Ethical approval
Animal surgery procedures and electrophysiological recordings followed the guidelines of the Canadian Council on Animal Care and were approved by the Institutional Animal Care and Use Committee of the University of Montreal. Animals were supplied by the Division of Animal Resources of the University of Montreal. The experiments were conducted in accordance with the Guide for Care and Use of Laboratory Animals of the National Institutes of Health (USA).
Animal surgery
Briefly, electrophysiological recordings and optical imaging were performed within layer II/III of V1 area of adult anaesthetized cats (Felis catus). Eight adult cats (2.5–3.5 kg, age 12–24 months) of either sex were used for this study. General anaesthesia was maintained by artificial ventilation with a mixture of N2O/O2 (70:30) supplemented with 0.5 % isoflurane (AErrane, Baxter, Toronto, ON, Canada) for the duration of the experiment. The following parameters were monitored throughout the experiment: the EEG, the expired CO2, the temperature and the heart rate. At the end of each experiment, euthanasia was achieved with a lethal dose of pentobarbital sodium (Somnotol, MTC Pharmaceuticals, Cambridge, ON, Canada; 100 mg kg−1) by intravenous injection. Details are described in Bachatene et al. [5].
Electrophysiology
Visual stimuli were generated with a VSG 2/5 graphic board (Cambridge Research Systems, Rochester, England) and displayed on a 21-in. monitor (Sony GDM-F520 Trinitron, Tokyo, Japan) placed 57 cm from the cat’s eyes, with 1024 × 768 pixels, running at 100 Hz frame refresh. Stimuli were drifting sine-wave grating square patches (~2°–5°) covering the excitatory RF (unidirectional movement). Patches characteristics were set to evoke optimal responses: contrast at 80 %, mean luminance at 40 cd/m2, optimal spatial and temporal frequencies set within the 0.1–0.5 cycles/deg. and 1.0–2.0 Hz range, respectively. In all cases the above parameters were chosen with the aim of evoking the maximal discharges. After manual RF characterization, nine oriented stimuli centered on the preferred orientation were selected and used for the rest of the experiment. Test orientations were applied in random order. Each oriented stimulus was presented in blocks of 25 trials lasting 4.1 s each, with a random inter-trial interval (1.0–3.0 s) during which no stimuli were presented. Thus, a recording session lasted for 25–30 min. Peri-stimulus time histograms were recorded. Once control orientation tuning curves were characterized, an adapting non-preferred stimulus was presented continuously for 3 or 12 min and 24 min in one experiment. The adapting stimulus was a drifting grating whose orientation was randomly selected in the range 22.5°–67.5° off of the neuron’s preferred orientation. All other stimulus parameters were kept constant, at control values, throughout the recordings. Neurons were isolated from multi-unit activity using autocorrelograms, principal component analysis, spike wave-shapes and cluster separation. Details are described in Bachatene et al. [5].
Optical imaging
Detailed account of intrinsic optical imaging is available in Cattan et al. [12]. Intrinsic optical imaging allows assessment of the activity of a large population of cells. Thus, we used this technique to visualize the range of shift propagation following adaptation.
After craniotomy, the dura mater was removed, a round chamber (15 mm in diameter) was fixed with dental cement above one hemisphere’s area 17 and the chamber was filled with mineral oil and closed with a cover glass.
Achromatic gratings were to stimulate all cortical area within the imaging window to obtain orientation maps in control and post-adaptation sessions (contrast: 75 %; generated by VSG software; Cambridge Research Systems, Rochester, UK), presented randomly in order to avoid stimulus-order bias with rotation in eight different orientations from 0° to 157.5°, spatial frequency: 0.3 cycles/°, temporal frequency: 1 Hz). Each trial started with the presentation of a black screen for 15 s, and this was followed by the presentation of every orientation (12 s). For each presented orientation spanning 12 s, the grating was kept stationary during the first 6 s to remove the cortical activity resulting from the stimulus onset, and this was followed by drifts in one direction for the next 3 s and then in the reverse direction for 3 s to maximize cortical responses. The stimulation was full-screen. From these recordings, we generated control polar orientation maps. Then we presented a patch as an adapter for 12 min (full screen stimulus). Immediately after adaptation, we stimulated again all cortical area within the imaging window by presenting full screen stimuli as in the control phase. The polar map obtained post-adaptation was compared with the control map to evaluate the spatial spread of local adaptation.
Data acquisition and processing
The cortex was illuminated with 630-nm light. Cortex images were captured with a CCD camera (Dalsa 1 M60P; Dalsa, Waterloo, Ontario, Canada), composed of two 50-mm f1.2 lenses arranged in tandem, and focused 500 µm below the cortical surface. Images were digitised with Imager 3001 (Optical Imaging, Germantown, NY, USA), with a spatial resolution of 1024 × 1024 pixels (binned 2 × 2), and a temporal resolution of 20-ms frame duration. Image analysis was performed with MATLAB programs (MathWorks, Natick, MA, USA).
Thirty images were recorded for every orientation. As the last images showed more activity than the initial images, the average of the last 10 images (21–30) was divided by the average of the first 20 images (1–20). This calculation was performed to remove the non-specific activity in initial images, while preserving the specific activity recorded mostly in the last frames. Then, the generalised indicator function method [22] was applied. In short, this method extracts the frames that account for as much of the signal as possible by using principal component analysis, and optimises the differences between signal and noise.
Pixel shifts
To quantify changes in the orientation of pixels between control and post-adaptation polar maps, the amplitude of the shift in orientation was calculated from pairs of pixels located at the same position in the two maps.
$$ s_{i,j} = { \hbox{min} }\left( {\left| {p2_{i,j} - p1_{i,j} } \right|,180 - \left| {p2_{i,j} - p1_{i,j} } \right|} \right) $$
where s is the shift-amplitude associated with the pair of pixels at the map position (i, j); p1 the pixel-orientation in the first map and, p2 the orientation of the same pixel in the second map. The shift-map comprised of all the shifts calculated for each position.
Connectivity strength, crosscorrelograms and shift predictor
Crosscorrelograms (CCG’s) were performed in order to compare the spike distribution of each neuron of the pair within a time-frame; one neuron is set as reference and the second as target; this allows us to show the firing of the target neuron at a specific time-spread in relation to the firing of the reference neuron.
Time axis is divided into bins. The first bin is defined as: XMin, XMin + Bin. The next bin is XMin + Bin, XMin + Bin*2, etc. We calculated the distances from each spike to all spikes of the spike train as follows:
$$ d\left[ i \right] = ts\left[ i \right] - ref\left[ k \right] $$
where ts[i] represents the spike train, and ref[k] is each timestamp.
Bin counts were then divided by the number of reference events to normalize the counts per bin into probabilities. 95 % statistical threshold for the significance of the bins was used. Each bin-width was set at 1 ms. The connectivity strengths were calculated from the counts/bin as follows:
where F is the neuron frequency and b represents the bin size of the calculated firing of the neuron.
The neuron frequency F was calculated as follows:
where T represents the total time interval and N the number of spikes within this interval.
The 95 % confidence limit was calculated assuming that the expected bin count (EBC) has a Poisson distribution:
$$ EBC = CS \times Nref $$
where Nref is the number of reference events.
The 95 % confidence limit is calculated as follows:
$$ Low\,Conf. = x \,such\, that \,Prob \left( {S < x} \right) = 0.005 $$
$$ High\, Conf. = y\, such \,that \,Prob \left( {S > y} \right) = 0.005 $$
where S represents a random variable which has a Poisson distribution with parameter EBC.
Raw CCG’s were corrected by subtracting a shift-predictor algorithm in order to eliminate the putative significant peaks due to the simultaneous stimulation of both cells during each trial.