{"id":"979029bb-7504-4528-8c66-246bab897155","arxiv_id":"2608.12914","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Inter-residue contact distances in simulations reproduce the relaxation timescales of transient infrared spectra, allowing each kinetic step to be assigned to a specific local contact network.","lead":"This paper links the relaxation times seen in time-resolved infrared spectroscopy of proteins to specific local motions in molecular dynamics simulations. If the link holds, infrared kinetic steps can be read as reports of which contact networks inside a protein are moving.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Timescale matching is generic under transfer-operator theory, so the claim that IR \"directly reports\" contact networks is underdetermined; the paper never computes IR spectra.","rationale":"The reader's conditional verdict focused on sampling convergence and visual comparison. I agree that undersampling is a real issue, but the more fundamental problem is inferential: the theoretical framework invoked by the authors makes timescale agreement a weak test of specificity. Even perfect sampling would not discriminate contact networks from other collective variables, because Markov state model theory implies that all observables of the same dynamical process share the same relaxation timescales. The paper's own admission that amplitudes differ, and in PDZ2L show opposite trends, removes the only information that could have distinguished contact networks from alternative descriptors. The mechanistic conclusion therefore rests on an assertion about electrostatics rather than on a computed IR response. This does not negate the value of the empirical timescale correlation, but it means the strongest claim in the Conclusions is not yet supported by the presented evidence. Since the reader already assigned a CONDITIONAL verdict, I keep that verdict unchanged while sharpening the required condition: the \"direct tracking\" claim should be tested either by computing IR spectra from the trajectories or by showing that a generic collective variable fails to reproduce the experimental dynamical content.","tokens_in":11823,"tokens_out":5896,"duration_ms":66205,"concrete_test":"Run a single control analysis on the existing 10 microsecond trajectories: compute the timescale spectrum D_control for a deliberately generic collective variable such as the total number of native contacts or the backbone RMSD from the initial state, using the same maximum-entropy regularization and time window as D_MD, and compare the peak positions with D_IR. If D_control reproduces the experimental peaks as well as D_MD does, the specificity claim is falsified; if D_control misses multiple experimental peaks while D_MD hits all of them, the contact-network interpretation is supported. An even more direct complementary check would be to compute a simulated amide I transient from the trajectories with an electrostatic frequency map and confirm that its dynamical content and cluster attribution match D_IR.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing step is the inference from matched peak positions to a causal, mechanistic link. The paper itself invokes transfer-operator theory (Introduction, refs. 18-19) to justify comparing timescales, stating that \"all observables of a dynamical system share the same relaxation timescales, differing only in their amplitudes.\" If that premise is correct, the agreement between D_MD and D_IR in peak locations (Figs. 1c, 4b-c, 5b) is a generic property of any observable of the same kinetic process, not a fingerprint of contact networks. The only discriminating information is amplitudes, and the paper repeatedly finds amplitude mismatches, including \"opposite trends\" in PDZ2L, and explicitly says amplitudes \"generally may differ.\" Consequently the assignment of specific experimental relaxations to clusters (200 ns to C6, 800 ns to alpha3 realignment) is read off from MD alone; no measured IR feature is shown to be uniquely sensitive to those contacts. The proposed electrostatic mechanism is asserted in the Discussion (contact rearrangements alter the electric field at amide I C=O groups) but is never tested by computing amide I shifts or spectra from the trajectories. Thus the central conclusion \"transient IR response directly tracks the dynamics of localized contact networks\" is not established by the presented evidence. This concern is separate from sampling quality: even with perfectly converged trajectories, timescale matching alone cannot identify the molecular origin of IR relaxation.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a general framework for interpreting time-resolved infrared (TRIR) spectra of proteins in terms of specific local structural motions. The authors decompose transient IR signals and nonequilibrium MD trajectories into multi-exponential timescales, construct a 'dynamical content' D(τ), and compare the MD-derived D_MD (from inter-residue contact distances) with the experimental D_IR for three PDZ-domain constructs (PDZ3, PDZ2S, PDZ2L). They report agreement in the locations of characteristic peaks and use MoSAIC correlation clustering to assign each experimental relaxation to a specific contact cluster, e.g., the 200 ns step to allosteric propagation to the β1–β2 loop and the 800 ns step to α3 helix realignment. The central claim is that the transient IR response 'directly tracks' the dynamics of localized contact networks.","tokens_in":12073,"tokens_out":3077,"duration_ms":35194,"significance":"If the central claim were fully established, this would be a valuable framework: it would turn TRIR from a reporter of global 'rugged landscape' relaxation into a residue-level mechanistic probe, and it offers experimentally testable predictions (e.g., for time-resolved X-ray studies). The paper has several strengths: D_IR and D_MD are computed independently from experiment and simulation with no fitted parameter inserted to force the match; the analysis pipeline uses openly available software (Contacts, MoSAIC); and the transfer-operator justification is explicitly acknowledged. However, the load-bearing inference from matched peak positions to a causal, contact-specific mechanism is underdetermined, as discussed in the major comments. The mechanistic assignments and the 'directly tracks' language therefore go beyond what the present evidence supports, though the work constitutes a plausible and useful consistency framework.","major_comments":[{"comment":"The paper invokes transfer-operator theory (refs. 18–19) to justify comparing timescales, yet this same theory implies that all observables of a dynamical system share the same relaxation timescales and differ only in amplitudes. Under that premise, the agreement in peak positions between D_MD and D_IR (Figs. 1c, 4b–c, 5b) is a generic consequence of both observables reporting the same kinetic process, not a fingerprint of contact networks. The only discriminating information is the amplitudes, and the paper repeatedly finds amplitude mismatches, including 'opposite trends' in PDZ2L (Fig. 5b), and explicitly states that amplitudes 'generally may differ' (Discussion). Consequently, the central claim that 'the transient IR response directly tracks the dynamics of localized contact networks' (Conclusions) is not established by the presented evidence. The authors should either compute amide I shifts or spectra from the trajectories to test the proposed electrostatic mechanism, or substantially reframe the claim as one of consistency/compatibility rather than direct reporting, and provide a quantitative assessment of which observables can actually distinguish contact clusters.","section":"Introduction; Discussion; Conclusions"},{"comment":"The agreement between D_IR and D_MD is assessed visually. No quantitative metric (e.g., log-timescale distance between peaks, cross-correlation, or an uncertainty estimate from the maximum-entropy fit or from trajectory subsampling) is provided, and no error bars appear in any dynamical-content figure. Because the entire assignment scheme rests on the identity of peak positions, the absence of a quantitative measure and of propagation of sampling uncertainty is a load-bearing gap. A bootstrap over experimental fits and MD trajectories, or at least a table with peak positions and confidence intervals, should be added.","section":"Comparison of IR and MD response (PDZ3); Figs. 1, 4, 5"},{"comment":"The paper itself flags sampling limitations that undermine the comparison: the C8 peak is 'expected to disappear with improved MD sampling' (Fig. 2c), and for PDZ2L 'the simulations of ligand rebinding were insufficiently sampled for a quantitative comparison' (p. 5), yet those data are still used to claim agreement and to assign kinetic steps. If undersampling shifts MD peak positions, the matched assignments could be coincidental. The authors should provide convergence checks (e.g., split-half or block analyses of the trajectories) and, where convergence is inadequate, either exclude the affected features from the comparison or flag them as non-quantitative in all figures and claims.","section":"MD modeling of PDZ3; PDZ2L comparison; Figs. 2c and S5"},{"comment":"The proposed molecular mechanism linking contact rearrangements to the amide I IR response is asserted but never tested: the paper states that contact rearrangements alter the electric field at backbone C=O groups and thereby shift amide I frequencies, but no amide I frequency map, electrostatic calculation, or spectral simulation is performed. Without such a computation, the causal statement that IR 'directly reports' contact-network dynamics remains a hypothesis. At minimum, a test using an established amide I frequency map (e.g., refs. 16–17) on the simulated trajectories would show whether the identified clusters actually produce IR-detectable signals with the claimed selectivity.","section":"Discussion (electrostatic mechanism)"}],"minor_comments":[{"comment":"The sentence 'providing an direct experimental measure' should read 'a direct experimental measure'.","section":"Results, PDZ3"},{"comment":"There is a misplaced period in 'the dynamical content D_IR of the experiment was obtained (Fig. 4b.)' The closing period should be outside the parenthesis.","section":"Fig. 4 caption / text"},{"comment":"The manuscript uses inconsistent spacing in phrases like 'Cαcontacts' and 'α 3 helix'; please use consistent notation throughout (e.g., 'Cα contacts', 'α3 helix').","section":"Various"},{"comment":"The statement says trajectories are available 'upon reasonable request' but does not mention deposition in a public repository. Given the emphasis on reproducibility, consider archiving representative trajectories or at least the processed contact-distance time series.","section":"Data Availability Statement"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope and the framework is potentially useful, but the central claim overreaches the evidence: the timescale-matching argument is generic under the very theory the authors cite, and the mechanistic assignments rely on amplitude-free peak matching plus an untested electrostatic mechanism. The sampling caveats (C8, PDZ2L rebinding) further weaken the quantitative comparison. These issues are addressable in revision, so I recommend major revision rather than rejection; the authors should either add direct spectral calculations or substantially temper the causal language and provide quantitative uncertainty-aware comparisons."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The genuinely new thing here is the systematic comparison across PDZ3, PDZ2S, and PDZ2L: contact-cluster dynamical contents from nonequilibrium MD reproduce the peak locations of the experimental IR dynamical contents, and each kinetic step is assigned to a specific cluster. That is a useful integrative result for people trying to interpret transient IR in terms of structure. The paper is honest about several weaknesses, which counts in its favor: it flags the C8 peak as under-sampled, admits the PDZ2L ligand-rebinding simulations were not converged, and openly discusses that amplitudes generally differ and in PDZ2L show opposite trends.\n\nThe main soft spot is not sampling, it is the inference. The paper itself invokes transfer-operator theory: all observables of the same kinetic process share the same relaxation timescales, differing only in amplitudes. If that is true, then matching peak positions between D_IR and D_MD is generic—any reasonable observable of the correct dynamics would show the same peaks. The only discriminating information is amplitudes, and those consistently do not match. So the conclusion that \"the transient IR response directly tracks the dynamics of localized contact networks\" is not actually established by the evidence shown. What is established is that the contact-cluster representation reproduces the timescales, and that clusters can be used post hoc to assign kinetic steps.\n\nA second issue is that the comparison is visual. No quantitative agreement metric, no error bars on the peaks, and the exclusions (N terminus, sub-0.5 ns, >10 µs, the C8 cluster) are post hoc. The 200 ns feature in PDZ3, for instance, is barely visible in D_IR but becomes prominent only in the difference signal D_ΔIR; calling that a perfect agreement is generous.\n\nDespite these problems, the paper deserves a serious referee. The framework is promising, the cluster assignments are testable hypotheses, and the authors are unusually candid about limitations. A good referee should push them to (a) provide a quantitative overlap measure with uncertainties, (b) temper the causal language, or better, test the electrostatic mechanism by actually computing amide I shifts from the trajectories, and (c) show that the cluster assignments survive when amplitudes are weighted differently. If revised along those lines, this could be a solid contribution. I would take it to reading group for the transfer-operator discussion.","headline":"Useful cross-validation of MD contact-cluster dynamics against transient IR timescales across three PDZ systems, but the central claim that IR 'directly reports' contact networks is undercut by transfer-operator theory and the paper's own admission that amplitudes differ.","tokens_in":12634,"tokens_out":1603,"would_cite":true,"duration_ms":18883,"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 argues that transient IR spectra of proteins report the dynamics of localized inter-residue contact networks, so each observed relaxation time can be assigned to a specific structural motion.","keywords":["time-resolved infrared spectroscopy","protein dynamics","nonequilibrium molecular dynamics","inter-residue contacts","contact networks","allostery","PDZ domains","dynamical content"],"falsifier":"Run dramatically longer or more replicated nonequilibrium MD for PDZ3 and check whether the 200 ns peak remains assigned to cluster C6 at the $\\beta_1$–$\\beta_2$ loop after the N-terminal contacts are removed; if the peak shifts, splits, or vanishes, the structural assignment is an undersampling artifact. Alternatively, experimentally stabilize or disrupt the $\\beta_1$–$\\beta_2$ loop contacts and see whether the 200 ns allosteric step in the transient IR response moves or disappears.","tokens_in":11586,"feed_emoji":"🧬","tokens_out":9164,"duration_ms":82267,"temperature":0.7,"pith_summary":"Time-resolved infrared (IR) spectra of proteins are usually read as a generic signature of relaxation on a rough energy landscape, with peaks that carry no direct structural meaning. The paper tries to establish that these peaks are specific: each relaxation time reports the motion of a localized network of inter-residue contacts. It compares the multiexponential timescale decomposition of experimental IR transients with the same analysis applied to inter-residue contact distances from nonequilibrium molecular dynamics of photoswitchable PDZ domains, and finds that the 'dynamical contents' $D(\\tau_k)$ of experiment and simulation line up at 1 ns, 20 ns, 200 ns, 800 ns and 2–3 µs. Correlation-based clustering then assigns each peak to a particular contact network, such as detachment of the $\\alpha_3$ helix near 1–20 ns and allosteric propagation to the $\\beta_1$–$\\beta_2$ loop at 200 ns. If true, transient IR becomes a direct window onto the molecular choreography of allosteric signaling and ligand binding.","feed_headline":"IR signals trace specific protein motions, simulations show","feed_subtitle":"In PDZ domains, each relaxation step from helix detachment to 200-ns allosteric signaling matches a local contact network.","key_machinery":"The central objects are the dynamical content $D(\\tau_k)=\\sqrt{\\sum_j |a_{jk}|^2}$ — the total amplitude of exponential relaxation on timescale $\\tau_k$ extracted from a multiexponential fit — and MoSAIC correlation clusters of inter-residue contact distances, groups of contacts whose distances move together. The dynamical content lets experiment and simulation be compared without modeling the full IR spectrum, and the clusters let each experimental timescale be assigned to a specific set of contacts. Heavy-atom contact distances (minimum heavy-atom distance below 4.5 Å, population above 10%) turn out to be the representation that reproduces the experimental timescales; C$\\alpha$ distances miss the 800 ns realignment, and backbone dihedral angles yield almost no structure.","core_discovery":"The central claim is that the characteristic relaxation times measured by transient IR spectroscopy are fingerprints of the dynamics of localized contact networks, not just of diffusion on a rough landscape. The authors construct the dynamical content $D(\\tau_k)=\\sqrt{\\sum_j |a_{jk}|^2}$ from multiexponential fits to the experimental transients, compute the same quantity for heavy-atom contact distances along nonequilibrium MD trajectories, and obtain matching peak positions within the 0.5 ns to 10 µs window. In PDZ3, cluster C6 at the $\\beta_1$–$\\beta_2$ loop accounts for the 200 ns allosteric step, while 1–20 ns and 800 ns features report detachment and realignment of the $\\alpha_3$ helix; in PDZ2S the early peaks track opening of the binding pocket, and in PDZ2L the dominant ~4 µs peak reflects the concerted rearrangement after ligand release. The paper further argues that the physical link is electrostatic: forming or breaking polar contacts changes the electric field at the backbone C=O groups that determine amide I frequencies.","pith_inferences":["Inference: If the mapping is as clean as claimed, transient IR could serve as a standalone experimental readout of contact-network dynamics once reference signatures for a fold family are established, without needing a simulation for every new construct.","Inference: The paper's view that rate-limiting steps are entropic coincidences of several contact rearrangements predicts weak temperature dependence of the relaxation times but strong dependence on solvent viscosity, both of which are testable.","Inference: Isotope-labeled or site-specific amide I probes placed on residues belonging to a single cluster, such as C6, could spatially resolve the 200 ns allosteric signal in experiment, an experiment the paper does not propose.","Inference: The one-peak-per-decade pattern may be the generic signature of sequential, structurally localized network rearrangements, connecting these PDZ results to the broader family of photoactive proteins beyond the systems studied here."],"forward_implications":["Each experimentally resolved relaxation step in a PDZ domain can be assigned to a specific contact network, turning a one-dimensional kinetic trace into a structural description of the transition.","The 200 ns feature provides a direct experimental measure of allosteric signal propagation from the photoswitch to the $\\beta_1$–$\\beta_2$ loop, and the 800 ns feature predicts a previously unrecognized realignment of the $\\alpha_3$ helix.","Inter-residue contact distances, rather than backbone dihedrals, are the structural variables that carry the IR-visible dynamics, shifting attention to side-chain packing and tertiary contacts in interpreting protein spectra.","Because contact rearrangements change the electrostatic environment of backbone C=O groups, mutations that alter polar contacts should have larger effects on amide I spectra than mutations that only move backbone $\\phi,\\psi$ angles.","The same contact-cluster workflow should transfer to other photoactive proteins whose transient IR shows the hierarchical one-peak-per-decade pattern, giving those systems the same structural assignments."],"supporting_citations":[{"why":"Supplies the PDZ3 transient IR data, including the bound/unbound difference whose 200 ns peak defines the allosteric signaling speed.","marker":"[8]"},{"why":"Supplies the 112 nonequilibrium MD trajectories of PDZ3 and the initial assignment of functional contact networks.","marker":"[15]"},{"why":"Supplies the PDZ2S transient IR spectra whose ~300 ns and 4 µs peaks the MD analysis must reproduce.","marker":"[6]"},{"why":"Supplies the PDZ2S nonequilibrium MD trajectories used to build contact clusters and compute dynamical content.","marker":"[14]"},{"why":"Supplies the PDZ2L transient IR data and the photoinduced ligand-unbinding scenario modeled here.","marker":"[7]"},{"why":"Provides the PDZ2L nonequilibrium MD trajectories and contact-cluster model, and flags the insufficient sampling of ligand rebinding.","marker":"[26]"},{"why":"Introduces the MoSAIC correlation-based feature selection used to identify the localized contact clusters.","marker":"[25]"},{"why":"Establishes the nonequilibrium dynamical-content formalism and multiexponential timescale analysis on which the experiment–simulation comparison rests.","marker":"[21]"}],"fun_headline_variants":["IR relaxation peaks encode local protein contact dynamics","Simulations tie each IR kinetic step to a specific motion","Protein IR spectra decoded by contact network dynamics","PDZ domains: IR kinetics mapped to hierarchical relaxations","Local contact networks drive protein IR relaxation times"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the 10 µs nonequilibrium MD trajectories are sampled well enough that the simulated relaxation peaks are real molecular signals rather than undersampling artifacts—a premise the paper itself qualifies by noting that the N-terminal C8 peak is expected to disappear with better sampling and that the PDZ2L ligand-rebinding trajectories were too short for quantitative comparison.","fun_headline_variants_meta":{"raw":{"variants":["IR relaxation peaks encode local protein contact dynamics","Simulations tie each IR kinetic step to a specific motion","Protein IR spectra decoded by contact network dynamics","PDZ domains: IR kinetics mapped to hierarchical relaxations","Local contact networks drive protein IR relaxation times"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000235,"raw_usage":{"total_tokens":1492,"prompt_tokens":931,"completion_tokens":561,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":547,"completion_tokens_details":{"reasoning_tokens":489}},"tokens_in":547,"tokens_out":561,"duration_ms":5986,"temperature":1.0,"reasoning_tokens":489,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T20:41:56.954152+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run dramatically longer or more replicated nonequilibrium MD for PDZ3 and check whether the 200 ns peak remains assigned to cluster C6 at the $\\beta_1$–$\\beta_2$ loop after the N-terminal contacts are removed; if the peak shifts, splits, or vanishes, the structural assignment is an undersampling artifact. Alternatively, experimentally stabilize or disrupt the $\\beta_1$–$\\beta_2$ loop contacts and see whether the 200 ns allosteric step in the transient IR response moves or disappears.","supporting_citations":[{"cited_title":"The Journal of Chemical Physics , author =","cited_arxiv_id":null,"evidence_quote":"Supplies the PDZ3 transient IR data, including the bound/unbound difference whose 200 ns peak defines the allosteric signaling speed."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the PDZ2L transient IR data and the photoinduced ligand-unbinding scenario modeled here."}],"review_version":1}