Temporo-nasally biased moving grating selectivity in mouse primary visual cortex

Orientation tuning in mouse primary visual cortex (V1) has long been reported to have a random or “salt-and-pepper” organisation, lacking the structure found in cats and primates. Laminar in-vivo multi-electrode array recordings here reveal previously elusive structure in the representation of visual patterns in the mouse visual cortex, with temporo-nasally drifting gratings eliciting consistently highest neuronal responses across cortical layers and columns, whilst upward moving gratings reliably evoked the lowest activities. We suggest this bias in direction selectivity to be behaviourally relevant as objects moving into the visual field from the side or behind may pose a predatory threat to the mouse whereas upward moving objects do not. We found furthermore that direction preference and selectivity was affected by stimulus spatial frequency, and that spatial and directional tuning curves showed high signal correlations decreasing with distance between recording sites. In addition, we show that despite this bias in direction selectivity, it is possible to decode stimulus identity and that spatiotemporal features achieve higher accuracy in the decoding task whereas spike count or population counts are sufficient to decode spatial frequencies implying different encoding strategies. Significance statement We show that temporo-nasally drifting gratings (i.e. opposite the normal visual flow during forward movement) reliably elicit the highest neural activity in mouse primary visual cortex, whereas upward moving gratings reliably evoke the lowest responses. This encoding may be highly behaviourally relevant, as objects approaching from the periphery may pose a threat (e.g. predators), whereas upward moving objects do not. This is a result at odds with the belief that mouse primary visual cortex is randomly organised. Further to this biased representation, we show that direction tuning depends on the underlying spatial frequency and that tuning preference is spatially correlated both across layers and columns and decreases with cortical distance, providing evidence for structural organisation in mouse primary visual cortex.

orientation preference in mouse V1 was revisited, with some authors detecting more structure than 47 hitherto reported (Kondo, Yoshida, and Ohki 2016;Ringach, Mineault, Tring, Olivas, Garcia-Junco-48 Clemente, and Trachtenberg 2016), with preferred orientations being found to show some degree of 49 clustering both on cross-columnar and laminar scales. The median similarity in tuning preference 50 was found to be significantly higher between neurons in close proximity (<100 µm) than for neurons 51 separated by 200 µm (Ringach, Mineault, Tring, Olivas, Garcia-Junco-Clemente, and Trachtenberg 52 2016). Correspondingly, we hypothesise that orientation tuning preference and selectivity in mouse 53 V1 is more correlated across layers and cortical columns than suggested by the salt-and-pepper tenet.

54
To test this, we analysed multi-unit activity (MUA, the aggregate spiking of a local population 55 of cells, recorded by thresholding the signal on a single electrode site, but not spike-sorting), to 56 examine both cross-laminar and cross-columnar arrangement of signals. Using multi-shank laminar 57 electrodes, we recorded signals from electrode sites at different locations spaced 100 µm apart 58 in depth, and 200 µm laterally, showing activity across layers and columns, effectively creating 59 an electrophysiological "image" of the signals in the tissue, with the "pixel" size approximately 60 corresponding to the spatial scale on which the MUA signal changes. Because of this fine-scale 61 random structure (Ohki, Chung, Ch'ng, Kara, and Reid 2005), it might be expected that the MUA 62 signal in mouse visual cortex contains little information about the spatial structure of a stimulus 63 beyond retinotopy. 64 Here, we studied how visual information is processed in mouse V1. Analysis of the board and photo-sensor (LCM555CN), attached to the bottom left corner of the monitor. A small 139 rectangular field flashing at stimulus on-and offset was used as a synchronization pulse. 140 We high-pass filtered the electrophysiological data, and thresholded it at 4 standard deviations, 141 to obtain the signals. Direction tuning of sites was evaluated using the sum of two modified von 142 Mises functions (Gao, DeAngelis, and Burkhalter 2010;Swindale 1998;Gatto and Jammalamadaka 143 2007). SF tuning curves were fit with a Difference of Gaussiants (DoG) function (Grubb and 144 Thompson 2003;So and Shapley 1981;Rodieck 1965). From these fits we calculated each site's 145 peak and cut-off spatial frequency, as well as preferred direction. Goodness of fit was estimated with 146 the coefficient of determination, R 2 = 1 − sse/sstotal, with sse denoting the sum of squared errors, 147 and sstotal = (n − 1)var(x) the total variation. Sites with R 2 <= 0.9 for both SF and directional fit 148 were discarded from further tuning analysis, which were 14 out of 384 sites at 0.01 cpd (cf. Results).

149
For population tuning curve calculations, the site multi-unit firing rates (responses) were 150 normalised across directions and SFs to fall between 0 and 1 for each repetition, enabling us to 151 compare across channels while accounting for slow temporal changes in excitability and different site 152 firing rates. Tuning curves were estimated by fitting the trial-averaged responses for each direction 153 or SF. We then compared the fitted tuning functions across sites by calculating the pairwise Pearson 154 correlation coefficient (r signal ), as well as the noise correlation (r noise ), estimated as the Pearson 155 correlation of deviations of each trial from the mean response for that direction. Direction and 156 orientation selectivity were calculated using the Direction Selectivity Index (DSI) and Orientation 157 Selectivity Index (OSI) as described in (Mazurek, Kager, and Van Hooser 2014).
with R pre f as the preferred direction, R null the opposite direction, and R ortho± denoting the orthogonal As DSI and OSI are known to be positively biased (Mazurek, Kager, and Van Hooser 2014), we also 161 calculated measures related to circular variance L osi and L dir proposed by Mazurek et al.(Mazurek,162 Kager, and Van Hooser 2014): where circular variance is CirV ar = 1 − L OSI . Correspondingly, a quantity related to circular 164 variance in direction space via DirCircV ar = 1 − L dir , scaled to fall between 0 and 1 as in Mazurek 165 et al.: To investigate how stimulus information is encoded in the neural signal, and in particular,

167
which features tell us more about certain aspects of the stimuli, three neural features were evaluated.

168
As a basis, spikes of all sites were binned at 5 ms, binarised if multiple spikes occurred in one bin, 169 creating a spatiotemporal matrix M of size 32x200 for each 1 s trial. Then, we used the following Data and code of this study are available from the corresponding author upon reasonable request.

192
Some of the data reported here has been previously described in (Tolkiehn and Schultz 2015). The 193 current paper incorporates additional data and expands upon the analysis of the data.

195
We recorded multi-unit activity from the left primary visual cortex of 12 isoflurane-anaesthetised 196 mice. Using 4-shank silicon microelectrodes with 8 linearly arranged recording sites per shank 197 at 100 µm site and 200 µm shank spacing, activity was recorded from sites spanning the cortical 198 laminar depth. Visual stimuli consisting of monocular, full-field drifting gratings were presented to 199 the right eye ( Fig. 1 (A), and Methods). Visual response properties including directional or spatial 200 tuning were characterised and compared among sites in order to investigate visual information 201 processing across cortical layers.

202
Activity in V1 is highly modulated by drifting gratings 203 Multi-unit sites were strongly driven by the moving gratings, as illustrated in Fig. 1  repetitions. We rejected sites from further analyses if the R 2 of the tuning function fit fell below 0.9.

208
The number of sites rejected in this manner was dependent upon the spatial frequency presented. at each SF and found that the distributions of mean response maxima and minima were distinctly 249 affected by SF (Fig. 2 A, B). This directional bias towards leftward moving gratings was also found in the preferred 268 directions obtained by taking the peak of the von-Mises function fits, illustrated in Fig. 3 (D-E).

269
Plotting the preferred directions against Direction Selectivity Index (DSI) and Orientation Selectivity 270 Index (OSI) further illustrated this bias, showing it occurred in sites ranging from low to high 271 direction (D) and orientation (E) selectivity. Direction selectivity was further explored with the 272 more conservative stimulus selectivity estimates L osi and L dir (Fig. 2 (F-G)). The selectivity indices indicating opposing tuning curves (Fig. 3 (B)). Noise correlations were generally significantly lower 300 (one-sided Mann-Whitney U test, Fig. 3  (signal) correlation between direction tuning functions decreased with distance between sites ( Fig. 3   312 (C)), where distance was inferred by probe geometry and discretised into 100 µm bins.

313
These findings were consonant with the normalised DoG tuning functions for SFs, demon-314 strating high similarity over sites and shanks (Fig. 3 (D)). Correlation coefficients for SF tunings 315 appeared all high with a centre of mass near a correlation coefficient of 1, as evident from Fig. 3 (E).

321
This was also confirmed by the average Pearson (signal) correlations between spatial tuning 322 functions, Fig. 3 (F). Yet, the spatial tuning similarity decreased more slowly than the tuning 323 similarity of direction tuning with distance.

324
High decoding success in spatial frequency and direction decoding 325 Given the high correlation between both SF and direction tuning and the bias towards leftward 326 forward moving gratings particularly for low SFs, we investigated if and how well different features 327 of multi-unit activity could be used to decode the stimulus features.

328
To do that, we devised two multivariate decoding tasks using Naive Bayes decoder: a) SF Trachtenberg 2016) who found that the direction tuning similarity decreased with cortical distance, 380 this study presents a high tuning stimilarity between sites (both SF and direction) across different 381 locations, that was also found to decrease with cortical distance, in contrast to studies that did not 382 indicate clustering of orientation preference (Ohki, Chung, Ch'ng, Kara, and Reid 2005). In addition, 383 we found an overrepresentation of preferred responses to leftward moving gratings (equivalent to 384 an object moving from right temporal to nasal visual field) in left V1. We showed that preferred