🧠 neuroscience

Sensory recurrence and convergent integration in the C. elegans connectome

This study reveals that the *C. elegans* connectome exhibits significantly higher sensory-driven recurrence and reduced convergence compared to random networks, suggesting a structural basis for non-feed-forward signaling through a feedback gradient from sensory neurons to integrating interneurons.

Original authors: Moutuou, E., Benali, H.

Published 2026-11-03

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Moutuou, E., Benali, H.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

The nervous system of the tiny roundworm Caenorhabditis elegans has long served as a map for understanding how brains work. Scientists have known for decades exactly how every single nerve cell in this worm connects to every other, a complete wiring diagram called a connectome. For a long time, the prevailing view was that this wiring follows a simple, one-way street: sensory neurons at the surface detect the world, pass the message to middle neurons called interneurons, which then hand the signal off to motor neurons that drive movement. It seemed like a clean, forward-moving flow of information. However, recent experiments using light to activate specific neurons showed that signals do not always travel in a straight line. In fact, sensory neurons often respond to signals from many other parts of the brain, not just the ones they are supposed to be listening to. This created a puzzle: if the wiring diagram looks like a one-way street, why does the actual traffic behave like a busy intersection with cars turning around?

To solve this, researchers Elkaïoum M. Moutuou and Habib Benali looked deeper into the worm's wiring diagram, not just counting the direct connections between cells, but tracing the paths signals could take as they bounced from one neuron to another and eventually returned to where they started. They focused on a specific part of the worm's brain responsible for sensing temperature, a circuit known as thermotaxis. Using a mathematical approach that treats the network as a system of flowing paths, they calculated how much of the activity reaching a neuron actually originated from that same neuron. They compared this real wiring against thousands of random networks that had the same number of connections but were shuffled together without any biological structure. The goal was to see if the worm's brain had a special design for looping signals back, or if the loops were just a random byproduct of having many connections.

The researchers found that the worm's brain is far more full of loops than a random arrangement of wires would be. In fact, the real network contains about three times more of these returning paths than the random models. What was most surprising was where these loops were located. In a random network, the most looping activity happens at the busiest, most connected hubs. But in the worm, the most intense looping happens at the very beginning of the chain, in the sensory neurons that first touch the outside world. As the signal moves deeper into the brain toward the neurons that integrate information and control movement, this looping effect fades away. This creates a gradient: the sensory periphery is a place where signals fold back on themselves, while the deeper processing centers receive input from many different sources without sending much back.

This pattern held true across five different versions of the worm's wiring map, including maps from two completely different adult worms reconstructed from scratch. The researchers also checked if this was just a result of neurons simply having two-way connections with each other. They found that even when they removed the direct two-way links, the sensory neurons still showed a much higher tendency for signals to return to them than would be expected by chance. The loops were short, often involving just one or two steps back through a neighbor, but they were concentrated specifically in the sensory layer. Meanwhile, the neurons further down the line, which act as integrators, receive signals from a wide variety of sources, but those sources do not send signals back to them. This means the sensory neurons are uniquely embedded in a web of feedback, while the integrators act as true convergence points.

The study also looked at how this structural finding matches up with what happens when the worm is actually alive and responding to stimuli. Previous experiments had shown that sensory neurons often react to being stimulated, suggesting they are part of a feedback loop. The new analysis confirms that the physical wiring supports this behavior, but it also reveals a limit to what the current maps can tell us. While the structure shows a clear gradient of looping from the sensory edge inward, the functional data from living worms is too noisy to see this same clear pattern in individual cells. The researchers suggest that the difference might be because the living experiments only look at very quick, single-shot responses, whereas the structural loops they found might take a few steps to close and influence the brain over a slightly longer time.

Ultimately, this work reshapes how we see the worm's brain. It is not just a simple pipeline moving information from the outside in. Instead, the very first neurons to sense the world are caught in a dense net of returning signals, folding the sensory experience back on itself before it ever reaches the decision-making centers. The deeper parts of the brain, by contrast, gather information from many different places without sending much back, acting as a true sink for diverse inputs. This structural arrangement, where the sensory edge is rich in recurrence and the core is rich in convergence, offers a physical explanation for why signals in the worm's brain do not flow in a straight line. It suggests that the brain's ability to process the world relies on a specific architecture where the senses are constantly talking to themselves, while the rest of the brain listens to the whole world.

Technical Summary: Sensory Recurrence and Convergent Integration in the C. elegans Connectome

Problem Statement
While the global synaptic architecture of the Caenorhabditis elegans nervous system is traditionally described as feed-forward (sensory $\to$ interneuron $\to$ motor), recent whole-brain optogenetic mapping reveals that signal propagation does not strictly follow this direction. Functional connectivity shows that sensory neurons receive more incoming functional connections than they send, and signals often fail to propagate forward as predicted by structural models. A key unresolved question is whether the synaptic network itself contains significant recurrent pathways that return activity to the sensory periphery, and if so, how this recurrence is distributed relative to the convergence of inputs. Simple loop counting is insufficient, as highly connected neurons naturally accumulate more loops even in random networks, obscuring specific structural motifs.

Methodology
The authors analyze the merged somatic connectome of the C. elegans hermaphrodite (280 neurons, 12,071 contacts) and five independent reconstructions (including two independent adult animals). They employ the Kubo–Martin–Schwinger (KMS) framework for directed networks to define a feedback coefficient ( $f_b$ ).

The Metric: The method utilizes the resolvent matrix $R_\beta = (I - e^{-\beta A})^{-1}$ , where $A$ is the adjacency matrix and $\beta$ is an inverse temperature parameter set slightly above the critical value $\beta_c = \log r(A)$ (specifically $\beta_f = 1.05\beta_c$ ).
Definition: The feedback coefficient $f_b$ for a neuron $u$ is defined as the share of the total walk mass reaching $u$ that originates from $u$ itself. Mathematically, $f_b = \frac{(R_\beta)_{uu} - 1}{\sum_w (R_\beta)_{uw} - 1}$ . This normalizes the closed walks (recurrence) by the total incoming walk mass (convergence), distinguishing it from standard centrality measures like Katz centrality or simple loop counts.
Null Models: To determine statistical significance, the authors generate strength-preserving null networks (permuting targets while maintaining in/out-strength) and reciprocity-preserving null networks.
Complementary Measures: The study also calculates Integration Capacity (IC) and Receptance Entropy ( $H_{rec}$ ) to quantify convergence (the diversity of sources contributing to a neuron's input).
Validation: Results are compared against the "signal propagation atlas" (Randi et al.), which maps functional responses to optogenetic stimulation.

Key Contributions and Results

Excess Recurrence in Sensory Neurons:
The C. elegans connectome exhibits approximately three times more recurrence than strength-matched random networks. Crucially, this excess is not distributed according to connectivity degree. In random networks, recurrence concentrates on highly connected neurons; in the C. elegans connectome, the excess recurrence is concentrated in sensory neurons and weakly connected neurons. Sensory neurons show a 4.4-fold excess recurrence compared to null expectations, significantly higher than interneurons (2.8-fold) or motor neurons (2.6-fold).
The Recurrence-Convergence Gradient:
A distinct gradient exists along the thermotaxis pathway (from sensory neurons like AFD/AWC/ASI to integrating interneurons like RIA).
- Recurrence: Highest at the sensory periphery and decreasing toward integrating interneurons.
- Convergence: Sensory neurons receive input from significantly fewer sources than expected (low receptance entropy and low integration capacity relative to nulls), whereas interneurons receive input from many diverse sources.
- Correlation: There is a strong negative correlation between excess recurrence and the diversity of input sources; neurons with the highest recurrence receive their non-self input from the fewest sources.
Robustness and Specificity:
- The pattern persists across five different connectome reconstructions, including two independently reconstructed adult animals.
- The gradient holds even when gap junctions are removed (chemical synapses only) and when reciprocal contacts are preserved in null models, indicating the phenomenon relies on loops beyond simple reciprocal pairs.
- The ordering is dependent on path scale: it is strongest near the critical value $\beta_c$ (where long walks contribute) and vanishes when only short paths (1-2 steps) are considered.
Comparison with Functional Data:
When compared to the functional signal propagation atlas, sensory neurons show a weak bias toward incoming responses, consistent with the structural finding. However, the functional atlas does not resolve the structural recurrence gradient at the single-neuron level. The functional in/out balance correlates more strongly with the number of contacts than with the structural excess recurrence, and functional reciprocity aligns closely with structural reciprocity.

Significance and Claims
The paper claims to identify a specific structural substrate for the non-feed-forward signaling observed in functional studies. The findings suggest that the C. elegans wiring is organized such that the sensory periphery is "folded back" on itself, creating a dense local recurrence relative to the convergence of inputs. In contrast, integrating interneurons act as hubs that pool signals from many diverse sources without significant feedback to those sources.

The authors conclude that this structural organization—where sensory neurons have high recurrence relative to their convergence, and interneurons have low recurrence relative to high convergence—provides a candidate mechanism for how signals might return to the sensory periphery through synapses alone. They note that while the structural data supports a departure from feed-forward flow, the current functional atlas does not yet validate this at the single-neuron scale, likely due to the limitations of brief, single-neuron stimulation in engaging multi-step loops or the influence of extrasynaptic transmission. The study emphasizes that the specific normalization of recurrence against convergence (via the KMS framework) is essential for detecting this organization, as standard loop counts fail to reveal it.

Technical Summary: Sensory Recurrence and Convergent Integration in the C. elegans Connectome

More like this