TL;DR
This study directly compares brain imaging (fMRI) and direct neural recordings (electrophysiology) in macaques to understand what the fMRI signal actually represents. They found that fMRI-based "population receptive fields" (pRFs) most closely mirror the receptive fields measured from multi-unit spiking activity, suggesting fMRI reliably captures neuronal receptive field properties.
What Problem Does This Paper Address?
Functional Magnetic Resonance Imaging (fMRI) is a powerful tool for mapping brain activity, particularly the visual cortex. A common technique, "population receptive field" (pRF) modeling, estimates the area of the visual field that elicits a response from a given voxel. While these fMRI-derived pRFs look similar to receptive fields measured directly from single neurons in animals, it's been unclear what neuronal signal fMRI is actually reflecting. Is it the average spiking activity of neurons, or perhaps the slower, local field potential (LFP) signals? This ambiguity makes interpreting fMRI results challenging, especially when comparing findings across different studies or species. This paper aims to bridge this gap by directly comparing fMRI pRFs with various electrophysiological signals in the same animals.
Methods
The researchers used a multi-pronged approach:
- fMRI: They performed whole-brain fMRI on awake, fixating macaques while presenting moving bar stimuli. They then applied four different pRF models (linear, linear with negative responses allowed, difference-of-Gaussians, and compressive spatial summation) to the fMRI data.
- Electrophysiology: In separate experiments, they recorded from multi-unit spiking activity (MUA) and local field potentials (LFPs) in different frequency bands (theta, alpha, beta, low gamma, high gamma) from electrodes implanted in the visual cortex (V1 and V4) of other macaques. They also applied the same pRF models to these electrophysiological signals.
- Comparison: The core of the study involved directly comparing the pRFs derived from fMRI (BOLD signal) with those from MUA and LFP signals. They analyzed similarities in pRF size, location, and spatial summation properties.
Datasets
The study used data collected from four macaque monkeys. Two animals underwent fMRI scans, and two others were implanted with electrode arrays for electrophysiological recordings. The data collected included:
- Whole-brain fMRI BOLD time-series.
- Multi-unit spiking activity (MUA) recordings.
- Local field potential (LFP) power in five frequency bands.
The study did not introduce new public datasets but rather generated novel comparative data by combining fMRI and electrophysiology within the same species.
Key Insights and Ideas
- fMRI reflects spiking activity: The most striking finding is that fMRI-derived pRFs closely matched those derived from MUA, suggesting that the BOLD signal primarily reflects the aggregate spiking activity of neurons, rather than slower LFP signals.
- Subadditive spatial summation is widespread: The "compressive spatial summation" (CSS) model, which accounts for subadditive summation (where the response to combined stimuli is less than the sum of individual responses), provided the best fits for both fMRI and electrophysiological data across many brain areas. This indicates a fundamental property of neuronal processing.
- Negative pRFs reveal functional organization: The study identified negative pRFs (where stimulus presentation suppresses activity) in the default mode network (DMN) and as part of surround suppression in early visual areas. This highlights the role of inhibition in shaping neural responses.
- Retinotopy extends beyond visual cortex: The whole-brain fMRI data revealed retinotopic organization in a surprisingly wide range of cortical and subcortical areas, including frontal and subcortical regions, suggesting a more distributed visual processing network than previously appreciated.
Key Results
- MUA-pRFs are most similar to BOLD-pRFs: pRFs derived from BOLD fMRI were most similar to MUA-pRFs in V1 and V4. LFP gamma power also provided a good approximation.
- CSS model outperforms others: The compressive spatial summation (CSS) model consistently provided better fits than simpler linear models for both fMRI and electrophysiological data.
- pRF size increases with eccentricity: As expected, pRFs generally became larger further away from the fixation point, a trend observed across most brain areas.
- Negative pRFs in DMN: Negative pRFs were observed in areas associated with the default mode network, consistent with findings in humans.
- Subcortical retinotopy: Retinotopic maps were successfully identified in the LGN and pulvinar, though with more variability than in cortical areas.
Limitations and Future Work
- Stimulus limitations: The study acknowledges that checkerboard stimuli might not be ideal for driving responses in all frontal areas, which are better driven by more complex stimuli.
- V4 heterogeneity: Differences in the eccentricity-size relationship between V1 and V4 might be due to V4 being smaller and more heterogeneous, leading to voxels sampling from a larger, more varied neuronal population.
- Future work on negative pRFs: The authors suggest further investigation into the phenomenon of negative pRFs, particularly their potential link to attention shifts and their role in the alpha and beta LFP bands.
- Brain-wide retinotopy: The unexpected observation of retinotopy in higher areas like the anterior cingulate cortex warrants further detailed investigation into brain-wide retinotopic organization.
- Electrode placement: In some V1 arrays, unusually small pRFs suggested electrodes might have been in white matter, picking up thalamic afferents. These were excluded from size analyses.
Summary updated