{"id":"823ac46e-febd-40e7-b752-b7b803e71e02","arxiv_id":"2507.07693","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.5,"correctness_risk":"medium","formal_verification":"none","parameter_count":8,"one_line_summary":"A tripartite-synapse model with astrocyte modulation of multiple synapses suppresses self-sustained activity and enlarges the stimulus-responsive regime in recurrent circuits, with the strongest effect for internal synapses.","lead":"This paper proposes a computational model of how astrocytes, a type of brain support cell, modulate synapses and shows that this 'higher-order' regulation prevents runaway activity in a minimal recurrent circuit. The generalist reader might care because it suggests a concrete mechanism, beyond direct neuron-to-neuron wiring, that could stabilize neural networks and improve their responsiveness to input.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claim that low-order schemes cannot replicate higher-order modulation (§III.C, Fig. 5) is tested only over two astrocyte parameters (β, Ca_th); a broader low-order parameter search could recover the same adequate-response interval, collapsing the higher-order advantage to a scaling effect.","rationale":"Good faith reading: the paper builds a clearly specified ASN-STP model that reduces to Tsodyks-Markram and De Pittà–Brunel in limits, and demonstrates in a minimal recurrent circuit that astrocyte modulation shrinks the SSA region. The central result is plausible and the model is internally consistent. The most load-bearing concern is not the linearization of Eq. 11 per se (although it deserves a robustness check), but the paper's own falsification test for the higher-order interpretation: Section III.C claims low-order modulation 'cannot be replicated' after varying only β and Ca_th. That claim is the hinge between 'astrocytes help' and 'astrocyte higher-order topology is specifically special'. With 8+ free parameters and an imposed symmetry between the two low-order astrocytes, the two-parameter scan is insufficient to establish irreducibility. A concrete low-order configuration with tuned U_astro or heterogeneous parameters might match Figure 3e/4e; if so, the central comparative claim would be reduced to a parameter-scaling statement. This does not invalidate the sufficiency result, but it changes the paper's main interpretation. The reader's CONDITIONAL verdict already captures this via 'limited parameter scans', so my recommendation is UNCHANGED, pending the proposed search. Credit where due: the model reductions in the SI are analytically derived, and the SSA criterion is validated by removing stimulation. No machine-checked proof or code is provided, so parameter-space claims depend on reproducible simulations.","tokens_in":24456,"tokens_out":11875,"duration_ms":140460,"concrete_test":"Replace the two-parameter scan of §III.C with a systematic search over low-order astrocyte parameters for the two-astrocyte scheme (synapses 2→3 and 3→1). Scan U_astro in [0.05,0.5], τ_f,astro in [1000,10000] ms, ε in [0.005,0.05], β in [0.05,0.3], Ca_th in [0.01,0.06], and allow the two astrocytes to have independent values (e.g., recurrent-synapse astrocyte stronger/faster). For each of ~10^4 random or grid combinations, simulate the N=3 ring with stimulus frequencies 1–100 Hz, compute the adequate-response interval defined in §III.B, and quantify tuning smoothness as the median absolute deviation of ⟨ν_out⟩ around a monotone fit (as in Fig. 4). If the best low-order interval and smoothness reach or exceed the higher-order panel (e) values, the 'cannot be replicated' claim fails; if not, the higher-order advantage is confirmed as structural.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central qualitative claim is that higher-order astrocyte modulation is functionally special: Section III.C states that the benefits 'cannot be replicated' by low-order schemes, and Figure 5 supports this by increasing β and decreasing Ca_th in the two-astrocyte low-order configuration. This is the load-bearing test for the 'higher-order' interpretation. However, the scan covers only two of the astrocyte-related parameters (β∈{0.05,0.1,0.15}, Ca_th∈{0.02,0.04}), holding U_astro=0.1, τ_f,astro=5000 ms, ε=0.01, and all synaptic/neuronal parameters at their Table I defaults. The model has at least eight relevant parameters, and the two low-order astrocytes are forced to be symmetric. Nothing in the model forbids, for example, a low-order astrocyte on the recurrent synapse 3→1 with larger U_astro or smaller τ_f,astro than the astrocyte on 2→3, which could produce the asymmetric, slow modulation that makes the higher-order scheme optimal. If any low-order parameter combination (including heterogeneous ones) yields an adequate-response interval and tuning smoothness matching or exceeding Figure 3e/4e, then the observed advantage is not an effect of higher-order interaction topology but of effective astrocyte drive, directly undermining the central claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a tripartite-synapse short-term plasticity model (ASN-STP) in which astrocyte calcium dynamics and presynaptic activity jointly control the neurotransmitter release probability u. The model is shown to reduce to the Tsodyks-Markram and De Pittà-Brunel models in appropriate limits. The authors apply the model to a directed ring of three leaky integrate-and-fire neurons with one externally stimulated neuron, comparing configurations with no astrocyte, one astrocyte per synapse (low order), and one astrocyte modulating multiple synapses (higher order). They report that higher-order astrocyte modulation, especially when a single astrocyte controls the two internal synapses (2→3 and 3→1), suppresses self-sustained activity (SSA), enlarges the 'adequate response interval' of the read-out neuron, and produces smoother input-output tuning than low-order schemes. They further claim that these benefits cannot be reproduced by amplifying low-order astrocyte drive or by diffusive gap-junction coupling of low-order astrocytes, and they present supporting simulations for N=5 and N=20 cycles.","tokens_in":24742,"tokens_out":8231,"duration_ms":86496,"significance":"If the central claim is accepted, the paper offers a concrete mechanism by which astrocyte-mediated higher-order interactions stabilize recurrent excitatory circuits and preserve stimulus encoding, bridging higher-order network theory and glial biology. The model is clearly specified and the reductions to previous models are checked in the Supplementary Information; the SSA transition boundary is derived from an explicit criterion (Supp. Eq. S.4) and validated against simulations, which is a notable strength. The paper also generates falsifiable predictions, e.g., that astrocyte modulation of internal synapses but not the input synapse expands the responsive regime. However, the generality of the main quantitative claims is tempered by the narrow parameter search supporting the 'cannot be replicated' conclusion and by the informal definition of the adequate-response interval.","major_comments":[{"comment":"The claim that the benefits of higher-order modulation 'cannot be replicated' by low-order schemes rests on a parameter scan that varies only β∈{0.05,0.1,0.15} and Ca_th∈{0.02,0.04} for the two-astrocyte configuration, with U_astro=0.1, τ_f,astro=5000 ms, ε=0.01, and all synaptic parameters fixed at their Table I defaults. This covers only a small slice of the low-order parameter space and forces the two low-order astrocytes to be symmetric. The model does not prohibit, for example, a low-order astrocyte on synapse 3→1 with a larger U_astro or a smaller τ_f,astro than the astrocyte on 2→3, which could emulate the slow asymmetric modulation characteristic of the higher-order scheme. Unless the authors either (i) perform a systematic search over astrocyte and synaptic parameters, including heterogeneous low-order configurations, or (ii) provide a mechanistic argument (e.g., based on the number of integrated inputs) that rules out such emulation, the conclusion that the advantage arises from higher-order interaction topology rather than from effective astrocyte drive is not established. This is load-bearing because the Introduction and Discussion advance the higher-order interpretation as the paper's main result.","section":"§III.C, Fig. 5"},{"comment":"The 'adequate response interval' is defined qualitatively as the range of α for which the read-out neuron 'responds proportionally' to stimulus frequencies and for which the circuit is not in the SSA regime. Neither the proportionality criterion nor the interval endpoints are specified quantitatively, and Figure 4's 'smoothness' assessment relies on visual inspection. Because the central quantitative claim—that the higher-order internal-synapse scheme (Fig. 3e/4e) yields the largest adequate-response interval and the smoothest tuning—depends on this definition, I recommend providing an explicit algorithm or a quantitative measure (e.g., a linear-regression R² or coefficient-of-variation threshold) and reporting the computed interval for each panel.","section":"§III.B, Figs. 3–4"},{"comment":"The model is justified as a first-order Taylor expansion of u(γ_astro, γ_pre) around the steady state, and Eq. (11) then defines u as exactly this linear expression. The paper does not report the actual ranges of γ_astro and γ_pre reached in the simulations, nor does it test whether a saturating nonlinear functional form would alter the phase diagrams. Since the quantitative predictions (SSA boundary, adequate-response interval) are derived from this specific functional form, the linearity assumption is load-bearing. I suggest either adding a sensitivity analysis with a saturating alternative, e.g., u = U_SE + (ε−U_SE) γ_astro/(1+γ_astro) + (1−U_SE) γ_pre/(1+γ_pre), or explicitly discussing the limitations of the linear form in the Discussion.","section":"§II.B.1, Eqs. (5) and (11)"}],"minor_comments":[{"comment":"The phrase 'In resume' should be 'In summary' (or 'In conclusion'), and there is a typo 'triparte-synapse' in the second paragraph of the Introduction.","section":"Intro., end of Section I"},{"comment":"The caption lists 'a single astrocyte modulating three internal synapses (panel c); all four internal synapses (panel c)'—the second 'panel c' should refer to a different panel (likely panel b), and the panel letters in the caption should be checked against the figure.","section":"Fig. 8 caption"},{"comment":"The references to 'eq. II C' are ambiguous; the calcium dynamics are given in Eq. (17), and the text and caption should cite the equation number directly rather than the section label.","section":"§III.C and Fig. 5 caption"},{"comment":"The notation for the calcium threshold is inconsistent: 'Ca_th' in the main text and Table I, but 'Cathr' and 'Cath' appear in Supplementary Figures D and the text near Eqs. (8)–(9). Please unify the notation.","section":"Throughout"},{"comment":"In the text describing Figure 8, the sentence 'when the astrocyte modulates all synapses but this one (n=4 interacting synapses, panel b)' is contradicted by the caption's labeling; please correct the panel references so that the number of modulated synapses matches the stated scheme.","section":"§III.F, Fig. 8 panel labels"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read this if you care about astrocyte models or higher-order interactions. The paper builds a manageable model—ASN-STP—that puts Tsodyks-Markram and De Pittà-Brunel astrocyte modulation into one framework, with an alpha parameter splitting neurotransmitter between synapse and astrocyte. The reduction to both previous models is checked in the SI, and the SSA transition criterion in Eq. S.4 is derived and validated by simulation. That is real work. The main qualitative result is also defensible: in a three-neuron recurrent loop, a single astrocyte modulating two internal synapses expands the stimulus-responsive region and gives the smoothest input-output curve. The mechanism is intuitive—integration of IP3 from multiple synapses raises calcium faster and releases gliotransmitters synchronously—and Supp. Fig. D supports it.\n\nThe soft spots are in the rhetoric and in one specific claim. Section III.C says low-order modulation \"cannot be replicated\" by low-order schemes because changing beta and Ca_th in the two-astrocyte setup does not recover the higher-order adequate-response interval. But the scan only varies two parameters and holds all others, including U_astro, tau_f_astro, and symmetry between the two low-order astrocytes. A heterogeneous low-order arrangement—say, a stronger astrocyte on the recurrence synapse—could plausibly mimic the higher-order effect. So the stress-test concern is fair: the strong version of the higher-order-is-special claim is not established. The qualitative demonstration is fine, but \"cannot be replicated\" should either be backed by a wider parameter search or softened. The Discussion also overstates by saying the homeostatic effect \"only emerges\" with higher-order modulation; their own Figure 3 shows low-order schemes also expand the responsive interval substantially. The quantitative results lack error bars, and no code or data are shipped, so the phase diagrams are hard to scrutinize. The linear first-order expansion for u is a modeling assumption—reasonable near steady state, but unverified away from it.\n\nThis paper deserves a serious referee. It is a useful proof-of-concept for computational neuroscientists studying astrocyte function or higher-order network dynamics, and the model itself is a clean generalization of previous work. I would accept it for review, but I would ask the authors to either broaden the low-order search (including heterogeneous parameters) or retract the \"cannot replicate\" claim, and to bring the Discussion in line with their own results.","headline":"A clearly specified astrocyte-STP model with a plausible higher-order stabilization result, but the paper overclaims that low-order schemes cannot replicate it based on a narrow parameter scan.","tokens_in":25277,"tokens_out":2063,"would_cite":true,"duration_ms":26414,"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 single astrocyte controlling multiple internal synapses stabilizes recurrent circuits against self-sustained activity and preserves stimulus encoding.","keywords":["astrocyte","tripartite synapse","short-term plasticity","higher-order interactions","recurrent neural circuits","self-sustained activity","gliotransmission"],"falsifier":"Run the same three-neuron ring with a saturating nonlinear gliotransmission term (for example, one whose effect levels off at high calcium levels) in place of the linear expansion in Eq. (11). If the internal-synapse higher-order scheme no longer yields the largest adequate-response interval, the claimed advantage is an artifact of the linear approximation rather than a property of higher-order astrocyte control.","tokens_in":24194,"feed_emoji":"🧠","tokens_out":9352,"duration_ms":90641,"temperature":0.7,"pith_summary":"Astrocytes are proposed to act as higher-order controllers of synaptic transmission: one astrocyte monitors and modulates several synapses at once, coupling their short-term plasticity dynamics. In a minimal recurrent loop of three excitatory neurons, this higher-order modulation prevents the self-sustained runaway activity that otherwise makes the circuit insensitive to external stimuli. The paper builds a short-term plasticity model in which astrocyte gliotransmission and presynaptic facilitation jointly set the neurotransmitter release probability, generalizing standard dynamic-synapse models. Simulations show that the stabilizing effect is strongest when a single astrocyte modulates the two internal synapses, and that it cannot be reproduced by boosting calcium responses in low-order, one-synapse-per-astrocyte schemes. If true, this offers a plausible mechanism by which glial cells could keep recurrent circuits responsive without inhibitory feedback.","feed_headline":"Astrocytes can stop runaway excitation in recurrent circuits","feed_subtitle":"One astrocyte regulating several inner synapses widens the input range a recurrent loop can encode.","key_machinery":"The machinery is a coupled astrocyte-synapse-neuron short-term plasticity model (ASN-STP). The release probability at each synapse is written as a first-order expansion $$u_{ij}^{a} = U_{SE} + (\\varepsilon-U_{SE})\\$gamma^{{a}}$_{ij,\\mathrm{astro}} + (1-U_{SE})\\$gamma^{{a}}$_{ij,\\mathrm{pre}},$$ where $\\gamma_{ij,\\mathrm{astro}}$ and $\\gamma_{ij,\\mathrm{pre}}$ are the fractions of receptors activated by gliotransmission and by presynaptic facilitation. Astrocyte calcium rises through IP$_3$ produced at each adjacent synapse, and when it crosses $Ca_{\\mathrm{th}}$ the astrocyte continuously releases gliotransmitter, lowering release probability when $\\varepsilon