dt = u − kM. Different input and clearance combinations can yield the same sampled M. Our chosen examples both reach two arbitrary units on day 7; after input stops, their disappearance separates. This identifies a useful additional observation, not fitted mouse parameters or a human regimen.

Isolation can alter this stock. FEAST showed that warm tissue dissociation can induce new engulfment and adhesion can generate misleading signals; its sniffer-cell controls identify acquisition during processing. Guldner's inhibitor cocktail during isolation and sorting substantially addresses this concern. It also includes an inhibitor of lysosomal degradation, so resulting stock is not an in vivo acquisition rate. FEAST found residual signals in border-associated macrophages that some inhibition failed to remove. Sorting and in situ evidence need to be interpreted together. s6 s1

03Neuronal origin does not specify where degradation occurred

With N representing label still inside neurons, M label transferred to recipients and transfer rate τ, total label in a closed two-compartment example obeys d(N+M)/dt = −kN·N − kM·M. Transfer cancels directly from total mass balance, but can still alter total disappearance by changing the clearance environment in which material resides.

Increasing transfer makes neuronal label disappear faster. The identical neuronal trajectory leaves more total label when recipient clearance is slower. Disappearance from neurons and destruction of neuron-derived material therefore require different measurements. Real regions can also exchange material across boundaries, contain extracellular material and experience cell migration; the closed model cannot simply be inverted to estimate their rates.

04Greater persistence in aged regional samples, with precision limits

Turnover experiments used male mice aged 4, 12 and 24 months, with seven days of labeling followed by a chase. There were four mice per age/time point and two background controls per age. Author code sets the first sample to time zero, followed by days 2, 6 and 13. Measurements concern neuron-derived proteins recovered from regional tissue, rather than a uniquely identified intrinsic neuronal degradation constant. s1 s5

Recalculating from the half-life columns for protein groups detected at all three ages within each region gives generally longer aged estimates:

These describe protein features, not estimates from 730 or 507 independent mice. Means, medians, mean individual ratios and ratios of means answer different questions; a universal claim of halved degradation would collapse those distinctions. s3

Positive evidence also exists without extrapolating half-lives: the released 13-day integral columns show positive regional mean aged-minus-young persistence areas, about 1.26–2.46 fraction-days. Public 18-channel PRIDE files reproduce 15,112 trajectories before background subtraction and 3,602 background-corrected integrals in complete-channel subsets. The remaining 404 released integrals were not reconstructed by inventing missing channel values. Keeping Table 2a's unsubtracted layer separate from Table 2d's corrected layer avoids mistaking different processing stages for arithmetic errors. s3 s11

Four means fit either one or three rate parameters, with model selection by AIC. Two-level half-lives additionally use linear interpolation or extrapolation between fitted points. In hippocampus, 82 of 507 aged estimates exceed the last day-13 point, with one around 89 days. This is not an 89-day observation or corresponding animal-population precision. Every released Table 2a trajectory already satisfies the maximum adjacent rise of five percentage points; relaxing that table's threshold cannot retrieve trajectories removed upstream. s3 s5

A numerical audit changes some regional comparisons: all 507 hippocampal entries labeled aged/young in Table 2b equal aged/middle-aged. Recomputing a uniform aged/young denominator changes the descriptive correlation for 312 exact shared hippocampal/sensory-cortex protein groups from −0.044 to 0.159. Positive persistence directions elsewhere remain, while regional vulnerability requires reassessment with consistent denominators. We do not claim complete reproduction of the original plot or original parameter fits. s3

05Overlapping aggregation and transfer do not establish functional cost

The authors report 1,726 proteins in aged neuronal aggregates, with substantial overlap between aggregation, slower disappearance and microglial accumulation lists. This fits other cells receiving material that is difficult to handle; aggregation impeding clearance and slower clearance promoting aggregation also remain possible. Lists from different experiments do not establish causal order. s1

Using three disjoint intersections of 166, 160 and 224 against 1,027 age-enriched groups, the article reports about 53.5% associated with persistence or aggregation. Table 4h instead contains 223 rows in the relevant aggregate/microglial category and 1,012 rows across its four microglia-containing categories. Counts and ID correspondence differ from Table 4g, with their specific cause not fully reconciled. We retain “about half” as the authors' overlap summary, without claiming exact reproduction or converting it into damaged-cell or synapse proportions. s4

Secretion signals, exosome-database annotations and lysosomal list overlap suggest possible routes, without identifying a channel for each protein. A 2023 cell-line study directly imaged α-synuclein and mitochondria moving through nanotubes between neuron-like and microglial cells. It supports biological plausibility, rather than showing that these proteins primarily use that route in normally aging brains. s9

Functional interpretations run in both directions. Removing material may relieve its producer; prolonged recipient retention or engulfing functional synapses may impose costs. The focal mouse experiments did not selectively perturb this transfer pathway and then measure synaptic, cognitive and motor outcomes to distinguish those explanations. A 2022 bulk-brain study also proposed that retaining protective or expensive proteins may be adaptive, with some aged proteins instead becoming shorter-lived. A 2024 study found age, sex and compartment patterns that were not uniformly monotonic. Greater persistence alone does not make every protein change damaging. s7 s8

06Human brain dynamics are becoming measurable; intervention benefit remains separate

A September 2026 study obtained surgical biopsies from 18 people with suspected normal-pressure hydrocephalus; its isotope-labeling subset detected newly synthesized tau. Sparse sampling across people and a precursor-input model estimated a roughly 34-day half-life for one tau peptide group. The precursor trajectory used data from other participants, rather than complete serial biopsies of one person. This disease/surgical population, protein pool and model estimate do not validate normal-human neuron-to-microglia age transfer or establish healthspan benefits from changing it. s10

Separating acquisition from retention requires clearer producer input, recipient tracing and pathway perturbation. Distinguishing protection from harm also requires functional measurements of producers, recipients and synapses. Current evidence motivates those experiments, while providing no established human conclusion that increasing engulfment or degradation makes a brain younger.

The calculation package contains original code, descriptive outputs and chosen models. Full source-data replay additionally requires the public files identified in its manifest. Evidence was checked through 2026-10-07. The same Codex agent researched, translated and self-reviewed this article; there was no independent or professional medical review.