Reward evoked plasticity in the mouse primary visual
cortex
Guido T. Meijer
Cognitive and Systems Neuroscience, Center for Neuroscience, Swammerdam Institute, Cognitive Science Center Amsterdam, Faculty of Science, University of Amsterdam
Behaviorally important stimuli can induce plasticity in primary sensory areas of the neocortex. Cortical representations can change in size in the primary auditory cortex and the neurons in the primary visual cortex (V1) can adjust their tuning curves when exposed to a stimulus that is coupled with reward. Using this concept we created a novel methodological approach to induce plasticity that can be measured in V1 of the mouse cortex. Already after three weeks of training, changes can be observed in V1. The mouse is exposed to two stimuli, which are presented one at a time. Both gratings moving in equal direction. The stimuli are aligned vertically adjacent to each other; reward delivery only follows one of the two stimuli (CS+) but not the other (CS-). After the training procedure the subset of neurons that are closely tuned to the CS+ direction respond more strongly to either one of the two stimuli; they have increased their stimulus-selectivity significantly.
Introduction
In the last few decades is has been accepted that corti-cal maps are dynamic and can be modified by experience. The representation of sensory stimuli continuously adapts itself according to the behavioral relevance of the encoded stimulus (Buonomano and Merzenich, 1998). One pos-sible effect of this neuronal adaptation is that the per-formance at which attributes of a stimulus can be distin-guished improves with practice, this effect is called per-ceptual learning. An example of perper-ceptual learning in the visual domain is that the perceptual threshold of detecting the orientation of a Gabor target improves with repeated exposure (Goldstone, 1998).
Perceptual learning is specific to the the location on the visual field in which the stimulus is shown and the eye that is used during training (Schoups et al., 1995; Crist et al., 1997). Since higher visual areas are more invariant towards stimulus position on the visual field and receive input from both eyes (Hubel and Wiesel, 1962), it is a sen-sible assumption that the neuronal processes that underly perceptual learning take place in the more earlier stages of visual processing (Fahle, 2004). Indeed, a vast body of evidence shows that perceptual learning results in chances that can be observed in the primary visual cortex of mice (Frenkel et al., 2006), rats (Hager and Dringenberg, 2010), cats (Hua et al., 2010), primates (Ghose et al., 2002) and also in humans (Schwartz et al., 2002; Bao et al., 2010).
Until recently, the theory of perceptual learning stated that the learning effects could only take place if the stim-ulus was attended and consciously observed. However, psychophysics studies have shown that perceptual learn-ing even has an effect for task-irrelevant and unattended stimuli, as long as they are paired with task relevant tar-gets (Watanabe et al., 2001; Seitz and Watanabe, 2005). A follow-up study used stimuli that were not only unat-tended but also rendered completely invisible to the sub-ject by flashing contour-rich patterns in one eye. Purely the coupling of a stimulus with a reward was sufficient to
induce perceptual learning (Seitz et al., 2009).
Animal studies provide a possibility to study the effect of reward processes on primary sensory cortices in more detail. Although poorly understood, advances are being made in the understanding of the effect of stimulus-reward pairing on the functioning of neuronal circuits in the sen-sory neocortex. It has been shown that the coupling of stimulation of the ventral tegmental area (VTA) of the rat with a specific tone renders a larger cortical representation in the primary auditory cortex for the conditioned tone (Bao et al., 2001). Concerning the visual cortex, Shuler and Bear (2006) have shown that there are individual V1 neurons that time their action potentials to the delivery of an appetitive reward. They found neurons that showed a sustained activation until reward delivery and neurons that peak in their activity during reward delivery. Also, a higher visual evoked potential has been found in primates related to a stimulus which has been coupled with a reward (Frankó et al., 2010). Goltstein et al. (subm.) showed that the appetitive conditioning of mice to moving gratings in a certain direction (CS+) results in a specific effect on a subset of V1 neurons which are direction selective (Hubel and Wiesel, 1959), namely the group of neurons that were closely tuned to the CS+ direction. These neurons showed broader orientation tuning and sharper direction tuning.
In this paper we present a novel behavioral paradigm that results in plasticity that can be observed in the V1 of the mouse in a very short time course of only several weeks. Moreover, the observed chances in V1 neurons tell us more about the underlying neuronal mechanisms con-cerning the influence of reward processes on the plasticity in the primary visual cortex.
Head-fixed mice were presented with two adjacent stimuli; one in the top and one in the bottom of their vi-sual field, both stimuli consisted of moving gratings in the same direction. Only one of the two stimuli was followed by a reward (CS+). Expected was that the neurons that have a receptive field that overlaps both stimulus
loca-Guido T. Meijer • Plasticity of the Visual Cortex 2
tions would be subjected to plasticity whilst the neurons that have a receptive field that fully overlaps either the CS+ area or the CS- area would remain unchanged. We found that after training the subset of neurons that had a preferred direction similar to the CS+ direction showed a greater specificity in their response to either one of the two stimuli. This indicates that the neurons that responded to both stimuli before the training started responding to ei-ther the CS+ or the CS- after the training procedure.
Materials and Methods
All experiments were conducted with approval of the an-imal experiment committee (DEC) of the University of Amsterdam.
Animals and surgical procedures. 16 male wild-type C57BL/6 mice were used in this study, ages ranging be-tween 70 and 116 days at experiment onset. Their day-night rhythm was reversed. Water access was unrestricted. Preceding training the animals were food deprived for 6 to 8 hours. The animals were weighed regularly to assure weight stability.
Analgesia (Temgesic, 0.05 mg/kg bodyweight) was in-jected subcutaneously 30 minutes prior to surgery. Ani-mals were anesthetized with an initial dose of 3% isoflu-rane in 97% O2, during surgery anesthesia was kept at 2% isoflurane and lowered to 1% towards the end of the pro-cedure. The skin on top of the skull was removed with sharp scissors and Xylocaine was applied as a local anes-thetic. The location of V1 was estimated at ~4 mm cau-dal and ~2.5 mm lateral from bregma. Using a dentist drill three holes were drilled in the skull; two in the ante-rior/left skullplate and one in the posteante-rior/left skullplate. Skullscrews were fastened in the holes and a custom built titanium headmount was cemented to the skull using black dental cement. The headmount enabled the head restric-tion of the animal during training. The exposed skull was covered with Quicksil on which a coverglass was put, the coverglass was fastened to the headmount using cyanoacrylate glue.
Visual stimuliTwo squares of moving square-wave grat-ings were used as stimuli, 30 retinal degrees in size. The two stimuli were presented vertically aligned and adjacent to each other in the center of the screen on a gray back-ground (Fig. 1b). The movement direction of the two stimuli (90◦, 180◦, 270◦or 360◦) was the same for both stimuli, they only differed in location. A movement direc-tion was arbitrarily selected for every mouse. The grat-ings had a frequency of 0.05 cycles/deg and moved with a speed of two cycles/second. A 20 inch screen with a re-fresh rate of 60 Hz was used. The screen was positioned orthogonally from the midline of the right eye at an es-timated angle of 45◦ from the direction that the animal was facing. The midpoint of the screen was at the same height as the eye of the animal. All stimuli were writ-ten and presented using MATLAB (The Mathworks, Nat-ick, MA, United States) and the Psychophysics Toolbox (http://www.psychtoolbox.org).
Handling and pre-trainingThe mice were handled for five days. In this period the mice were also fed the reward sub-stance (vanilla pudding) either by syringe and/or a con-tainer with pudding was put in their homecage so they could habituate to eating the substance. In this period the animals were also exposed to the training environment so they could habituate to this context. After the surgery and a recovery period of four days the mice were pre-trained for four days. The pre-training consisted of head-fixation in a cranial window holder for 15 minutes facing a grey screen with reward freely accessible.
ConditioningThe animals were presented with two stim-uli; the CS+ and the CS-. The stimuli were identical and differed only in location on the screen. The CS+ was al-ways followed by a reward and the CS- never. Stimuli were shown for five seconds with a 20 second inter trial in-terval (ITI). During the ITI reward was delivered through a plastic tube that was moved towards the animal by a servo. One second after stimulus off-set the reward arm started moving and enabled the mouse to eat for five seconds, per trial the animal got ~20 µl of reward substance. Following CS- trials the reward arm would move to 1 cm in front of the mouse. The pump that pushed the reward substance through the tube would activate for half a second at every trial onset as to ensure that only the visual stimuli were predictive for reward.
The animals were conditioned for ten days doing 60 trials per day (Fig. 1d). CS+ and CS- trials were presented equally and pseudo-randomly. It was ensured that no more that three consecutive CS+ or CS- trials were presented in succession. A CCD camera (25 fr/sec) recorded the ani-mal’s behavior from the side. A red LED was lit during stimulus presentation to synchronize the videos with stim-ulus presentation (Fig. 1a).
Eye-tracking Because of the use of retinotopic relevant stimuli the position of the eye had to be monitored to check whether the animals exhibited large pupil displace-ments. In general head-fixed mice keep their gaze focused on the same point in the visual field (Andermann et al., 2010; Van Alphen et al., 2001). The frequency of sac-cades is low in mice (7.5 min−1), especially in the verti-cal direction (Sakatani and Isa, 2007). However, the eye movements of mice have never been quantified during a task that resulted in an attribution of value to a retinotopi-cally specific stimulus. A result of this conditioning could be that mice made more eye movements towards the CS+. The head of the mouse was recorded using a near infra-red CCD camera (CV-A50 IR, JAI) with a sampling rate of 25 frames/sec. The camera was placed 60◦ lat-eral to the body axis, looking down from an angle of 45◦. Using a high-magnification lens (MVL50M23, Navitar, Rochester, NY), with a minimum object distance of 200 mm and a field of view of 7.0◦- 5.2◦ , a high-definition video could be acquired of the head of the mouse (Fig. 1c).
Intrinsic optical imaging IOS measurements were made on three points in the experiment; before conditioning, af-ter five days and afaf-ter conditioning. Animals were kept
Figure 1: a, schematic drawing of the behavioral set-up used to condition the animals. Three cameras were used to record the animal’s behavior, reward was pumped through a tube and delivered using a servo. All components of the set-up were controlled by MatLab over (virtual) serial ports. b, the stimuli that were used, both CS+ and CS- were moving square wave gratings in equal direction. c, a frame taken from the video feed of the eye-tracking camera. d, timeline of the entire experiment, a two-photon measurement takes up a day per animal, the rest of the animals are still being conditioned during this time.
under light anesthesia (1 - 1.35% isoflurane) during imag-ing. An individual anesthesia level was determined for each animal, this level was set as the level at which the animal exhibited blinking responses upon the touching of the eye with a wet Q-tip. This level was maintained for every IOS measurement. Intrinsic signals were mea-sured through the intact skull and a coverglass which were highly illuminated with 810-nm light. Images were col-lected at a rate of two frames per second using a CCD camera attached to two Kodak camera lenses which were focused ~300 µm below the skull. A trial consisted of four seconds of pre-stimulus interval followed by four seconds of visual stimulation and concluded by a seven second post-stimulus interval. ITI was set to five seconds. Vi-sual stimuli were the CS+ and CS- and two identical mov-ing gratmov-ing stimuli that were shifted clock wise for 45◦in movement direction.
To increase the quality of the recorded images and the signal-to-noise ratio the gathered images were base-lined against the first eight frames grabbed during the pre-stimulus interval. Additionally, two areas were selected where there was no activation present, in every frame the signal from these areas was substracted from the signal area. Subsequently, a couple of additional steps were taken to be able to compare the activation elicited by the two stimuli. The images were tilted until the two blobs of activation, incited by the stimulus presentation on the two locations in the visual field (Fig. 2a), were vertically realigned to each other. For every frame, a vertical band was then cut out through the center of mass of the
acti-vation blob. To visualize the signal over time for every trial a response profile was created by placing the bands in chronological order. The signal frames were defined from frame 12 to frame 20, equal to two seconds after stimu-lus onset to two seconds after stimustimu-lus offset. For these frames a curve can be created by averaging the images over the first dimension. This curve depicts the size and amplitude of the activation blob.
With this data it can be inferred whether the training procedure has any large scale effects on the cortical rep-resentation of the CS+ or CS-. Also, whether the strenght of the population response changes due to the training can be determined. There are three measurement points so the learning effects over time can be detected.
Calcium imaging Two-photon calcium imaging was per-formed succeeding the last IOS measurement. The re-sults from the intrinsic imaging were directly used to de-termine the location on the cortex where the border area between the CS+ and CS- was. This location was marked by matching the real-time image of the skull with the ob-tained online maps from the IOS.
The animal was injected with 0.05 mg/kg bodyweight Temgesic 30 minutes prior to induction of anesthesia with 3% isoflurane. Directly following the conclusion of the intrinsic imaging a craniotomy was made above the esti-mated location of the overlap area of the two stimuli in V1. The exposed dura was left intact and kept moist with buffered artificial cerebral spinal fluid (Svoboda et al.,
Guido T. Meijer • Plasticity of the Visual Cortex 4
1999).
To visualize a large portion of the cortical area that represented the border region between the two stimuli in the visual field three injection sites were selected orthogo-nally aligned to the midline of the border. Multi-cell bolus loading was performed using the fluorescent calcium in-dicator Oregon Green BAPTA-1 AM (OGB) with DMSO containing 20% pluronic acid (Stosiek et al., 2003). As-trocytes were labeled using the red dye sulforhodamine-101 (Nimmerjahn et al., 2004). Injections were made with a glass pipette ( ~6 MΩ resistance) at a depth of 200 -300 µm below the surface of the cortex using a Luigs-Neumann SM-5 manipulator.
Two-photon laser scanning was performed with a Leica SP5 resonant laser scanning microscope and a Spectra-Physics Mai Tai High Performance Mode Locked Ti:Sapphire laser operating at 810 nm. Two photo-multiplier tubes were used to detect fluorescence emanat-ing from the brain tissue, one detected the green light from the Oregon Green BAPTA 1 at a wavelength of 525 nm and the other detected light emitted from the red fluores-cent Sulphorhodamine 101 at 585 nm. A region of be-tween 200 × 200 µm and 300 × 300 µm was imaged with a resolution fo 512 × 512 pixels at a sampling rate of 30 fr/sec. Batches of 8 frames were averaged together and saved to disk as a TIFF image, this reduced the effective sampling rate down to 2.5 Hz.
The gathered images were realigned with an algorithm that makes use of a single step discrete Fourier-transform (Guizar-Sicairos et al., 2008). Neurons, astrocytes and bloodvessels were detected semi-automatically using a custom written user interface ’Predator Cell Detection’, running on the MATLAB platform. Fluorescence values were computed for all measured time points per neuron, resulting in a fluorescence timeline for each neuron. From these timelines tuning curves were constructed (see Sup-plementaries).
During calcium imaging square wave stimuli were shown on the CS+ and CS- locations (see Visual stim-uli). On both locations moving square wave gratings were shown for 8 movement directions. As a reference a grey screen was shown for 5 seconds preceding the display of the moving grating stimulus. For the two stimulus loca-tions and for all 8 movement direcloca-tions 10 trials were pre-sented.
Whether a neuron was direction selective was deter-mined by constructing a tuning curve for each neuron. The response of a neuron to a movement direction was defined as the relative increase in fluorescence due to the move-ment of the stimulus divided by the non-moving reference stimulus (Equation 1). This resulted in a ∆FF value for each of the eight directions of movement for all 10 trials. It was determined whether a neuron exhibited significant tuning by performing a one-way ANOVA over the eight groups of ∆FF values, a neuron was considered to be di-rection selective upon a significant result from the one-way ANOVA (p < 0.05). The preferred direction of ev-ery direction selective neuron was then determined as the movement direction with the largest ∆FF value.
Equation 1: ∆F
F =
Fstimulus− Fref erence
Fref erence
Statistical analyses All presented data were derived by calculating the mean effect for every mouse and subse-quently pooling all mice together to get an overall mean and standard error. Only Figure 4 shows data that was cal-culated from all neurons together irrespectively of which mouse they were originally from. When comparing mul-tiple means a mulmul-tiple comparisons test was used with a Bonferroni correction to maintain the familywise error rate. All error bars depict standard error of means.
For the statistics done on the absolute response differ-ences (Fig. 3a), one mouse was excluded from the analy-sis. This mouse did not have any direction tuned neurons for 5 out of 8 movement directions, it also had a relatively low number of neurons in total (165) and a low percentage of neurons that were significantly tuned (12.7%).
Absolute response differenceIn order to assess changes in the response selectivity of neurons between the two stim-uli an unit of measurement called the absolute response difference was defined. Equation 2 shows the definition of the absolute response difference (ARD). Fstimulus1stands
for the relative amount of fluorescence (in ∆FF ) evoked by the presentation of the top stimulus in the movement direc-tion equal to the neuron’s preferred direcdirec-tion. Fstimulus2
is the same unit for the bottom stimulus. The ARD is cal-culated as the absolute from the substraction of Fstimulus1
and Fstimulus2.
Equation 2:
ARD = |Fstimulus1− Fstimulus2|
Results
No changes found in large scale population effects as measured with IOS
Intrinsic optical signal measurements were made before, half-way and after conditioning (Fig. 2). However, no significant changes in cortical representation were found. A Gaussian curve was fit over the horizontally collapsed images of the activation spots elicited by the CS+ stimu-lus (see Materials and Methods). The standard deviation of this Gaussian curve would be higher if the cortical area activated by the CS+ stimulus would increase over mea-surements, however, there was no significant change in standard deviaton between the pre -mid and post measure-ments (ANOVA F(2,24) = 0.60, p = 0.56). Also the size of the activation spots that resulted from the presentation of the CS stimulus showed no significant change (ANOVA F(2,24) = 0.71, p = 0.50).
The amplitude of the response is measured as the amount of light that is absorbed by the cortex, reflect-ing neuronal activity. The mean percentage change from the baseline condition is taken as the amount of activity elicited by the stimulus. The overall response to the CS+ stimulus did not change over measurements (ANOVA
Figure 2: Results from the intrinsic optical signal measurements. a, when neuronal tissue become active its optical properties change and more light is being absorbed, this results in a darkening of the image. Examples of spots of darkening of the cortex which is an indication for neuronal activity in that area. North on the image is anterior of the animal. b, The standard deviation of fitted Gaussian curves to the horizontally collapsed IOS images for pre, mid and post conditioning measurements. This standard deviation is a measurement of the size of the part of the cortex that becomes active during stimulus presentation. Error bars represent standard error of means. c, Amplitude of responses to CS+ and CS- stimulus shown for pre, mid and post measurements. Amplitude is depicted as a percentage change in light intensity from baseline, higher values representing lower light intensities.
F(2,24) = 0.36, p = 0.70; nor the response upon the CS-stimulus F(2,24) = 0.33, p = 0.73).
An assembly specific effect in stimulus response speci-ficity as measured with calcium imaging
We imaged a total of 9116 neurons from 11 conditioned mice with a mean of 829 ± 223 (s.e.m.) per mouse. From these imaged neurons 2912 had a tuning curve that showed significant tuning to a movement direction (ANOVA, p > 0.05), there were on average 265 ± 74 tuned neurons per mouse which summed up to an overall percentage of tuned neurons of 31.94%. All further analyses only make use of the neurons which exhibit significant tuning.
The specificity with which a neuron responds to either one of two stimuli can be quantified with an index which we dubbed the absolute response difference. This index is defined as the fluorescence (∆FF ) of the neuron when pre-sented with stimulus 1 subtracted from the fluorescence when presented with stimulus 2, subsequently the abso-lute is taken from the resulting value (see Materials and Methods). This results in an index which is high when the neuron responds strongly to either stimulus 1 or stimulus 2 and low when the neuron is ambivalent between the two. Per mouse, all neurons were classified on preferred di-rection and the mean absolute response difference was cal-culated for each subgroup (Fig. 3a). The assembly of neu-rons that have the direction of the CS+ to which the mouse was conditioned show a significantly greater absolute re-sponse difference (ANOVA F(7,72) = 6.76, p < 10−5; the CS+ vs CS+opp, p < 0.05; the rest, p < 0.001) compared to
all the other movement directions.
The effect size of this effect is defined as the abso-lute response difference for the CS+opp subtracted from
the CS+. Due to individual differences per mouse in the amount of training it received this effect size can be corre-lated to exposure (Fig. 3b). There is a significant
correla-tion between effect size and the number of trials an animal has performed (r(9) = 0.64, p < 0.05).
To quantify changes in the percentage of neurons that respond to the CS+ or CS- stimulus all the neurons were grouped according to whether they were tuned to the CS+ stimulus or the CS- stimulus. Neurons with a tuning curve for both stimuli were ignored. Within these two groups the percentages of neurons with a preferred direction to either one of four orientations was calculated. This was done for each animal seperately, the percentages that were calculated where then averaged over all mice (Fig. 4). No significant changes in percentages have been found for the neurons responding to the CS+ location, the CS- location or both locations (ANOVA, p = 0.54; p = 0.26; p = 0.21; respectively)
Currently no behavioral correlate has been found Using a top-down data exploration approach we tried to find correlates in the animal’s behavior to the condition-ing paradigm. As first described by Pavlov (1927) the behavior of an animal changes during a classical condi-tioning paradigm. As the animal learns the link between the conditioned cue and the reward it will begin to display anticipatory behavior as a result of the expectancy of re-ward. In the current experiment our expectations were that the mouse might exhibit more licking or sniffing behavior during or after the presentation of the CS+, compared to the CS- trials. In order to capture and quantify this behav-ioral change in the current experiment we focused upon the camera feed taken from the side of the animal. The videos were loaded into MATLAB, using a custom written user interface regions of interest (ROI) were drawn around the nose and mouth. These areas were cut out of the video and the absolute change in intensity for each pixel was cal-culated for every frame. This intensity change reflects the amount of movement in the ROI, enabeling an algorithm
Guido T. Meijer • Plasticity of the Visual Cortex 6
Figure 3: Results from the calcium imaging of 2912 significantly tuned neurons from 10 mice. a, the mean absolute response difference depicted for eight sub-groups of neurons, averaged over mice. The neurons are grouped by preferred direction relative to the CS+ direction used during conditioning. The response difference is calculated per animal and meaned over all animals (n=10). Error bars represent standard error of means. b, the relation between the effect size and the amount of trials an animal has performed, depicted in a scatterplot. The effect size is defined as the response of the CSoppsubstracted from the CS+. The line is drawn using
a linear least squares fit.
to detect licks and sniffs. The average amount of licks and sniffs was calculated for every session and separated in CS+ and CS- trials. However, no significant changes between the two trial types was found.
The eye tracking videos also supply a valuable source of information concerning the animal’s behavior. It pro-vides us with a high-resolution video feed of the animal’s head in which wisker and nose movement can be quanti-fied. Also pupil dilation could be a source of anticipatory behavior. However, these videos have not been analysed at the moment of writing.
Discussion
The current study has presented a novel method to induce plasticity in V1 of the mouse using a paradigm of appeti-tive classical conditioning. Within three weeks a group of six mice can be handled, trained and imaged which cre-ates opportunities to use this methodological approach in combination with a farmacutical intervention. After the training paradigm the subset of neurons that have the CS+ direction as their preferred direction show an increased absolute response difference, a reflection of stimulus re-sponse selectivity. Furthermore, these neurons have a rel-atively lower contribution to the representation of the CS+ stimulus location in contrast to the CS- stimulus loca-tion, suggesting that they have shifted their receptive fields away from the CS+ stimulus location and towards the CS-stimulus location.
In the auditory cortex the conditioning of a tone with a reward signal, elicited by stimulation of the VTA, results in a larger part of the cortex that responds to that tone (Bao et al., 2001). The effect is also reversible via extinction,
further supporting the notion that this is an effect based upon reinforcement learning (Bieszczad and Weinberger, 2012). Also pairing a tone with stimulation of the nucleus basalis, implicated with the distribution of acetylcholine (Mesulam and Geula, 1988), renders a larger amount of neurons in the primary auditory cortex that respond to said tone (Kilgard, 1998). Based upon these findings we hy-pothesised that after pairing of a visual stimulus with a reward the part of V1 that represents that portion of the vi-sual field would also increase in size. However, we found no such spatial increase in cortical representation of the rewarded stimulus. Apparently, the visual cortex responds in a different way to reward signals. The neurons in V1 show a retinotopic organisation in a way that neighbor-ing neurons also have receptive fields that are spatially close together (Smith and Häusser, 2010). It is possible that large scale changes in cortical representation of large portions of the visual field disturb this smooth retinotopic organisation and are therefore avoided in the encoding of reward signals. On the other hand, the reports concerning the auditory cortex depended upon coupling of a stimu-lus with stimulation of a brain region which is not a nat-uralistic reward. This method of conditioning might elicit plasticity in the auditory cortex that would otherwise never be able to occur. Another possibility is that the changes in size of the activated area of V1 are too small to be detected with intrinsic optical signal imaging, a technique depen-dent upon changes in bloodflow, which limits its spatial resolution.
Using IOS we were also unable to detect any changes in the amplitude of the response elicited by the used stim-uli. Merely upon the exposure to the visual stimuli an
in-Figure 4: Neurons are classified upon preferred orientation, depicted is the relative proportion of these four groups of neurons from the total in percentage. a, for all neurons that have a significant tuning curve for the CS+ stimulus. b, for all neurons that have a significant tuning curve for the CS- stimulus. c, for both groups taken together.
crease in amplitude would be expected as dictated by the effect of stimulus-selective response potentiation, which states that mere exposure to a stimulus enhances its evoked response in V1 (Frenkel et al., 2006). But also pairing of a stimulus with a reward increases visually evoked poten-tials in V1 (Frankó et al., 2010). A possibility is that the technique of IOS is not sensitive enough to replicate these findings. IOS relies upon changes in light absorption by the brain tissue for its signal but the observed values range in the 1-2% relative change compared to baseline, which might be too small to detect changes in amplitude. An-other possibility is that the mice did not receive enough training for this effect to exhibit itself. In the study per-formed by Frenkel et al. the mice were exposed to 500-2000 trials while in the current study 400-800 trails were excecuted, the macaques in the study by Frankó et al. were also excessively trained for 50 to 100 days.
A prominent hypothesis concerning the nature of en-coding in sensory cortices is the processing of information by cell assemblies (Hebb, 1949; Harris, 2005). These as-semblies of neurons share features, such as preferred di-rection in V1, and thereby inherently categorize informa-tion coming from the LGN. The behavior of these groups of neurons in a paradigm in which value is attributed to a specific stimulus is yet largely unknown. It is, however, possible to evoke an assembly specific effect in the cell assembly that encodes the same stimulus features of the stimuli that are used in a behavioral paradigm. It has been shown that stimulus-reward pairing is able to modify the
tuning curves of a specific cell assembly in V1 (Goltstein et al., subm.).
In the current study a cell assembly specific effect was found concerning the stimulus response specificity. Cell assemblies were defined by selecting those neurons which showed significant tuning to a specific direction of movement, subsequently these neurons were classified by preferred direction, resulting in eight groups of neu-rons which each share the feature of preferred direction. In the cell assembly which encodes the movement direc-tion equal to the direcdirec-tion in which the grating stimuli moved during conditioning a significantly higher stimulus response selectivity was found. After the training, the neu-rons in this cell assembly have become more selective in their response to either the CS+ or the CS-. At this point it cannot be said whether these neurons become more selec-tive towards the CS+ or the CS-. The assembly of neurons that have their preferred direction close to the orthogonal of the conditioned direction do not show this effect. A smaller but still significantl effect can also be observed in the assembly that encodes the opposite direction relative to the CS+ direction (p < 0.05).
Why do these neurons become more selective? Con-cider a neuron that has a receptive field on the border of the two stimuli, the amount information conveyed by this neu-ron regarding which stimulus is presented is very low; it activates during CS+ presentation but also during CS- pre-sentation. In the scenario in which one of the two stimuli predicts reward delivery, accurate information concerning
Guido T. Meijer • Plasticity of the Visual Cortex 8
which stimulus is presented is more essential than in a sce-nario in which this is not the case. It is therefore possible that plasticity is evoked specifically in these ’border neu-rons’, causing them to either change the position of their receptive field away from the border region or to suppress their direction sensitivity, effectively removing them from the equation altogether.
In order to selectively target border neurons for plas-ticity, a criterium has to exist by which a neuron is to be considered a border neuron. One can argue that these bor-der neurons have a higher prediction error, for half of the times they activate a reward is received but the other half of the times this does not happen. A neuron that has its receptive field completely overlapping the CS+ area has a very low prediction error because every time it detects a stimulus, a reward follows. It is possible that there is an overlapping homeostatic process which effectively minimizes prediction errors in V1 and thereby increasing the amount of information each neuron carries concerning stimulus-reward predictions. This could be advantageous for the effectiveness of the functioning of the visual sys-tem as a whole. If the presence of an object that could yield a benefit for the organism is already detected on the level of V1 this could streamline the processing steps fol-lowing up on V1 and result in more efficient way of pro-cessing.
The question remains what happens to the border neu-rons which are targeted for plasticity. One could conceive that these border neurons were to be actively inhibited be-cause their reward prediction is very poor. If this was the case there would be relatively fewer neurons responding to the CS-orientation compared to the other orientations. This is expected because the increase in selectivity is only observed in the group of neurons that are tuned to the CS-orientation. However, as can be observed in Figure 4, no significant changes in the composition of the general pop-ulation can be observed. All four orientations are repre-sented by a more or less equal percentage of neurons. If they are not inhibited in their response there are two other possible scenarios. The first possibility is that the border neurons shift their receptive fields away from the border region towards either the CS+ or the CS- area. The sec-ond option is that their receptive fields remain on the same location of the visual field but their response properties change in the sense that they stop responding to one of the two stimuli when it is present in their receptive field. Which of these scenarios is more likely cannot be said since no receptive field mapping has been done.
In summary, we have developed a novel method that resulted in plasticity effects that could be observed in a specific group of neurons in the primary visual cortex of the mouse. Our paradigm only requires three weeks of handling, training and imaging to observe changes in V1 that resulted from the coupling of reward with the presen-tation of a retinotopic specific visual stimulus. It is there-fore a valuable candidate amongst other methodological approaches to study plasticity in the visual cortex with its primary advantage of evoking plasticity in a more natural way than, for example, monocular deprivation. Contrary to findings in the auditory cortex, we report no large scale
changes in the retinotopic organization as a result of con-ditioning. On the scale of micro-circuitry, we found that the group of neurons that are tuned to the CS+ direction in-crease their stimulus response selectivity after training. A possible explanation for this phenomenon is that the neu-rons that used to respond to both stimuli, after the training only respond to one of the two stimuli. The advantage of this is that these neurons increase their stimulus-reward prediction.
References
Andermann M, Kerlin A, Reid R (2010) Chronic cellular imaging of mouse visual cortex during operant behavior and passive viewing. Front Cell Neurosci .
Bao M, Yang L, Rios C, He B, Engel S (2010) Perceptual learning in-creases the strength of the earliest signals in visual cortex. J Neurosci 30:15080–15084.
Bao S, Chan V, Merzenich M (2001) Cortical remodelling induced by activity of ventral tegmental dopamine neurons. Nature 412:79–83. Bieszczad KM, Weinberger NM (2012) Extinction reveals that primary
sensory cortex predicts reinforcement outcome. Eur J Neurosci no– no.
Buonomano DV, Merzenich MM (1998) Cortical plasticity: From synapses to maps. Ann Rev of Neurosci 21:149–186.
Crist R, Kapadia M, Westheimer G, Gilbert C (1997) Perceptual learn-ing of spatial localization: Specificity for orientation, position, and context. J Neurophysiol 78:2889 –2894.
Fahle M (2004) Perceptual learning: A case for early selection. J Vision 4:879–890.
Frankó E, Seitz A, Vogels R (2010) Dissociable neural effects of long-term stimulus-reward pairing in macaque visual cortex. J Cogn Neu-rosci 22:1425–1439.
Frenkel M, Sawtell N, Diogo A, Yoon B, Neve R, Bear M (2006) In-structive effect of visual experience in mouse visual cortex. Neuron 51:339–349.
Ghose G, Yang T, Maunsell J (2002) Physiological correlates of percep-tual learning in monkey v1 and v2. J Neurophysiol 87:1867. Goldstone R (1998) Perceptual learning. Annul Rev Psychol 49:585–
612.
Goltstein PM, Coffey EBJ, Roelfsema PR, Pennartz CMA (subm.) In vivo two-photon ca2+ imaging reveals selective reward effects on stimulus specific assemblies in mouse visual cortex .
Guizar-Sicairos M, Thurman ST, Fienup JR (2008) Efficient subpixel image registration algorithms. Opt Lett 33:156–158.
Hager A, Dringenberg H (2010) Training-induced plasticity in the visual cortex of adult rats following visual discrimination learning. Learn Memory 17:394–401.
Harris K (2005) Neural signatures of cell assembly organization. Nat Rev Neurosci 6:399–407.
Hebb DO (1949) The organization of behaviour. Wiley, New York, USA. Hua T, Bao P, Huang C, Wang Z, Xu J, Zhou Y, Lu Z (2010) Perceptual learning improves contrast sensitivity of v1 neurons in cats. Curr Biol 30:887–894.
Hubel DH, Wiesel TN (1959) Receptive fields of single neurones in the cat’s striate cortex. The Journal of Physiology 148:574 –591. Hubel DH, Wiesel TN (1962) Receptive fields, binocular interaction and
functional architecture in the cat’s visual cortex. J of Physio 160:106 –154.
Kilgard M (1998) Cortical map reorganization enabled by nucleus basalis activity. Science 279:1714–1718.
Mesulam M, Geula C (1988) Nucleus basalis (Ch4) and cortical choliner-gic innervation in the human brain: Observations based on the distri-bution of acetylcholinesterase and choline acetyltransferase. J Comp Neurol 275:216–240.
Nimmerjahn A, Kirchhoff F, Kerr JND, Helmchen F (2004) Sulforho-damine 101 as a specific marker of astroglia in the neocortex in vivo. Nat Meth 1:31–37.
Pavlov IP (1927) Conditioned Reflexes: an investigation of the physio-logical activity of the cerebral cortex. Oxford University Press, Lon-don.
Sakatani T, Isa T (2007) Quantitative analysis of spontaneous saccade-like rapid eye movements in C57BL/6 mice. Neurosci Res 58:324–
331.
Schoups A, Vogels R, Orban G (1995) Human perceptual learning in identifying the oblique orientation: retinotopy, orientation specificity and monocularity. J Physiol 483:797–810.
Schwartz S, Maquet P, Frith C (2002) Neural correlates of perceptual learning: a functional MRI study of visual texture discrimination. PNAS 99:17137–17142.
Seitz A, Watanabe T (2005) A unified model for perceptual learning. Trend Cogn Sci 9:329–334.
Seitz A, Kim D, Watanabe T (2009) Rewards evoke learning of uncon-sciously processed visual stimuli in adult humans. Neuron 61:700– 707.
Shuler M, Bear M (2006) Reward timing in the primary visual cortex. Science 311:1606–1609.
Smith S, Häusser M (2010) Parallel processing of visual space by neigh-boring neurons in mouse visual cortex. Nat Neurosci 13:1144–1149. Stosiek C, Garaschuk O, Holthoff K, Konnerth A (2003) In vivo two-photon calcium imaging of neuronal networks. PNAS 100:7319 – 7324.
Svoboda K, Helmchen F, Denk W, Tank D (1999) Spread of dendritic excitation in layer 2/3 pyramidal neurons in rat barrel cortex in vivo. Nat Neurosci 2:65–73.
Van Alphen A, Stahl J, De Zeeuw C (2001) The dynamic characteristics of the mouse horizontal vestibulo-ocular and optokinetic response. Brain Res 890:296–305.
Watanabe T, Nanez J, Sasaki Y (2001) Perceptual learning without per-ception. Nature 413:844–848.
Guido T. Meijer • Plasticity of the Visual Cortex 10
Figure 5: In vivocalcium imaging enabled the monitoring of many neurons consecutively. a, averaged image taken during one session. Neurons are colored green, astrocytes yellow and bloodvessels are black. Raw data is shown for two neurons; the neuron encircled with red is a direction tuned neuron and the neuron encircled with blue is not. b, tuning curves for the two neurons, individual responses are shown in thin lines. The red neuron is clearly tuned to movement in the 45◦direction, the opposite (225◦) also elicits a large response. The blue neuron does not show any tuning to a specific movement direction. c, raw fluorescence trace of the two neurons. Vertical lines represent the presentation of moving grating stimuli in the preferred direction of the red neuron (45◦) and the opposite (225◦). As can be seen from the timeseries, the red neuron consistently responds to the stimuli with a calcium transient whilst the blue neuron does not.
