In a groundbreaking study published in Nature, researchers have unveiled sophisticated predictive models linking the complex architecture of individual neurons in the mouse primary visual cortex (VISp) to their specific projection targets across the brain. This advance stems from integrating multimodal data encompassing transcriptomic signatures, morphological features, and precise cortical location, offering unprecedented insight into the neural circuits underlying sensory processing.
The visual cortex is a well-organized, topographically mapped region that transforms visual inputs into detailed neuronal signals. Understanding how neurons in this area wire to various target regions is crucial for deciphering brain-wide communication and function. Prior studies have shown that certain transcriptomic profiles relate to general axonal shapes; however, the precise prediction of where neurons send their signals—a concept known as their projectome—remained elusive.
Leveraging an extensive dataset of whole-neuron morphologies (WNMs) from the VISp, the team applied logistic regression models enriched with sparse Reduced Rank Regression (RRR)-derived latent factors representing morphological features. These latent factors effectively distilled complex neuronal shapes and branching patterns into quantifiable variables. Critically, the models also incorporated the exact cortical surface location of each neuron, reflecting the established visuotopic organization of the visual cortex.
The models were stratified according to well-defined neuronal subclasses within the VISp, including Layer 2/3 Intratelencephalic (L2/3 IT), Layer 4 and Layer 5 Intratelencephalic (L4 and L5 IT), Layer 5 Extratelencephalic (L5 ET), and Layer 6 Corticothalamic (L6 CT) cells. Importantly, the performance of each model was assessed against a null baseline using the corrected Akaike information criterion (AICc), which adjusts for model complexity to prevent overfitting.
Notably, morphological latent factor-based models yielded superior predictive power for L4 and L5 IT neurons and L5 ET neurons compared to L2/3 IT and L6 CT neurons. This heterogeneity in prediction accuracy underscores the intrinsic diversity of neuronal wiring within the VISp and suggests that certain subclasses’ axonal architectures are more tightly coupled to their projection targets.
Beyond morphology, the spatial location of neurons within the VISp proved a potent predictor for their projection destinations, particularly to cortical areas and certain subcortical structures. The pronounced visuotopic mapping meant that neurons situated nearer a target region were far likelier to project there. For example, medial VISp neurons showed greater projection probabilities to the posteromedial visual area (VISpm), while those in anterolateral zones favored projections to rostral lateral areas (VISrl and VISl).
Combining morphological latent factors and cortical position enhanced model performance, revealing a synergistic relationship in forecasting neuronal wiring. This dual-factor approach outperformed simpler models, illustrating that structural features and spatial context jointly inform neuronal connectivity.
Detailed examination of latent spaces revealed nuanced predictions for distinct neurons. Among L4 and L5 IT cells, latent factors resembling L4 IT profiles correlated with reduced projection likelihood across targets, whereas factors aligned with L4/L5 IT and L5 IT-2 profiles predicted differential projection affinities, such as stronger connections to the caudoputamen but weaker to retrosplenial cortex subdivisions.
Mapping predicted projection probabilities across the VISp further illuminated functional organization. Central VISp locations exhibited reduced propensity to project broadly, whereas lateral and anterior sectors manifested heightened contralateral VISp projections, reflecting specialized hemispheric communicative roles. This topographic precision highlights how cortical position encodes vital wiring information.
For L5 ET neurons, morphology wielded more influence on subcortical projections than cortical projections. Specific subcortical targets varied in their dependence on distinct latent factors, emphasizing the complexity of these neurons’ integrative roles. For instance, projections to the dorsal lateral geniculate nucleus (LGd-co) were predominantly predicted by one latent factor, whereas the lateral posterior nucleus (LP) depended on a composite of multiple factors.
Similarly, VISp location impacted L5 ET projection probabilities to cortical and subcortical regions. Anterior VISp neurons projected more robustly to caudoputamen and pontine grey, highlighting spatially nuanced functional specializations within the L5 ET subclass. However, in general, L5 ET neurons demonstrated lower cortical projection probabilities than L4 and L5 IT neurons at comparable locations.
Critically, the study validated model predictions at the single-cell level. By fitting models excluding each probed neuron, researchers predicted that neurons with high probability scores frequently projected to those regions in reality. While inherently probabilistic, these findings confirm the utility of integrating transcriptomic-related morphology and cortical position in elucidating the brain’s intricate connectome.
This research bridges molecular identity, single-cell morphology, and spatial architecture to decode the wiring blueprint of the visual cortex. It challenges traditional notions that neuron connectivity is random or solely genetically scripted, instead portraying a sophisticated system where morphology and location co-determine projection specificity.
The implications extend beyond basic neuroscience, providing a framework for understanding developmental circuit assembly, experience-dependent plasticity, and potential pathologies arising from miswired cortical circuits. Future work may leverage these predictive models to influence targeted neuronal manipulation or regeneration strategies, especially relevant in sensory disorders.
In summary, the multi-modal integrative approach crystallized here offers a transformative vantage point into neural connectivity. By quantitatively predicting where neurons send their signals based on latent morphological and positional features, the study opens new horizons in connectomics, blending data science and biological insight. This fusion not only advances our fundamental understanding of brain organization but also lays the groundwork for novel therapeutic avenues.
As researchers refine these models and incorporate additional data modalities such as electrophysiology or gene expression dynamics, the precision of predicting neural circuitry will undoubtedly improve. This work exemplifies the power of computational neuroscience married to experimental data, heralding a new era of brain mapping that is both comprehensive and predictive.
The elucidation of projection probabilities linked to accessible morphological and spatial features democratizes connectomic analysis, making it feasible to infer difficult-to-measure circuitry properties from more readily available datasets. This paradigm shift enhances our ability to chart the brain’s wiring at single-cell resolution and paves the way for mechanistic insights into neural computation and cognition.
Subject of Research:
Single-cell transcriptomes and morphological features predict projection targets in the mouse visual cortex.
Article Title:
Connecting single-cell transcriptomes to projectomes in the mouse visual cortex.
Article References:
Sorensen, S.A., Gouwens, N.W., Wang, Y. et al. Connecting single-cell transcriptomes to projectomes in the mouse visual cortex. Nature (2026). https://doi.org/10.1038/s41586-026-10424-8
Image Credits: AI Generated
