{"id":"a2589776-e4a1-418c-a8ca-09d6ec532fca","arxiv_id":"1908.08482","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":8,"one_line_summary":"Mucus-driven upward advection in a one-dimensional airway model keeps influenza infection above the initial virus deposition depth and makes infection appear to spread downward over time.","lead":"A new mathematical model tracks flu virus moving along a one-dimensional human airway, swept upward by mucus and spreading by diffusion. It finds that upward mucus flow dominates, so infection rarely spreads below where the virus first lands and the upper airway is infected first.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Central localization claim rests on molecular-diffusion-only PCF transport; no sensitivity test for effective dispersion from cilia, cough, or airway geometry is provided, so the 'no spread below xd' result is not yet robust.","rationale":"The reader's weakest_assumption identifies the same load-bearing issue: the model restricts PCF transport to molecular diffusion and uniform upward advection, ignoring other mixing mechanisms. My reading agrees and sharpens it: the localization claim is a direct consequence of the ratio of advection to any down-gradient transport, and the paper provides no sensitivity analysis that would show the result holds for effective dispersion values plausible in the human airway. The concern does not refute the model's internal logic; it questions the model's applicability to the real respiratory tract. Since the reader already conditioned acceptance on such uncertainties, the verdict remains CONDITIONAL. I add credit where due: the advection scheme is exact (vaΔt/Δx = 1, Methods 7.1), the qualitative knockout comparisons are a reasonable sanity check, and the paper explicitly notes the limitations of its empirical immune response. The concrete test is a one-dimensional parameter sweep that would directly determine whether the localization claim is robust or an artifact of the molecular-diffusion assumption.","tokens_in":27702,"tokens_out":6583,"duration_ms":81148,"concrete_test":"Repeat the simulations underlying Figure 3(d-f) with DPCF replaced by D_eff = 10^-10, 10^-9, and 10^-8 m^2/s, keeping all other parameters fixed, and additionally with va reduced to 20 µm/s to bracket uncertainty in mucus transport. Record the deepest depth x at which the target-cell fraction T falls below 0.95 by 7 dpi. If this depth exceeds xd = 15 cm by more than ~1 cm for any tested combination, the advective-barrier claim is not quantitatively robust; if the depletion boundary remains at xd for all tested values, the concern is settled.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The model's central prediction—that advection prevents infection from disseminating below the deposition depth xd—is a consequence of the Péclet number set by Table 1: DPCF = 10^-12 m^2/s and va = 40 µm/s in Eqn. (1). At these values, the 7-day diffusive length scale is sqrt(2 D t) ≈ 1.1 mm, so the conclusion that advection dominates is unsurprising and internally consistent. However, the claim is load-bearing on the assumption that molecular diffusion is the only down-gradient transport in the periciliary fluid. In the living airway, ciliary beating, cough-induced flows, surface tension gradients, and airway geometry can create effective axial dispersion far larger than DPCF. The paper's defense of its transport model (Section 2.2, Figure 4c) only compares advection with molecular diffusion within the model; it does not estimate or bound an effective dispersion coefficient in vivo. A modest effective dispersion of D_eff = 10^-10 m^2/s already yields a diffusive length of roughly 1 cm over 7 days, which is larger than the 0.5-mm Gaussian inoculum and could seed infection below xd; D_eff = 10^-8 m^2/s would erase the barrier. Because no sensitivity analysis over DPCF is presented, the paper has not shown that the localization result survives the most plausible physiological perturbation to its transport assumptions.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper develops a one-dimensional partial differential equation model of influenza A virus infection in the human respiratory tract (HRT), coupling a standard target-cell–eclipse–infectious–virus kinetic scheme with diffusion of virions in the periciliary fluid (PCF) and upward advection driven by the mucociliary escalator. The model is used to study how these spatial transport processes alter infection kinetics and localization. The main claims are that advection dominates over molecular diffusion, that advection prevents infection from disseminating below the depth of initial virus deposition, and that the apparent downward spread of infection is an artifact of faster kinetics in the upper (downstream) airway. The model is then extended with target-cell regeneration and empirical immune-response components (IFN, antibodies, CTLs), and applied to construct hypothetical time courses for seasonal and avian H5N1 influenza infections and to evaluate neuraminidase inhibitor therapy.","tokens_in":28022,"tokens_out":6168,"duration_ms":65361,"significance":"If the central claims are sustained, the model offers a mechanistic explanation for spatial patterns of influenza infection, suggests that non-spatial models may underestimate the virus production rate, and provides a modular platform for exploring antiviral interventions. The paper is clearly written, the numerical implementation is explicitly described, and the spatial results follow from the stated equations. Its main strengths are the transparent formulation and the demonstration that adding just two spatial parameters can generate qualitatively novel infection kinetics, including the apparent reversed direction of spread. However, the significance is conditional: the localization prediction rests on a single assumed value of the PCF diffusion coefficient with no sensitivity analysis, and the seasonal/avian and immune-response portraits are based on hand-fitted empirical curves and parameter shifts rather than inference from data. The platform is a useful contribution, but the quantitative claims go beyond what the current evidence supports.","major_comments":[{"comment":"The central claim that advection prevents infection from disseminating below the deposition depth is load-bearing but is not robustly established, because the only downward transport mechanism in the model is molecular diffusion with DPCF = 10^-12 m^2/s, and no sensitivity analysis over this parameter is provided. The comparison in Figure 4(c) shows only that molecular diffusion is negligible relative to advection at the chosen value of DPCF; it does not bound the effect of physiologically plausible enhanced dispersion from ciliary beating, cough-induced flows, surface tension gradients, or airway geometry. The authors should either estimate an effective axial dispersion coefficient for the PCF in vivo or demonstrate that the qualitative conclusion persists for effective dispersion coefficients up to at least 10^-10 m^2/s, for which the 7-day diffusive length scale is about 1 cm and could seed infection below the deposition site. Without such an analysis, the 'no spread below xd' prediction is a direct consequence of the assumed transport parameters rather than a demonstrated physiological property.","section":"Section 2.2, Table 1, Eq. (1)"},{"comment":"The '10-fold increase' in the virus production rate is a calibration, not a prediction: the value p = 11×pBaccam is selected so that the spatial model with advection reproduces the 2–3 dpi viral titer peak of the non-spatial Baccam model. The subsequent claim that non-spatial models underestimate the virus production rate is therefore contingent on the chosen target peak time and on the assumed values of DPCF and va. The authors should explicitly state that this is a fitting procedure and should provide a sensitivity analysis showing how the inferred p changes when DPCF or va is varied over a plausible range. The current presentation, which lists p as a fixed parameter in Table 1, risks being read as a model prediction rather than a fitted quantity.","section":"Section 2.2, p = 11×pBaccam"},{"comment":"The seasonal and avian infection 'portraits' are constructed by manually shifting parameters to match pre-specified, hypothesized time courses, and the immune response components are empirical curves whose parameters are chosen by eye to match experimental data in Figure 7. Consequently, the statements that the model 'can capture' these infections, and the subsequent predictions about NAI treatment efficacy in Section 2.5, Figure 11 and Table 3, are demonstrations of model expressiveness under assumed parameter sets rather than validated quantitative predictions. The authors should reframe these sections to make the conditional nature explicit, and ideally perform a sensitivity analysis over the shifted parameters (β, p, τE, τI, f50, A0, kC) to identify which parameter changes drive the different infection outcomes. The current narrative overstates the support for the seasonal-versus-avian conclusions.","section":"Section 2.5, Table 2, Figure 10, and Section 2.4, Figure 7"}],"minor_comments":[{"comment":"The abstract states a '10-fold increase' in the virus production rate, while the text and Table 1 use 11×pBaccam; please make these consistent.","section":"Abstract and Section 2.2"},{"comment":"The axis labels in panel (b) contain '10□2' and '10□1', which appear to be a rendering artifact; they should be formatted as superscripts.","section":"Figure 4"},{"comment":"The boundary conditions are described inconsistently: a reflective boundary condition is used for the diffusion step at the bottom of the HRT (V_{Nx+1} = V_{Nx}), whereas the advection step sets V_{Nx} = 0, i.e., no virus beyond the end of the HRT. The authors should clarify the combined boundary treatment or note explicitly that the bottom boundary is not physically relevant once advection is present.","section":"Section 7.1"},{"comment":"The model assumes a uniform advection speed and constant diffusion coefficient along the entire 30 cm HRT, but the nose, trachea, and bronchi have very different mucus properties and geometry. This simplification is not discussed in the limitations; a sentence acknowledging it would be helpful.","section":"Section 2.1 and Table 1"},{"comment":"The paper does not mention the availability of simulation code or data. Providing the code would improve reproducibility, as the numerical scheme involves a nontrivial operator-splitting approach with specific time-step constraints.","section":"Section 7.3"}],"recommendation":"major_revision","confidential_remarks":"The paper is a well-executed modeling study with a clear and useful platform, but the main quantitative claims are not yet supported by the presented analyses. The localization result needs a sensitivity analysis over the effective dispersion coefficient, and the fitted/qualitative nature of the p, immune, and seasonal/avian parameters should be acknowledged more prominently. With those additions, the paper could be a solid contribution to within-host influenza modeling."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague — spend an hour on this one only if you care about spatial structure in within-host influenza models. The paper grafts diffusion and advection onto the standard Baccam target-cell model, representing the HRT as a 1D track with stationary cells and virus moving in the periciliary fluid. The qualitative punchline is that upward advection dominates infection kinetics, making the infection appear to travel downward while actually preventing spread below the deposition site, and forcing a 10-fold increase in virus production to match observed peak timing.\n\nWhat's genuinely new: this is, as far as I know, the first model to combine diffusion and advection in a 1D HRT geometry. The numerical treatment is clean — Crank-Nicolson for diffusion, exact translation for advection with CFL=1, grid-convergence check. Physical parameters (diffusion coefficient, mucus speed) are well sourced. The apparent downward spread is a nice consequence of spatially homogeneous kinetics plus transport, and it matches the direction of spread seen in mouse imaging. The point that non-spatial models may underestimate p, and hence resistance risk, is worth making.\n\nThe soft spots are real but not fatal. The 'no spread below deposition depth' result is a Péclet-number story: with D=10^-12 m^2/s and v=40 µm/s, the 7-day diffusive length is ~1 mm, so of course advection wins. The paper does not test sensitivity to effective dispersion from ciliary beating, cough, or airway geometry. A modest D_eff of 10^-10 m^2/s would give a diffusive length of ~1 cm and could seed infection below xd. So the barrier is a conditional prediction, not an established fact. The authors should either bound effective dispersion or soften the abstract's 'prevents' phrasing. The 11x p is tuned to match peak timing, without uncertainty quantification. The immune and seasonal/avian portraits are explicitly hand-fit — the paper is honest about that, but the reader should not mistake them for validation. No code is released, though the equations are fully specified.\n\nBottom line: the model is a useful platform, the core qualitative insight about advection-dominated kinetics is likely robust, and the paper deserves serious peer review and a conditional accept with requested sensitivity analysis on transport. I'd bring it to reading group for the modeling discussion, and I'd cite it if I were working on spatial within-host dynamics.","headline":"A useful spatial extension of the standard IAV target-cell model showing advection dominates and yields apparent downward spread, but the 'no spread below deposition depth' claim rests on an untested molecular-diffusion-only transport assumption.","tokens_in":28570,"tokens_out":2940,"would_cite":true,"duration_ms":31830,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["92C50","35K57"],"pacs":[],"model":"deepseek-v4-flash","headline":"Upward mucus flow in the airways, not random diffusion, determines where influenza A infection spreads, confining it above the depth of initial virus deposition.","keywords":["influenza A virus","mathematical model","advection-diffusion","periciliary fluid","mucus escalator","infection localization","respiratory tract","viral kinetics"],"falsifier":"Measure the effective spread of inert fluorescent particles or virus-sized tracers deposited at a known depth in the airway surface liquid of living airway epithelium: if observed dispersion exceeds $D_{\\mathrm{PCF}}=10^{-12}\\ \\mathrm{m^2/s}$ by orders of magnitude, or if tracers move downward against the mucus flow, the confinement prediction fails. Alternatively, imaging infection initiated from a single well-localized deposit and finding infected cells below the deposit depth would contradict the claim.","tokens_in":27493,"feed_emoji":"🦠","tokens_out":6256,"duration_ms":60185,"temperature":0.7,"pith_summary":"This paper argues that the upward sweep of mucus in the airways, not random diffusion, controls where and how fast influenza A infection spreads in the human respiratory tract. Modelling the tract as a one-dimensional tube and adding virus advection to the standard infection equations, the authors find that advection is so dominant that infection effectively cannot travel below the depth where the virus first lands. Because the upper tract receives virus carried up from below, it becomes infected, peaks, and clears sooner, which makes infection look as though it is moving downward even though it is not. The authors also show that reproducing observed viral kinetics requires a roughly 10-fold higher virus production rate than non-spatial models estimate, and that a model with simple immune components can reproduce seasonal and avian strain time courses.","feed_headline":"Upward mucus flow walls off flu infection below its landing site","feed_subtitle":"A 1-D respiratory-tract model says advection, not diffusion, sets flu kinetics, so spread only looks downward.","key_machinery":"The load-bearing object is a one-dimensional advection-diffusion equation for virus concentration in the periciliary fluid, coupled to a spatially distributed target-cell--eclipse--infectious-cell infection model. The tract runs from the nose ($x=0$) to a depth of 30 cm; virus diffuses with coefficient $D_{\\mathrm{PCF}}$ and is advected upward at speed $v_a$, with an absorbing top and a reflective bottom. Infection starts as a Gaussian inoculum at depth $x_d$. The advection term is the mechanism that changes the model's behaviour: at $v_a=40\\ \\mu\\mathrm{m/s}$, a virion would traverse the 30 cm tract in about 12 minutes unless it infects a cell first, so upward transport dominates viral spread and confines infection above the deposition site.","core_discovery":"The central claim is that upward advection of the periciliary fluid is the dominant transport mechanism for influenza A virus in the respiratory tract and acts as a physiological barrier to downward dissemination. When both diffusion ($D_{\\mathrm{PCF}}=10^{-12}\\ \\mathrm{m^2/s}$) and advection ($v_a=40\\ \\mu\\mathrm{m/s}$) are included in a one-dimensional spatial model, infection remains confined to depths at or above the initial deposition site, with no target-cell consumption below it. Since virions released by infected cells are carried upward, cells higher in the tract are exposed to more virus and progress through infection faster; the resulting gradient in infection timing creates an illusion of downward spread. The paper further claims that fitting this spatial picture to the timing of influenza virus shedding requires a virus production rate around 11 times that fitted by a non-spatial model, and that including target-cell regeneration and simplified interferon, antibody, and cytotoxic-T-lymphocyte responses lets the model capture the time courses of seasonal and avian influenza infections and their response to neuraminidase inhibitors.","pith_inferences":["If effective dispersion in the periciliary fluid is much larger than molecular diffusion (from ciliary beating, coughing, or mucus heterogeneity), the predicted barrier to downward dissemination would weaken; this is testable with particle-tracking experiments.","The 10-fold production-rate correction implies that the within-host supply of viral mutants, and thus the probability of resistance emergence, may be substantially larger than well-mixed model estimates.","The same advection-confinement logic could apply to other respiratory viruses that replicate in the airway surface liquid, and to multi-drop deposition, where the confinement boundary would be set by the deepest droplet rather than a single inoculum.","High-resolution spatiotemporal imaging of a single localized influenza infection could distinguish the model's predicted stationary, confined infection from true downward spread."],"forward_implications":["If advection dominates, the depth of initial virus deposition largely determines the extent of respiratory-tract involvement: depositions higher than about 15 cm give lower and later viral peaks, while deeper depositions change little.","Non-spatial models fitted to viral-titer data would underestimate the true virus production rate, and therefore the rate at which mutations, including drug-resistance mutations, are generated.","The apparent downward progression of infection seen in imaging studies is reinterpreted as a timing illusion caused by the mucus escalator, not as actual directional spread.","Neuraminidase inhibitor treatment is predicted to be most effective when started early for seasonal-strain infections, but for a slowly peaking avian-strain course even delayed treatment can still substantially reduce viral load."],"supporting_citations":[{"why":"Supplies the baseline non-spatial infection parameters and human viral-titer time course that the spatial model must reproduce.","marker":"[1]"},{"why":"Provides the measured periciliary-fluid advection speed of about 40 µm/s that drives the confinement result.","marker":"[47]"},{"why":"Supplies the Stokes-Einstein estimate of the virus diffusion coefficient in the periciliary fluid.","marker":"[37]"},{"why":"Describes the mucus layer and mucociliary escalator, the physical basis for upward advection.","marker":"[26]"},{"why":"GFP reporter-virus imaging in mice cited as consistent with the model's upper-to-lower appearance of spread.","marker":"[46]"},{"why":"Real-time imaging of an influenza reporter virus cited as consistent with the predicted spread pattern.","marker":"[73]"},{"why":"Provides the seasonal versus avian H5N1 pharyngeal viral-load data used to build the strain-specific infection portraits.","marker":"[19]"}],"fun_headline_variants":["Mucus escalator keeps flu high, stalling downward spread","Advection, not diffusion, decides flu's reach in airways","Flu's downward spread is just a mirage of timing","Upward mucus flow halts flu's invasion deeper into lungs","Mucus escalator confines flu to upper airways, model shows"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The confinement result rests on treating the periciliary fluid as quiescent apart from a uniform 40 µm/s upward flow and molecular diffusion of $10^{-12}\\ \\mathrm{m^2/s}$, with infection starting from a single localized droplet; if additional mixing is substantial or deposition is more distributed, the predicted barrier to downward spread may not hold.","fun_headline_variants_meta":{"raw":{"variants":["Mucus escalator keeps flu high, stalling downward spread","Advection, not diffusion, decides flu's reach in airways","Flu's downward spread is just a mirage of timing","Upward mucus flow halts flu's invasion deeper into lungs","Mucus escalator confines flu to upper airways, model shows"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000741,"raw_usage":{"total_tokens":3377,"prompt_tokens":1086,"completion_tokens":2291,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":702,"completion_tokens_details":{"reasoning_tokens":2216}},"tokens_in":702,"tokens_out":2291,"duration_ms":16725,"temperature":1.0,"reasoning_tokens":2216,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:38:00.924826+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the effective spread of inert fluorescent particles or virus-sized tracers deposited at a known depth in the airway surface liquid of living airway epithelium: if observed dispersion exceeds $D_{\\mathrm{PCF}}=10^{-12}\\ \\mathrm{m^2/s}$ by orders of magnitude, or if tracers move downward against the mucus flow, the confinement prediction fails. Alternatively, imaging infection initiated from a single well-localized deposit and finding infected cells below the deposit depth would contradict the claim.","supporting_citations":[{"cited_title":"Matsui, S","cited_arxiv_id":null,"evidence_quote":"Provides the measured periciliary-fluid advection speed of about 40 µm/s that drives the confinement result."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the Stokes-Einstein estimate of the virus diffusion coefficient in the periciliary fluid."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Describes the mucus layer and mucociliary escalator, the physical basis for upward advection."},{"cited_title":"Manicassamy, S","cited_arxiv_id":null,"evidence_quote":"GFP reporter-virus imaging in mice cited as consistent with the model's upper-to-lower appearance of spread."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Real-time imaging of an influenza reporter virus cited as consistent with the predicted spread pattern."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the seasonal versus avian H5N1 pharyngeal viral-load data used to build the strain-specific infection portraits."}],"review_version":1}