{"id":"31df76f6-7e77-4d1a-88c8-6b607bb43c17","arxiv_id":"1908.05122","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"PTMs have type-dependent effects on protein backbone flexibility: N-glycosylation has little effect, CDK2 phosphorylation rigidifies the site and destabilizes two distant phosphorylation regions, and actin methylation increases local deformation.","lead":"The authors used a structural alphabet of 16 protein blocks to compare X-ray structures of proteins with and without post-translational modifications. They find that N-glycosylation leaves the backbone unchanged, while phosphorylation in CDK2 stiffens the modified loop and loosens two distant regulatory regions, suggesting an allosteric effect.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"CDK2 pT160 vs non-pT160 comparison is confounded by unmatched cyclin species, ligand, and construct differences; the reported Neq changes at Thr160, Tyr15, and Thr39 may not be caused by the phosphorylation.","rationale":"The paper's most specific and medically interesting result is the CDK2 observation that T160 phosphorylation is associated with local rigidification and increased flexibility at two distal phosphorylation sites. For that causal claim to hold, the compared structure sets must differ in phosphorylation status without systematic differences in other factors. The reader's weakest assumption identifies exactly this vulnerability. My reading of Table 2 strengthens the concern by showing an explicit species imbalance between the pT160 and non-pT160 cyclin A2 complexes, plus unstated variation in ligands and constructs. This is not an internal inconsistency in the method; the Neq computation and the use of external PDB structures are transparent and reproducible. The problem is external validity of the causal attribution. I do not see a reason to move the verdict: the observation is interesting and testable, but the current comparison does not rule out confounders. A matched or covariate-adjusted reanalysis could settle it, so CONDITIONAL remains appropriate.","tokens_in":21953,"tokens_out":5780,"duration_ms":63379,"concrete_test":"Recompute the CDK2 Neq profiles using only human cyclin A2 complexes (UniProt P20248), only ATP-bound structures without inhibitors, and matching construct boundaries and resolution ranges between the pT160 and non-pT160 sets. If the Neq differences at position 160, residues 14-15, and residue 39 shrink below the per-position sampling error or lose their reported separation (e.g., +1 Neq 5.27 vs 2.40; region 8-18 Neq 3.89 vs 6.52), the allosteric claim is not supported. A complementary check is to fit a per-residue mixed-effects model with phosphorylation status as the fixed effect and cyclin species, ligand identity, resolution, and construct as covariates.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central CDK2 result in 'Local and global backbone diversity compared between modified and unmodified proteins' compares 98 pT160 chains with 46 non-pT160 chains, all described as cyclin A2 complexes. But Table 2 shows the pT160 set is 70 human, 24 bovine, and 4 mouse cyclin A2, while the non-pT160 set is entirely human. Ligand types, ATP-analog occupancy, inhibitors, construct boundaries, and crystal forms are not matched or controlled. Supplementary Figure S12 only reports no visible change in ligand binding pattern, which cannot exclude systematic differences in ligand chemistry or crystal contacts. Neq is a per-position ensemble statistic, so any covariate correlated with phosphorylation state can produce the local rigidification at position 160 and the increased flexibility near positions 14-15 and 39. The authors themselves concede this class of confounders in the Discussion and use it to withhold a global actin conclusion ('the lack of consistency in the number, the types and the binding modes of the ligands... make it difficult to conclude'). The same standard is not met for CDK2, so the causal reading that T160 phosphorylation rigidifies the activation loop and allosterically destabilizes two distal regions is not yet secured.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript investigates how post-translational modifications (PTMs) affect protein backbone conformation using the PTM-SD structural database and the Protein Blocks (PB) structural alphabet. The authors first analyze local backbone diversity around N-glycosylation and phosphorylation sites in non-redundant datasets, and then focus on four proteins: renin endopeptidase (N-glycosylation), liver carboxylesterase (N-glycosylation), cyclin-dependent kinase 2 (CDK2, phosphorylation of Thr160), and actin (methylation of His75). They compute Neq, an entropy-like measure of local backbone diversity, from ensembles of X-ray structures, and compare Neq profiles between modified and unmodified proteins, using Mann-Whitney-Wilcoxon tests. The central reported finding is that CDK2 Thr160 phosphorylation rigidifies the activation loop locally while increasing backbone flexibility near Thr14/Tyr15 and Thr39, leading the authors to propose a coupled stiffening-flexibility allosteric mechanism. The manuscript also reports a local destabilizing effect of actin His75 methylation and concludes that N-glycosylation does not alter backbone flexibility in the two examples studied.","tokens_in":22164,"tokens_out":7131,"duration_ms":72897,"significance":"If the CDK2 allosteric claim is correct, it would be a valuable structural observation linking a single phosphorylation event to a specific distal regulatory response in a medically important kinase. The study uses a curated database (PTM-SD) and an established conformational alphabet (PBxplore), and it provides the list of PDB entries used, which supports reproducibility of the dataset. The comparative design, measuring Neq profiles in modified versus unmodified structures, does not fit parameters to the target conclusion, so circularity is not a concern. However, the causal interpretation of the CDK2 and actin comparisons is currently undermined by uncontrolled covariates and by the statistical treatment of the data.","major_comments":[{"comment":"The CDK2 comparison is confounded by differences in the cyclin A2 partner species and by ligand chemistry. As the authors correctly note, all chains are cyclin A2 complexes, but Table 2 shows that the pT160 set contains 70 human, 24 bovine, and 4 mouse cyclin A2 chains, whereas the non-pT160 set contains only human cyclin A2. In addition, the manuscript does not report whether the ligand types (ATP, ADP, ATP analogs, inhibitors) are matched between the two sets; Supplementary Figure S12 only states that no visible change in binding pattern is observed, which cannot exclude systematic differences in ligand chemistry or crystal contacts. Because Neq is a per-position ensemble statistic, any covariate correlated with phosphorylation state could produce the reported local rigidification at position 160 and the increased flexibility near positions 14-15 and 39. The authors should perform a human-only (or species-matched) sensitivity analysis and, if feasible, a ligand-matched analysis to establish that the effect is attributable to Thr160 phosphorylation.","section":"Results: 'Local and global backbone diversity compared between modified and unmodified proteins'; Table 2"},{"comment":"The Mann-Whitney-Wilcoxon tests in Table 1 treat individual sequence positions as independent samples. The PTM-region test uses n=21 positions and the whole-protein test uses n≈300 positions, but neighboring Neq values along a polypeptide are spatially correlated rather than statistically independent. Moreover, the 'Neq 0 PTM vs Neq 1 PTM' comparisons are, for a given protein, paired by residue position, yet an unpaired Mann-Whitney-Wilcoxon test is applied; a paired Wilcoxon signed-rank test or a permutation test respecting the spatial structure would be more appropriate. No correction for multiple testing is made across the eight MWW comparisons reported in Table 1 and Supplementary Table S4. These issues mean the reported p-values (e.g., CDK2 p=0.0132, actin p=0.0007) do not provide reliable statistical evidence as presented. The authors should re-analyze the data with an appropriate paired or block-resampling procedure and report effect sizes or at least the unadjusted versus adjusted results.","section":"Table 1 and 'Various analyses' in Methods"},{"comment":"The manuscript over-interprets non-significant results as evidence of absence when concluding that N-glycosylation 'does not impact the intrinsic flexibility' of renin endopeptidase and liver carboxylesterase. The local comparison is based on only 21 positions in the PTM region, and no power analysis or equivalence testing is provided. A non-significant Mann-Whitney-Wilcoxon test with such a small sample cannot distinguish a true null effect from an underpowered test. Because this null conclusion is part of the paper's general claim that PTM effects depend on PTM type, the authors should either adopt an equivalence testing framework or explicitly describe the conclusion as limited by low statistical power.","section":"Results, N-glycosylation section; Discussion, 'For the two examples of N-glycosylation'"},{"comment":"The definition of the PTM region is not specified. Table 1 consistently shows n=21 for the PTM-region sample, and the text refers to 'the PTM site and its neighboring positions,' but the window size, the exact set of positions, and how the Neq values are pooled across chains or structures are not described in Methods. Without this information, the local MWW test is not reproducible, and it is unclear whether the window is the same for all four proteins. Please provide the precise construction of the PTM-region sample, including the flanking window length and any residue-position exclusion criteria.","section":"Methods and Table 1"}],"minor_comments":[{"comment":"The Methods state '92 phosphorylations on 76 structures', while the Results and Supplementary Table S1 report 75 chains; please reconcile this inconsistency.","section":"Methods, 'Dataset'"},{"comment":"Equation (2) is malformed as printed: 'Neq=exp(−fxx=116∑lnfx)' should read Neq = exp(−Σ_{x=1}^{16} f_x ln f_x). Please correct the typesetting.","section":"Equation (2) in Methods"},{"comment":"The RMSDA formula in Equation (1) is garbled in the manuscript; it should be presented as a sum over the eight dihedral-angle differences, properly formatted.","section":"Equation (1) in Methods"},{"comment":"The legend for Figure 2 indicates the red line as 'the amount of data' but does not define what 'data' means; the figure legend should explicitly state that the red curve is the percentage of ordered residues used to compute each Neq value.","section":"Figure 2 legend"},{"comment":"The table caption would be clearer if it explicitly stated that the sample sizes n are the number of sequence positions included in each Neq comparison.","section":"Table 1 caption"}],"recommendation":"major_revision","confidential_remarks":"The manuscript has a promising descriptive dataset and the CDK2 observation could be interesting, but the central causal claim is not yet secured because of the confounding factors detailed in the major comments. I would ask the authors to re-analyze the CDK2 comparison with species-matched and ligand-matched subsets, and to redo the statistical tests using paired/block-resampling methods. If the CDK2 rigidity-flexibility pattern survives those checks, the paper could be suitable for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's the short version: the paper applies the Protein Blocks/Neq machinery to quantify local backbone diversity around PTM sites, and the interesting output is the CDK2 observation — pT160 is associated with local rigidification of the activation loop and increased flexibility near Thr14/Tyr15 and Thr39, two other regulatory phosphorylation sites. That specific observation is new relative to the RMSD-based work of Xin and Radivojac and the disorder-focused work of Gao and Xu. The catch is that the causal framing runs ahead of the evidence.\n\nWhat the paper does well: PTM-SD is a curated, usable resource; the non-redundant filtering is sensible; the dataset and PDB lists are shipped; and the four-protein design is honest about its own limits — for actin the authors explicitly withhold a global conclusion because of ligand variability. Using Neq rather than RMSD to measure per-position conformational diversity is a genuine plus, and the N-glycosylation null result for renin and carboxylesterase is credible.\n\nThe soft spots are real but fixable. First, the CDK2 comparison is confounded exactly as the stress-test note says: the pT160 set is 70 human, 24 bovine, and 4 mouse cyclin A2 complexes, while the non-pT160 set is entirely human; ligands, inhibitors, and construct boundaries are not matched. Saying the ligands bind in the same pocket is not the same as matching them. The allosteric reading is plausible, not secured. Second, the Mann-Whitney-Wilcoxon tests treat sequence positions as independent samples and no multiple-testing correction is applied; the headline CDK2 p-value of 0.013 would not survive Bonferroni across the four proteins. Third, the N-glycosylation 'no effect' conclusion leans partly on absence of evidence — the PTM-region test uses 21 positions — though the chain-level Neq profiles are actually similar, so the null is reasonable.\n\nCircularity is not a concern: the analysis compares measured Neq profiles between modified and unmodified structures, no parameter is fitted to the conclusion, and the PTM-SD/PBxplore self-citations are tooling. Citation pattern is fine.\n\nWho this is for: structural bioinformaticians working on PTM-structure relationships. It is a moderate, workmanlike contribution with one observation worth following up. It deserves a serious referee: the flaws are addressable and the core idea is useful. I'd send it to review with a request for species-matched CDK2 controls and corrected statistics.","headline":"A workmanlike PB/Neq study of PTM backbone effects; the CDK2 allosteric observation is plausible but not yet secured by the species/ligand-matched comparison.","tokens_in":22779,"tokens_out":5765,"would_cite":false,"duration_ms":53035,"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":"A structural survey argues that post-translational modifications alter backbone flexibility in type-specific ways, with CDK2's Thr160 phosphorylation stiffening its activation loop and increasing flexibility at two other regulatory…","keywords":["post-translational modifications","protein backbone conformation","Protein Blocks","Neq structural entropy","phosphorylation","N-glycosylation","methylation","CDK2"],"falsifier":"Find a matched set of CDK2 crystal structures that use the same cyclin A2 species, similar ligands, and comparable crystal forms and differ only in whether Thr160 is phosphorylated, then recompute $N_{eq}$ at position +1, positions 8-18, and around Thr39. If the rigidification at +1 or the flexibility increase at the two distant regions disappears in the matched comparison, the paper's central allostery claim is not supported.","tokens_in":21742,"feed_emoji":"🧬","tokens_out":9964,"duration_ms":96636,"temperature":0.7,"pith_summary":"The paper sets out to determine whether post-translational modifications (PTMs) change the local and global shape of a protein's backbone, using a 16-letter structural alphabet to compare many X-ray crystal structures of the same protein with and without a modification. Across a non-redundant screen, N-glycosylation and phosphorylation sites appear in highly diverse backbone contexts rather than in one special conformation. In focused comparisons, the authors find that modification effects are type-specific: N-glycosylation in two disulfide-stabilized proteins leaves backbone diversity essentially unchanged, while phosphorylation of CDK2 at Thr160 lowers local backbone diversity at the modified loop and raises it near two other regulatory phosphorylation sites. Methylation of actin's His75 raises local backbone diversity near the site, though heterogeneous ligands block a global conclusion. If right, this means a single PTM can act as both a local stabilizer and a distant destabilizer, providing a structural route for PTM crosstalk in a medically important kinase.","feed_headline":"One phosphate reshapes CDK2's backbone at three sites","feed_subtitle":"Crystal-structure comparison shows Thr160 phosphorylation stiffens CDK2's activation loop and flexes two regulatory regions.","key_machinery":"Protein Blocks (PBs) are a 16-letter structural alphabet for the backbone: each letter represents the local conformation of a five-residue fragment through eight backbone dihedral angles, so every crystal structure becomes a one-dimensional string of letters. The central measure is $N_{eq}$, defined as $N_{eq} = \\exp(-\\sum_x f_x \\ln f_x)$ over the 16 PB frequencies at a position, ranging from 1 (always the same PB) to 16 (all conformations equally frequent). $N_{eq}$ quantifies local backbone deformability across aligned copies of the same protein, and it is compared between PTM regions and the rest of the protein, and between modified and unmodified structure sets, using normalized B-factors and a nonparametric rank test.","core_discovery":"The paper's central discovery is that PTMs affect backbone diversity in a type- and position-dependent way. In a non-redundant screen, neither N-glycosylation nor phosphorylation sites have a single privileged local conformation, although Ser/Thr phosphorylation prefers the beta-strand core Protein Block d in about 40% of cases. In the four case studies, only phosphorylation produces a statistically significant difference between the PTM region and the rest of the protein: CDK2's Thr160 phosphorylation lowers $N_{eq}$ at position +1 from 5.27 to 2.40 and lowers normalized B-factors, rigidifying the activation loop, while raising $N_{eq}$ around positions 8-18 (near Thr14/Tyr15) and near Thr39, two other regulatory phosphorylation sites. The paper interprets this as a coupled stiffening-flexibility pattern indicative of underlying allostery. N-glycosylation in renin endopeptidase and liver carboxylesterase changes no $N_{eq}$ profile, while actin H75 methylation increases local $N_{eq}$ around the site but cannot sustain a global claim because of ligand heterogeneity.","pith_inferences":["A testable extension would replace Thr160 of CDK2 with a phosphomimetic aspartate or a nonphosphorylatable alanine and compare crystal structures; if the two distant regions do not become more flexible, the observed coupling is specific to the phosphate group or to crystallographic conditions rather than a general phosphoregulatory mechanism.","The same $N_{eq}$ strategy could be applied to other kinases with a single well-resolved phosphorylation site to see whether 'stiffen one's own loop, loosen other regulatory loops' is a recurrent motif of kinase activation rather than a CDK2 peculiarity.","Because the actin conclusion is ligand-confounded, a cleaner test would compare methylated and unmethylated actin structures holding the bound nucleotide and drug constant; local differences near His75 would then be attributable to methylation with higher confidence."],"forward_implications":["The paper's results imply that a single PTM can have opposite structural effects inside one protein: phosphorylation of CDK2's Thr160 both rigidifies the activation loop and increases backbone diversity at two distant regulatory regions.","The non-redundant survey implies that N-glycosylation and phosphorylation do not require a particular backbone context, so PTM sites should not be assumed to sit in special or conserved local structures.","The two N-glycosylation case studies imply that some glycans, particularly those solvent-exposed on disulfide-stabilized proteins, can be structurally silent, so detecting a PTM in a crystal does not by itself indicate a conformational role.","The local and distant pattern in CDK2, together with the known opposing roles of Thr160 and Thr14/Tyr15 phosphorylation, implies a structural route for PTM crosstalk in cell-cycle regulation."],"supporting_citations":[{"why":"Builds PTM-SD, the curated structure database from which all modified and unmodified structure sets are drawn.","marker":"Craveur et al. 2014"},{"why":"Defines $N_{eq}$, the per-position backbone diversity measure used throughout the study.","marker":"de Brevern et al. 2000"},{"why":"Defines the 16 Protein Blocks and their assignment to local backbone fragments.","marker":"Etchebest et al. 2005"},{"why":"Provides the prior systematic comparison of modified versus unmodified structures that this paper extends.","marker":"Xin and Radivojac 2012"},{"why":"Reports disorder-to-order transitions for phosphothreonine and other PTMs, the main comparison point for the CDK2 rigidity result.","marker":"Gao and Xu 2012"},{"why":"Supplies the B-factor normalization procedure used for mobility comparisons.","marker":"Smith et al. 2003"},{"why":"Computationally predicted conformational changes in the CDK2 activation loop upon phosphorylation, supporting the rigidity interpretation.","marker":"Groban et al. 2006"},{"why":"Explains how Tyr15 phosphorylation inhibits CDK2-cyclin A activity, giving functional relevance to the flexibility change near Thr14/Tyr15.","marker":"Welburn et al. 2007"},{"why":"Links Thr39 phosphorylation to CDK2 localization and apoptosis, providing functional context for the second distant flexible region.","marker":"Maddika et al. 2008"},{"why":"Establishes the opposing regulatory roles of Thr160 and Tyr15 phosphorylation in CDK2, which the allosteric interpretation relies on.","marker":"Gu et al. 1992"}],"fun_headline_variants":["Phosphorylation stiffens CDK2 loop, flexes regulatory sites","PTM backbone effects are type-specific, CDK2 shows allostery","CDK2 Thr160 phosphorylation shifts local backbone flexibility","N-glycosylation has no backbone effect, phosphorylation does"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the modified and unmodified crystal-structure sets differ only in the PTM; ligand variation, crystal contacts, additional modifications, and construct boundaries are not fully controlled, and for actin the authors concede this heterogeneity precludes a global conclusion.","fun_headline_variants_meta":{"raw":{"variants":["Phosphorylation stiffens CDK2 loop, flexes regulatory sites","PTM backbone effects are type-specific, CDK2 shows allostery","CDK2 Thr160 phosphorylation shifts local backbone flexibility","N-glycosylation has no backbone effect, phosphorylation does"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000198,"raw_usage":{"total_tokens":1416,"prompt_tokens":1039,"completion_tokens":377,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":655,"completion_tokens_details":{"reasoning_tokens":304}},"tokens_in":655,"tokens_out":377,"duration_ms":4484,"temperature":1.0,"reasoning_tokens":304,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:22:15.394025+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find a matched set of CDK2 crystal structures that use the same cyclin A2 species, similar ligands, and comparable crystal forms and differ only in whether Thr160 is phosphorylated, then recompute $N_{eq}$ at position +1, positions 8-18, and around Thr39. If the rigidification at +1 or the flexibility increase at the two distant regions disappears in the matched comparison, the paper's central allostery claim is not supported.","supporting_citations":[],"review_version":1}