{"id":"40fb9d08-0b5d-4bdc-804c-a74ec00fb2d3","arxiv_id":"2411.18557","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":10,"one_line_summary":"HT-NaBS uses high-Q pixelated nanoantennas, acoustic droplet printing, and spectral imaging to measure antibody affinity, kinetics, specificity, and epitope binning on a single chip.","lead":"This paper combines high-quality-factor silicon nanoantennas, acoustic bioprinting, and hyperspectral imaging into a label-free biosensor chip that measures antibody binding to multiple antigens in parallel. If the platform scales as projected, it could accelerate antibody drug discovery by testing many candidate antibodies quickly and with very small sample volumes.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Kinetic constants rest on Langmuir fits to surface-immobilized antigen; orientation heterogeneity and the unvalidated sub-pM LOD are the load-bearing soft spots.","rationale":"The reader identified the same core weakness: random antigen immobilization and surface heterogeneity undermine the assumption that Langmuir kinetics reflect intrinsic 1:1 binding, and the sub-pM LOD is extrapolated rather than measured. My stress test adds that the reported kinetic parameters (Fig. 4c) lack error bars and no density-dependence control is shown. The paper contains real engineering advances: high-Q VINPix-derived sensors, acoustic printing of picoliter droplets, 50-sensor multiplexed readout, and epitope binning demonstrations. These support the qualitative claims of specificity and multiplexed detection. However, the quantitative claims (KD values, four-order dynamic range, 45 fM LOD) require the homogeneous-surface assumption and direct low-concentration validation. A CONDITIONAL verdict with the density-dependence and low-concentration checks as conditions is appropriate. I do not see a reason to reject the paper; the concerns are testable and addressable.","tokens_in":22172,"tokens_out":1674,"duration_ms":14456,"concrete_test":"Re-run the kinetics and LOD experiments for at least one antigen-antibody pair (e.g., RBD/anti-SARS-CoV-2) at three immobilized antigen densities (e.g., 100 nM, 1 µM, 10 µM printing concentration) and confirm whether extracted KD stays constant within error. Also measure the blank signal at 50 fM and 500 fM antibody concentrations directly (not extrapolated), with at least 30 sensors, to support the 45-fM LOD claim. If KD shifts by more than expected errors across densities, the Langmuir-derived rate constants are surface-density-dependent and should be reported as apparent values, not intrinsic molecular constants.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim is that HT-NaBS yields true binding kinetics and affinities (ka, kd, KD) and a sub-pM LOD, enabling high-throughput antibody screening. The most load-bearing premise is that measured resonance shifts report a homogeneous 1:1 antigen-antibody interaction consistent with the Langmuir model. The authors themselves note that variation in antibody binding is influenced by steric hindrance, stochastic walking of antibodies, thermodynamic effects, surface repulsion, and unfavorable epitope orientation (p.4, Fig. 2d discussion). Random epoxysilane immobilization can leave capture antigens in multiple orientations or partially denatured, so not all immobilized antigen is competent for antibody binding. This creates a heterogeneous binding population: a high-affinity subpopulation and a low-affinity or nonbinding fraction. A Langmuir fit to the mass-weighted average then yields effective, not intrinsic, rate constants. The reported tabulated values in Fig. 4c (ka ~ 2.6-4.0e6/M/s, kd ~ 1.9-3.0e-4/s, KD ~ 0.07-0.08 nM) have no error bars, and no controls (e.g. varying immobilized antigen density) are shown to test whether extracted KD depends on surface density. A second load-bearing issue is the claimed 45 fM LOD: Fig. 4d states a linear dynamic range from 70 pM to 400 nM, and the LOD is an extrapolation from blank plus 3 sigma. The lowest validated concentration is 1 pM (Fig. 4b), and the 45 fM value is below the lowest tested concentration by more than an order of magnitude, so the sub-pM claim is not backed by direct measurements. These two issues jointly undermine the headline 'sub-pM LOD within 30 minutes' and the quantitative affinity claims, while leaving the qualitative multiplexed specificity demonstration intact.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript presents HT-NaBS, a label-free antibody screening platform that combines high-Q pixelated silicon nanoantennas (adapted from the authors' prior VINPix design), acoustic droplet bioprinting for site-specific sensor functionalization, and hyperspectral imaging for parallel readout. The authors characterize the optical performance (Q-factor versus asymmetry parameter ΔL, sensing figure of merit ≈ 353 RIU⁻¹), demonstrate multiplexed detection of antibodies against SARS-CoV-2 RBD and Influenza A/B hemagglutinin, extract binding kinetic parameters (ka, kd, KD) and EC50 values, show specificity controls, report a claimed sub-picomolar limit of detection (≈45 fM) with a near-four-decade linear dynamic range (70 pM–400 nM), and apply the platform to epitope binning of an H5N1 HA antibody panel and of glycoengineered Cetuximab variants. The headline claim is that the platform can characterize antibody affinity, kinetics, and epitope coverage at high throughput, with low sample consumption and a 30-minute assay time.","tokens_in":22565,"tokens_out":13575,"duration_ms":116353,"significance":"If the kinetic and sensitivity claims are fully validated, this is a genuine advance: it integrates a previously reported high-Q nanoantenna pixel (ref. 32) with a mature acoustic-printing method into an end-to-end screening workflow, and the multiplexed spectral readout of dozens to hundreds of sensors per field of view is a step beyond conventional 96/384-well platforms. The optical characterization is careful and coherent — the Q-versus-ΔL scaling, the FOM measurement against NaCl standards, and the use of 50-antenna statistics give the photonic core of the paper a solid footing — and the specificity and epitope-binning demonstrations are concrete and useful. The principal uncertainties concern whether the extracted rate constants are intrinsic molecular parameters and whether the sub-picomolar LOD is experimentally supported; both issues are addressable with additional controls and revised claims, so the paper's core contribution is defensible in revision.","major_comments":[{"comment":"The claimed 45 fM LOD is an extrapolation: the lowest measured concentration standard shown is 1 pM (Fig. 4b), the stated linear dynamic range begins at 70 pM (Fig. 4d(iii)), and no data point below 1 pM is presented. The IUPAC blank-plus-3σ procedure yields a resonance-shift threshold, but converting that threshold to 45 fM requires the calibration model to be valid more than an order of magnitude below the lowest validated standard. The factor of roughly 20 between the claimed LOD and the lowest measured concentration, and the factor of roughly 1500 between the LOD and the start of the linear range, make the abstract's 'sub-picomolar LOD' claim unsupported as stated. Please measure standards at 50–500 fM and report recoveries, or revise the claim to an extrapolated LOD and clearly separate detection limit from quantitation range.","section":"Fig. 4d and Abstract"},{"comment":"The kinetic parameters in Fig. 4c are reported as point values without confidence intervals, and the underlying sensorgrams are collected at only three concentrations (1 pM, 1 nM, 1 µM) that sparsely bracket the fitted KD of roughly 70–80 pM; for the 1 pM condition, equilibrium is approached only near the 25-minute mark, leaving little of the association phase for fitting ka, and mass-transport limitation is not discussed. It is therefore not demonstrated that ka and kd are individually identifiable from these data. In addition, the reported EC50 values (2.35–5.72 nM) exceed the corresponding KD values by about two orders of magnitude, which is difficult to reconcile with a 1:1 equilibrium-occupancy response and suggests that either the response model or the interpretation of the dose–response curves is more complex than stated. Please provide per-replicate fits with uncertainties, a statement of the fitting model with an identifiability check, and an explanation of the EC50/KD discrepancy.","section":"Fig. 4b–c"},{"comment":"The Langmuir 1:1 interpretation rests on the premise of a homogeneous, orientationally uniform, and fully accessible antigen layer. The authors themselves enumerate the competing effects — steric hindrance, stochastic walking of antibodies, surface repulsion, and unfavorable epitope orientation — in the Fig. 2d discussion, and the capture antigens are printed from 1 µM solutions, conditions that favor high surface density and possible bivalent-avidity stabilization of IgG binding. Under these conditions, the fitted constants are apparent, density-dependent values rather than intrinsic molecular constants, and the paper provides no control in which the immobilized antigen density is varied (or monovalent Fab fragments are used) to show that the extracted KD is density-independent. Please add such a control, or explicitly qualify the tabulated ka, kd, and KD as apparent values and discuss the avidity contribution.","section":"Fig. 2d discussion, pp. 4–5, and Fig. 4 caption"},{"comment":"There is an internal inconsistency in the epitope-binning results. The text states that 'antibodies 1 and 3 share the overlapped epitope' in the Fig. 5d(ii) discussion, yet the heat map in Fig. 5e reports 143 pm and 181 pm for the (1,3) and (3,1) pairs — values comparable to the clearly non-blocking pairs — while the only mutually blocking pair is (2,3)/(3,2) with 36 pm and 32 pm. Either the text or the heat map is mislabeled. Please correct the statement (if the intended claim is that antibodies 2 and 3 overlap) and state the explicit shift threshold, defined relative to the measurement noise, that separates blocking from non-blocking pairs.","section":"Fig. 5e and text on p. 6"},{"comment":"The conclusion that all five glycoengineered Cetuximab variants share a single epitope is an absence-of-signal result: the reported pairwise shifts range from about −22 to +24 pm and appear comparable to the measurement noise, and no positive control (for example, a known non-overlapping anti-EGFR antibody) is shown to demonstrate that a spatially distinct epitope would be detected as a significant shift in this assay format. Without such a control, 'negligible shift' cannot be distinguished from 'assay unable to resolve this pair,' and the claim that Fab/Fc glycosylation does not affect antigen binding is weaker than presented. Please add a positive control or calibrate the blocking threshold against the measured noise floor.","section":"Fig. 5f"}],"minor_comments":[{"comment":"The FWHM values '0.31 µm to 0.78 µm' appear to be a unit error: at λ ≈ 1600 nm and Q ≈ 2000–5000 the expected linewidths are 0.3–0.8 nm, not micrometers; please correct.","section":"p. 3, Fig. 2a discussion"},{"comment":"The statement that the 1.5 nm redshift is '1.5x the FWHM of the high-Q resonators' is inconsistent with Q ≈ 5000, which implies a FWHM of roughly 0.3 nm; please reconcile the factor and state the operating Q used for this comparison.","section":"Fig. 3c"},{"comment":"The abstract and main text claim deposition rates up to 25,000 droplets per second, but the described custom printer is driven at a continuous repetition frequency of 1 kHz in the Methods; please clarify whether 25 kHz is a demonstrated capability of this system or a general ADE limit, and align the claims with the settings actually used.","section":"Abstract, p. 4, and Methods (Acoustic bioprinting)"},{"comment":"The abstract and conclusion state Q-factors exceeding 5000, but the measured maximum average Q reported in Fig. 2a is approximately 4200; please attribute the 5000 figure to simulations or report the measured maximum.","section":"Abstract, Fig. 2a, and Conclusion"},{"comment":"The comparison of the 45 fM LOD to 'gold-standard affinity-based assays such as ELISA' citing ref. [48] is not well supported, since the cited reference is a review of COVID-19 antibody test limitations; a direct quantitative benchmark (for example, a table of LODs from SPR or ELISA for comparable antibodies) would be more informative.","section":"Fig. 4d discussion"},{"comment":"The heat-map colorbar scales as rendered (a '2500' scale for Fig. 5e and a '30/−30' scale for Fig. 5f) do not match the reported shift magnitudes (up to about 220 pm and about ±24 pm, respectively); please clarify the units, the scaling factors mentioned in the caption, and the color assignment for blocking versus non-blocking.","section":"Fig. 5e–f captions"}],"recommendation":"major_revision","confidential_remarks":"The nanoantenna platform is carried over from the authors' prior VINPix publication (ref. 32), and the incremental contribution is the bioprinting functionalization, the hyperspectral assay workflow, and the biological demonstrations; the editor may wish to consider novelty relative to that prior work in assessing scope. The validation in Fig. 4c relies on a literature comparison (ref. [47]); the editor should verify that the commercial antibodies (e.g., AcroBiosystems SAD-S35) are the same clones characterized in that reference, since the manuscript does not state this. There is also a pattern of headline numbers exceeding demonstrated values (Q > 5000 versus measured ≈4200; 25 kHz printing versus 1 kHz in Methods; 45 fM LOD versus the 1 pM lowest standard), and I would urge the authors to align all claims with directly measured quantities in the revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis is a credible engineering demonstration. The genuinely new part is the integration: taking the VINPix high-Q nanoantennas, the acoustic bioprinter, and hyperspectral imaging and running real antibody-binding assays on one chip. The epitope-binning data on glycoengineered Cetuximab variants are also new and useful.\n\nThe optics are the strongest part. The Q-factor measurements, the FOM around 350, and the layer-by-layer resonance shifts all hang together, and the simulation-experiment comparison is reasonable. The specificity panels are clean—target antibody shifts, non-target antibodies do not—and the H5N1 binning logic is coherent. The paper is also honest in the discussion about variability sources: steric hindrance, stochastic walking, unfavorable epitope orientation. That is a point in its favor.\n\nThe soft spots are real, but they are addressable rather than fatal. The 45 fM LOD is an extrapolation: the lowest measured concentration is 1 pM, and the stated linear range starts at 70 pM. Claiming sub-pM sensitivity without a single sub-pM data point is not defensible. The kinetic constants in Fig. 4c come from Langmuir fits to randomly immobilized antigen, so the quoted ka, kd, and KD are likely population averages over heterogeneous orientations, not intrinsic molecular constants. The authors acknowledge this heterogeneity themselves. They should report error bars, test antigen-density dependence, and ideally show that the extracted KD is independent of surface density. The demonstrated multiplexing is 50 sensors; the abstract says hundreds, which is not yet shown experimentally. The roadmap to 30,000 is plausible but still a roadmap.\n\nOn citation pattern: the heavy self-citation to VINPix and the bioprinter is appropriate—they are building on their own prior work, and those papers are the right references.\n\nThis deserves a serious referee. The engineering is credible, the assay logic is sound, and the weaknesses are fixable with more data. I would recommend sending it to peer review and asking for raw sensorgrams, blank measurements, error propagation, and a direct LOD validation at sub-pM concentrations. With those, the quantitative claims could be made to stick.","headline":"A credible integrated nanophotonic antibody-screening platform with solid optics but overclaimed LOD and kinetics that need tightening before the quantitative results can be trusted.","tokens_in":23251,"tokens_out":2178,"would_cite":false,"duration_ms":22272,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that a single nanophotonic chip functionalized by acoustic bioprinting can characterize antibody libraries—binding kinetics, affinity, specificity, and epitope bins—in 30 minutes at sub-picomolar concentrations.","keywords":["antibody screening","high-throughput biosensing","high-Q nanoantennas","acoustic bioprinting","epitope binning","label-free detection","guided mode resonances","Langmuir kinetics"],"falsifier":"Immobilize the same H5N1 hemagglutinin through a site-specific tag so all copies present the same epitope orientation, repeat the kinetic and epitope-binning measurements, and compare the $K_D$ and heat maps to the random-epoxy results; if they differ materially, the reported constants are population averages, not intrinsic molecular affinities.","tokens_in":21979,"feed_emoji":"🔬","tokens_out":10512,"duration_ms":88192,"temperature":0.7,"pith_summary":"This paper claims that a single nanophotonic chip can characterize antibody libraries at a scale and speed that well-plate assays cannot match. The platform pairs high-quality-factor silicon nanoantennas, patterned at over one million sensors per square centimeter, with an acoustic bioprinter that deposits picoliter droplets of capture antigens onto individual sensors. The authors demonstrate selective detection and full kinetic fits for antibodies against SARS-CoV-2, Influenza A and B, sub-picomolar limit of detection, and a four-order linear dynamic range, all within 30 minutes. They also use the same readout to bin antibodies by epitope overlap for H5N1 hemagglutinin and for glycoengineered Cetuximab antibodies against EGFR. If the platform works as described, large synthetic and natural antibody repertoires—and eventually de novo designed proteins—could be screened with far less sample and time than current methods.","feed_headline":"A million-sensor chip detects antibodies at sub-picomolar levels","feed_subtitle":"In one 30-minute run, the chip returns binding kinetics, affinity, and epitope bins without labels.","key_machinery":"The load-bearing mechanism is the very-large-scale-integrated nanoantenna pixel (VINPix): a truncated one-dimensional array of symmetry-broken silicon nanoblocks supporting guided mode resonances, bracketed by photonic-crystal mirrors that shrink the mode volume and produce a Gaussian field envelope. It converts antibody capture into a resonance wavelength shift that can be read from free space, and because each pixel is small, more than one million sensors fit on a square centimeter. Around that transducer, the assay rests on three supporting mechanisms: digitized acoustic droplet ejection prints picoliter droplets of capture antigens at up to 25,000 droplets per second; an epoxy-silane self-assembled monolayer covalently anchors the printed antigens and is backfilled with m-PEG-amine to block nonspecific binding; and spectro-microscopy sweeps a tunable near-IR laser while a CCD captures frames, building a data cube from which every sensor's spectrum and resonance shift are reconstructed.","core_discovery":"The central discovery is that independently addressable high-Q nanoantennas can be turned into a massively multiplexed, label-free antibody assay by combining them with digitized acoustic bioprinting and hyperspectral readout. Each 15 µm × 3 µm silicon nanoantenna supports a guided mode resonance with Q above 5000, an electric near-field enhancement above 40-fold, and roughly 34 percent of the field energy exposed to the surface, so molecular binding shifts the resonance measurably. The authors report a limit of detection near 45 fM and a linear dynamic range from 70 pM to 400 nM for the antibody-antigen pairs tested. Real-time sensorgrams fit the Langmuir adsorption model, giving association and dissociation rates and $K_D$ values near 0.07 nM for SARS-CoV-2, Influenza A, and Influenza B. Tandem epitope binning separates four H5N1 antibodies into overlapping and non-overlapping epitopes, while five glycoengineered Cetuximab variants all block a common EGFR epitope.","pith_inferences":["The reported $K_D$ values are likely population averages: if capture antigens were immobilized in a fixed orientation, the platform's variance would drop and the constants would better approximate intrinsic molecular values.","The same sensor array could map general pairwise protein-protein interactions, with each printed sensor address encoding one interaction pair, not just antibody-antigen complexes.","A direct side-by-side comparison of HT-NaBS and solution-phase affinity measurements for the same antibody pairs would reveal how much surface immobilization biases the kinetics.","If the acoustic printer is combined with live-cell capture of secreted antibodies, the platform could become a high-throughput functional screen for antibody-secreting cells."],"forward_implications":["With simultaneous imaging of tens of thousands of sensors, a 1 cm² chip could carry out the work of over 10,000 96-well plate assays.","One 30-minute run yields association and dissociation rates, $K_D$, and epitope bins, so lead selection can be based on affinity and epitope diversity rather than endpoint titer.","Because the printing step is nozzle-free and probe-agnostic, the same platform can be extended to nucleic acids, aptamers, metabolites, and other proteins.","Higher quality factors, in the hundreds of thousands to millions, could push sensitivity toward near-single-molecule detection while keeping free-space excitation and readout.","High-content kinetic and affinity data could close the loop in computational protein design by providing fast empirical screening of designed antibody libraries."],"supporting_citations":[{"why":"supplies the pixelated high-Q nanoantenna architecture (VINPix) that forms the sensing array.","marker":"[32]"},{"why":"supplies the acoustic droplet ejection hardware and method used to print capture antigens onto individual sensors.","marker":"[42]"},{"why":"supplies the imaging-based spectro-microscopy readout that lets hundreds of resonators be read simultaneously.","marker":"[20]"},{"why":"supplies the epoxysilane self-assembled-monolayer chemistry that covalently anchors printed antigens to the sensor surface.","marker":"[35]"},{"why":"provides the Langmuir adsorption model and kinetic framework used to fit association, dissociation, and $K_D$ values.","marker":"[46]"},{"why":"provides reference kinetic and affinity values for SARS-CoV-2 antibodies against which the platform's results are validated.","marker":"[47]"},{"why":"provides the chemoenzymatic glycoengineering method used to make the Cetuximab variants tested in epitope binning.","marker":"[60]"}],"fun_headline_variants":["Nanophotonic chip screens antibodies in minutes without labels","Million-sensor nanophotonics array finds antibodies at femtomolar","Bioprinted nanoantennas enable rapid, label-free antibody screening","High-Q sensors and bioprinting speed antibody discovery","Chip with a million sensors detects antibodies in 30 minutes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that capture antigens randomly printed onto the epoxy surface keep their native shape and are equally reachable by antibodies, so the observed resonance shifts and Langmuir fits reflect true one-to-one binding kinetics rather than a mixture of orientations and steric effects.","fun_headline_variants_meta":{"raw":{"variants":["Nanophotonic chip screens antibodies in minutes without labels","Million-sensor nanophotonics array finds antibodies at femtomolar","Bioprinted nanoantennas enable rapid, label-free antibody screening","High-Q sensors and bioprinting speed antibody discovery","Chip with a million sensors detects antibodies in 30 minutes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000354,"raw_usage":{"total_tokens":1985,"prompt_tokens":1067,"completion_tokens":918,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":683,"completion_tokens_details":{"reasoning_tokens":828}},"tokens_in":683,"tokens_out":918,"duration_ms":7399,"temperature":1.0,"reasoning_tokens":828,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:05:57.057634+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Immobilize the same H5N1 hemagglutinin through a site-specific tag so all copies present the same epitope orientation, repeat the kinetic and epitope-binning measurements, and compare the $K_D$ and heat maps to the random-epoxy results; if they differ materially, the reported constants are population averages, not intrinsic molecular affinities.","supporting_citations":[{"cited_title":"Preiner, N","cited_arxiv_id":null,"evidence_quote":"supplies the acoustic droplet ejection hardware and method used to print capture antigens onto individual sensors."},{"cited_title":"Krasnok, M","cited_arxiv_id":null,"evidence_quote":"supplies the imaging-based spectro-microscopy readout that lets hundreds of resonators be read simultaneously."},{"cited_title":"Dolia, H","cited_arxiv_id":null,"evidence_quote":"supplies the epoxysilane self-assembled-monolayer chemistry that covalently anchors printed antigens to the sensor surface."},{"cited_title":"Elrod, B","cited_arxiv_id":null,"evidence_quote":"provides the Langmuir adsorption model and kinetic framework used to fit association, dissociation, and $K_D$ values."},{"cited_title":"Dholakia, B","cited_arxiv_id":null,"evidence_quote":"provides reference kinetic and affinity values for SARS-CoV-2 antibodies against which the platform's results are validated."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"provides the chemoenzymatic glycoengineering method used to make the Cetuximab variants tested in epitope binning."}],"review_version":1}