Spectral decomposition of local field potentials uncovers frequency-tuned gain modulation of working memory in primate visual system

Reviewer #1 (Public review):

Working memory affects sensory processing. Observers make faster and more accurate perceptual decisions at remembered locations, and corresponding regions of retinotopic visual cortex display enhanced response gain and modulations in oscillatory activity and spike-phase coupling.

Roshanaei et al investigate the relationship between working memory, oscillatory activity, and response gain by reanalyzing extracellular laminar probe recordings from area MT of rhesus monkeys performing a spatial working memory task. During the memory period, visual probes were flashed in the receptive field of the recorded neurons, allowing a comparison of visual responses when memory overlapped with this receptive field (IN) or a location in the opposite hemifield (OUT). They first replicate a range of findings, including increased power in lower frequency bands (theta and alpha/beta) and increased visually-evoked responses in the IN condition. The authors next deployed a spectral technique (MODWT) to decompose the local field potential on single trials into 6 non-arbitrary component frequency bands. This approach allows the authors to observe shifts in peak spectral frequencies across IN and OUT trials. Finally, these single-trial spectral decompositions allowed the authors to relate frequency band power and response gain. This analysis revealed that response gain tended to increase with power in lower (alpha, beta, and theta) frequency bands, and this effect minimally interacted with the remembered location.

Together, these interesting results provide correlational evidence that the effect of working memory on response gain may be mediated by oscillatory power. As the authors note, these results are also consistent with theories positing that lower frequency oscillatory activity primarily reflects working-memory related feedback signals from prefrontal and parietal cortex.

These findings also suggest opportunities for further exploration. From a methodological perspective, it's not clear if the particular spectral decomposition highlighted here is necessary for obtaining these results, or if applying more standard approaches to single trials (as in Lundqvist et al., 2016) would have provided similar sensitivity. Additionally, although the relationship among working memory, oscillatory power, and response gain explored here is necessarily correlational, it could be of interest to subject these factors to a mediation analysis in this or future studies. Finally, the careful analysis of oscillatory phenomena reported here can ideally be used to inform large-scale circuit models and constrain the underlying mechanism.

Reviewer #2 (Public review):

Summary:

Roshanaei et al investigate how working memory (WM) modulates neural activity in the primate visual system by examining local field potentials (LFPs) and spiking activity recorded in area MT. This work is an extension and the reuse of the dataset of the group's prior manuscript, Bahmani et al, Neuron 2018. The animals perform a spatial working memory task where they need to remember the location of a probe stimulus presented within (IN condition) or outside (OUT condition) the neuron's mapped receptive field (RF).

As the first step, the authors replicate the findings in their Neuron 2018 paper by showing:
(1) Significant modulation of the LFP power in αβ band during the working memory period in IN vs OUT conditions. This effect was absent in the gamma band.
(2) A significant increase in phase-coded mutual information for probe location for the IN condition compared to the OUT condition.

The authors then apply the Maximal Overlap Discrete Wavelet Transform (MODWT) to decompose LFP signals at the single-trial level, an approach that allows them to identify oscillatory components without imposing pre-defined frequency bands. They find that the precise frequencies of low-frequency oscillations (theta, alpha, and beta) correlate with the visually evoked firing rates of MT neurons.

Strengths:

The work addresses an important question: how cognitive states such as working memory modulate sensory processing in the visual cortex. More specifically, as we are expanding our understanding of the role of feedback in the brain, a me role of oscillations.

The application of MODWT to single-trial LFPs represents a methodological advance over traditional bandpass filtering, which typically relies on trial-averaged power and may miss fine-grained frequency variability.

The work aligns with ongoing efforts to understand how feedback and oscillatory dynamics contribute to top-down modulation in the brain.

Weaknesses:

(1) Several early results (e.g., increases in alpha/beta power and phase coding) closely replicate previous work from the same group and may be better placed in the Supplementary Information or omitted entirely. The novelty of the current paper lies mainly in the single-trial decomposition and frequency-rate relationship. However, the manuscript fails to expand the prior findings using the traditional methods, or at least offer a more mechanistic insight into the role of top-down modulation of the MT area during working memory tasks. Single-trial analysis can offer new avenues for mechanistic insight. For example, authors could have investigated the relationship of Cross-frequency coupling (CFC) with trial-by-trial behavior of the animal (Voytek et al., 2010) or transient synchronous oscillations for memory maintenance (Buschman et al, 2012).

(2) The statistical methods require greater transparency. Details such as whether tests were one- or two-sided, how multiple comparisons were controlled, and how correlations among nearby electrodes were handled are not fully reported.