Abstract
Local recurrent circuits in the cerebral cortex are hypothesized to implement signal amplification through excitatory feedback. However, the functional organization of these circuits remains difficult to probe due to the spatial intermingling of neurons with diverse tuning. By combining computational modeling with targeted photoablation in the mouse somatosensory cortex, this study demonstrates that recurrent excitation specifically amplifies sensory signals within functionally similar subnetworks. Notably, these circuits exhibit a high degree of fragility: removing fewer than 5% of neurons within a specific functional group significantly degrades the sensory representation of the remaining population. Our findings suggest that cortical circuits prioritize signal amplification at the cost of redundancy.
Introduction
The majority of synapses in the mammalian cortex are local and excitatory. These recurrent connections are thought to underpin fundamental computations such as pattern completion and signal amplification. Previous research has identified "subnetworks"—groups of neurons with similar receptive fields and higher-than-average connectivity. Yet, understanding how these subnetworks function during active behavior is challenging because neurons encoding different information are often physically adjacent and distributed across large volumes. This study investigates whether these recurrent circuits drive input-specific amplification and assesses their robustness to the loss of individual components during a tactile discrimination task.
Methods
The researchers employed a "model-predict, experiment-verify" approach to study Layer 2/3 (L2/3) of the vibrissal somatosensory cortex (vS1):
- Computational Modeling: The team developed a network model consisting of 1,700 excitatory and 300 inhibitory integrate-and-fire neurons. They varied the "subnetwork connectivity"—the probability that neurons with similar tuning are connected—to observe the effects on signal processing and robustness.
- In Vivo Volumetric Imaging: Using two-photon microscopy and the calcium indicator GCaMP6s, they recorded the activity of approximately 8,000 neurons per mouse across 18 imaging planes while mice performed a pole-localization task.
- Targeted Photoablation: After identifying neurons selective for "touch" or "whisking," the researchers used high-intensity multiphoton excitation to precisely ablate a small, specific subset of these neurons (10–20 cells). They then observed the functional impact on the remaining, spared neurons to determine the necessity of the ablated population for network-wide encoding.
Key Insights
The Amplification-Fragility Tradeoff
The computational model revealed three distinct regimes of connectivity. Low-connectivity networks are robust to damage but fail to amplify signals. Conversely, high-connectivity networks amplify signals effectively but are "brittle"; losing just a few members causes the entire subnetwork's performance to decline. The experimental data closely matched the regime of elevated subnetwork connectivity (0.4), suggesting that real cortical circuits are tuned for high amplification despite the resulting vulnerability.
Input-Specific Recurrence
The cortex does not function as a single monolithic amplifier. Instead, it appears to be a collection of overlapping, stimulus-specific amplifiers. In experiments, ablating "touch" neurons only degraded the "touch" representation among spared neurons, leaving "whisking" representations unaffected. This specificity confirms that recurrent interactions are constrained by functional similarity.
The Similarity and Spatial Rules
The impact of ablation followed a clear logic: spared neurons that were most similar in their tuning to the ablated neurons suffered the greatest loss in encoding quality. Furthermore, the effect of losing a neighbor decayed over a distance of approximately 87 µm. This distance matches the known physical scale of local excitatory connections in the rodent cortex, providing strong evidence that the observed degradation was due to the loss of local recurrent support.
Results
- Significant Impact from Minimal Loss: Ablating a tiny fraction of neurons (10–20 cells, or <5% of the total) representing touch caused a significant decline in the touch encoding scores ($R_{touch}$) of the remaining neurons, dropping from approximately 0.123 to 0.100.
- Functional Specificity: While touch representation was degraded by touch-cell ablation, whisking representation remained stable ($P = 0.820$). Conversely, ablating whisking neurons or "silent" neurons had no significant effect on the touch representation.
- Model Validation: The experimental results validated the hypothesis that recurrent excitation among neurons with similar selectivity drives input-specific amplification during behavior.
Discussion and Future Work
This research suggests that the brain prioritizes signal clarity through specific subnetwork amplification, even if it makes the system more vulnerable to the loss of individual neurons than previously assumed.
One intriguing finding was the "Whisking Paradox": ablating whisking neurons did not significantly degrade the whisking representation. The authors hypothesize that because whisking input is "asynchronous" (neurons fire at different phases), it may not trigger the same level of recurrent amplification as the highly synchronized "pop" of a touch event. Future work should explore whether homeostatic mechanisms allow the cortex to rewire and recover lost amplification over time, and how different inhibitory interneuron types (such as PV, SOM, and VIP cells) modulate these recurrent dynamics.
Summary updated