Imagine recording from a single neuron in the primary visual cortex (V1). When a visual stimulus appears, the neuron fires rapidly. Traditional feedforward models of neuroscience treat this neuron as a simple filter, mapping raw sensory inputs to downstream processors.
But what happens when the stimulus disappears, and the neuron keeps firing because the animal is holding that image in working memory? Or what if the neuron fires more intensely simply because the animal is paying attention to that region of space?
A single scalar value—like a neuron's firing rate—is fundamentally ambiguous. An increase in activity could mean the local circuit has computed something novel, or it could mean the circuit is inheriting a signal computed elsewhere and transmitted via feedback. This conundrum is known as signal origin ambiguity. To map the true functional architecture of the brain, systems neuroscience must move beyond observing what information a brain region contains and figure out where that information was actually generated.
The Cognitive Blackboard
For decades, early sensory cortices were viewed as passive relays. However, a growing body of evidence suggests they actively participate in higher-order cognitive processes.
Consider how the brain handles subjective perception. Niclas Kilian-Hütten et al. demonstrated in Auditory Cortex Encodes the Perceptual Interpretation of Ambiguous Sound that early auditory regions encode what a listener hears rather than the objective physical sound wave. Similarly, Lars Riecke et al. showed in Hearing Illusory Sounds in Noise: Sensory-Perceptual Transformations in Primary Auditory Cortex that primary auditory cortex activity tracks the subjective reconstruction of a sound during the auditory continuity illusion.
This phenomenon is not isolated to hearing. In Learning a New Selection Rule in Visual and Frontal Cortex, Chris van der Togt et al. found that selection signals in the early visual cortex reflect a subject's internal choice even when that choice is a total guess.
To explain these patterns, Floris P. de Lange proposed a compelling framework in Early Visual Cortex as a Multiscale Cognitive Blackboard: early sensory areas act as high-resolution "blackboards" where higher-level cognitive areas write task-relevant information via feedback connections.
[Higher-Order Executive Areas] -- (Top-Down Feedback) --> [Early Sensory Cortex]
(Computes Choice/Memory) ("Cognitive Blackboard")
While this framework explains why cognitive signals appear in early cortices, it highlights the core ambiguity: how do we prove the blackboard isn't writing on itself?
Structural Clues: Layers and Latencies
To resolve where a signal originates, researchers exploit the physical constraints of neural architecture: time (latency) and space (cortical layers).
The Temporal Signature of Feedback
Feedforward signals travel quickly from the periphery upward. Top-down feedback or local recurrent loops take time to propagate. Therefore, if a cognitive modulation appears with a distinct delay, it points away from pure feedforward inheritance.
For example, Ulf H. Schnabel et al. showed in Figure-ground perception in the awake mouse and neuronal activity elicited by figure-ground stimuli in primary visual cortex that mouse V1 neurons exhibit enhanced activity for figures compared to backgrounds, but this enhancement is driven by delayed synaptic inputs. On a macro scale, Kohitij Kar et al. demonstrated in Evidence that recurrent circuits are critical to the ventral stream’s execution of core object recognition behavior that primate object recognition requires a ~30ms temporal delay to solve challenging, occluded images—a delay that feedforward artificial networks cannot replicate.
The Spatial Signature of Layers
The neocortex is organized into six distinct layers. Crucially, feedforward inputs primarily enter the middle layer (Layer 4), whereas top-down feedback signals predominantly target the superficial (Layers 1–3) and deep (Layers 5–6) layers.
Cortical Layers:
┌────────────────────────────────────────┐
│ Layers 1–3 (Superficial) <── Feedback │
├────────────────────────────────────────┤
│ Layer 4 <── Feedforward
├────────────────────────────────────────┤
│ Layers 5–6 (Deep) <── Feedback │
└────────────────────────────────────────┘
By recording across these layers simultaneously, researchers can isolate the direction of information flow. In Layer-specificity in the effects of attention and working memory on activity in primary visual cortex, Timo van Kerkoerle et al. used laminar probes in macaque V1 to demonstrate that attention and working memory signals specifically target the superficial and deep layers, leaving Layer 4 unaffected.
To pinpoint this further, Fang Liu utilized high-resolution dendritic imaging in Spatiotemporal functional organization of excitatory synaptic inputs onto macaque V1 neurons to reveal that the apical dendrites of V1 neurons—which reach up into the topmost layers—serve as the explicit structural hubs for these delayed feedback signals.
Mathematical Separation in State Space
Anatomy provides physical clues, but mathematical modeling allows us to disentangle mixed signals within a single population of neurons. When a network mixes multiple types of information (like "what" an object is and "when" it appeared), can we separate internally generated activity from external driving forces?
In Functional, But Not Anatomical, Separation of “What” and “When” in Prefrontal Cortex, Christian K. Machens et al. addressed this by developing a "Difference-of-Covariances" (DOC) method. They proved that the highly heterogeneous, "messy" activity of prefrontal neurons can be mathematically projected into distinct functional components. This decomposition revealed that memory ("what") is maintained by internal, recurrent network loops, while timing ("when") is driven by external inputs.
Similarly, Evan D. Remington et al. explored this dynamic in Flexible sensorimotor computations through rapid reconfiguration of cortical dynamics. By analyzing neural trajectories in a state space, the authors showed that the brain achieves behavioral flexibility by using persistent external inputs to shift the network's starting state, while letting local, internal dynamics govern how that trajectory evolves over time.
A Ubiquitous Challenge Across Modalities
Signal origin ambiguity is not a quirk of the visual system; it is a fundamental property of neural hierarchies across all modalities.
- The Motor System: In The Largest Response Component in the Motor Cortex Reflects Movement Timing but Not Movement Type, Matthew T. Kaufman et al. discovered that the most dominant signal in the motor cortex is a massive "trigger" signal that dictates movement timing. Yet, looking at the motor cortex alone cannot determine whether this trigger is generated locally or driven by subcortical structures.
- The Olfactory System: When an animal sniffs rapidly, the physical concentration of an odor changes. Rebecca Jordan et al. showed in Sniffing Fast: Paradoxical Effects on Odor Concentration Discrimination at the Levels of Olfactory Bulb Output and Behavior that the olfactory bulb must account for this action-driven change. However, whether this requires top-down motor feedback or local sensory adaptation remains an open question.
- The Thalamus: Long considered a simple feedforward gatekeeper, the thalamus also exhibits complex cognitive signals. Sami El‐Boustani et al. identified decision-related signals within thalamic nuclei in Anatomically and functionally distinct thalamocortical inputs to primary and secondary mouse whisker somatosensory cortices, leaving it unresolved whether the thalamus computes these choices or inherits them via feedback loops from the cortex.
Moving Beyond Correlation
Why does solving signal origin ambiguity matter so much? Because mapping a signal across the brain tells us where information is distributed, but it does not tell us how it got there.
As David J. Ottenheimer et al. demonstrated in A stable, distributed code for cue value in mouse cortex during reward learning, reward-predicting value signals are remarkably stable and broadcast across the entire frontal cortex. Without resolving the causal hierarchy, we cannot distinguish between a central hub broadcasting this value upward or a distributed network computing it through bottom-up integration.
Observational methods—even high-resolution fMRI or large-scale electrophysiology—primarily establish correlations. To truly resolve neural signal provenance, future paradigms must combine layer-specific recordings and state-space modeling with targeted causal interventions, such as optogenetically silencing higher-order cortical areas to see exactly what information vanishes from the sensory blackboard.