{"id":"99252e13-d168-4b15-bea2-0d090ef0eadc","arxiv_id":"2607.13966","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"When the water around yeast frataxin freezes into ice Ih, the protein stays folded but explores far fewer conformations and keeps a denser, more persistent hydration shell.","lead":"Computer simulations grew ice around the yeast protein frataxin and tracked what the protein does while its water freezes. The ice cage didn't unfold the protein, but it trapped the protein in fewer shapes and left a denser, longer-lived water layer on its surface.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Final 200 ns window not shown to be stationary for the protein; the authors' own dPCA analysis reports 'progressive confinement' continuing after 500 ns, so the observed contraction could be kinetic trapping rather than an ice-induced equilibrium restriction.","rationale":"The reader identified the same load-bearing weakness: the final 200 ns window is asserted to be quasi-stationary based on solvent observables, but the protein's own dynamics may still be relaxing, making the observed conformational restriction a sampling artifact. I agree with this assessment. The manuscript's own text in Section 3.2 ('progressive confinement') is direct evidence that the protein ensemble is not demonstrated to be stationary, and the use of full-trajectory dPCA/FEL compounds the issue. This does not necessarily invalidate the paper—if the intended claim is explicitly kinetic confinement, longer simulations and multiple liquid control replicas could rescue it—but as written, the thermodynamic and 'stability' interpretation is not supported. Since the reader's verdict is CONDITIONAL and this concern is exactly what makes the result conditional rather than confirmed, no change in verdict is needed.","tokens_in":15898,"tokens_out":6598,"duration_ms":75474,"concrete_test":"Split the final 200 ns of each ice replica into two 100 ns blocks and recompute the dPCA projection, SASA, and Rg distributions for each block. If the Jensen–Shannon divergence between the first and second block is comparable to the ice-versus-Tm difference, or if mean PC1/PC2 occupancy drifts monotonically across the 200 ns window, then the protein ensemble is not stationary and the confinement effect is a kinetic artifact rather than a steady-state property.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires comparing ice and liquid ensembles in a quasi-stationary regime. The paper justifies the final 200 ns window using plateaus in solvent density and potential energy (Section 3.1), but it never establishes that the protein's conformational observables are stationary in this window. Section 3.2 explicitly states that during the first ~500 ns 'significant conformational transitions' occur and are 'followed by progressive confinement within specific regions of the essential subspace,' implying the protein may still be drifting during the 500–1000 ns interval from which the final 200 ns are drawn. Moreover, the dPCA free-energy landscapes in Figure 5 are computed from the full 1 µs trajectories (Section 2.3), so they mix the ice-growth transient with the purportedly stationary regime; applying the Boltzmann relation to a non-equilibrium trajectory yields a histogram, not an equilibrium free energy. The four ice replicas appear to start from the same protein structure (PDB 2GA5), so they demonstrate reproducibility of a relaxation pathway, not independent equilibrium sampling, and each liquid control has only one trajectory. If the final 200 ns is still evolving, the observed contraction in conformational space and the 'stability' language in the title/conclusions overreach what the simulations establish.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper uses ~1 µs atomistic MD simulations of yeast frataxin (Yfh1) in TIP4P/2005 water to compare the protein's behavior when the solvent crystallizes into ice Ih (four ice-seeded replicas, R1–R4) with single liquid-water controls at 298 K (Ta), 278 K (TMD), and near the model melting point (Tm). Ice formation is validated by solvent density, potential energy, and local bond-order parameters W4/W6. The authors then use PCA, dPCA, and free-energy landscapes built from the 1 µs trajectories, together with SASA, radius of gyration, hydrogen-bonding, and hydration-density analyses restricted mostly to the final 200 ns (or final 100 ns for hydration persistence), to conclude that ice formation does not drive global unfolding but instead restricts conformational sampling, discretizes the conformational landscape, and preserves/densifies the interfacial hydration layer. The central claim is that solvent structure, not temperature alone, governs protein behavior under freezing conditions.","tokens_in":16239,"tokens_out":7088,"duration_ms":69050,"significance":"If established, the result would address an experimentally difficult regime—protein behavior in a growing ice lattice—and would have implications for cryopreservation and for understanding cold denaturation. The paper has several strengths: the explicit ice-growth setup with four ice replicas, the multi-observable validation of the liquid–solid transition (density, energy, W4/W6), and the direct liquid controls at three temperatures. However, the central equilibrium interpretation is currently not secured: the analysis window is justified by solvent stationarity rather than protein stationarity, the free-energy landscapes are computed from non-stationary full trajectories, and each liquid condition rests on a single trajectory with no run-to-run error bars. These issues affect the main conclusion and require re-analysis rather than mere editing.","major_comments":[{"comment":"The paper justifies the final-200 ns analysis window using plateaus in solvent density and potential energy (§3.1), but the protein's conformational observables are not shown to be stationary in that window. §3.2 explicitly says that the first ~500 ns contain 'significant conformational transitions' and are 'followed by progressive confinement within specific regions of the essential subspace.' The dPCA projections in Fig. 5 are shown over the full 1 µs and continue to gate into discrete minima toward the end. A 'progressive confinement' that continues past 500 ns means the 800–1000 ns window may still be part of a slow relaxation after ice growth, not a quasi-stationary ensemble. The observed contraction of conformational space could then be kinetic trapping rather than an ice-induced equilibrium restriction. Please quantify protein stationarity inside the final 200 ns (block analysis o","section":"§3.1, §3.2, Fig. 5"},{"comment":"Eq. (1) applies the Boltzmann relation to the joint distribution of PC1/PC2 from the full 1 µs trajectory. Because the trajectory includes the liquid-to-ice transition and the slow protein drift discussed in §3.2, the resulting histogram is not a Boltzmann equilibrium population; barrier heights and basin depths in the 'free energy landscapes' of Fig. 5 are not thermodynamically interpretable. Please compute the landscapes only from quasi-stationary portions of the trajectories (or use an equilibrium sampling scheme), and consider reporting them as conformational density plots rather than free energies.","section":"§2.4, Eq. (1)"},{"comment":"Table 1 lists only one liquid-control trajectory for each of Ta, TMD, and Tm, and each liquid system is analyzed as a single trajectory. Figures 6 and 7 therefore compare temporal fluctuations within one trajectory (ice: four replicas) with temporal fluctuations in one liquid trajectory; no run-to-run or bootstrapped error bars are provided. The four ice replicas all start from the same PDB structure 2GA5, so they demonstrate reproducibility of one relaxation pathway rather than independent equilibrium samples. At a minimum, add multiple independent liquid simulations or block-bootstrap uncertainties, and state explicitly whether the R1–R4 replicas differ in protein initial conditions or only in solvent/seed arrangement.","section":"Table 1, §2.2, Figs. 6–7"},{"comment":"The title and Conclusions use the word 'stability,' but no stability free energy is computed; the paper measures conformational sampling, SASA, Rg, and hydrogen bonding. The last paragraph of §3.3 itself notes that current force fields are parameterized to maintain the native fold over accessible timescales, so the preservation of secondary structure does not establish thermodynamic stability. I recommend either computing a direct free-energy/stability metric (e.g., from equilibrium sampling or reweighting) or restricting the conclusions to 'restriction of conformational sampling' rather than 'stability.'","section":"§3.3, Conclusions"}],"minor_comments":[{"comment":"Typo: 'The Computations were carried out...' should read 'Computations were carried out...'.","section":"§2.2"},{"comment":"The text says residues 1–19 were 'excluded from all structural analyses,' but Fig. S3 and parts of the SASA/Rg analysis appear to report values for the full protein. Please clarify which analyses use the globular domain only and which use all atoms.","section":"§2.3 / Figs. S3–S4"},{"comment":"The inset panel is described as showing W6 vs W4, but the axis labels are 'W4 (A.U.)' and 'W6 (A.U.)'. Please make the axis labeling and legend consistent, and state whether the W4/W6 values in the main panel come from the final 200 ns window.","section":"Fig. 3"},{"comment":"Minor notation issue: the text reports Tm Rg ≈ 1.414 nm while the figure y-axis is labeled generically; a uniform significant-figures convention would improve readability.","section":"§3.3 / Fig. 6E"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern largely lands: the central claim is currently supported by non-stationary trajectories and single liquid controls. The four ice replicas and order-parameter validation are genuine strengths, and I believe the paper is worth a major revision rather than rejection. I would ask the editor to ensure the authors quantitatively distinguish kinetic trapping from equilibrium confinement, and to address the lack of error bars on the liquid controls."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The genuinely new thing here is applying the Kuiper ice-seeded model to a cold-labile protein, Yfh1, and showing that when the surrounding water crystallizes the protein's conformational landscape becomes discretized while its first-shell hydration layer stays liquid-like and denser. That is a real observation, not previously in the literature, and the phase transition itself is well validated: density, potential energy, and W4/W6 bond-order parameters all converge consistently across four ice replicas. The hydration-density normalization is thoughtful, and the distinction between the quasi-liquid interfacial layer and bulk ice is clearly drawn. Credit where due: this is a careful simulation study of a regime that is experimentally hard to reach.\n\nThe soft spots are real but not fatal. Each liquid control is a single trajectory with no run-to-run error bars, so the comparison rests on four ice replicas against one liquid sample per temperature. More importantly, the final 200 ns window is justified by plateaus in solvent density and potential energy, not by stationarity of the protein's conformational observables. The authors themselves report 'significant conformational transitions' during the first ~500 ns and 'progressive confinement' afterward, which suggests the protein may still be drifting through 500–1000 ns. The dPCA free-energy landscapes are computed from the full 1 µs trajectories, so they mix the ice-growth transient with the allegedly stationary regime; applying the Boltzmann relation to a non-equilibrium trajectory yields a histogram, not an equilibrium free energy. And because the four replicas start from the same PDB structure, they demonstrate reproducibility of a relaxation pathway, not independent equilibrium sampling. The authors even mention kinetic trapping in the conclusions, yet the title and abstract frame the result as 'stability.' That overreaches the measurements.\n\nNone of this is disqualifying. The observed confinement is plausible and the hydration effect is interesting; the main claims are conditional, not wrong. The paper would benefit from more liquid replicas, longer sampling, or an explicit test of whether the final 200 ns is stationary for the protein's own observables (e.g., block-averaged RMSD or dPCA projections). No code or data are shipped, which limits independent checking but is common for this type of MD study.\n\nWho is this for? People working on cryopreservation, freeze-drying, or protein–water interactions at low temperature will want to read it as a first computational look at a protein embedded in ice Ih. It deserves a serious referee: the question is timely, the model choice is defensible, and the flaws are addressable rather than load-bearing. I would send it out, with the clear request that the authors either strengthen the equilibrium case or soften the 'stability' language.","headline":"A plausible, novel MD observation that ice Ih confinement restricts Yfh1's conformational ensemble and densifies its hydration shell, but the equilibrium interpretation is undercut by single liquid controls, a post hoc analysis window, and possible kinetic trapping.","tokens_in":16678,"tokens_out":1547,"would_cite":true,"duration_ms":17087,"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":"Freezing water confines a protein's shape instead of unfolding it, and densifies its hydration layer.","keywords":["protein stability","ice nucleation","confinement","hydration layer","molecular dynamics","free energy landscape","frataxin","cryopreservation"],"falsifier":"If a frozen Yfh1 sample showed the same broad conformational ensemble as supercooled liquid at the same temperature, the confinement claim would fail; in simulation, extending the frozen trajectories well beyond 200 ns and observing the protein cross between the discrete minima would indicate kinetic trapping, not thermodynamic confinement.","tokens_in":15804,"feed_emoji":"❄️","tokens_out":5356,"duration_ms":51080,"temperature":0.7,"pith_summary":"Low-temperature protein studies have mostly focused on supercooled liquid water, but what happens when the solvent actually crystallizes into ice is largely invisible to experiment. This paper uses molecular dynamics simulations that grow hexagonal ice around yeast frataxin (Yfh1) to ask whether ice formation drives, suppresses, or merely constrains protein conformational change. The authors claim that ice does not drive global unfolding: instead, the crystalline lattice acts as a geometric constraint on the solvent, shrinking the protein's accessible conformational space and fragmenting its free-energy landscape into discrete, kinetically confined states. At the same time, the hydration layer within a few ångströms of the protein stays liquid-like but becomes denser per unit area and far more persistent. If correct, this means protein behavior at low temperature is governed by the structural state of water—ice versus liquid—not by temperature alone, which has direct implications for cold denaturation and cryopreservation.","feed_headline":"Ice confines proteins instead of unfolding them","feed_subtitle":"Simulations show freezing water densifies the hydration layer while restricting protein shape changes.","key_machinery":"The central mechanism is the ice lattice itself as a geometric constraint. The simulations use a fixed hexagonal ice seed surrounded by liquid water; below the model's melting point the seed grows into an extended ice Ih lattice around the protein, while leaving a liquid-like interfacial layer. This bulk crystallization is validated by solvent density, potential energy, and local bond-order parameters. The load-bearing analyses are principal component analysis (PCA) and dihedral PCA (dPCA) of backbone motion, which map the free-energy landscape, together with surface hydration density—water molecules normalized by solvent-accessible surface area—and hydrogen-bond counts.","core_discovery":"The paper reports that as the aqueous environment of yeast frataxin crystallizes into hexagonal ice, the protein's conformational ensemble contracts. Principal component analysis and dihedral principal component analysis of 1-microsecond trajectories show a shift from a continuous, connected free-energy landscape in liquid water to a discretized landscape with well-separated minima in the frozen replicas. The radius of gyration and secondary-structure content indicate partial expansion of the protein without global unfolding. Interfacial water within about 5.6 Å of the protein retains liquid-like hydrogen bonding, but its surface density (water molecules per unit solvent-accessible surface a","pith_inferences":["If the confinement is genuinely geometric, the accessible substates should depend on the shape and orientation of the ice cavity around the protein; varying the seed geometry or protein orientation relative to the crystal planes could test this directly.","The densification and persistence of interfacial water under ice may be a general feature of ice-confined solutes, not specific to Yfh1; the same mechanism could apply to other proteins, nucleic acids, or nanoparticles in frozen environments.","The restriction of the analysis to the final 200 ns means the discrete minima could be kinetic traps rather than equilibrium states; longer simulations or enhanced-sampling methods would distinguish true confinement from slow relaxation after ice growth."],"forward_implications":["Low-temperature protein behavior should be described by solvent phase, not just temperature: the same protein at the same temperature shows a broad liquid ensemble or a confined crystalline ensemble depending on whether water is liquid or ice.","Cryopreservation by freezing may protect proteins by reducing conformational sampling and trapping substates, rather than by immobilizing the native fold; the hydration shell remains dense and liquid-like at the interface.","Cold-denaturation mechanisms derived for supercooled liquid water do not transfer directly to frozen systems, because ice adds a mechanical/geometric constraint that can suppress the unfolding that supercooling alone would promote.","Simulations of proteins at low temperature should explicitly model ice growth; treating the solvent as a supercooled liquid misses the dominant effect of solvent structure."],"fun_headline_variants":["Ice cages protein shapes, not unfolding them","Freezing water densifies hydration layer, curbs protein motion","In ice, proteins lose motion but keep their structure","Ice traps proteins in conformations without unfolding","Frozen water packs protein surface, reduces flexibility"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The claim rests on treating the final 200 ns of each trajectory as a quasi-stationary ensemble for the protein, even though the protein undergoes significant conformational transitions during the first ~500 ns of ice growth; if those states are still relaxing, the observed contraction of conformational space reflects slow non-equilibrium trapping rather than a thermodynamic consequence of ice.","fun_headline_variants_meta":{"raw":{"variants":["Ice cages protein shapes, not unfolding them","Freezing water densifies hydration layer, curbs protein motion","In ice, proteins lose motion but keep their structure","Ice traps proteins in conformations without unfolding","Frozen water packs protein surface, reduces flexibility"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00067,"raw_usage":{"total_tokens":2912,"prompt_tokens":787,"completion_tokens":2125,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":531,"completion_tokens_details":{"reasoning_tokens":2052}},"tokens_in":531,"tokens_out":2125,"duration_ms":15971,"temperature":1.0,"reasoning_tokens":2052,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T03:10:11.743173+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"If a frozen Yfh1 sample showed the same broad conformational ensemble as supercooled liquid at the same temperature, the confinement claim would fail; in simulation, extending the frozen trajectories well beyond 200 ns and observing the protein cross between the discrete minima would indicate kinetic trapping, not thermodynamic confinement.","supporting_citations":[],"review_version":1}