Ripples in macaque V1 and V4 are modulated by top-down visual attention

Significance Short-lived, high-frequency episodes in the electrical activities of mammalian hippocampus, called ripples, occur during sleep and quiescence, and are linked to memory formation. To which extent they occur during active behaviors outside hippocampus is poorly understood. We detected ripples in visual areas V1 and V4 of macaque monkeys during top-down spatial attention. Ripples occurred at frequencies not dissimilar to rates reported in the hippocampus. Critically, the ripple rate was modulated by attention cued to the receptive field, by stimulus size, and by the size of the attentional focus. Moreover, reaction times were modestly reduced on trials where ripples occurred.

( ) = . µα+( σ 2 α 2 2 )−α . ( , µ + σ 2 α , σ) + . ( , µ , σ) The parameters of µ and σ are the mean and standard deviation of the cumulative Gaussian, which are determined by onset response time. The parameter α is the dissipating rate and c and d act as weighting factors for the response magnitude and dissipation terms. For our data, we acquired 150 ms of MUAE signal after stimulus onset, which was z-scored to baseline activity (before stimulus onset). The function f(t) was fit to the z-scored activity. The latency of visual responses (lat33) was determined as a point in time where the fitted function reached 33% of its maximum (1, 2).

Current source density (CSD) analysis
We used the inverse CSD (iCSD) toolbox to compute CSDs. CSD analysis was calculated by applying the spline method (3). LFPs were averaged over trial repetitions. Then, the iCSD, which is the second spatial derivative voltage (Ф) of LFPs, can be approximated using the following formula: ( ) = -σ Ф( + ℎ) − 2Ф( ) + Ф ( − ℎ) ℎ 2 Where z is the depth at which the CSD is estimated, and h is the space between two electrodes (150µm). With this toolbox we used a Spline fitting method to interpolate Ф smoothly between electrode contacts. In our computation we assumed a tissue conductivity (σ) and a cortical column radius of 0.4 s/m and 500µm respectively (4,5).

Laminar alignment
The CSD was used for alignment of probe contacts to the cortical layers with reference to layer IVc. Previous studies (6)(7)(8) established that an early sink in the CSD profile corresponds to input layers, with associated current inflow. First, we computed the CSD across all stimulus presentations of each session (Supplementary figure S7A) and visually determined the contact that featured the early sink (35-55 ms after stimulus onset; Supplementary figure S7B). Additionally, we determined the channel/contact with the shortest latency stimulus-evoked response MUAE response (details in supplementary materials and supplementary figure S7C, D). Using these criteria, we assigned the reference contact as input layers (presumably layer IVc in V1, and layer IV in V4) and signals from other contacts were assigned to superficial, input and deep layers depending on their distance from the reference contact. For area V1, channels at 0.25 mm above and 0.25 mm below the reference channel were labelled as input layer (presumably layer IV), channels at 0.25 mm to 1 mm above the reference were labeled as superficial layer (presumably I, II, III) and channels below the reference channel at 0.25 mm to 0.75 mm were labeled as deep layer (presumably V, VI) (9, 10). For V4, contacts less than 0.1 mm above and below the reference contact were identified as input, 0.1 to 1 mm above the reference contact were identified as superficial, and those 0.1 mm to 0.75 mm below the reference channels were labelled as deep layers (Supplementary figure S7, C, D). Channels outside these ranges were excluded from further analyses.

Cross-correlation
The temporal relationship between ripples that occurred in V1 and V4 was explored using cross-correlation analysis. The cross-correlation (CC) was performed using the xcorr function in Matlab, according to: Where x and y are vectors representing ripple occurrence in area V1 and V4. To compute the crosscorrelation, the envelope of the LFP was extracted. For each trial, the LFP of the sustained period was filtered at ripple band (100-200 Hz, Butterworth, 4 th order) and rectified to convert negative values to positive. Using the Hilbert transform, the envelope of the signal was calculated and smoothed by a 4 th order Butterworth filter (1-20 Hz). The cross-correlation coefficient of V1 and V4 trials was computed and corrected by the shuffling of V4's trials. M and T denotes the number of trials and discrete time bins respectively and t represents the time lag. The area under the cross-correlation curve above the shuffle predictor was calculated for the left and the right part of the cross correlogram relative to time zero (AUC), and it was used to examine whether ripple occurrence between V1 and V4 was on average simultaneous, or whether it was systematically shifted in time for one of the two areas.

Ripple rate and RF overlap
To assess whether V1 and V4 RF overlap, and therefore stimulus placement, affected the ripple rate, we classified recording sessions into two categories. Namely, where V1 RFs were completely or almost completely covered by all of the V4 RFs (these are labelled as 'fully overlapped', examples in figure S9) and sessions where the overlap was less complete ('partial overlap', example in figure S9). Then, we quantified ripple rate for sessions with full and partial RF overlap. A one-way ANOVA revealed that there was no difference in ripple rate between sessions with full and partial overlap. This was the case for both V1 and V4 (F (1,732) = 1.57, p = 0.21, pooled data and figure S10 panel A). Furthermore, we explored whether ripple rate modulation by stimulus size would change depending on RF overlap between V1 and V4. The reasoning here is that if small stimuli are centered on V1 RFs, which do not fully overlap with V4 RFs, V4 RFs would not, or only partially be driven by the small stimulus. This would not be the case if V1 and V4 have full overlap. We explored the ripple rate among the fully overlapped sessions (n=18) in different task conditions. Corroborating the results reported in the main manuscript (where data were pooled across all session), small stimuli elicited a higher ripple than large stimuli in both areas. For these 18 recordings ripple rate for small and large stimuli were 0.15 ES and 0.09 ES in V1. In V4, rates were 0.11 ES and 0.05 ES for small and large stimuli (V1: Z = -4.2, p <0.001. V4: Z = -7.01, p<0.001, Wilcoxon's signed rank test, figure S10 panel B). In addition, attention to RF enhanced ripple rate when compared to cue away conditions among these sessions. Cue RF and away conditions in V1 triggered 0.13 ES and 0.10 ES respectively. In V4 cue RF conditions triggered ripple rates of 0.12 ES while cue away conditions elicited 0.05 ES (V1: Z = 2.08, p = 0.03. V4: Z = 8.11, p <0.001, figure S10 panel C). We conducted a 3-way repeated measures ANOVA, to determine main effects of stimuli, attentional location and attentional focus, and possible interactions on ripple rate. ANOVA revealed that stimulus size (p=0.005) and attention location (p=0.02) had main effect on ripple rate frequency in V1 data. There was a significant attention location*size interaction (p=0.04) and attention location*attentional focus interaction (p=0.02) in V1. The 3-way ANOVA in V4 showed that stimulus size (p=0.0009) and attention location (p<0.001) significantly increased ripple rate. We found a stimulus size and attention location interaction on ripple rate (p=0.0003) in V4.            Middle and bottom) Ripple rate plotted against RF eccentricity for V1 and V4 and cued RF and cued away conditions. RF eccentricity was calculated as the vector length distance between the fixation spot and the centre of the RF (described in 'receptive field mapping'). R-values (r) and associated significances (p) indicate Pearson correlations between the respective variables. Only for V4 a small systematic relationship between RF area and ripple rate was apparent.   Fig S.14. Firing rates associated with small (purple/pink) and large stimuli (green), as well as cue RF and cue away conditions in V1 and V4. In both areas, small stimuli elicited higher firing rates than large stimuli. In addition, cue RF conditions elicited higher firing rates than cue away conditions during the sustained stimulus period.

Fig. S15. Reaction times associated with wide and narrow foci of attention and with valid and invalid cueing.
During narrow blocks of attention (red bars and histograms) targets were presented in ~10% or ~7% (M1 or M2) of the trials at far positions (invalid; possible target locations indicated by black dots inside the grey circles which represent small and large stimuli respectively). Conversely during wide blocks of attention targets were presented in ~10% or ~7% (M1 or M2) of the trials at the centre position (invalid). Thus, we would expect for a given target location that validly cued targets yield shorter reaction times than invalidly cued targets. While this was the case in both monkeys if the targets appeared at the centre, we only found that pattern for far targets in monkey 1. In monkey 1 narrow foci of attention always yielded faster reaction times for a given target location, compared to wide foci of attention. Thus, we found a significant main effect of focus and position, but no interaction on reaction times in that monkey (2-factor ANOVA). However, in monkey 2 we found a significant interaction between target position and focus of attention, in line with both predictions.  3-way repeated measures ANOVA to determine whether ripple rate in area V1 during the sustained period depended on stimulus size (small/large), locus of attention (cue RF/away) and attentional focus size (narrow/wide). Table denotes sum of squares (SS), degrees of freedom (df), mean squares (MS), F-value (F), and p-value (P). Interaction terms are indicated by * sign. 3-way repeated measures ANOVA to determine whether ripple rate in area V4 during the sustained period depended on stimulus size (small/large), locus of attention (cue RF/away) and attentional focus size (narrow/wide). Table denotes sum of squares (SS), degrees of freedom (df), mean squares (MS), F-value (F), and p-value (P). Interaction terms are indicated by * sign. Linear regression model to predict V1 firing rate from occurrence of ripple, attention to RF vs away, size of focus of attention (narrow/wide), and size of stimulus (small/large). The digits in front the predictors represent the categories assigned to fixed effects into the model. Ripple (0 is without and 1 with ripple), attention (1 attention to RF 2 and away), focus (1 narrow and 2 wide focus of attention) and size (1 large and 2 small stimuli). The column of β reports the coefficients of each predictor estimated by model and column T is the corresponding t-statistic. Asterisks indicate significance.  Using a linear regression model to predict V4 firing rate from occurrence of ripple, attention to RF vs away, size of focus of attention (narrow/wide), and size of stimulus (small/large). The digits in front the predictors represent the categories assigned to fixed effects into the model. Ripple (0 is without and 1 with ripple), attention (1 attention to RF 2 and away), focus (1 narrow and 2 wide focus of attention) and size (1 large and 2 small stimuli). The column of β reports the coefficients of each predictor estimated by model and column T is corresponding t-statistic. Asterisks indicate significance.