Integrated analysis of ultra-deep proteomes in cortex, cerebrospinal fluid and serum reveals a mitochondrial signature in Alzheimer’s disease

A clinician holds three vials. One contains a sliver of cortex tissue, one contains cerebrospinal fluid drawn from the spine, and one contains serum from a blood draw. All three came from the same patient. All three are windows into the same disease. The question Wang and colleagues asked is: what if you looked through all three at once and deeper than anyone had looked before? Alzheimer's disease has long been defined by two pathological signatures — amyloid plaques and tau tangles — and biomarker research has faithfully followed that lead. Levels of amyloid-beta, the amyloid-beta 42 to 40 ratio, total tau, and phosphorylated tau isoforms are all being developed or already used as cerebrospinal fluid and blood markers. Positron emission tomography imaging of amyloid and tau is highly accurate, but it's expensive and inaccessible for most patients, which is why there's enormous pressure to find protein markers in biofluids that could be deployed more widely. The problem is that mass spectrometry-based proteomics, which can in principle detect more than twelve thousand proteins, runs into a brutal biological reality in biofluids. Protein concentrations span at least ten orders of magnitude. Albumin sits at roughly fifty milligrams per milliliter in blood. Interleukin six sits at about four point two picograms per milliliter. When you're trying to detect a faint disease signal buried under a mountain of albumin, most proteins simply don't make it into your dataset. The result: many unbiased proteomic screens of cerebrospinal fluid and blood have been attempted, but their outputs are fragmented and hard to translate into reliable candidates for clinical validation. Wang and colleagues set out to fix this with what they call an ultra-deep tandem-mass-tag liquid chromatography, liquid chromatography, mass spectrometry and mass spectrometry platform. TMT stands for tandem-mass-tag, a chemical labeling strategy that lets you run up to sixteen samples together in a single mass spectrometry experiment — critical for consistency across a large study. Liquid chromatography fractionation then separates peptides more thoroughly before they hit the instrument, so the mass spectrometer can "see" proteins that would otherwise be drowned out by the abundant ones. Rather than depleting those abundant proteins with antibodies — a standard but imperfect approach — the team relied on the raw separation power of the platform itself. The payoff is concrete. They quantified thirteen thousand eight hundred thirty-three proteins from human cortex, five thousand nine hundred forty-one from cerebrospinal fluid, and four thousand eight hundred twenty-six from serum. Then they integrated ten independent proteomic datasets, assembling a combined resource of seventeen thousand five hundred forty-one proteins corresponding to thirteen thousand two hundred sixteen genes across three hundred sixty-five Alzheimer's, mild cognitive impairment, and control cases. That scale — both in depth per sample and breadth across cohorts — is what made what came next possible. The central finding is a mitochondrial signal hiding in plain sight. In the cerebrospinal fluid discovery set, comparing eleven Alzheimer's cases to nine controls, the team identified three hundred fifty-five differentially expressed proteins. When they applied a very stringent cutoff — a Z-score greater than five and a false discovery rate below one percent — sixty-eight proteins rose to the top. Sixty-seven of those sixty-eight were mitochondrial proteins. And they were all decreased in Alzheimer's disease. Let that land for a moment. Sixty-seven out of sixty-eight top hits pointing to the same organelle, all going in the same direction. These proteins are linked to energy metabolism, mitochondrial biogenesis, reactive oxygen species reduction, and mitochondrial DNA repair. Their coordinated reduction points to a failure of cellular energy support — exactly the kind of thing you'd expect to devastate neurons, which are among the most energy-hungry cells in the body. But here's why the signal was missed before: the mitochondrial proteins in cerebrospinal fluid are low abundance. Their median abundance rank in this dataset was two thousand nine hundred sixty out of nearly six thousand proteins. When the team simulated what shallower proteomic depths would have found, a typical coverage of around five hundred proteins — common in earlier studies — detected many previously reported cerebrospinal fluid biomarkers but missed the mitochondrial changes entirely. You needed to go past four thousand proteins before the majority of this signature emerged. Ultra-deep profiling wasn't a methodological luxury here. It was a prerequisite. The signal didn't stop at cerebrospinal fluid. To confirm it was genuinely cross-compartmental, the team applied order statistics — a method that combines ranked protein lists across datasets — and gene set enrichment analysis across cortex, cerebrospinal fluid, and serum together. Mitochondrial functions emerged as a consistently enriched pathway. Individual proteins like SUCLG2, PRDX3, CPT2, HSD17B10, ALDH6A1, GATM, and SOD2 ranked high across the integrated list. And when the team looked at the thirty-seven proteins that were differentially expressed in all three compartments — cortex, cerebrospinal fluid, and serum simultaneously — twenty-two of those thirty-seven were mitochondrial. The cross-species validation added another layer. Using the five times FAD mouse model of amyloidosis, the team profiled cerebrospinal fluid from thirty-two mice pooled into six five times FAD groups and five wild-type groups, aged nine to twelve months. They quantified one thousand fifty-six proteins and found eighty-five differentially expressed between groups. Eleven of those overlapped with human cerebrospinal fluid findings. Six of those eleven were mitochondrial. An independent species, an independent experiment — same mitochondrial story. Against this backdrop, the team also pulled out a shortlist of six cerebrospinal fluid proteins that appeared consistently altered across at least two independent human datasets: SMOC1, C1QTNF5, OLFML3, SLIT2, SPON1, and GPNMB. Appearing in two or more independent datasets was the filter they used to separate reproducible signal from dataset-specific noise — a reasonable bar given they were integrating ten cohorts across three tissues. Worth noting: when they looked at their ultra-deep cerebrospinal fluid analysis of twenty human cases alone, most previously reported biomarker candidates didn't reach statistical significance. Only SMOC1 and TGFB2 did. That's a useful corrective — many candidates in the literature may not be as solid as their individual studies suggest. Validation of the shortlisted candidates involved two orthogonal approaches. GPNMB was tested by enzyme-linked immunosorbent assay in seven Alzheimer's cases versus seven healthy controls, confirming its increase in Alzheimer's disease cerebrospinal fluid. For mitochondrial proteins, the team used TOMAHAQ, a targeted mass spectrometry workflow that uses synthetic peptides labeled with TMT0, spiked into labeled samples, and measured at the MS3 level for precise quantification. Two mitochondrial proteins, AK2 and PCK2, were confirmed as decreased in Alzheimer's disease cerebrospinal fluid by TOMAHAQ, with Pearson correlations validating agreement between discovery and targeted measurements. The three-compartment design matters beyond just confirming individual proteins. If you see the same protein changes in cortex tissue and in cerebrospinal fluid, that's evidence the cerebrospinal fluid is actually reporting what's happening in the brain — the foundational assumption any cerebrospinal fluid biomarker has to satisfy. Wang and colleagues found exactly that consistency for both mitochondrial proteins and top-ranked candidates like SMOC1 and tau. Blood is the harder case: the dynamic range problem in serum is severe, and the team is candid that brain-specific signals are harder to isolate there. But the shared mitochondrial signal across all three compartments gives the finding a cross-tissue credibility that single-tissue studies simply can't claim. What does this mean going forward? The six cerebrospinal fluid candidates — SMOC1, C1QTNF5, OLFML3, SLIT2, SPON1, and GPNMB — are prioritized leads, not established clinical tests. They need large-scale validation in independent cohorts before anyone draws a diagnostic line around them. The authors are clear on that. But the mitochondrial signature is a different kind of result. It's not a single protein to measure — it's a pathway-level signal, consistent across species, across tissues, across datasets, and only visible at a depth that most previous studies never reached. If energy failure is a consistent, cross-compartmental feature of Alzheimer's disease, it's also theoretically something you can target — not just measure. That's speculative, and the authors treat it as such. But the mitochondrial signal points toward biology rather than just biomarkers, which is where the most durable clinical leverage tends to come from. The lesson of this study isn't just about what they found. It's about what was always there, waiting for someone to look deep enough. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.