When a vaccine is injected into the arm, it raises antibodies in the blood but often leaves the wet lining of the nose — the entry point for most human pathogens — nearly undefended. This is the mucosal immunity gap, and the field can now measure the front-line defense, secretory IgA in nasal lining fluid, with validated assays and reference standards. But the measurement is reported as a single scalar: micrograms of antibody per milliliter of fluid wrung from a sponge that sat in the nostril for five minutes. A separate literature, in brain imaging, has documented in detail what happens when you summarize a convoluted three-dimensional biological surface with one scalar number: focal shrinkage in one place cancels swelling in another, and the summary reads "normal" while the structure is severely remodeled. Nasal antibody measurement currently commits exactly this error. The sponge absorbs fluid from an unregistered patch of a highly individual, folded mucosal surface, and the resulting concentration conflates how much antibody a person secretes with where and how large their contact patch happened to be.
The proposal is to close that gap by pairing per-subject shape models of the nasal airway with the newly standardized antibody sampling workflow, and to report an areal mucosal antibody density — total recovered secretory IgA divided by the actual mucosal surface area the sampler contacted — instead of a bulk fluid concentration. The core hypothesis is that a meaningful share of the between-person variability in nasal antibody measurements is anatomical rather than immunological, and that removing it will sharpen group contrasts and improve test-retest reproducibility, a prerequisite for ever defining a protective threshold.
Core Claim
The standardized nasal antibody pipeline — expanding polyvinyl alcohol sponge sampling, the IgA-specific monoclonal antibody W2, and the CS2 freeze-dried nasal reference standard — solved the analytical problem but left a geometric one. Two papers built that pipeline: Exploring the standardized detection and sampling methods of human nasal SARS-CoV-2 RBD IgA benchmarked three sampling methods in 154 participants and found the expanding sponge detects RBD-specific IgA in 95.5% of participants versus 88.3% for a standard swab and 68.8% for a deep nasopharyngeal swab; Exploring the standardization of human nasal antibody measurements showed that using serum-derived standards misestimates nasal potency by up to 10-fold and validated CS2 across seven laboratories. Both treat the nose as a well-mixed bucket: neither records which mucosal region the sponge touched, nor the surface area of that region, nor the participant's airway geometry. No participant in either study underwent structural imaging.
Meanwhile, the shape-measurement literature supplies both the warning and the remedy. In a cohort of 208 Alzheimer's patients and 717 controls from OASIS3, Quantifying Hippocampal Shape Asymmetry in Alzheimer's Disease Using Optimal Shape Correspondences showed that gross volumetric asymmetry completely failed to separate disease from control ($p = 0.80$, Mann-Whitney U) while point-wise shape morphometry on 512 optimized surface correspondences was highly significant ($p < 0.001$, Hotelling's $T^2$), with 177 of 512 points showing focal effects. Classification of Alzheimer's disease based on hippocampal multivariate morphometry statistics extended this to classification: multivariate morphometry reached AUCs of 0.94, 0.93, and 0.92 across diagnostic contrasts, while volume-plus-age models managed only 0.76, 0.66, and 0.58. The lesson generalizes: a scalar summary of a folded manifold discards the signal. We claim it discards signal in the nose too, and that the fix — point-wise correspondence models of the mucosal surface, with the sampling device registered onto them — is now technically feasible because a public upper-airway imaging dataset exists: NasalSeg provides 130 annotated 3D CT scans of the nasal cavity and paranasal sinuses.
The quantity we propose to standardize is
$$ \rho \;=\; \frac{M_{\mathrm{sIgA}}}{A_{\mathrm{contact}}}, $$
where $M_{\mathrm{sIgA}}$ is the total mass of RBD-specific IgA recovered from the sponge eluate (measured by the existing W2 ELISA against CS2) and $A_{\mathrm{contact}}$ is the mucosal surface area within the sponge's registered contact zone, computed on a per-subject point distribution model of the nasal airway boundary.
Research Program
The program has three phases. Phase one is purely computational and uses only existing data. We build a point distribution model of the nasal airway on the 130 NasalSeg CT scans, adapting the particle-based correspondence optimization from the hippocampal asymmetry work — dense correspondence points distributed across the air-tissue boundary, with deformation vectors projected onto the surface normal to avoid registration sliding artifacts. The immediate obstacle is segmentation quality: the self-configuring nnU-Net baseline on NasalSeg achieved an average Dice Similarity Coefficient of only 58.55% across the five labeled structures, with fold-level performance ranging from roughly 50% to 66%, which is too noisy for clean surface meshes. We will adapt the skeleton-supervised approach of Skeleton Supervised Airway Segmentation — which reached 88.79% DSC on the pulmonary Binary Airway Segmentation benchmark using only 1.96% of the voxel annotation workload, against a fully supervised U-Net upper bound of 90.53% — from the bronchial tree to the nasal cavity, using annotated centerlines through the turbinate meatuses and nasopharynx to regularize the thin boundaries that voxel-only learning breaks. The phase-one deliverable is a quantified answer to a question nobody has answered: how much does anterior nasal mucosal surface area vary across individuals? No reported result found on Lacuna for population variance of human nasal mucosal surface area. If the inter-individual spread turns out to be small, the premise of the whole proposal weakens, and we would report that honestly and stop; this is a genuine go/no-go gate.
Phase two is the paired cohort. We recruit approximately 60 adults across the exposure strata used in the RBD-IgA study — unexposed, post-infection, and recipients of intranasal or inhaled vaccines — and perform, in the same visit, a low-dose CT of the sinuses (or reuse clinically indicated scans where available) and M3 expanding-sponge sampling with the W2 ELISA calibrated against CS2. To register the contact zone, the sponge is tagged with a thin radiopaque filament and a second scan is acquired with the sponge in situ, so the contact patch is localized directly on the per-subject surface model rather than inferred from insertion depth. Each participant is sampled twice, one week apart, to separate biological from anatomical variance. The primary outcome is the between-subject geometric coefficient of variation of $\rho$ versus bulk concentration, and the within-subject test-retest intraclass correlation of each measure. The secondary outcome is discrimination AUC between unexposed and mucosally vaccinated participants for bulk concentration, for volume-normalized measures, and for $\rho$ — an explicit structural analogue of the morphometry-versus-volume contrast in the Alzheimer's classification study.
Phase three adds a longitudinal vaccination arm: participants receiving an intranasal vaccine are imaged and sampled before and after vaccination. Here we apply the point-wise machinery in its full form, tracking localized turbinate thickening and swelling as normal-direction displacement fields, because the volumetric-measurement literature predicts that post-vaccination localized edema in one turbinate with compensatory luminal change elsewhere can cancel in a scalar summary. The same principle was shown in a different organ by One-Month Changes in Choroidal Vascularity Index of Medium Vessel Layer in Children with Myopia Wearing Orthokeratology Lenses, where total choroidal volume increased while the medium-vessel-layer vascularity index significantly decreased ($p < 0.05$) and the large-vessel layer was unchanged — aggregate volume hiding layer-specific physiology.
We are deliberate about what this program does not claim. It cannot establish a protective threshold, because as the RBD-IgA paper itself notes, the field lacks any correlate of sterilizing mucosal protection. And it does not touch the central nervous system safety question raised for intranasal nanocarriers by Nanotechnology-driven advances in intranasal vaccine delivery systems against infectious diseases; we only note that a registered contact-zone map is a natural future safety instrument, since the olfactory neuroepithelium occupies a specific, mappable region of the superior nasal vault.
Baselines
The naive reference point is the historical practice: bulk nasal concentration measured against serum-derived calibration, the exact procedure that Exploring the standardization of human nasal antibody measurements showed introduces systematic errors of up to 10-fold, on top of the swab methods that detect IgA in only 68.8% (nasopharyngeal) to 88.3% (standard swab) of participants. It matters because it defines the floor the field climbed from, and any areal measure must at minimum retain the 95.5% detection rate the expanding sponge achieves.
The strongest existing competitor is the complete current pipeline: M3 expanding-sponge sampling, the W2 monoclonal ELISA with intermediate precision below 17%, relative bias below ±4%, and a concordance correlation coefficient of 0.87 against electrochemiluminescence assays, calibrated against the CS2 freeze-dried national standard, all from Exploring the standardized detection and sampling methods of human nasal SARS-CoV-2 RBD IgA and its companion standardization study. This is the strongest competitor because it already removes essentially all analytical and calibration variance; what remains is exactly the anatomical and sampling-geometry variance we claim to address. To keep the comparison fair, every baseline runs on the same eluates, the same assay plates, and the same CT scans, differing only in the normalization applied afterward — no baseline is disadvantaged by sampling or assay differences. Beating it means three things together: a materially lower between-subject geometric coefficient of variation for $\rho$ than for CS2-calibrated bulk concentration, a higher one-week test-retest intraclass correlation, and a higher AUC for separating unexposed from mucosally vaccinated participants, while retaining at least the 95.5% detection rate.
The key ablation isolates the shape ingredient by replacing the per-subject contact-area denominator with a scalar alternative: total nasal cavity air volume from the NasalSeg-style segmentation, the direct analogue of the volume-plus-age models that scored AUC 0.58 to 0.76 where morphometry scored 0.92 to 0.94 in the Alzheimer's study. A second ablation replaces the per-subject contact area with the population-average contact area from the phase-one template, isolating whether individualized geometry — rather than any area normalization at all — is what earns the improvement. If either ablation matches $\rho$, the point distribution model machinery is not paying for itself and the simpler normalization should be adopted.
Why not just…?
Why not just measure serum antibodies instead? Serum draws are easier and better standardized, but the mucosal-systemic disconnect is documented throughout the file: nasal lining fluid is dominated by dimeric secretory IgA (about 87% of nasal immunoglobulin) while serum is dominated by monomeric IgG, and the RBD-IgA study found nasal IgA levels do not track serum levels closely — though no reported correlation coefficient was found on Lacuna for the nasal-versus-serum association, which is itself evidence of how under-measured this relationship is. More decisively for the biology, systemic delivery of the 4-1BB boost in Intrinsic 4-1BB signals are indispensable for the establishment of an influenza-specific tissue-resident memory CD8 T-cell population in the lung completely failed to seed lung-resident memory T cells, while the intranasal route produced a population stable at roughly 10% of CD8 T cells for over 300 days and protected mice against lethal heterotypic influenza challenge 9 to 11 months later. Blood does not read out the mucosal compartment.
Why not just normalize the eluate by total nasal cavity volume from a standard segmentation? This is the most serious alternative, and it is exactly the ablation we run rather than argue around. The counter-evidence is structural: cavity air volume is not mucosal surface area — two noses with identical air volumes can have very different turbinate folding and hence different secreting surface and different sponge contact patches — and the volumetric-measurement literature shows scalar summaries of folded manifolds cancel focal effects, as when hippocampal volumetric asymmetry gave $p = 0.80$ where point-wise morphometry gave $p < 0.001$. But if volume normalization performs as well as areal normalization in phase two, the honest conclusion is that the simpler measure suffices, and the proposal is designed to surface that result rather than bury it.
Why not just keep the bulk measurement and recruit a larger cohort? Averaging over anatomical confounds with more participants reduces the variance of a group mean but leaves every individual measurement confounded, which blocks the field's stated goal of defining an individual-level protective threshold for nasal IgA. The sampling-method comparison already shows device geometry matters enormously — 95.5% versus 68.8% detection across methods differing only in contact area and duration — so the contact geometry is demonstrably not a negligible factor in what the assay recovers. That said, if phase one finds anterior mucosal surface area varies only trivially across people, larger-cohort bulk measurement would be the right answer, and the go/no-go gate exists precisely to catch this.
Proposed Experiments and Decisive Checks
The decisive checks, in order of cost. First, on NasalSeg alone: quantify inter-individual spread of anterior nasal mucosal surface area and contact-zone geometry across the 130 scans; a small spread (say, a coefficient of variation under roughly 10%) argues the anatomical confound is too weak to matter and the program should stop before human sampling. Second, the segmentation check: our skeleton-supervised upper-airway model must materially beat the 58.55% average DSC nnU-Net baseline on NasalSeg under identical cross-validation, since surface meshes extracted from noisy boundaries would contaminate $A_{\mathrm{contact}}$ with segmentation error. Third, the paired-cohort test: two sponges one week apart in 60 participants, comparing the test-retest intraclass correlation and between-subject geometric coefficient of variation of bulk concentration versus $\rho$, with the volume-normalization and template-area ablations run on the identical data. Fourth, the discrimination test: AUC for unexposed versus mucosally vaccinated participants across all normalizations. Fifth, the longitudinal arm: point-wise normal-direction displacement maps before and after intranasal vaccination, testing whether localized turbinate remodeling is visible in point-wise morphometry while total cavity volume shows no significant change — a direct replication of the structural-cancellation phenomenon in a new organ, and the first time these shape-measurement tools leave the brain.
References
- Exploring the standardized detection and sampling methods of human nasal SARS-CoV-2 RBD IgA
- Exploring the standardization of human nasal antibody measurements
- Quantifying Hippocampal Shape Asymmetry in Alzheimer's Disease Using Optimal Shape Correspondences
- Classification of Alzheimer's disease based on hippocampal multivariate morphometry statistics
- NasalSeg: A Dataset for Automatic Segmentation of Nasal Cavity and Paranasal Sinuses from 3D CT Images
- Skeleton Supervised Airway Segmentation
- One-Month Changes in Choroidal Vascularity Index of Medium Vessel Layer in Children with Myopia Wearing Orthokeratology Lenses: a Predictor for One-Year Changes in Axial Length
- Nanotechnology-driven advances in intranasal vaccine delivery systems against infectious diseases
- Intrinsic 4-1BB signals are indispensable for the establishment of an influenza-specific tissue-resident memory CD8 T-cell population in the lung