IP3R Structural Simulator

An interactive 3-D model of the IP3 receptor — the endoplasmic reticulum's IP3- and Ca²⁺-gated Ca²⁺-release channel — driven by models with a citable basis rather than by animation, and an instrument that illustrates, demonstrates and independently re-derives the results of the ip3r_genes publication project. It is the IP3R counterpart of the PIEZO1 simulator, whose renderer, reader and conventions it reuses.

Human ITPR3 with IP3 bound (PDB 6DQN), cytosolic cap up. Blue/cyan/green: the IP3-binding β-trefoil, MIR and RIH domains; violet: the pore domain; gold spheres: the ten residues within 4.5 Å of IP3 at each of the four sites.

What it does

See the channel. Nine curated depositions — every ITPR reference and gating-state structure ip3r_genes S11 selected, plus 6DQN — drawn as cartoon, tube, spheres or sticks, coloured by functional element, subunit, secondary structure, B-factor or per-residue conservation (the S17 tables, four layers, on a fixed scale with unscored residues grey). Residue annotation is painted only when the deposit is verified to be in that paralog's human numbering (rat 7LHF is not, and is left grey). Beside them, under More IP3R deposits, 17 more: every other full-length EM tetramer of a human ITPR or rat ITPR1 in the PDB at ≤ 4 Å, wild type, chosen by stated rules (scripts/curate_ip3r.py, make extended). They include ITPR2's resting and preactivated states (9YKY, 9YLI), ITPR3's pre-active B/C, inactive and Ca²⁺ classes, and rat ITPR1 in a nanodisc and under Ca²⁺/IP3/ATP. They can be viewed, measured, superposed and morphed, but no panel or check uses them, so no publication number depends on them.

Work in it. A click selects a residue (gold); shift-click adds one. The right-click menu selects the same residue on every subunit, a chain, centres on an atom or starts a distance (Ctrl+M measures: two clicks, a rod labelled in Å). A HUD carries the deposit, a scale bar exact at the pivot and a gnomon with the four-fold axis's cytosolic end. An amber line shows whenever AlphaFold residues are drawn. View → Sequence shows each chain's whole construct in the deposit's numbering, unresolved residues dimmed. It is tracked by element, deep JSD or resolution, with the sites underlined, and a drag there selects on the model. The Analyses menu runs the command-line science (selectivity, protonation, the 3-D wall, the salt bridge, the image cost, charge–space, shortfall, …) in its own process, stamped with the deposit and the parameter set. Panels are movable docks (View → Reset layout). Full screen (F11, View → Full screen or the right-click menu; Esc leaves) keeps only the 3-D view and its HUD, and puts every panel back as it was. Help → Guide (F1) explains all of it.

Measure it. The four-fold axis is found by superposing each subunit on its neighbour; the pore profile, the selectivity filter (GGGVGD) and the gate are located; every bound IP3 is found and its contacts listed. python -m ip3r states measures the whole ITPR3 state panel the same way — and shows the gating transition at the pore:

PDB	state	gate radius (Å)
8TKH	labile resting	1.95
8TLA	higher-order inhibited	2.44
7T3P	preactivated	2.53
6DQN	IP3-bound	2.55
6DQJ	apo	2.69
8TKG	resting	2.73
8TKF	activated	5.85

states --extended adds the 11 more ITPR3 deposits: every one is shut (gate 1.57–2.59 Å; inactive 7T3U the narrowest), so 8TKF (and the 7T3T control) remain the only open IP3R pores in the PDB.

Move it. An elastic-network model of the tetramer gives its collective modes, each labelled by its C4 irreducible representation: A (all four subunits alike — the only kind that can couple to IP3 binding at all four sites and to a symmetric pore opening), B, or the degenerate E pair.

Watch it open. The Transition tab (and python -m ip3r transition) puts two states of one paralog on a residue-matched basis — 2,194 residues on each of the four subunits of resting 8TKG and activated 8TKF — superposes them on the pore domain, and morphs between them with peptide Cα–Cα distances restrained (an interpolation, labelled as one). Side chains are interpolated too, atom by atom, so the last frame is 8TKF itself and the gate plotted along the path (2.73 → 5.85 Å, half-way at 0.42) is the gate on screen; the rigid-side-chain shortcut, drawn dashed, ends 0.7 Å short. Each residue is coloured by how far it moves, on a fixed 0–25 Å scale: the cytosolic RIH and MIR domains move 17–22 Å on average, while the pore domain moves 2.5 Å and the filter 0.8 Å. The tab then asks whether the resting state's elastic network points towards the activated one. It does: the collective A modes together overlap the observed displacement at 0.67, where a random direction of the same symmetry scores 0.04 over all 20 modes. No single mode is the answer. The lowest A mode (#5) gives 0.39 and #10 gives 0.50, and how the move splits among them depends on the 15 Å cutoff (from 12 to 21.6 Å the lowest falls 0.48 → 0.24 while the A modes together hold at 0.66–0.68; transition --cutoff-scan). So the tab's headline plot is the A subspace: the collective A modes added lowest first, beside the same-symmetry null and the √(A share) ceiling. The displacement is 100 % A-symmetric, but that is inherited from C4-imposed reconstruction and is not a finding. The network keeps every second Cα: at every third, the strand after the unresolved 77–85 loop (residue 86) hangs on too few springs and produces five spurious local modes. Any low-collectivity mode that remains is named where it sits (the Modes tab and the report say so).

Gate it. The De Young–Keizer receptor as reduced by Li & Rinzel: the bell-shaped Ca²⁺ dependence of open probability, IP3 relieving Ca²⁺ inhibition, whole-cell Ca²⁺ oscillations (measured window 0.36–0.63 µM IP3), and stochastic clusters in which Ca²⁺ coupling turns independent blips into cooperative openings. The Gating panel also offers the Hill-type model Mak, McBride & Foskett (1998) fitted to single IP3R-1 channels, in which IP3 tunes Ca²⁺ inhibition alone. From 33 nM to 10 µM IP3 it moves the half-inhibition point 6.2× and half-activation 1.016×; De Young–Keizer moves them 2.8× and 2.0× (python -m ip3r gating --model mak). The park/drive receptor's stationary bell is there too (--model pd). It moves them 43× and 1.09×, so it passes the same test. "The three models side by side" draws all three bells at 33 nM and at 10 µM IP3.

The Puffs panel can fill the same cluster with park/drive receptors (Siekmann et al. 2012, with the gating variables of Cao et al. 2013; every constant read from the authors' code). Nearly all of these are parked at rest, so the cluster stays quiet until one receptor enters drive mode. Then its Ca²⁺ pulls the others in. Both receptors are measured with one ruler. Over 30 s at 0.2 µM IP3, the De Young–Keizer cluster reaches half its channels in at most 1 event at any coupling from 0 to 2 µM per open channel. The park/drive cluster does so 10–17 times at 0.09–0.32 µM (Fano 2.8, against ≤ 1.32), and its event sizes split into blips and puffs with a valley between (python -m ip3r puffs --scan).

python -m ip3r microdomain puts the park/drive cluster in Cao et al.'s microdomain: Ca²⁺ pools that fill and drain, fluo-4, and a store that can deplete. Puffs are then read from F/F0. As the h42 recovery rate rises from 0.1 to 5 s⁻¹, the puff rate rises 5.5× and the inter-puff-interval CV climbs from 0.79 toward 1, as in Cao 2013. With the release set so that the mean blip matches Cao's, fluorescence amplitude bends at about 12 receptors while Ca²⁺ bends less, because the dye saturates. The mean-field cluster's sustained 9 % open state survives the microdomain's kinetics, and depleting the store makes it more active, not less. That state belongs to the receptor model (docs/SCIENCE_PUFF_DOMAIN.md). In the Puffs panel, the park/drive receptor has a "Simulate in the microdomain" box. It draws F/F0 with the puffs marked, the number open, and the inter-puff intervals against Thurley's refractory density and an exponential with the same mean.

The Channel tab turns each ITPR3 deposit's pore into a K+ conductance by drift-diffusion (ported from PIEZO1), with and without the charges of the side chains that line it. Only activated 8TKF conducts. It gives 65 pS uncharged (25–150 pS over the unmeasured diffusivity and ion radius) against 358–545 pS measured, so the continuum model falls 2.4× short even at its most generous. Its own lining charges lower the conductance rather than raising it. Cancelling the salt-bridged ones (the filter's D2478 is paired with R2471 of the next subunit) lowers it further, to 23 pS (python -m ip3r unitary).

Most of that 2.4× is the model's geometry (python -m ip3r shortfall). The 1-D model reduces each slice to its inscribed circle. The same electrolyte in the voxelised lumen, solved in 3-D, conducts 1.3–2.0× more, because the circle leaves out the lumen's corners. It is not a substate: Schmitz et al. 2022's independent active-state deposit 7T3T reads the same as 8TKF (85 against 106 pS). Nor is it the exit window, which moves the answer by 2 %. At bulk diffusivity and a 1 Å K+ exclusion, RyR1's open deposit gives 787 pS against 801 measured, and ITPR3 gives 261–278 pS against 358–545.

The corners change how much current flows, but not where the voltage falls (python -m ip3r lumen; Channel panel → "Draw the lumen"). In 8TKF, 7T3T and RyR1's 9HEO, the 3-D potential puts the half-drop point within 1.3 Å of the 1-D model's, and the filter's share within 4 points (8TKF: 32 % against 35 %). The viewer draws the lumen coloured by that potential, and the panel plots both curves.

The lining charges' effect does not survive the move to 3-D (python -m ip3r wall3d). Solved in the lumen's real shape (the ions' equilibrium, then one Laplace solve per species), 8TKF's full wall changes g by ×0.80, ×1.24 or ×3.1, depending on whether the charge fills the cross-section, sits at each group's own centre, or is screened by Poisson–Boltzmann. The 1-D model's ×0.46 holds only in a cylinder, and one unmeasured smoothing width moves the 3-D reading 0.4–5.7×. With the filter's D2478 salt bridges paired, the wall lowers g under every closure at the registered width (×0.45–0.59). RyR1's charge mutants cannot tell the closures apart: all of them miss D4899Q (×0.20 measured) by 4×.

The viewer draws that charge on the lumen (Channel panel → "Wall charge", "Colour by"): the equilibrium wall potential on a fixed ±5 kT/e scale, or where K+'s drop falls through the charged pore. The charge moves the drop, not only its size. 8TKF's filter holds 32 % of the neutral drop, but only 1–18 % with its full wall, because the D2478 ring is a cation well that carries almost none of it. With D2478's salt bridges paired the filter takes 23–38 % again. So where the voltage falls in the filter depends on the same open question as the conductance: whether D2478 is charged.

That question is now answered (python -m ip3r bridge, docs/SCIENCE_BRIDGE.md). D2478 is ionised by both pKa routes (network pKa 2.0–2.2 with its R2471′ partner counted, 4.0–4.8 without it; PROPKA 5.2), and the arginine never titrates. So the pair is two charges, a radial dipole behind the filter wall, and neither the full wall nor the paired one is a physical state. A new closure solves Poisson–Boltzmann over the whole box with the protein at ε 4 and every charged group at its own centre. It puts 8TKF's wall at ×3.18: between the partner omitted (×5.30) and the pair omitted (×1.79), and raising g in every reading once the rest of the protein's charge is counted (7T3T ×3.47). The protein's permittivity (2–20) moves this < 3 %. RyR1's D4899Q is still missed (×0.79 against ×0.20), so no placement of point charges explains it.

The lumen box offers that closure too ("dielectric"; python -m ip3r lumen 8TKF --charge dielectric [--paired]). Unticked, "salt bridges paired" is the dipole; ticked, the pair omitted. It takes two or three minutes. In 8TKF the dipole gives the same K+ conductance as the full wall under PB (×5.07), but not the same profile. The full wall moves the steepest drop 25 Å cytosolic and leaves the filter 18 %. The dipole keeps it at the filter, with 26 % there (neutral 32 %, pair omitted 41 %). A conductance cannot tell these readings apart; the drawn drop can (docs/SCIENCE_BRIDGE.md §4).

The one electrostatic term every closure left out was the image force: an ion near a wall of lower permittivity than the water is pushed away by its own polarisation (python -m ip3r born, docs/SCIENCE_BORN.md). Its cost, solved voxel by voxel on the deposit, is about 1 kT on the axis at the filter (8TKF 1.19, 7T3T 1.54, 9HEO 0.77; four times that for Ca²⁺). It cuts a neutral pore's K+ conductance to ×0.20–0.28. It costs the charged ITPR3 pore nothing (the dipole ×3.38 → ×3.42 in 8TKF): where the wall charge dominates, neutrality pins the counter-ion density and the potential pays the cost. RyR1's wall does not pin it everywhere, so 9HEO falls ×3.86 → ×2.18. It moves D4899Q from ×0.79 to ×0.48, but it moves D4938N further (×0.74 → ×0.28 against ×0.65 measured), so the mutants' summed error is unchanged. The image cost is not what singles out D4899. Ion size at the crowded filter is the candidate left.

That candidate is charge–space competition: ions with size, in a crowded, charged fluid that screens a divalent best (hard spheres + the mean spherical approximation, as in Nonner 2000 and Gillespie 2008; python -m ip3r csc, docs/SCIENCE_CSC.md). RyR1's filter now binds Ca²⁺ as Gillespie's model does: at 150 mM K⁺ and 1 mM Ca²⁺ it holds 10 M Ca²⁺ against 0.9 M K⁺, with a screening advantage of 4.2 kT and an excluded-volume advantage of 0.9 kT (his ~4 and ~0.5–1). But P_Ca:P_K only rises from 0.46 to 0.64, against 7.0. The uncharged gate stretch is in series with the filter and holds 72 % of Ca²⁺'s resistance, and on its own it reads 0.54 by hand (the diffusivity × area ratio). D4899Q stays mild (×0.85 against ×0.14). What the model lacks is the field at the gate, not the filter's physics.

In 3-D that field is there (python -m ip3r sel3d, docs/SCIENCE_SEL3D.md). P_Ca:P_K is read from the charged lumen's linear response, P = D × a Boltzmann-weighted Laplace conductance, with the csc fluid in every voxel. Under Poisson the wall's field reaches the gate, and the filter still binds Ca²⁺ (8.6 M against 1.2 M K⁺ in 9HEO). But the ratio reaches only 1.08 in RyR1 (7.0 measured; 1-D 0.87), and 1.39 / 1.79 in ITPR3's 8TKF / 7T3T (15.2 measured; 1-D 0.34 / 0.63). 9HEO's gate still holds over half of Ca²⁺'s resistance. The one constant that moves it is how far each charge is spread toward the gate (smoothing 6 Å: 2.7). Xu's mutants still do not single out D4899Q (×0.85 under pb + csc against ×0.14; D4938N ×0.83 against ×0.47).

The lumen box counts it too (dielectric → "+ image"; python -m ip3r lumen 8TKF --charge dielectric --image), on a 1 Å grid, from the cache in about 20 s. The surface can be coloured by W itself. With W, a cation's well is u + W. The potential u alone deepens to −16 kT/e at the wall, but K⁺'s well gets shallower (8TKF −6.2 → −5.1 kT). The image moves ITPR3's K⁺ drop back to the filter: 8TKF's dipole leaves it 29 % (24 % without W), near the neutral pore's 30 %. RyR1 9HEO's gate carries the drop with or without it (docs/SCIENCE_BORN.md §4).

"Colour by → K+ energy" paints u + W, the energy a cation actually feels, on the wall potential's fixed ± scale (u alone without the image). That is the well the summary names.

Selectivity is a ratio, so it does not depend on the unmeasured diffusivity and tests the wall charge directly. Vais et al. 2010's own solutions were run through the same pore (python -m ip3r selectivity). No reading of 8TKF's wall comes near the measured P_Ca:P_K of 15.2: 0.17 uncharged, 0.00 with its lining charges, 0.69 with only the acidic rings. The model's i_Ca is at most 0.046 against 0.30 pA/mM. Under a local Donnan partition, the lining lysines act as Ca²⁺ barriers. The calibration shows that charge in discrete rings cannot make a continuum pore Ca²⁺-selective; a charged tract can (docs/SCIENCE_PERM.md). Solving Poisson–Boltzmann across each slice instead of assuming local Donnan (selectivity --closure radial) screens the vestibule's K2529 ring, but lifts the charged reading only to 0.04 (0.01–0.16 over ε 80–10), so the closure is not the Ca²⁺ barrier.

Nor is protonation (python -m ip3r protonation [8TKF|9HEO] [--corners]). Two independent pKa routes, a Tanford–Kirkwood network titrated by Monte Carlo and PROPKA 3, both keep every lining group of 8TKF charged at Vais's pH 7.3. The lysine rings stay charged even at a protein-like permittivity of 4, where the acid rings lose up to half their charge. No combination of rings formal or neutral exceeds P_Ca:P_K 0.69, and with the lysines charged none exceeds 0.05. RyR1's open deposit, run through Xu et al. 2006's own protocol as the control, fails the same way: 0.46 against 7.0 under every pKa reading. The model also misses the order of Xu's charge mutants. The missing piece is therefore the continuum model, not the IP3R wall. Xu's E4900N also shows that PROPKA wrongly buries E4900 (pKa 8).

The gate is not the cap (Round 7.21). python -m ip3r gate widens a deposit's gate by up to 4 Å (a radial, C4-preserving move, calibrated on synthetic atoms) and reads the pore again. On 9HEO the gate's share of Ca²⁺'s resistance falls from 53 % to 6 %, yet P_Ca:P_K moves only from 1.08 to 1.27 against 7.0. On 8TKF it moves from 1.39 to 1.50 against 15.2, and on 7T3T it falls. K⁺ conductance gains at most 24 %. Neither the conductance shortfall nor the selectivity gap is the gate's shape (docs/SCIENCE_GATE.md).

Nor is the protocol (Round 7.23). python -m ip3r reversal runs Xu's and Vais's bi-ionic experiments through the 3-D charged lumen: steady Poisson–Nernst–Planck on the voxels (Gummel), the reversal found as the root of the net current. It is calibrated on Planck's junction, Teorell–Meyer–Sievers and an excluded NMDG⁺'s Donnan jump. At reversal, P_Ca:P_K lies within 15 % of the linear-response reading: 9HEO 0.90 against 7.0, 8TKF 1.15 and 7T3T 1.54 against 15.2. Meanwhile the charged wall shuts out Cl⁻ (P_Cl:P_K 0.003 against Vais's 0.27). Scaling 8TKF's wall charge from 0 to 2× never gives both measured ratios: P_Cl:P_K halves by a twentieth of the charge, and P_Ca:P_K never reaches 1 (docs/SCIENCE_REVERSAL.md).

The viewer shows that state (Round 7.24; lumen box → Steady state → "at reversal", or python -m ip3r reversal 8TKF --lumen). The lumen is coloured by each ion's concentration or electrochemical drop at the experiment's reversal, Cl⁻ included. Under pb + csc the wall gathers Ca²⁺ to 6–8 M just luminal of the filter. That well carries only 3–7 % of Ca²⁺'s drop. Its resistance lies in the uncharged stretch cytosolic of it: 60 % at 9HEO's gate, 31 % at 8TKF's.

No mean-field wall gives both ratios (Round 7.25). python -m ip3r wallsearch scores wall models on P_Cl:P_K and P_Ca:P_K together. Take each ratio over the uncharged pore's. For point ions in series, Hölder's inequality then gives (P_Cl:P_K)(P_Ca:P_K)² ≤ 1 for any potential profile. Vais's pair needs about 3,200. The two exits open to a charged wall are small. The reversal's non-linearity reaches 1.2. Opposite-charge ring pairs reach 1.01 in 8TKF's lumen and 7.9 only in 7T3T's wide gate, and that gain is in Cl⁻. A Ca²⁺-only well leaves Cl⁻ untouched, but it peaks at P_Ca:P_K 4.5 (8TKF) / 6.9 (7T3T) at reversal, spanning the whole membrane at 4 kT. Deeper wells fill with Ca²⁺ and repel. The measured pair needs an interaction this continuum does not carry, such as Ca²⁺ occupancy blocking K⁺ (docs/SCIENCE_WALLSEARCH.md).

The candidates can be drawn (Round 7.26). In the lumen box, Steady state → "at reversal, candidate wall" offers the span well, that well with the deposit's charge, and the best ring pair. Each is coloured by each ion's concentration at reversal. The deposit's own reading, if drawn first, is plotted dashed beside it. Inside the well, Ca²⁺'s resistance is at the well's edge, where it has to climb out. 7T3T's ring pair keeps some of its parallel-path gain at reversal (B ≈ 4.6), but that gain is Cl⁻'s.

A Ca²⁺ site that blocks K⁺ reaches the pair (Round 7.27). python -m ip3r casite adds a saturable Ca²⁺ site over the span, four sites deep d. It can be compensated (−2e fixed per bound Ca²⁺) and can let its occupancy block K⁺. Read at Vais's reversal, affinity alone falls short. Uncompensated sites peak near 5 / 7 (8TKF / 7T3T), and compensated ones plateau at 9.1 / 13.3 when full. Compensated and blocking, the site crosses 15.2 at d = 4.4 / 4.0 kT (K_d 3–6 mM, 8TKF's half occupied), with P_Cl:P_K still the uncharged pore's 0.29 / 0.37. The measured pair needs Ca²⁺ occupancy that blocks K⁺, the anomalous-mole-fraction mechanism (docs/SCIENCE_CASITE.md).

The site can be drawn (Round 7.28). The last candidate wall in the lumen box is the compensated, blocking site at 8TKF's crossing depth, coloured by its occupancy θ or by K⁺'s block energy. On 8TKF it holds 2.0 Ca²⁺ (θ up to 0.89). K⁺'s drop is steepest inside the occupied band, and Ca²⁺'s at the band's cytosolic edge, where it leaves the site.

Ryanodine receptors. Rabbit RyR1 loads beside the IP3Rs: six deposits chosen from the PDB by stated rules (scripts/curate_ryr.py: full-length, wild type, activators only, ≤ 4 Å, best per stated state), one each closed, primed, open, inactivated and closed-inactivated, plus a primed deposit from the open state's own paper for the morph (9R8O → 9HEO). They go through the same measurement: every one is in P11716 numbering, the shut states gate at I4937, and only open 9HEO widens (5.05 Å). Its modelled K+ conductance is 136 pS neutral and 180 pS charged, against 801 pS measured (Xu et al. 2006), so the continuum model is short on both receptors. RyR1 also allows a stronger test than one number. Five charge-neutralising mutants were measured, and the model makes each one on the structure (python -m ip3r mutants). It gets the direction right for all four lining residues and no effect for E4955Q, which does not line the pore. It misses the largest effect: D4899Q cuts the conductance to 0.20× but the model predicts 0.90×. And D4899, like ITPR3's homologous D2478, is salt-bridged, so cancelling ion pairs predicts no effect at all. The bridge does not neutralise that charge.

RyR1 gating and sparks. Dynamics → Gating → "RyR1" draws Stern et al. 1997's two-gate scheme against Murayama et al. 2015's measured rabbit-RyR1 bell. The scheme activates where RyR1 does (3.9 vs 4.4 µM) but inactivates 6.7× too readily (48 vs 320 µM). Puffs → "RyR1 sparks" runs a 30-channel cluster with the Ca²⁺ coupling derived from one channel's current at 30 nm. Uncoupled it gives only blips. Coupled it gives 1.5 sparks per second that recruit nearly the whole cluster, read with the same ruler as IP3R puffs. With a single cluster Ca²⁺ the sparks last ~120 ms, against ~6 ms measured. "RyR1 sparks in the cleft" puts the same channels in Stern's junctional cleft (two rows, 60 × 15 nm, edges leaking), where each channel sees its own Ca²⁺ from a steady diffusion solve. Sparks there end by local inactivation after ~20 ms: six times shorter, still 3× the measurement, and no geometric uncertainty closes the gap. The Gating tab also draws the scheme fitted to both measured flanks (dashed: Ka 4.9, Ki 249 µM), and "RyR1 in the cleft, gating fitted to Murayama 2015" runs it. Once a spark starts, it never ends. With inactivation made as weak as the measured bell says, the tens of µM in the cleft cannot shut the array at any inactivation rate. So what ends a real spark is missing from a Ca²⁺-only scheme. Adding the fibre's 1 mM free Mg²⁺ (spark-mg; in the GUI, Puffs → Mg²⁺ and "Triggered sparks vs Mg²⁺", and the dash-dot bell in Gating), from open sources, supplies it. Competing at the activation site, Mg²⁺ shuts triggered sparks with no channel inactivated: the array's feedback falls below one. How fast depends on the site's Mg²⁺ affinity. Meissner et al. 1997 measured it in the assay behind the fitted bell. Carried into that assay's 0.17 M NaCl, where Na⁺ already holds the same site, it gives K_Mg,A ≈ 770 µM (range 345–1,600 µM), and the activation site alone ends a triggered spark in 32 ms at best (435 ms at the central value), against 6.3 ms measured. With Mg²⁺ also at the inhibitory site, which holds ~80 % of channels shut before any trigger, sparks end in 6 ms under every reading. The V channels, the half of the couplon opposite the voltage sensors, are now simulated (ec; Rios 1993's allosteric model, Stern's rates). With Stern's constants the couplon reproduces his release under voltage clamp: a peak, a plateau, and a stop when the membrane repolarises. With the C channels fitted to the measured bell, no Mg²⁺ arrangement gives a peak. Either release goes on after repolarisation, or the C channels hardly open. Emptying the SR (ec --depletion) restores a peak only by releasing more of the store than a fibre loses. Nor does a two-site inactivation gate fitted to the bell's slope (ec --two-site) help. It inactivates less at cleft Ca²⁺ than the one-site gate, and control is lost further.

Check the publication. The Findings tab re-derives 45 results from the six ip3r_genes papers and its structural baseline, by three routes of different strength: recomputed from coordinates with code the two projects do not share, rederived from the publication's input tables with this project's arithmetic, and read (the table read, the prose tested).

Residue-keyed findings can be drawn on any structure in human numbering. Paper 6's two modules, for example, are shown as Cα traces: the ligand core in green and the pore module less the luminal loop in magenta.

Colour by distance to IP3. "Distance to IP3 (S22 shells)" paints every residue by its all-atom distance to the IP3 bound on its own subunit, in S22's four shells (contact < 4.5 Å, then 8, 11.5 and 15 Å). Residues beyond 15 Å, and subunits with no IP3, are grey. The shell checks' exhibit plots conservation against that distance for all three paralogs.

Read the tree. The Tree tab draws Paper 2's RyR-rooted maximum-likelihood tree (134 proteins) from the committed rooted.nwk, parsed by this project's reader. Each paralog's whole clade is boxed with its size and SH-aLRT/UFBoot support, and so is the RyR outgroup. Hagfish and lamprey tips are yellow diamonds, labelled with the support of their clade and of the node where it joins. A white dot marks each node that clears both support bars (80/95). "Vertebrates" zooms to the 57 vertebrate tips with their labels. "Show" on any tree check opens this tab. "Beside --bnni" draws the model-violation re-search next to it, both rooted on RyR, and lists what the guard did to each clade claim: 9 of 10 held, and the ITPR1 core (never well supported) fell from 47.8/95 to 47.5/73.

See every genome. The Genomes tab draws Papers 3 and 4 as a grid: a row for each of the 309 assemblies in the retention sweep and a column for each cell (ITPR1–3 and the RyR control). Each row sits beside a contig-N50 strip on a fixed log scale. You can colour the cells by what the sweep found, by the known genes it missed, by the S15a evidence state, by how a protein search could reach the gene, or by lesions: whether the cell carries more frameshifts and stops than its own genome's identity-matched siblings (S15b §8, rebuilt from the per-locus table). Rows sort by N50 (the contiguity bar is drawn), by class, or by name, and can be filtered to a class or to assemblies above the bar. Clicking a row names the genome and lists its four cells. "Show" on a P3 or P4 check opens this tab on the right layer. P3.lesion_strata opens it on the birds, where ITPR3's excess (25 genomes to 2) sits mostly below the bar.

See where the receptor is. The Range tab draws Paper 1 as one bar per clade of the 6,928-proteome sweep. Each bar shows the fraction of swept proteomes carrying an IP3 receptor call, on a fixed 0–1 scale, coloured by supergroup; archaea and bacteria are collapsed to one row each. A red cross marks a clade whose absence held in controlled genome assemblies (a small one where it held for a class inside the clade). You can hide small clades or expand the prokaryotes. Clicking a row lists its genome-level absences. Double-clicking a clade (or View → "Genomes (S23)") draws its S23 genome assemblies one by one: each genome's control verdict, what the copy ledger found, whether its contig N50 reaches its own contiguity bar, and its complete gene models on a fixed 0–20 scale. "Show" on any P1 check opens this tab.

See the variants, and where the uncertain ones sit. The Variants tab lists the S17 harvest for one paralog and class. "Draw on structure" puts a sphere on every variant residue of that class on all visible subunits: P/LP red, B/LB blue, conflicting violet, VUS amber. It draws only on a deposit in that paralog's human numbering; on any other it says why and draws nothing. Choose a layer under "VUS by layer" and each VUS is placed against its own gene's labelled medians on that layer, as in Paper 5 §8. A VUS at or above the P/LP median is pathogenic-like (pale red), one at or below the B/LB median is benign-like (pale blue), and an unscored one is grey. The table gains a stratum column and the plot shows the three classes with both medians. This is a stratification, not a call.

Each median is itself uncertain, so the plot shades its exact 95 % interval (a distribution-free interval from order statistics). A VUS that a threshold inside an interval could move is drawn hollow and marked "near the P/LP (or B/LB) median" in the table. A class with five positions or fewer cannot bound its median: ITPR2's P/LP median is one position and ITPR3's is five. The panel says so, and no VUS of theirs counts as firmly pathogenic-like.

See what the map leaves out, as a prediction. Representation → Completeness adds the AlphaFold model's residues where a deposit has none. "+ AlphaFold gaps" fills the internal stretches, each fitted on resolved residues on both sides. "+ gaps and ends" also adds the termini, which have one side to fit on. The fill is drawn in AlphaFold's own pLDDT colours whatever the deposit is coloured by. Each seam is a bond from the deposit to the fill: pale where it closes, red where it does not. The fill follows a morph or mode frame. It is only drawn: every measurement in the application still runs on the deposit. The model is chosen by measuring which downloaded prediction is in the deposit's numbering, or failing that, which one is the same protein through an alignment of the deposit's construct. So rat 7LHF is filled from rat ITPR1 isoform 8. The two splice segments the isoform lacks are left empty, and the panel says so. 9YKK (ITPR2) has no model and is refused, and the panel shows why. On 8TKG, 1,592 residues are filled over the four subunits, at a mean pLDDT of 38. AlphaFold is least sure exactly where the map is empty. Filling 16 stretches that 8TKG does resolve (but another deposit does not) lands at a median 1.1 Å, against 5.5 Å for a straight line (python -m ip3r graft 8TKG --calibrate). Length does not break a fill, but low confidence does (python -m ip3r graft 8TKH --long). Hidden 60-residue windows fill to 1.5 Å where AlphaFold is confident. The one resolved stretch below pLDDT 50, 8TKH 926–943, was placed 58 Å off, further than a straight line. So the summary counts the residues below pLDDT 50 (1,308 of 8TKG's 1,592) and says they are not positions.

Compare two states by eye. Representation → Superpose draws another deposit of the same paralog on the one shown, in orange. It uses the Transition tab's residue-matched fit (on the pore domain, or global), with no morph built. Every atom of the second deposit is moved by that fit's transform, so side chains and IP3 are its own. Its subunits are relabelled to the ones they were matched to, so the Subunits toggles hide the pair together. 8TKF on 8TKG: 3.14 Å over the pore, 16.11 Å overall, the cytosolic cap swinging while the membrane domain stays put. A session keeps the choice and refits it on restore.

The checks, as of Round 7.7

52 checks: 50 confirmed, 2 discrepancies. Both discrepancies are genuine, and neither touches a published paper's headline:

• P6.contacts_heavy_atom. S22's positive control — S0's ten IP3 contacts recovered in all six IP3-bound depositions — is reproduced exactly with S22's rule, which counts hydrogens. Under S0's own heavy-atom definition, Arg503 is 4.78 Å (8TKG) and 4.83 Å (8TKH) from IP3, so the control holds in 4 of 6 depositions, and in the other two only through a hydrogen.
• P5.report_both_metrics. The S17 report says the gate and filter are the most constrained elements "on both metrics and in all three paralogs". On the JSD the ITPR1 top two are RIH-associated and gate, and the ITPR2 top two are gate and the β-trefoil. The paper's own Results state the narrower, correct version, which P5.gate_filter_most_conserved confirms.

The table of every check, its kind and what it re-derived is in docs/SCIENCE_CHECKS.md.

Every check is calibrated: tests/test_checks_calibration.py runs each one on a copy of only the tables it declares, then plants a change and requires the verdict to flip.

Installing

bash scripts/create_env.sh          # or: conda create -n ip3r_sim --clone piezo1
conda activate ip3r_sim
make fetch                          # the registry's 16 mmCIF files, ~46 MB, into ref/

Clone ip3r_genes beside this repository (or set IP3R_GENES_DIR) for the findings checks; everything else runs without it.

Running

./run_app.command                   # the GUI (activates ip3r_sim; double-click in Finder)
python -m ip3r                      # the GUI, from an activated environment
python -m ip3r --session view.json  # the GUI, reopened on a saved session
python -m ip3r checks [--paper constraint] [--figures out/]
python -m ip3r states               # the ITPR3 gating states at the pore
python -m ip3r states --extended    # ... and every other full-length ITPR3 deposit
python -m ip3r unitary              # their K+ conductance, vs 358/545 pS
python -m ip3r shortfall [--scan]   # every open deposit in 1-D and 3-D vs the measurement
python -m ip3r lumen [8TKF ...] [--charge dielectric [--paired] [--image]]  # where the voltage falls: 3-D vs 1-D, neutral or charged
python -m ip3r wall3d [--scan] [--mutants]  # the lining charges in 3-D: three closures, RyR1 mutants
python -m ip3r bridge [PDB] [--scan] [--mutants]  # the lining salt bridge: pKas, then its field with the protein in it
python -m ip3r born [PDB] [--scan] [--mutants]    # the image cost of the low-eps wall on K+ g and Xu's mutants
python -m ip3r csc [PDB] [--scan]                 # charge-space competition: filter binding, Xu's six P_Ca:P_K, Vais's ratios
python -m ip3r reversal PDB --lumen [READING ...] [--experiment Ca2+|Cl-]  # each ion on the lumen at reversal
python -m ip3r sel3d [PDB] [--mutants]            # P_Ca:P_K from the 3-D charged lumen, point ions and with the csc fluid
python -m ip3r selectivity          # 8TKF's P_Cl:P_K, P_Ca:P_K, i_Ca vs Vais 2010
python -m ip3r protonation [9HEO] [--corners]  # lining pKas (network, PROPKA) and selectivity under each
python -m ip3r states --paralog RYR1   # the curated RyR1 states (also unitary)
python -m ip3r mutants              # RyR1 charge mutants: model vs Xu 2006
python -m ip3r transition 9R8O 9HEO # RyR1 primed -> open
python -m ip3r ryr-gating | sparks --scan   # RyR1 bells; sparks over the coupling band
python -m ip3r sparks --cleft      # sparks with each channel's own Ca2+ in the cleft
python -m ip3r sparks --cleft --fit  # the same with Ka, Ki fitted to Murayama's bell
python -m ip3r spark-termination --scan fit|ki|rate|use|ratio|fraction|low-activity|speed  # what ends a cleft spark
python -m ip3r spark-mg [--scan] [--reading measured|selectivity|meissner]  # Mg2+ ends a triggered spark
python -m ip3r ec [--scan] [--reading R] [--use]  # the couplon under voltage clamp (V + C channels)
python -m ip3r ec --two-site       # ... with a two-site inactivation gate fitted to the bell's slope
python -m ip3r ec --depletion [--pool-scan]  # ... with the SR emptying (Stern's Fig. 20 pool)
python -m ip3r info 8TKF            # one deposit, measured
python -m ip3r modes 6DQN           # normal modes with C4 irreps
python -m ip3r transition 8TKG 8TKF # morph, displacement, mode overlap (--gate: pore per frame;
                                    #  --cutoff-scan, --stride-check: the network against itself)
python -m ip3r graft 8TKG --calibrate   # AlphaFold fills, seams, and how good they are
python -m ip3r graft 8TKH --long        # ... on long windows and islands; 7LHF by alignment
python -m ip3r gating | oscillate --window | puffs --ip3 0.2
python -m ip3r puffs --model park-drive | puffs --scan   # the two receptors
make help

Mouse: left-drag rotate, shift-drag pan, wheel zoom, click to identify a residue (element and conservation shown). Ctrl+1 side view, Ctrl+2 down the pore, Ctrl+3 one IP3 site (its 15 Å pocket, the tetramer in front of it clipped away), Ctrl+0 fit to view, Space spin. Until you move the camera, the view keeps the molecule filling the viewport as the window or the visible subunits change. "Show on structure" for the IP3-contact and ligand-shell checks opens the site view. The deposition list is grouped by family, with RyR1 collapsed.

Changing a parameter. Help → Parameters… (Ctrl+Shift+P) lists every registered number with its default, bounds and source; double-click a value to change it (out-of-range values are clamped, and the clamp is reported). While anything differs from its default an amber banner runs across the top of the window, and the findings checks report not run rather than confirm. Nothing typed is remembered after you quit. "Export…" writes the set as JSON, which IP3R_PARAMETERS=file.json python -m ip3r … reproduces headless. A control whose starting value is a parameter (the Puffs cluster size and coupling, the microdomain run length) follows an edit unless you have typed your own value there. Drawings nobody asked for (the Gating plot, the displacement and ligand-shell colours and their legend) are redrawn. Results you ran (a puff simulation, a channel measurement, modes, a transition) keep the values they were computed with until you run them again.

Saving where you were. File → Save session… (Ctrl+Shift+S) writes the view as JSON: the deposit, style, colouring, layer, subunits, marked sites, pore, camera, open tab, any transition built (end, fit, method, frame), the normal mode animating and its amplitude, the Dynamics settings (gating model, oscillation, puffs, microdomain) and the Variants view (paralog, class, layer, drawn). A mode animation is recomputed and restarted; a simulation is not re-run, only its settings come back. File → Open session… (Ctrl+O) or --session puts it back. A session holds no coordinates and no results: it is re-derived from the same inputs on opening. It does record the parameter overrides in force when it was saved. If they differ from the current set, you are asked whether to apply them (the deposit is then re-measured) or keep your own.

How it is built, and how it checks itself

See INTERFACE.md for the module map and docs/SCIENCE.md for the models (the ryanodine receptor in docs/SCIENCE_RYR.md, and the couplon in docs/SCIENCE_EC.md). Every number a calculation uses is a registered parameter with a unit, bounds and a source (python -m ip3r params); curated data imported from ip3r_genes records the SHA-256 of its source tables (make sync-check). Files stay under 500 lines. What is next is in ROADMAP.md.

The GUI is tested by driving the real application (make screenshots), which also writes the screenshots here. Its steps are in named groups, so a change can be checked by the groups it touches: make screenshots STEPS=lumen,extras, and make screenshot-groups lists them. make screenshots-full also runs every findings check in the GUI first.

References

De Young & Keizer 1992 (PNAS 89:9895); Li & Rinzel 1994 (J Theor Biol 166:461); Bezprozvanny, Watras & Ehrlich 1991 (Nature 351:751); Mak, McBride & Foskett 1998 (PNAS 95:15821); Swillens et al. 1999 (PNAS 96:13750); Shuai & Jung 2002 (Biophys J 83:87); Siekmann et al. 2012 (Biophys J 103:658); Cao et al. 2013 (Biophys J 105:1133); Cao et al. 2014 (PLoS Comput Biol 10:e1003783); Smith & Parker 2009 (PNAS 106:6404); Paknejad & Hite 2018 (NSMB 25:660); Atilgan et al. 2001 (Biophys J 80:505); Yang, Song & Jernigan 2009 (PNAS 106:12347); Tama & Sanejouand 2001 (Protein Eng 14:1); Brüschweiler 1995 (J Chem Phys 102:3396); Kabsch 1976 (Acta Cryst A32:922); Hanley & McNeil 1982 (Radiology 143:29); Barlow & Thornton 1983 (J Mol Biol 168:867); Xu et al. 2006 (Biophys J 90:443); Stern, Pizarro & Ríos 1997 (J Gen Physiol 110:415); Murayama et al. 2015 (PLoS One 10:e0130606); Ríos et al. 1999 (J Gen Physiol 114:31). Full entries: ip3r/resources/references.json.