{"id":"0bae453c-09a9-446c-8124-a2f6add69aa6","arxiv_id":"2501.13774","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"In a 10,000-patient in silico trial, alternating temozolomide cycles with CAR-T cell injections produced the longest median survival for modeled malignant gliomas.","lead":"A mathematical model of brain tumor growth under combined chemotherapy and CAR-T immunotherapy suggests that alternating the two treatments gives the longest survival in simulated patients. The study offers a framework for comparing treatment schedules in silico before clinical trials.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The alternating-protocol recommendation depends on excluding double-resistant tumor cells; if such a population exists, the predicted synergy may disappear.","rationale":"The reader's weakest_assumption already identifies the missing double-resistant population, and my independent reading agrees: it is the single condition that, if false, most directly undermines the protocol recommendation. The paper's own Section II A acknowledges the omission. The claim that alternating TMZ and CAR-T is optimal depends on each therapy controlling the resistant clone left by the other; a double-resistant clone breaks that complementarity. Other concerns—such as the post hoc restriction of ρ4 to 0.1 or the incomplete proof of Proposition 4—are real but secondary: they affect quantitative effect sizes or mathematical exposition, not the existence of a mechanism for synergy. The proposed extension is a direct falsifiable check: it perturbs the exact structural assumption on which the conclusion rests, using the paper's own parameter ranges. Because the paper already has a CONDITIONAL verdict, and because the missing compartment is a model limitation rather than an internal inconsistency, I recommend no change to the reader's verdict. I do not see grounds for rejection: the model is coherent, the numerical experiments are systematic, and the authors are explicit about the limitation.","tokens_in":36063,"tokens_out":6131,"duration_ms":59300,"concrete_test":"Extend system (1)–(5) with a fourth tumor compartment D such that dD/dt = r3 D (1 - (S + RC + RE + D)/K), optionally adding a TMZ-induced conversion term ε2 E(S + RC) or a small initial fraction δ3. Keep all other equations unchanged and re-run the same in silico trials on the same 10,000 virtual patients, sampling r3 from {r1/2, r1, 2r1} and δ3 from [10^-4, 0.1], for the 10T, 5T2C5T, 1C5T1C5T, 2C10T, and 2C protocols. If the median survival advantage of the alternating protocols over 10T shrinks to less than about 30 days under any plausible δ3, the no-double-resistant assumption is confirmed as load-bearing; if the ordering is preserved, the central recommendation survives this test.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central protocol recommendation (5T2C5T and 1C5T1C5T, median survival near 650 days, Section III B 3 and Discussion E) rests on a two-sided resistance structure: TMZ controls the CAR-T-resistant clone RC and CAR-T controls the TMZ-resistant clone RE. The authors state this explicitly in Section II A: 'we do not include populations resistant to both TMZ and CAR-T therapy.' That exclusion is load-bearing, not a routine simplification. If a double-resistant compartment D exists—either pre-existing at low fraction or induced by treatment—then neither therapy removes it, and the tumor will eventually reach the fatal threshold K/5 driven by D, regardless of how the two drugs are sequenced. In that case the survival differences among 10T, 5T2C5T, and 1C5T1C5T would compress toward the time needed for D to expand, and the 'evolutionary double-bind' rationale for alternating schedules loses its force. The in silico cohort samples δ1 and δ2 but has no δ3 for double resistance, so the 10^4 virtual patients cannot reveal this failure mode. Proposition 6's tumor-free equilibrium also disappears once D is added. The paper is transparent about the assumption, but transparency does not make the conclusion robust to it.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript introduces a five-dimensional ODE model of malignant glioma growth under combined temozolomide (TMZ) and CAR-T cell therapy. The state variables are TMZ-sensitive/CAR-T-sensitive cells (S), CAR-T-resistant but TMZ-sensitive cells (RC), TMZ-resistant but CAR-T-sensitive cells (RE), CAR-T cells (C), and normalized TMZ efficacy (E). The authors prove non-negativity of solutions, identify invariant surfaces, analyze isolated equilibria, and give explicit conditions for local stability of a tumor-free equilibrium under constant daily administration of both therapies. They then simulate a cohort of 10^4 virtual patients with parameters sampled uniformly from literature-based ranges and compare several finite impulsive protocols: TMZ monotherapy, CAR-T monotherapy, and six combined schedules. The central numerical claim is that alternating schedules, specifically 5T2C5T and 1C5T1C5T, give the best median survival, approximately 650 days for r2=r1/2, and that the main survival determinants are tumor growth rate r1, TMZ efficacy alpha1, and tumor-induced CAR-T inactivation rho4.","tokens_in":83,"tokens_out":7951,"duration_ms":140353,"significance":"If the conclusions are robust, the paper provides a useful in silico framework for comparing TMZ/CAR-T scheduling in malignant glioma, with the notable strength of an explicit analytic condition for the tumor-free equilibrium under constant treatment (Proposition 6) and a fully reproducible virtual-patient protocol whose code is stated to be available on GitHub. The correlation analyses between survival and model parameters are clearly presented and could inform design of future experimental or clinical studies. However, the central protocol recommendation depends on several assumptions whose robustness is not established: the exclusion of double-resistant tumor cells, a post hoc restriction of the rho4 range, and unverified parameter identifications such as alpha3=alpha1. These assumptions are load-bearing because the alternating-protocol synergy relies on the two resistance mechanisms being complementary and on CAR-T cells surviving TMZ exposure; if either assumption fails, the predicted ranking of protocols could change substantially.","major_comments":[{"comment":"The manuscript explicitly excludes tumor cells resistant to both TMZ and CAR-T therapy (Section II A: 'we do not include populations resistant to both TMZ and CAR-T therapy'). This exclusion is load-bearing for the main protocol recommendation. The alternating-protocol benefit arises precisely because TMZ controls the CAR-T-resistant clone RC and CAR-T controls the TMZ-resistant clone RE; if a double-resistant clone D exists, neither treatment removes it, all tested protocols converge to the outgrowth time of D, and the 5T2C5T/1C5T1C5T advantage over other schedules may disappear. The virtual cohort samples initial fractions delta1 and delta2 but has no delta3 for double resistance, so the 10^4-patient trials cannot reveal this failure mode, and Proposition 6's tumor-free equilibrium no longer exists once D is present. Please add a sensitivity analysis that includes a double-resistant compartment (or an explicit, evidence-based argument for why such a population cannot arise in malignant glioma) and state how the protocol ranking changes with the initial fraction and growth rate of D.","section":"Section II A, Eqs. (1)-(3); Section III B 3; Table VI"},{"comment":"The upper bound of rho4 is set to 0.1 in all subsequent simulations 'because the median survival time drops sharply if we increase this boundary further (see Fig. 6)'. This is a post hoc model-selection step based on the survival endpoint itself, and it has direct consequences for the reported efficacy of CAR-T-containing protocols: if the biologically plausible upper bound of rho4 were larger, the absolute and relative benefits of CAR-T and combined protocols would shrink substantially (as Fig. 6 shows for CAR-T monotherapy). The authors should either justify the 0.1 cap with independent data on tumor-induced immunosuppression in gliomas or report the sensitivity of the protocol ranking and median survival times to the chosen rho4 range.","section":"Section II D and Fig. 6"},{"comment":"Proposition 6 establishes local stability of the tumor-free equilibrium for constant daily administration of both therapies (constant inputs ar v and ar E0). However, Section IV B states that administering doses 'above the critical thresholds' a finite number of times drives the system 'towards a tumor-free equilibrium'. This is not supported by the analysis: after the last impulsive dose, E decays to zero and C decays or is inactivated, and the tumor regrows, consistent with Proposition 5 and with the simulation results, which measure survival time to the fatal threshold K/5 rather than eradication. The finite-protocol simulations should be described as delaying tumor progression, not as steering the system to the tumor-free equilibrium; otherwise the quantitative claim in the discussion overstates what Proposition 6 and the in silico trials demonstrate.","section":"Section III A (Proposition 6) and Section IV B"},{"comment":"The assumption alpha3 = alpha1 for TMZ killing of CAR-T cells is stated without direct experimental support and is not subjected to sensitivity analysis. Since alpha3 controls the survival of CAR-T cells during TMZ cycles, it directly affects the relative performance of alternating protocols versus TMZ-first or TMZ-last protocols: if TMZ kills CAR-T cells more effectively than tumor cells, the benefit of alternating schedules could be reduced or reversed, whereas if alpha3 is much smaller than alpha1, the benefit could be amplified. The main protocol recommendation therefore depends on an unquantified parameter. Please add a sensitivity analysis over alpha3/alpha1 in a plausible range, or provide direct evidence justifying the identification.","section":"Section II B 2 (Table I) and Section III B 3"}],"minor_comments":[{"comment":"In the proof of Proposition 3, the text writes f4(S,RC,RE,0,E)=v, but the system (1)-(5) has no v term in Eq. (4); for the original system this value is 0. The formula presumably refers to the impulsive or constant-treatment variant, and should be corrected for clarity.","section":"Section III A, proof of Proposition 3"},{"comment":"The initial total tumor size T0 appears in many correlation tables (Tables II-V, VII-X, XII, XIV-XVII) but is not listed in Table I and no sampling distribution or reference value is specified for it. Please add T0 with its range and rationale so that the virtual-patient protocol is reproducible.","section":"Table I and Section II B 3"},{"comment":"The expression for lambda2 repeats the term rho2 S1/(g1+S1) twice; it should evidently contain the corresponding RE term rho3 RE/(g2+RE). The inequality that follows suggests this is a typographical error, but it makes the displayed formula incorrect as written.","section":"Proposition 5, Eq. (13)"},{"comment":"In the discussion of CAR-T monotherapy correlations, the text refers to 'the fraction of tumor cells resistant to CAR-T therapy, delta1'; the correct parameter is delta2, the initial fraction of RC cells. Also, Table XVII lists delta1 twice with different correlation values; the second row should presumably be delta2.","section":"Section IV D"},{"comment":"The difference between the two leading protocols, 5T2C5T and 1C5T1C5T, is 1-2 days in median survival (652 vs. 653 days for 2v=10^9; 689 vs. 688 days for 2v=2e9). This difference is far smaller than the spread within each protocol, so statements that these two are 'the best' should be accompanied by a statistical comparison (e.g., confidence intervals for the median or a paired test) to show that their ranking relative to each other and to 2C10T is meaningful.","section":"Section III B 3, Table VI"},{"comment":"Pearson correlation p-values are reported as 0.00 in several tables; please report them as p<0.001 or with the actual numerical values, and note that a p-value of exactly zero for a finite Monte Carlo sample is not statistically meaningful.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is clearly written and the mathematical analysis is a reasonable starting point, with the explicit tumor-free stability condition and reproducible in silico cohort being genuine strengths. My main concern is not the internal consistency of the model but its robustness for the central protocol recommendation: the exclusion of double-resistant cells and the post hoc rho4 cap are both load-bearing for the claim that alternating TMZ/CAR-T is optimal. These issues are fixable within the scope of the paper by adding a double-resistant compartment, re-running the protocol comparison with different rho4 ranges, and performing sensitivity analysis on alpha3. If the authors can show that the alternating recommendation survives those perturbations, I would be willing to support publication; otherwise the conclusions should be substantially softened. The journal's mathematical audience may also appreciate more complete justification of Proposition 4, since the proof as written checks only selected coordinate subsets, although my reading suggests the omitted cases may be excluded by a short argument that should be made explicit."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper deserves a serious referee. It builds a five-compartment ODE model of TMZ and CAR-T therapy for glioma, with TMZ-induced resistance, and it is upfront about what it does and does not include. The main deliverable is a systematic in silico comparison of six combination protocols on 10,000 virtual patients, with 5T2C5T and 1C5T1C5T giving the best median survival (~650 days). That protocol ranking is new, and the correlation analysis pointing to tumor growth rate, TMZ efficacy, and immunosuppression as survival drivers is a reasonable use of the model.\n\nWhat the paper does well: the math is mostly sound, with explicit non-negativity, invariant surfaces, and a clear stability condition for the tumor-free equilibrium under constant treatment (Proposition 6). The authors are transparent about their assumptions and state plainly that the results are qualitative rather than quantitative. Code is on GitHub. The discussion connects the alternating recommendation to the existing evolutionary double-bind literature, which is honest and puts the claim in context.\n\nWhere the soft spots are: the central recommendation depends on excluding a double-resistant tumor population. The authors say this explicitly in Section II A, but the point is load-bearing. If cells resistant to both TMZ and CAR-T exist at any meaningful frequency, neither drug controls them, and the advantage of alternating schedules would shrink to the time it takes that double-resistant clone to grow to the fatal threshold. That is not a fatal flaw for a modeling paper, but it is a robustness gap that should be addressed or at least discussed more thoroughly. Proposition 4's proof only checks selected coordinate subsets and is incomplete as written; the post hoc cap on rho4, chosen after seeing survival outcomes in Figure 6, is another real weakness. The parameter choices r2 = r1/2 and alpha3 = alpha1 are assumed, and the model is not calibrated to patient data, so the 650-day median should be read as a model output, not a prediction.\n\nThe stress-test note is right: transparency about excluding double-resistant cells does not make the conclusion robust to that exclusion. But the paper itself claims only qualitative insight, and the alternating concept is independently supported in the literature, so this is a conditionally useful modeling contribution rather than a hollow one. I would send it out, asking for a sensitivity analysis with a double-resistant compartment or a clear biological argument against its early emergence, a completed proof of Proposition 4, and a discussion of how the rho4 cap affects the protocol ranking.\n\nThis is a paper for mathematical oncology researchers and for clinicians willing to read it as hypothesis-generating. With those revisions, it would be a solid contribution to the CAR-T modeling literature.","headline":"Plausible in silico case for alternating TMZ/CAR-T, but the argument depends on excluding double-resistant cells; worth refereeing, with robustness checks.","tokens_in":36849,"tokens_out":1532,"would_cite":true,"duration_ms":16318,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["34A37","37N25","92B05"],"pacs":[],"model":"deepseek-v4-flash","headline":"A five-variable model of malignant glioma predicts that alternating CAR-T injections with temozolomide cycles more than doubles median survival in virtual patients.","keywords":["malignant glioma","CAR-T cell therapy","temozolomide","combination therapy","drug resistance","in silico clinical trial","impulsive dynamical system","mathematical oncology"],"falsifier":"Grow tumor cells under alternating TMZ and CAR-T exposure in vitro or in an animal model and measure whether a double-resistant subpopulation emerges while both treatments are present. If such double-resistant clones appear and proliferate at rates comparable to sensitive cells, the model's predicted median survival near 650 days would not be reproduced when the same protocol is simulated with that compartment added.","tokens_in":35842,"feed_emoji":"🧠","tokens_out":5619,"duration_ms":49833,"temperature":0.7,"pith_summary":"The paper builds a mathematical model of malignant glioma growth under combined temozolomide chemotherapy and CAR-T cell immunotherapy, treating drug administrations as instantaneous pulses. It proves that a single round of both therapies only sends the tumor toward its carrying capacity, while constant administration above explicit dose thresholds can stabilize a tumor-free state. The central message from virtual trials is that alternating CAR-T injections between TMZ cycles, specifically the 5T2C5T and 1C5T1C5T protocols, yields the longest median survival, nearly 650 days, compared with 268 days untreated and 558 days with TMZ alone. If correct, this would argue for testing alternating schedules in clinical trials rather than delivering all of one therapy first.","feed_headline":"Alternating CAR-T and TMZ more than doubles modeled glioma survival","feed_subtitle":"Virtual-patient trial of 10,000 gliomas finds alternating schedules beat either therapy alone, with median survival near 650 days.","key_machinery":"The engine is a five-dimensional compartmental ODE system with variables $S$ (cells sensitive to both therapies), $R_C$ (CAR-T-resistant, TMZ-sensitive cells), $R_E$ (TMZ-resistant, CAR-T-sensitive cells), $C$ (CAR-T cells), and $E$ (normalized TMZ efficacy decaying at rate $\\mu$). Treatment is inserted as impulsive jumps $E \\to E+E_0$ and $C \\to C+v$, turning the continuous system into a hybrid model. The mathematical analysis uses invariant surfaces, including the carrying-capacity plane $K-S-R_C-R_E=0$ when no treatment acts, and the eigenvalues of the tumor-free equilibrium to extract dose thresholds. The same model then feeds population-scale virtual trials in which parameters are sampled uniformly and survival is measured as time to $10^{12}$ tumor cells.","core_discovery":"On the paper's own terms, the central discovery is that the sequence and interleaving of chemotherapy and immunotherapy matter more than total dose: alternating TMZ cycles with CAR-T injections lets each therapy control the tumor population the other cannot, and this alternating concept is proposed as the optimal combined-treatment strategy. In silico trials over 10,000 virtual patients give median survival of 652–653 days at $10^9$ CAR-T cells and 688–689 days at $2\\cdot 10^9$ for the two best alternating protocols, against 558 days for ten TMZ cycles alone. The paper also derives explicit threshold inequalities, $\\bar{E}_0(\\alpha_1+\\epsilon_1)>r_1\\mu$ and $\\bar{v}> r_2(\\bar{E}_0\\alpha_3+\\mu\\rho_1)/(\\mu\\alpha_2)$, under which constant treatment makes the tumor-free equilibrium stable, and it shows that if TMZ-resistant cells grow twice as fast as sensitive cells, TMZ becomes counterproductive and CAR-T monotherapy becomes the best option.","pith_inferences":["A direct extension not pursued in the paper: the same alternating logic should be tested in other solid tumors with heterogeneous antigen expression, where chemotherapy can reduce the antigen-negative population and CAR-T can clear the chemo-resistant one.","One could test the model's mechanism by adding a fourth tumor compartment resistant to both treatments; if double-resistant cells arise at clinically relevant rates, the predicted superiority of alternating schedules would likely shrink or disappear.","The protocol-equivalence analysis, in which roughly 61–75% of virtual patients achieve similar survival under any protocol, suggests that only a minority of patients need personalized sequencing; identifying those patients by low immunosuppression and high growth rate could be a prospective stratification.","A testable corollary is that the timing gap between CAR-T injections matters less than dose or sequence, since the model shows only a slight survival decline as the gap increases."],"forward_implications":["An alternating schedule, starting with CAR-T and alternating with TMZ cycles, should be prioritized in future trial designs over giving all TMZ first or all CAR-T first.","Constant daily administration of both therapies, with TMZ efficacy and CAR-T dose above the thresholds in Proposition 6, would suffice for tumor eradication in the model.","Tumor proliferation rate, TMZ killing efficacy, and tumor-induced immunosuppression are the survival biomarkers that most influence outcomes, with immunosuppression being the main factor limiting CAR-T monotherapy.","For tumors whose TMZ-resistant cells grow as fast as or faster than sensitive cells, adding TMZ to CAR-T can reduce survival relative to CAR-T alone, so protocols should be stratified by resistance phenotype."],"supporting_citations":[{"why":"Supplies the TMZ resistance-transition rate $\\epsilon_1$, parameter ranges, and the fatal-volume survival endpoint used throughout the model.","marker":"[56]"},{"why":"Provides the CAR-T population dynamics terms and the killing and stimulation rate parameters ($\\alpha_2$, $\\rho_2$, $\\rho_3$, $g_1$, $g_2$).","marker":"[62]"},{"why":"Defines the standard Stupp TMZ schedule of five days on and 23 days off that the paper's TMZ cycles follow.","marker":"[2]"},{"why":"In vivo evidence that combining TMZ with engineered T cells outperforms monotherapies, used to support the alternating-protocol conclusion.","marker":"[26]"},{"why":"Baseline mathematical CAR-T therapy model for glioblastoma whose CAR-T dynamics the present model extends.","marker":"[20]"},{"why":"Dual CAR-T cell therapy mathematical model supplying parameters and treatment comparison methodology.","marker":"[65]"},{"why":"Clinical observation that about half of glioblastoma patients do not respond to TMZ, cited to validate the scenario with fast-growing TMZ-resistant cells.","marker":"[78]"}],"fun_headline_variants":["Alternating CAR-T and chemo beats each alone in virtual glioma trial","Schedule trumps dose: alternating CAR-T and TMZ extends virtual gliomas","Alternating therapy boosts virtual glioma survival to 650 days","CAR-T and chemo: order matters in 10,000-patient simulation"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The model assumes there are no tumor cells resistant to both TMZ and CAR-T, and that CAR-T-resistant cells grow and respond to TMZ exactly like sensitive cells; if double-resistant cells emerge, the alternating strategy loses its complementary control.","fun_headline_variants_meta":{"raw":{"variants":["Alternating CAR-T and chemo beats each alone in virtual glioma trial","Schedule trumps dose: alternating CAR-T and TMZ extends virtual gliomas","Alternating therapy boosts virtual glioma survival to 650 days","CAR-T and chemo: order matters in 10,000-patient simulation"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000264,"raw_usage":{"total_tokens":1657,"prompt_tokens":1054,"completion_tokens":603,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":670,"completion_tokens_details":{"reasoning_tokens":527}},"tokens_in":670,"tokens_out":603,"duration_ms":5131,"temperature":1.0,"reasoning_tokens":527,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T15:36:16.549625+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Grow tumor cells under alternating TMZ and CAR-T exposure in vitro or in an animal model and measure whether a double-resistant subpopulation emerges while both treatments are present. If such double-resistant clones appear and proliferate at rates comparable to sensitive cells, the model's predicted median survival near 650 days would not be reproduced when the same protocol is simulated with that compartment added.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the TMZ resistance-transition rate $\\epsilon_1$, parameter ranges, and the fatal-volume survival endpoint used throughout the model."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the CAR-T population dynamics terms and the killing and stimulation rate parameters ($\\alpha_2$, $\\rho_2$, $\\rho_3$, $g_1$, $g_2$)."},{"cited_title":"Schmidts , author A","cited_arxiv_id":null,"evidence_quote":"In vivo evidence that combining TMZ with engineered T cells outperforms monotherapies, used to support the alternating-protocol conclusion."},{"cited_title":"Castellarin , author K","cited_arxiv_id":null,"evidence_quote":"Baseline mathematical CAR-T therapy model for glioblastoma whose CAR-T dynamics the present model extends."},{"cited_title":"Li , author R","cited_arxiv_id":null,"evidence_quote":"Dual CAR-T cell therapy mathematical model supplying parameters and treatment comparison methodology."}],"review_version":1}