TRN: deep kinetic retune to remove cold-start depol-block on rising LTS edge

[JCorson]

Background

Follow-up to #347 (closed as won't-fix). #347 documented a transient depol-block-recovery sequence in the first ~15 ms of REPETITIVE_FIRING (+3 µA/cm² × 200 ms from a fresh v_rest = −80 mV cold start): one full Na⁺ spike at +46.6 mV, then 5–8 suppressed/aborted spikes (V max −7…+7 mV) while Na inactivation recovers, then resumed firing.

The pattern is dt-invariant (verified at 40/80/160 kHz) and is therefore biophysical, not a numerical artifact. It is also phase-dependent on where in the ~10 Hz spontaneous limit cycle the step lands — with pre_stimulus_duration = 50 ms (vs the current 10 ms), the block disappears.

Why a deeper retune is required

The #347 sweep ruled out every minimal-change candidate:

• g_Na trim (Candidate A) — makes the block worse.
• g_T trim (Candidate B) — clears the block at ≤2.4 mS/cm² but breaks the rebound burst and raises spontaneous AP peak above +40 mV.
• sNa slow Na inactivation (Candidate C, the fix(channels): opt-in slow Na inactivation for INaP (STN depol block) #325/Cortical Pyramidal: enable slow Na inactivation (INaP sNaP, fast-Na sNa) for biological accuracy #327/CA1 Pyramidal: enable slow Na inactivation (INaP sNaP, fast-Na sNa) for biological accuracy #328/Purkinje: enable slow Na inactivation (INaP sNaP, fast-Na sNa) for biological accuracy #329/Dopaminergic SNc: add slow Na inactivation to make_snc_inap_channel and fast Na #330 precedent) — doesn't engage on the 15 ms transient timescale.
• g_KCa bump (Candidate D) — Ca needs accumulation time.
• 2D (g_T × g_K) — no combination clears all 5 gates.
• ICaT ft V½ shift — at ≤−85 mV the cell stops pacing entirely.

So a clean fix needs more than a one-line parameter change. The constraint set (rebound-burst preserved, in-band tonic AP shape, in-band Ca peak, 1–15 Hz pacing, and no LTS-rise depol-block) over-determines the (g_Na, g_K, g_T, g_KCa, g_h, g_NaL, g_KL) space.

Scope

Comparable to #308/#346: a coordinated kinetic-and-conductance retune. Possible levers identified during #347:

1. Faster Na h-gate recovery — shift TRN h V½ or τ_h to let Na recover from the first big spike before the LTS pulls Vm too high to escape.
2. Add an Im (Kv7) current — TRN literature (Cueni 2008, Bessaïh papers, Crandall/Avanzini) supports a slow K⁺ current that contributes to inter-spike repolarisation. Would help break the +5 mV plateau.
3. Reshape ICaT inactivation — sigmoid τ_ft (issue TRN: implement TRN-specific ICaT factory with sigmoid-shaped tau for HP92 rebound burst (follow-up to #286) #295) holds the LTS plateau open ~200 ms; a less-flat plateau may resolve the block while preserving the burst.
4. Combined: small g_Na trim + small g_T trim + Im + h V½ shift, jointly tuned.

Each lever needs literature support; this is not a free-tuning exercise.

Acceptance criteria (carried from #347)

1. No depol-block — every Vm peak in the first 25 ms post-step-onset of REPETITIVE_FIRING (+3 µA/cm² × 200 ms from cold start at v_rest) reaches ≥ +10 mV (HP92 floor).
2. Rebound burst — 5–15 spikes, 200–600 Hz, exactly 1 burst at −4 µA/cm² × 500 ms (test_trn_step_release_produces_hp92_rebound_burst).
3. Spontaneous tonic shape — peak +10/+40 mV, AHP −75/−55 mV, half-width 0.4–1.2 ms, threshold −65/−40 mV.
4. Spontaneous tonic rate — 1–15 Hz.
5. Calcium peak — 5–12 µM under +3 µA/cm² × 200 ms (test_strong_stim_peak_ca_in_band).
6. New regression test in tests/integration/test_trn.py pinning gate (1).

Diagnostic infrastructure

#347 left four diagnostic scripts under scratch/ (gitignored) that exercise these gates against a TRN variant builder. Reusable for the deeper retune:

• trn_lts_rise_dt_sweep.py — dt invariance check
• trn_lts_depol_block_sweep.py — single-axis A/B/C/D
• trn_lts_2d_sweep.py — 2D (g_T × g_K)
• trn_lts_ft_vhalf_sweep.py — ICaT ft V½ shift
• trn_lts_validate_winner.py — full 5-gate evaluation

Priority

Low — visible only on cold start before pacing equilibrates; macroscopic burst-then-tonic phenotype is correct. UX impact is a 15 ms transient on the user's first run from preset defaults.

References

• Closed predecessor: TRN: transient depol-block recovery on rising LTS edge during depolarizing step #347 (depol-block diagnosis)
• Closed predecessor: fix(presets): retune TRN for in-band HP92 tonic AP shape #346 / TRN: tonic AP peak voltage and AHP depth outside HP92 literature ranges #308 (TRN tonic AP shape retune that this builds on)
• Slow Na inactivation precedent (different mechanism, sustained block): fix(channels): opt-in slow Na inactivation for INaP (STN depol block) #325, Cortical Pyramidal: enable slow Na inactivation (INaP sNaP, fast-Na sNa) for biological accuracy #327, CA1 Pyramidal: enable slow Na inactivation (INaP sNaP, fast-Na sNa) for biological accuracy #328, Purkinje: enable slow Na inactivation (INaP sNaP, fast-Na sNa) for biological accuracy #329, Dopaminergic SNc: add slow Na inactivation to make_snc_inap_channel and fast Na #330
• Huguenard & Prince (1992), J. Neurosci. 12:3804 (TRN biophysics)
• Cueni et al. (2008), Nat. Neurosci. 11:683 (TRN dendritic Ca²⁺ during LTS)