Continuous flashing suppression of neural responses and population orientation coding in macaque V1

Using two-photon calcium imaging in awake macaques, this study demonstrates that continuous flash suppression substantially diminishes V1 neuronal responses and population orientation coding, suggesting that while coarse orientation discrimination remains possible, the severely degraded neural signals are insufficient to support high-level visual and cognitive processing.

The Gist

The Big Question: Can Your Brain "See" What You Can't?

Imagine you are watching a movie, but someone is constantly flashing bright, chaotic, colorful patterns right in front of your other eye. So bright and distracting is this flashing light that you completely lose track of the movie playing on the screen. You can't see the actors or the plot. This is called Continuous Flash Suppression (CFS).

Scientists have used this trick for years to study "subconscious vision." The big debate has been: If you can't consciously see the movie, is your brain still watching it? Does your brain understand the plot (high-level thinking), or is it just seeing blurry shapes (low-level features)?

To answer this, the researchers in this paper went straight to the source: the brain's "first camera," a part called V1.

The Experiment: Peeking Behind the Curtain

The researchers didn't just ask monkeys what they saw (monkeys can't talk!). Instead, they used a high-tech "night-vision camera" (two-photon calcium imaging) to watch thousands of individual neurons in the monkeys' brains light up while they looked at these flashing patterns.

Think of the V1 neurons as a massive crowd of security guards at the entrance of a stadium. Their job is to check the tickets (visual information) before letting the guests (the image) into the VIP section (the rest of the brain for complex thinking).

The Setup:

1. The Target: A simple striped pattern (a grating) shown to one eye.
2. The Masker: A chaotic, flashing noise shown to the other eye.
3. The Result: The monkey's brain "sees" the noise, but the stripes disappear from conscious awareness.

The Findings: The Security Guards Are Overwhelmed

The researchers found that the "flashing noise" didn't just hide the stripes; it crippled the security guards.

1. The "Noise-Loving" Guards: Some guards only listen to the eye seeing the flashing noise. When the noise flashed, these guards stopped talking about the stripes entirely. They were completely silenced.
2. The "Both-Eyes" Guards: Most guards listen to both eyes. These guys were almost completely silenced too. The noise was so loud it drowned out the signal.
3. The "Stripe-Loving" Guards: A few guards only listened to the eye seeing the stripes. They kept talking, but their voices were much quieter and fuzzier than usual.

The Analogy: Imagine trying to hear a whisper (the stripes) while someone is screaming in your ear (the flashing noise). Even if you are straining to listen, the whisper is barely audible. The brain's "signal-to-noise ratio" has crashed.

The Test: Can the Brain Reconstruct the Image?

To see what this meant for the rest of the brain, the researchers used two types of "AI detectives" (machine learning models) to analyze the data from the security guards.

Detective 1: The "Rough Sketch" Artist (Classification)

• The Task: Can the AI tell if the stripes are vertical or horizontal?
• The Result: Yes. Even with the noise, the AI could still guess the general direction of the stripes about 80-90% of the time.
• Meaning: The brain can still do simple, low-level tasks like "Is it vertical or horizontal?"

Detective 2: The "Photographer" (Reconstruction)

• The Task: Can the AI rebuild the actual picture of the stripes from the brain signals?
• The Result: No. When the AI tried to draw the image based on the suppressed brain signals, the result was a blurry, unrecognizable mess. It was like trying to paint a masterpiece using only a few drops of watered-down paint.
• Meaning: The brain has lost the detailed information needed to recognize what the object is.

The Conclusion: What Does This Mean for Us?

This study solves a major mystery in psychology.

• The Old Theory: Maybe the brain is secretly processing complex thoughts (like recognizing a face or a tool) even when we can't see it.
• The New Reality: The brain is not getting enough information to do that. The "first camera" (V1) is so suppressed that the image is too broken to be useful for complex tasks.

The Final Metaphor:
Imagine you are trying to read a book, but someone is shining a strobe light in your face.

• You might still be able to tell that the lines are horizontal (low-level processing).
• But you cannot read the words or understand the story (high-level processing) because the letters are too blurry and broken.

Why does this matter?
It suggests that when we feel like we are "subconsciously" reacting to something we can't see, we might actually just be reacting to very basic, blurry fragments (like "it's long" or "it's vertical"), not the full, complex meaning of the object. The brain needs a clearer picture to do the heavy lifting of thinking and understanding.

Technical Summary

1. Problem Statement

Continuous Flash Suppression (CFS) is a paradigm used to render a visual stimulus invisible by presenting a dynamic masker to one eye while a target stimulus is shown to the other. While CFS has been widely used to study subconscious high-level visual and cognitive processing (e.g., semantic priming), recent debates question whether these effects rely on intact low-level feature processing or if they are artifacts.

A critical, unresolved issue is the extent to which CFS suppresses neuronal activity in the primary visual cortex (V1), where binocular inputs first merge. If V1 responses are severely degraded, the "subconscious" high-level processing observed in behavioral studies may not be supported by sufficient neural information. Previous fMRI studies yielded conflicting results regarding V1 activity under CFS, often due to experimental designs that compared monocular vs. dichoptic masking rather than isolating the pure suppressive effect of CFS. Furthermore, the differential impact of CFS on neurons with varying ocular dominance (preferring the target eye, masker eye, or both) had not been quantified at the cellular level.

2. Methodology

The study employed two-photon calcium imaging in awake, fixating macaques to record large populations of V1 neurons at cellular resolution.

• Subjects: Two adult rhesus macaques (Monkey A and Monkey B).
• Imaging Setup: Customized two-photon microscope recording from superficial layers of V1 (and V2 in Monkey A) using GCaMP5G as a calcium indicator.
• Stimuli:
  • Target: A drifting square-wave grating (0.45 contrast, 12 orientations, 2 spatial frequencies).
  • Masker: A flashing white noise pattern (10 Hz) presented dichoptically to the opposite eye.
  • Conditions: Binocular, Monocular (baseline), CFS (dichoptic), and Masker-only.
• Data Analysis:
  • Ocular Dominance Index (ODI): Calculated for each neuron to categorize them into three groups: preferring the grating eye, preferring the masker eye, or binocular.
  • Population Tuning: Orientation tuning functions were fitted with Gaussian models to analyze amplitude and slope changes.
  • Information Theory: Fisher Information was calculated to quantify the precision of orientation encoding.
  • Machine Learning Decoding:
    • Classification: Support Vector Machines (SVM) were trained to classify 12 orientations (coarse discrimination).
    • Reconstruction: Transformer models were trained to reconstruct the grating image from neural activity (orientation recognition).
  • Modeling: An ocular-dominance-dependent gain control model was developed to simulate the observed suppression patterns.

3. Key Contributions

• Cellular Resolution Mapping: Provided the first large-scale, single-neuron resolution mapping of how CFS affects V1 orientation tuning, specifically distinguishing between neurons based on their ocular preferences.
• Quantification of Information Loss: Demonstrated that CFS does not merely reduce signal-to-noise ratio but fundamentally alters the population code, severely degrading Fisher information for orientations near the neuron's preference.
• Decoupling Discrimination from Recognition: Showed that while coarse orientation discrimination remains possible under CFS, the ability to reconstruct the stimulus (a proxy for high-level recognition) is severely compromised, especially when suppression is strong.
• Mechanistic Modeling: Proposed a gain control model incorporating interocular inhibition and binocular summation to explain the differential suppression across ocular dominance groups.

4. Key Results

A. Suppression of V1 Orientation Responses

• Global Suppression: CFS caused profound reductions in population orientation tuning.
  • Monkey A: Amplitude decreased by 84.18%, slope by 91.31%.
  • Monkey B: Amplitude decreased by 60.78%, slope by 71.50%.
• Ocular Dominance Dependence:
  • Masker-Eye Preferring Neurons: Orientation tuning was completely wiped out (flattened curves).
  • Binocular Neurons: Tuning was nearly completely abolished (Monkey A) or substantially abolished (Monkey B).
  • Grating-Eye Preferring Neurons: Tuning was the least affected but still significantly suppressed (Amplitude reduced by ~42-78%).

B. Fisher Information and Encoding Precision

• Fisher information (sensitivity to small orientation changes) dropped significantly under CFS.
• For orientations within 15° of the preferred direction, Fisher information fell to 29.1% (Monkey A) and 43.4% (Monkey B) of baseline levels. This indicates a loss of precision in encoding orientation information.

C. Machine Learning Decoding Outcomes

• Orientation Classification (Coarse Discrimination): SVM classifiers maintained high accuracy under CFS (Monkey A: ~81.7%, Monkey B: ~90.4%), suggesting that coarse orientation discrimination is still supported by the residual neural signal.
• Stimulus Reconstruction (Recognition): Transformer models showed a stark contrast:
  • Monkey A (Strong Suppression): Reconstruction quality collapsed (SSIM dropped from ~0.52 to 0.16), indicating the stimulus could not be recognized.
  • Monkey B (Weaker Suppression): Reconstruction remained relatively intact (SSIM dropped from ~0.61 to 0.53).
• V2 Findings: Similar suppression patterns were observed in V2, with Fisher information dropping to 33.1% of baseline in Monkey A.

D. Modeling

The proposed gain control model successfully replicated the experimental data ($R^2 > 0.98$), suggesting that CFS acts via a combination of interocular inhibition (reducing amplitude) and binocular summation (elevating baseline noise), with the strength of inhibition dependent on the neuron's ocular dominance.

5. Significance and Implications

• Limits of Subconscious Processing: The study suggests that while CFS allows for low-level feature discrimination (e.g., distinguishing "vertical" from "horizontal"), the degradation of V1 population coding is often too severe to support high-level visual recognition (e.g., identifying specific objects or semantic categories) unless the suppression is weak.
• Re-evaluating CFS Literature: The findings provide a neural basis for why some studies report high-level CFS effects while others do not. High-level effects may rely on residual low-level feature differences (e.g., elongated shapes) rather than intact category representations.
• Individual Variability: The study highlights significant inter-individual variability in suppression strength (Monkey A vs. B), which may explain heterogeneity in human CFS behavioral studies.
• Dorsal-Ventral Hypothesis: The results constrain the "dorsal-ventral dissociation" hypothesis. If V1 input is too degraded to reconstruct the stimulus, downstream ventral stream areas likely lack the necessary information for complex category processing, regardless of pathway specificity.

In conclusion, CFS severely compromises the fidelity of orientation information in V1 in an ocular-dominance-dependent manner. While this residual signal supports coarse discrimination, it is often insufficient for the stimulus reconstruction required for high-level visual and cognitive processing.