When a vaccine protects someone whose blood contains no detectable antibodies at all, immunologists call it the humoral-cellular gap: antibody titers, the easiest thing to measure with a plate-based enzyme-linked immunosorbent assay (ELISA), systematically miss the protection that T cells quietly provide. A landmark trial analysis found that vaccinated people with completely unquantifiable neutralizing antibodies still kept roughly 63% protection against severe COVID-19 (Lindsay N. Carpp et al.). Alzheimer's disease (AD) research has the mirror-image blind spot. In transgenic mice engineered to build human-like amyloid and tau pathology, every intervention study in the literature measures brain-resident innate inflammation — microglia, astrocytes, inflammasomes — but nobody measures the two arms of peripheral adaptive immunity, antibodies versus T cells, in the same animals whose brains are being scored. The result is that we cannot tell whether humoral titers, cellular responses, or both, track how much pathology accumulates.
The opportunity is sharpened by a specific unresolved mechanism. You Jung Kang et al. showed that interferon-gamma (IFN$\gamma$), the signature cytokine of T cells, flips microglia into a neurotoxic state ($\text{Nrf2}^{\text{low}}\text{CD86}^{\text{high}}$) that accelerates tau phosphorylation in 5XFAD mice and in 3D human mini-brains — but they attribute the IFN$\gamma$ to local astrocytes and neurons and never profile the peripheral T-cell compartment. Meanwhile Taylor R. Heald-Sargent et al. showed across 36 human diseases that an NKG2A+ bias in CD8+ T cells — cells resembling tissue-resident memory T cells — is an anti-inflammatory correlate of protection. Nobody has connected these two observations in a living transgenic AD brain.
Core Claim
In a transgenic AD model, a correlate-of-protection analysis that measures both arms of peripheral adaptive immunity — anti-amyloid and anti-tau antibody titers by ELISA, and antigen-specific T-cell responses by IFN$\gamma$ ELISpot plus ELISA of T-cell-derived cytokines — will explain brain pathology and cognitive decline better than the antibody-only readout that current AD studies implicitly assume, and a meaningful fraction of the IFN$\gamma$ driving the microglial Nrf2 switch will prove to come from infiltrating peripheral T cells rather than brain-resident glia. This is the humoral-cellular gap, transplanted from viral vaccinology into neurodegeneration, with ELISA serving as the shared quantitative backbone that lets both arms be measured in matched plasma, cerebrospinal fluid, and brain homogenate samples.
Research Program
The program has three linked studies in the 5xFAD mouse model, with a validation arm in the TAPS cross-breed line. The first study is a longitudinal natural-history cohort: mice followed from two to eight months of age with monthly peripheral sampling. At each time point we run, in parallel, indirect ELISA for circulating IgG against A$\beta$42 and phosphorylated tau, and sandwich ELISA for IFN$\gamma$, IL-2, IL-6, IL-1$\beta$, and TNF-$\alpha$ in plasma — the same dual-epitope sandwich architecture used to quantify cytokines in the 5XFAD brain and in rat endotoxemia (You Jung Kang et al.; Cynthia Shannon Weickert et al.). On the cellular side we run IFN$\gamma$ ELISpot against pooled A$\beta$ and tau peptides, and flow cytometry for the NKG2A+ CD8+ bias that Taylor R. Heald-Sargent et al. identified as protective. Terminal endpoints replicate the established battery: A$\beta$ plaque burden, tau phosphorylation at AT8 and AT100 epitopes, Iba-1 and GFAP staining, and Morris Water Maze and Novel Object Recognition performance, as used in the ibrutinib and Edaravone-Dexborneol studies (Hyunju Lee et al.; Rui Mao, Shu Shu et al.).
The analysis applies the correlates-of-risk machinery built for vaccine trials. Following the modeling logic of David S. Khoury et al. and Youyi Fong et al., we fit, per animal,
$$\text{logit}\,P(\text{severe pathology}) = \beta_0 + \beta_1 \log(\text{titer}) + \beta_2\,\text{ELISpot} + \beta_3\,\text{NKG2A bias},$$
where "severe pathology" is a pre-registered threshold on plaque burden plus AT8 intensity, and each marker is lagged so peripheral measurements predict later brain outcome rather than co-occur with it. The decisive quantity is the improvement in prediction when the cellular terms are added to a humoral-only model. The sharpest test echoes the Carpp design directly: among mice whose anti-A$\beta$ titers are at or below the ELISA limit of quantification, do T-cell measures still stratify pathology and cognition? If they do, antibody-negative animals are not immune-naive animals, and mouse studies that sort or stratify by titer alone are misclassifying them.
The second study tests the origin question causally. We quantify IFN$\gamma$ by ELISA simultaneously in plasma, cerebrospinal fluid, and brain homogenate, and use T-cell receptor sequencing of sorted infiltrating CD8+ cells to establish whether brain-infiltrating clonotypes are shared with the periphery. We then intervene: CD8+ depletion, IFN$\gamma$ neutralization, and — in the arm we most want — NKG2A agonism to push the CD8+ population toward the protective bias. The readout is whether the $\text{Nrf2}^{\text{low}}\text{CD86}^{\text{high}}$ microglial phenotype identified by You Jung Kang et al. is rescued, and whether tau phosphorylation falls. The third study repeats the correlate analysis in the TAPS hybrid line (Luana Cristina Camargo et al.), which combines pyroglutamate-modified A$\beta$ with standard plaque deposition and shows faster, more aggressive decline than either parental line — an aggressive phenotype that, despite its growing use, has never been profiled immunologically.
Baselines
The do-nothing reference point is the untreated 5xFAD natural history, as characterized qualitatively in the ibrutinib study (Hyunju Lee et al.); for exact quantitative anchors we rely on the reported values from the wild-type LPS rat model — cortical microglial density 1.9 times control at 24 hours post-insult, soluble A$\beta$42 at 2 times control, and p-tau doubled (Cynthia Shannon Weickert et al.) — with the caveat that these come from acute endotoxemia, not a transgenic line, and the file reports no equivalent numeric natural-history values for 5xFAD itself. For our induced-pathology endpoints we treat these multipliers as the reference magnitudes a systemic-immune perturbation must produce or exceed to be considered biologically meaningful.
The strongest existing competitor is the Kang et al. pipeline (You Jung Kang et al.): ELISA of IFN$\gamma$ and IL-6, flow cytometry, immunofluorescence, and qPCR in 5XFAD mice, human AD tissue, and 3D mini-brains, arriving at the Nrf2/Keap1 microglial switch. It is the closest prior work because it already uses our model, our ELISA readouts, and our mechanistic endpoint. The fair comparison is to run their exact brain-side measurement suite on our cohort, at matched age, sex balance, and sample size, with the same tissue-processing protocols, and then ask whether their brain-intrinsic account — IFN$\gamma$ produced locally by astrocytes and neurons — predicts pathology as well as a model that also knows the peripheral T-cell state. We beat their account only if the combined model predicts individual-animal pathology significantly better than the brain-intrinsic model, and if CD8 depletion or IFN$\gamma$ neutralization changes the microglial phenotype that their pipeline reads out. The file gives no reported numeric value for their effect sizes, so we will treat our replication of their qualitative phenotype as a prerequisite before testing them.
The key ablation isolates the proposal's central ingredient: a humoral-only arm in which anti-A$\beta$ and anti-tau IgG ELISA titers are the sole predictors. If titers alone predict pathology and cognition as well as the combined model, the humoral-cellular gap does not matter in AD mice and the proposal's premise is falsified. We pre-register this as the primary negative control, run on the same animals and the same assays, so the comparison is a pure feature-subset test rather than a different experiment.
Why not just…?
Why not just use wild-type rats given LPS instead of transgenic mice? The LPS model is cheaper and faster, and it does produce quantifiable pathology — the 1.9-fold microglial increase and doubled p-tau at 24 hours (Cynthia Shannon Weickert et al.) make it an excellent assay-validation sandbox, and we will use it that way to debug the paired ELISA panels before touching transgenic cohorts. But it produces only diffuse plaques lacking dense cores, its A$\beta$ and tau are rodent-sequence proteins that cannot form the invariant human filament folds that determine seeding behavior (Michel Goedert et al.; Alysa Kasen et al.), and it models an acute innate shock rather than the months-long exposure over which an adaptive humoral-versus-cellular correlate can even be defined. Antibody titers and T-cell memory take weeks to diverge; a 7-to-9-day endpoint cannot contain the question.
Why not just measure antibody titers by ELISA, as standard practice does? This is the real threat, because serology is cheap and reproducible, and we cannot rule out that in mice — unlike humans — titers track pathology perfectly. The file's evidence argues otherwise: the ~63% protection retained by antibody-negative vaccinated people (Lindsay N. Carpp et al.), the systematic undercounting of immune control in people who never seroconvert (David W. Eyre et al.), and the clinical observation that amyloid-directed antibody strategies clear plaques without reliably halting cognitive decline (Allan I. Levey et al.). But that is an argument, not a result, which is why the humoral-only ablation arm is pre-registered as the primary test rather than waved away: if titers alone predict mouse pathology, the honest conclusion is that the gap is a human-specific phenomenon and the proposal should be reframed as establishing that negative fact.
Why not just use the 3D human mini-brain? The Kang et al. mini-brain replicates the IFN$\gamma$/Nrf2 switch in a fully human genetic context and is genuinely superior for the cell-intrinsic mechanism. But it has no circulating peripheral immune system — no T cells, no B cells, no lymphoid tissue — so it cannot contain a humoral-versus-cellular comparison by construction, and it cannot run Morris Water Maze or Novel Object Recognition. We therefore use it in the follow-up role it is best suited for: a confirmation chamber in which sorted peripheral T-cell populations from our mice are perfused through the microfluidic mini-brain to test whether they, alone, reproduce the microglial switch that local astrocytes were previously blamed for.
Why not just use RT-QuIC for the pathology readout? The K12 RT-QuIC assay detects tau seeds at $10^{-8}$ to $10^{-9}$ dilutions within 20 to 50 hours and discriminates AD-type from Pick-type folds (Michael A. Metrick et al.), and we will adopt it as a sensitive secondary pathology endpoint. But it measures seeded fibril assembly, not soluble proteins; it cannot quantify IFN$\gamma$, IL-2, or antibody titers, which are the very quantities the proposal is about. ELISA and RT-QuIC answer different questions here, and the proposal needs both rather than either.
Proposed Experiments and Decisive Checks
Experiment 1 delivers the correlate result: the combined immunity model versus the humoral-only ablation on the longitudinal 5xFAD cohort, with the Carpp-style split — pathology stratification among mice with unquantifiable anti-A$\beta$ titers — as the headline analysis. Success is a statistically significant improvement in predicting severe pathology (pre-registered as plaque burden above the cohort median plus AT8 intensity above median) from the added ELISpot and NKG2A terms, and clear stratification in the antibody-negative subgroup. Experiment 2 delivers the origin result: paired plasma/CSF/brain IFN$\gamma$ ELISA with T-cell receptor clonotype sharing between blood and brain, followed by CD8 depletion and IFN$\gamma$ neutralization, with success defined as loss of the $\text{Nrf2}^{\text{low}}\text{CD86}^{\text{high}}$ microglial phenotype and reduced AT8/AT100 signal relative to isotype-treated littermates. Experiment 3 repeats the Experiment 1 analysis in TAPS mice (Luana Cristina Camargo et al.) to test whether the correlate structure survives a more aggressive, multi-pathology phenotype. As a benchmarking anchor for effect sizes worth chasing, the intervention literature in this corpus reports that Edaravone-Dexborneol reduced the fraction of APP/PS1 mice declining on Y-maze from 80% to roughly 27% (Rui Mao, Shu Shu et al.) — a behavioral effect magnitude that any immune-directed manipulation would need to approach to be therapeutically interesting. The file reports no numeric hazard ratios or titer thresholds for the CoVPN 3008 correlates study (Bo Zhang et al.), so the correlate-model coefficients from Experiment 1 will be the first quantitative values of this kind in a neurodegeneration model, and we will report them whether or not they favor the cellular arm.
References
- Neutralizing antibody correlate of protection against severe-critical COVID-19 in the ENSEMBLE single-dose Ad26.COV2.S vaccine efficacy trial
- Neurotoxic Microglial Activation via IFNγ‐Induced Nrf2 Reduction Exacerbating Alzheimer's Disease
- Integrative systems biology reveals NKG2A-biased immune responses correlate with protection in infectious disease, autoimmune disease, and cancer
- Lipopolysaccharide endotoxemia induces amyloid-β and p-tau formation in the rat brain.
- Chronic oral administration of ibrutinib prevents long-term memory deficits and reduces AD pathology and neuroinflammatory responses in a mouse model of AD
- Edaravone-Dexborneol slows down pathological progression and cognitive decline via inhibiting S100A9 in APPswe/PS1dE9 mice
- Correlates of Protection, Thresholds of Protection, and Immunobridging among Persons with SARS-CoV-2 Infection
- Analysis of Antibody Markers as Immune Correlates of Risk of Severe COVID-19 in the PREVENT-19 Efficacy Trial of the NVX-CoV2373 Recombinant Protein Vaccine
- PEAβ Triggers Cognitive Decline and Amyloid Burden in a Novel Mouse Model of Alzheimer’s Disease
- Tau and neurodegeneration
- Driving Tau Pathogenicity Seeding ‐ How Fibril Structure and Post‐Translational Modifications Alter Seeding Capacity
- Stringent thresholds in SARS-CoV-2 IgG assays lead to under-detection of mild infections
- A phase II study repurposing atomoxetine for neuroprotection in mild cognitive impairment
- A single ultrasensitive assay for detection and discrimination of tau aggregates of Alzheimer and Pick diseases
- Neutralizing and binding antibodies are a correlate of risk of COVID-19 in the CoVPN 3008 study in people with HIV