Skip to content

DD027: Multicompartmental Neuron Models

  • Status: Proposed
  • Author: OpenWorm Core Team
  • Date: 2026-02-24
  • Supersedes: None
  • Related: DD001 (Neural Circuit Architecture), DD005 (Cell-Type Specialization), DD010 (Validation Framework), DD017 (Hybrid ML), DD020 (Connectome Data Access), DD024 (Validation Data Acquisition)

TL;DR

Some C. elegans neurons require multicompartmental cable equation models — single-compartment (isopotential) approximations miss compartmentalized calcium signals (RIA) and all-or-none action potentials (AWA). This DD specifies how to build Level D multicompartmental models for a subset of neurons, starting with 5 representative neurons (AWC, AIY, AVA, RIM, VD5) and scaling to all 302, using existing NeuroML morphologies and ion channel libraries.

Quick Action Reference

Question Answer
Phase Phase 2 (Stage 1: 5 neurons), Phase 4-5 (Stage 2: all 302)
Layer Neural Architecture Extension
What does this produce? Multicompartmental NeuroML neuron models with per-segment channel densities and spatially resolved synapses
Success metric Level D neurons reproduce published I-V curves and compartmentalized calcium dynamics (Hendricks 2012, Nicoletti 2019)
Where is the code? openworm/c302 (parameters_D.py framework, morphologies/, channel_models/)
Quick start See Getting Started below

Biological Motivation

Experimental evidence shows that single-compartment (isopotential) models are insufficient for a subset of C. elegans neurons:

  1. Compartmentalized calcium dynamics (RIA). Hendricks et al. (2012) demonstrated that calcium dynamics in the RIA interneuron are compartmentalized across distinct segments of the neurite, encoding head movement direction through spatially separated signals within a single cell.

  2. All-or-none action potentials (AWA). Liu et al. (2018) showed that AWA olfactory neurons fire calcium-mediated all-or-none action potentials — a fundamentally different signaling mode from the graded potentials assumed by Level C1.

These findings indicate that model complexity must vary among neurons: some are well-described by the single-compartment approximation, while others require multicompartmental representations that capture signal propagation along neurites.

Technical Approach

NeuroML Multicompartmental Support

NeuroML 2 natively supports multicompartmental morphologies. The <cell> element can contain a <morphology> with multiple <segment> elements organized into <segmentGroup> definitions, with per-segment channel density assignments. This means Level D can be implemented within the existing NeuroML/LEMS framework without a new file format — the same jnml -validate pipeline applies, and the same NEURON simulator backend can execute multicompartmental cells alongside single-compartment ones in the same network simulation (Cannon et al. 2014; Gleeson et al. 2018).

Feasibility (Zhao et al. 2024)

Zhao et al. (2024) showed that the "representative neuron" strategy makes multicompartmental modeling tractable at scale: build detailed models for a small set of representative neurons (one per functional group), fit them to published electrophysiology, then propagate fitted parameters to all neurons in the same functional class. Using this approach with 5 representative neurons (AWC, AIY, AVA, RIM, VD5), they produced 136 multicompartmental neurons whose I-V curves matched experimental recordings. Nicoletti et al. (2019) earlier demonstrated a similar multicompartmental approach for AWCon with multiple ion channel types. This establishes that Level D is achievable with current data — it does not require waiting for new experimental techniques.

Code Reuse Opportunity

The BAAIWorm repository (github.com/Jessie940611/BAAIWorm, Apache 2.0 license) contains NMODL ion channel files and SWC neuron morphology reconstructions. These can be converted to NeuroML format using pyNeuroML's NMODL→NeuroML converter, providing a head start on the channel library expansion and morphological models.

Implementation Pathway

Stage 1 (Phase 2 — Proof of Concept)

  1. Select 5 representative neurons with published morphological reconstructions AND published electrophysiology: AWC (sensory), AIY (interneuron), AVA (command interneuron), RIM (interneuron), VD5 (motor neuron) — the same set validated by Zhao et al. (2024)
  2. Obtain morphologies from EM reconstructions (Witvliet et al. 2021; Cook et al. 2019) or from BAAIWorm SWC files; convert to NeuroML <morphology> elements with segments < 2 μm
  3. Assign per-segment channel densities from the Extended Channel Library (14 classes), guided by CeNGEN expression profiles (DD005) and functional group membership
  4. Optimize passive parameters (axial resistance, membrane capacitance) and channel densities using automated fitting (DD017 differentiable backend or NEURON's built-in optimizer) to match published I-V curves and current-clamp responses
  5. Propagate fitted parameters to all neurons in the same CeNGEN functional class, scaling channel densities by expression level (DD005)

Stage 2 (Phase 4-5 — Scale to Full Circuit)

  1. Extend to all 302 neurons using the representative-neuron approach
  2. Incorporate subcellular molecular data from expansion microscopy (Alon et al. 2021; Shaib et al. 2023) as it becomes available
  3. Apply spatially resolved synapse placement (see section below)
  4. Infer parameters using data-constrained fitting methods including RNN-based approaches (Linka et al. 2023)

OpenWorm Extensions Beyond Zhao et al.

(a) We target all 302 neurons, not 136; (b) we use NeuroML standard format enabling multi-simulator support and community sharing; (c) we integrate with CeNGEN transcriptomics for principled parameter propagation rather than purely functional-group-based assignment; (d) our models include neuropeptidergic modulation (DD006) and organ systems (DD007, DD009, DD018) that the locomotion-only circuit does not capture.

Spatially Resolved Synapse Placement

For the single-compartment models (Levels A-C, C1), synapses are abstract neuron-to-neuron connections with no spatial structure — all inputs sum at the single compartment. However, for multicompartmental neurons (Level D), the location of synapses along neurites matters because it determines signal propagation delays, spatial input integration, and the degree to which nearby synapses interact nonlinearly.

Zhao et al. (2024) demonstrated a practical approach: for each connection in the Cook et al. (2019) adjacency matrix, assign a distance along the neurite drawn from an inverse Gaussian distribution fitted to experimental synapse centroid distance measurements from serial-section EM (Witvliet et al. 2021). Each synapse is then placed on the neurite segment closest to the assigned distance. This produces spatially realistic clustering of synapses along neurites, matching the biological organization observed in EM.

OpenWorm will adopt this approach with one improvement: quantitative validation that the constructed distributions match the experimental distributions (as in Zhao et al. Fig. 4B-C), integrated into DD010 Tier 1 as a non-blocking structural validation.

Applies only when: neural.level: D and neural.spatial_synapses: true. For Level C1, synapse placement is irrelevant and this feature is disabled.

Data requirement: Synapse centroid distances from Witvliet et al. 2021, to be acquired per DD024 (Validation Data Acquisition Pipeline). See also DD020 for ConnectomeToolbox data access.

Quality Criteria

  1. Backward compatibility: Level D neurons must coexist with Level C1 single-compartment neurons in the same network simulation
  2. Biological fidelity: Per-neuron I-V curves reproduce published electrophysiology within ±15%
  3. Compartmentalized dynamics: RIA model shows spatially separated calcium signals matching Hendricks et al. 2012
  4. Structural validation: Synapse placement distributions match Witvliet et al. 2021 EM measurements (DD010 Tier 1)
  5. Standard format: All models in NeuroML2 format, passing jnml -validate

Validation

Level D neurons must pass all DD010 tiers. Individual cell models should additionally reproduce published I-V curves and compartmentalized calcium dynamics where available (e.g., RIA spatial signals per Hendricks et al. 2012, AWC responses per Nicoletti et al. 2019).

Getting Started

Prerequisites: Familiarity with DD001 (Neural Circuit Architecture) — especially the c302 framework, Level C1 baseline, and the Hodgkin-Huxley formulation.

Key resources:

  1. c302 Level D framework: c302/parameters_D.py — existing Level D infrastructure with ChannelDensity support and Species (Ca); currently uses placeholder channels
  2. Existing morphologies: CElegansNeuroML/CElegans/generatedNeuroML2/ — all 302 neurons as multicompartmental NeuroML2 cells (also copied into c302/NeuroML2/)
  3. Ion channel library: Nicoletti et al. 2019 NeuroML2 channels at openworm/NicolettiEtAl2019_NeuronModels/NeuroML2/ (31 channels, validated)
  4. EM morphology data: Witvliet et al. 2021 and Cook et al. 2019 for synapse centroid distances
  5. BAAIWorm reference: Per-neuron conductance JSONs at eworm/components/param/cell/*.json

First contribution: Start with DD027 Draft Issues — Issue 14 (evaluate existing morphologies) is the entry point.

Existing Code Resources

Resource Repository What It Provides
Level D framework openworm/c302 (parameters_D.py) Multicompartmental infrastructure, ChannelDensity, placeholder channels
302 neuron morphologies openworm/CElegansNeuroML All neurons as multicompartmental NeuroML2
31 ion channels (NeuroML2) openworm/NicolettiEtAl2019_NeuronModels AWCon (16 ch) + RMD (15 ch), validated
22 NMODL channel files openworm/NicolettiEtAl2024_MN_IN Motor neurons + interneurons
17+ NMODL channels + per-neuron conductances Jessie940611/BAAIWorm Full channel library + JSON conductance params
Morphology data (SWC) Jessie940611/BAAIWorm HOC files for every neuron

Integration Contract

Inputs / Outputs

Inputs (What This Subsystem Consumes)

Input Source Variable Format Units
Single-compartment neuron models DD001 Level C1 network NeuroML —
Cell-type-specific conductances DD005 Per-class channel expression CSV TPM → g_max
Neuron morphologies (EM) DD020 / Witvliet 2021 3D segment coordinates NeuroML <morphology> µm
Synapse centroid distances DD024 / Witvliet 2021 Distance distributions CSV µm
Differentiable fitting backend DD017 Parameter optimizer Python API —

Outputs (What This Subsystem Produces)

Output Consumer Variable Format Units
Multicompartmental neuron models DD001 network Level D cells NeuroML <cell> —
Fitted channel densities DD010 Tier 1 Per-segment g_max NeuroML <channelDensity> S/cm²
Spatially placed synapses DD001 network Synapse locations NeuroML <connection> with segment refs µm
Validation metrics DD010 I-V curve fits JSON —

References

  1. Hendricks, M., et al. (2012). "Compartmentalized calcium dynamics in a C. elegans interneuron encode head movement." Nature, 487:99-103. doi:10.1038/nature11081
  2. Liu, Q., et al. (2018). "C. elegans AWA olfactory neurons fire calcium-mediated all-or-none action potentials." Cell, 175:57-70. doi:10.1016/j.cell.2018.08.018
  3. Nicoletti, M., et al. (2019). "Biophysical modeling of C. elegans neurons: Single ion currents and whole-cell dynamics of AWCon and RMD." PLoS ONE, 14:e0218738. doi:10.1371/journal.pone.0218738
  4. Zhao, B., et al. (2024). "MetaWorm: an integrative data-driven model of C. elegans." Nature Computational Science, 4:978-990. doi:10.1038/s43588-024-00738-w
  5. Cannon, R. C., et al. (2014). "LEMS: a language for expressing complex biological models in concise and hierarchical form." Frontiers in Neuroinformatics, 8:79. doi:10.3389/fninf.2014.00079
  6. Gleeson, P., et al. (2018). "c302: a multiscale framework for modelling the nervous system of C. elegans." Phil. Trans. R. Soc. B, 373:20170379. doi:10.1098/rstb.2017.0379
  7. Witvliet, D., et al. (2021). "Connectomes across development reveal principles of brain maturation." Nature, 596:257-261. doi:10.1038/s41586-021-03778-8
  8. Shaib, A. H., et al. (2023). "Expansion microscopy at the nanoscale." Nature Biotechnology. doi:10.1038/s41587-024-02431-9
  9. Linka, K., et al. (2023). "A new family of constitutive artificial neural networks towards automated model discovery." Acta Biomaterialia. doi:10.1016/j.actbio.2023.01.055
  • Approved by: Pending (awaiting founder review)
  • Implementation Status: Not started
  • Next Review: After Phase 2 kickoff