|Home | About | Journals | Submit | Contact Us | Français|
The structure of neurones changes during development and in response to injury or alteration in sensory experience. Changes occur in the number, shape and dimensions of dendritic spines together with their synapses. However, precise data on these changes in response to learning are sparse. Here, we show using quantitative transmission electron microscopy that a simple form of learning involving mystacial vibrissae results in about 70% increase in the density of inhibitory synapses on spines of neurones located in layer IV barrels that represent the stimulated vibrissae. The spines contain one asymmetrical (excitatory) and one symmetrical (inhibitory) synapse (double-synapse spines) and their density increases 3-fold due to learning with no apparent change in the density of asymmetrical synapses. This effect seems to be specific for learning as pseudoconditioning (where the conditioned and unconditioned stimuli are delivered at random) does not lead to the enhancement of symmetrical synapses, but instead results in an up-regulation of asymmetrical synapses on spines. Symmetrical synapses of cells located in barrels receiving the conditioned stimulus show also a greater concentration of γ-amino-butyric acid (GABA) in their presynaptic terminals. These results indicate that the immediate effect of classical conditioning in the ‘conditioned’ barrels is rapid, pronounced and inhibitory.
Neurones of the cerebral cortex are plastic, i.e. they are able to modify their structure and function. Plasticity occurs during development (LeVay et al., 1980; Stern et al., 2001; Desai et al., 2002), in response to injury (Darian-Smith and Gilbert, 1995; Kaas et al., 2008) and due to alterations of sensory experience (Wiesel and Hubel, 1965; Glazewski and Fox, 1996; Trachtenberg et al., 2002), including learning (Wilson and McNaughton, 1994; Kleim et al., 1996). Various reports show that the morphology of all major components of the neurone are malleable, i.e. the axons (Portera-Cailliau et al., 2005; De Paola et al., 2006), cell bodies (von der Ohe et al., 2006) and also the dendrites (Holtmaat et al., 2005; Knott and Holtmaat, 2008), together with their spines, where the majority of brain synapses are located (Knott and Holtmaat, 2008). There is now unequivocal evidence that dendritic spines are at the centre of the mechanisms underlying plasticity of the neuronal circuits (Harris and Woolsey, 1981; Mahajan and Desiraju, 1988; Engert and Bonhoeffer, 1999; Grutzendler et al., 2002; Trachtenberg et al., 2002). It has been shown that spine density, morphology, and motility are experience- and activity-dependent (Lendvai et al., 2000; Trachtenberg et al., 2002; Holtmaat et al., 2005). While spines have the most capacity for change during early development, it seems that they also change in adulthood (Knott et al., 2002; Holtmaat et al., 2006).
It is commonly assumed that the mechanisms underlying learning and memory are based on changes in synaptic weights and connectivity between neurons, guided by neuronal activity and supported by molecular cues (Bailey and Kandel, 2008; Howland and Wang, 2008). While several molecules are known to specifically interfere with memory, the detailed anatomical, physiological and molecular mechanisms underlying memory in a mammalian brain are still not known (Costa-Mattioli and Sonenberg, 2008). A part of the somatosensory cortex of rodents that represents the mystacial vibrissae - the barrel cortex - offers an excellent model to study the mechanisms of learning and memory because of its clearly defined structure, connectivity and the ease of inducing plasticity via a learning paradigm (Woolsey and Van der Loos, 1970; Siucinska and Kossut, 1996; Fox, 2002). The neurones of the barrel cortex have the ability to change their receptive field properties due to alterations in sensory experience (Armstrong-James and Fox, 1987). For example, sensory deprivation of selected vibrissae causes expansion of the cortical representations of the spared vibrissae (Glazewski and Fox, 1996). Similarly, stimulation of a row of vibrissae paired with a tail shock (classical conditioning), results in a short lasting enlargement of the cortical representation of this row, mainly in layer IV (Siucinska and Kossut, 1996). The main aim of the present study was to investigate the effects of whisker conditioning on symmetrical and asymmetrical synapse number. Whisker conditioning resulted in selective increase in the density of inhibitory synapses on double-synapse spines in the ‘trained’ barrels. Neither whisker stimulation alone nor pseudoconditioning produce a similar effect.
The experiments were performed on Swiss Webster mice aged 6-7 weeks raised in standard conditions (Siucinska and Kossut; 1996). All experiments were compliant with the European Communities Council Directive of 24 November 1986 (86/609/EEC) and were approved by the Animal Care and Use Committees of the Polish Academy of Sciences and the Jagiellonian University.
From 34 experimental mice, 24 were trained and 10 served as controls. From the trained mice 10 were conditioned (CS-UCS), 7 pseudoconditioned (Pseudo) and 7 had whiskers only stimulated (CS). Control animals were split into two groups from which one group was habituated in a homemade restrainer along with all mice awaiting the training and the second group left non-habituated. During the habituation period mice spent 10 min per day for 3 weeks in the restrainer. The restrainer holds the mouse’s neck stationary, while leaving the rest of the body, including the head, free. After a period of habituation, mice were conditioned using a classical (Pavlovian) delay conditioning paradigm. A stroke of the selected vibrissae (B-row - CS) on one side of the snout was paired with a mild tail shock (UCS) (Siucinska and Kossut, 1996). The pairing procedure comprised 3 strokes lasting 3 s each, applied to the row of whiskers, repeated at a frequency of 4 times a minute for 10 min. This procedure was applied for 3 consecutive days. The UCS consisted of weak 0.5 mA electric current applied to the tail for 0.5 s at the end of the last stroke in the series. In the case of pseudoconditioned (random pairing of CS and UCS) and whisker stimulated (only strokes to the selected row of vibrissae applied) animals the number and frequency of stimuli applied were identical to those used during conditioning protocol.
Animal behaviour was assessed during training with the aim to provide evidence that animals actually alter their behaviour (learn) in response to conditioning. The number of times the head turned towards the stimulating brush (CS) was counted during the first and last session of training of 5 CS-UCS, 5 CS and 5 pseudoconditioned animals as an indication of learning.
24 hours after each experiment, animals were deeply anesthetized with pentobarbitone and perfused through the heart with 20 ml of phosphate buffered solution containing 2.0% freshly dissolved paraformaldehyde and 0.2% of glutaraldehyde followed by 100-150 ml of phosphate buffered solution of 2% paraformaldehyde and 2.5% glutaraldehyde. Immediately after perfusion the brains were removed and left overnight in phosphate buffered solution of 2% paraformaldehyde and 2.5% glutaraldehyde at 4°C. Next day, slices of 60 μm of thickness were cut tangentially to the surface using a vibratome. Slices containing the barrel field (layer IV) were prepared using our standard procedure for transmission electron microscopy (TEM) (Kirov, 1999; Knott et al. 2002). Briefly, the slices were washed in 0.1 M sodium cacodylate buffer (3 × 5 min), postfixed at 4°C in 1% osmium tetroxide in 0.1 M sodium cacodylate buffer (2 × 1 h, the first change also containing 1.5% potasium ferrocyanide), washed in distilled water (3 × 5 min), incubated at 4°C in 70% alcohol containing 1% uranyl acetate, dehydrated in increasing concentrations of ethanol (50%, 70%, 90%, 96% - 5 min each) and 100% (3 × 5 min) and propylene oxide (2 × 10 min). Slices were next incubated in increasing concentrations of propylene oxide and embedded between silicon-coated glass slides in Epon (Poliscience). Embedded slices were photographed (×5) under Nikon Optiphot using a computer assisted Nikon DXM 1200 F digital camera. The images were stacked together with use of the Adobe Photoshop CS and the barrel field reconstructed. The B2 barrel was identified from the pattern of barrels drawn from under the microscope together with the blood vessels pattern characterising its location within the barrel field. The embedded slices containing the B2 barrel were then trimmed with help of the drawings to blocks encompassing exclusively B1 and B2 or B2 and B3 barrels. Trimmed slices were cut in 30-50 series of ultra-thin sections at 60-70 nm using an ultramicrotome (Ultracut, Reichert). The ultra-thin sections were collected on formvar-coated copper-palladium slots, counterstained with 1% lead citrate and photographed at 7,000-10,000X using a JEOL 100SX TEM aiming for the central part of the B2 barrel where cell bodies are sparse.
The tissue of 3 conditioned and 3 control animals was fixed and embedded as above with the omission of uranyl acetate and osmium/potassium ferrocyanide steps and replacing Epon with Durcupan resin. Ultra-thin sections of the same region of B2 as before were cut and mounted onto 200 mesh nickel grids. Post-embedding immunogold reaction with GABA antibody (anti-GABA 990, kindly provided by O.-P. Ottersen and J. Storm-Mathisen, Oslo, Norway) was used to identify inhibitory synapses according to the procedure of Ottersen (Ottersen, 1989; Mahendrasingham et al., 2004). Briefly, grids were successively incubated in drops of freshly made 1% aq. sodium metaperiodate, Tris-buffered 1% human serum albumin (TBHSA), anti-GABA antibody diluted 1:1,000 diluted in TBHSA (primary antibody), washed and then incubated in goat anti-rabbit secondary antibody conjugated to 10 nm gold particles (BioCell, UK) diluted 1: 100 in TBHSA. The anti-GABA antibody was pre-incubated with glutaraldehyde conjugated β-alanine, glutamic acid, glycine and taurine to eliminate unwanted immunoreactivity with other amino acids. Finally, grids were washed in TBHSA and then distilled water, and counterstained with 2% aq. uranyl acetate (Mahendrasingam et al., 2004). Grids were viewed in a JEOL 100CXII TEM and images taken at a magnification of 29000 – 48000X using an Olympus/SIS systems megaview III digital camera. Special care was taken to process the tissue from all experimental animals in the same way and to standarise staining conditions. The animals were perfused on the same day. Their brains were embedded together and sliced on the same day. All tissue samples were processed for immunochemistry in parallel and in identical conditions. The comparison between GABA content in the terminals of symmetrical and asymmetrical synapses was performed using the same set of sections. The only difference was that 3 animals have been trained and 3 left untreated.
The density of symmetrical and asymmetrical synapses located on dendritic shafts and single- and double-synapse spines together with the density of single- and double-synapse dendritic spines were estimated using the serial disector method (Weibel, 1979; Gundersen, 1986; Fiala and Harris, 2001). Axosomatic synapses were then omitted from the analysis. Similarly, only synaptically connected spines have been included. The total density of synapses in this study indicates the number of synapses on dendritic spines and shafts counted in the defined area of the centre of the B2 barrels. The centres of the barrels were chosen for counting synapses because the lower density of cell bodies minimises obscuration of synapses on finer neuronal elements, where most of them are located (White, 1976; Knott et al., 2002). Synapses and spines have been defined and counted according to Knott et al. (2002). Briefly, counting was done by placing a sample rectangle over the sequence of serial sections and counting each structure only once through the series; only structures fully within the rectangle or intersecting the left and the upper sides of the rectangle were included. The counting was done blind concerning the experimental group being counted.
To compare the effects of learning on synapse density across the experimental groups we counted the number of synapses in about 100 μm3 volume of the B2 barrel hollow within each animal, averaged them across groups of treatment and compared animal means using one-way ANOVA with post-hoc Tukey test after testing for normality and homogeneity. In the case of immunogold experiments the data were compared using Kolmogorov-Smirnov and Mann-Whitney U statistics. In behavioural studies the Mann-Whitney U-test has been used. Both control groups (see above) appeared to be identical (Mann-Whitney U-test) and were pooled together. The data was expressed as means ± standard deviation (SD) throughout the paper, unless if stated otherwise.
All experimental procedures were authorised by the Local Ethical Committee at the Nencki Institute and Keele University in accordance with the 1986 European Committee Council Directive (86/609/EEC).
In this study we compared the density of synapses in layer IV barrels of mice subjected to a brief learning paradigm involving mystacial vibrissae with that of untreated mice. Moreover, to test whether learning was the only cause of synaptic changes, we measured whether whisker stimulation alone or pseudoconditioning (where the conditioned stimulus and the reinforcement are presented randomly) were enough to produce similar effects. With the aim of recognising the neuronal elements where synaptic changes take place, we counted synapses on dendritic shafts and spines independently, characterising them as symmetrical (most likely inhibitory) or asymmetrical (most likely excitatory) (Fig. 1A). During the course of the study, we noticed that the symmetrical synapses were the only pool of synapses which changed density specifically due to learning. Therefore, we also asked whether the level of GABA changed in the terminals lying presynaptic to the symmetrical synapses.
The synapses and dendritic spines were counted in the following volumes of the tissue: control group - 803.53 ± 73.05 μm3 (average per animal, 114.80 ± 35.10 μm3), conditioned group - 667.39 ± 60.67 μm3 (average per animal, 95.34 ± 46.3 μm3) whisker stimulated group - 794.94 ± 72.27 μm3 (113.56 ± 24.18 μm3) and pseudoconditioned group - 675.52 ± 61.41 μm3 (average per animal, 96.50 ± 26.86 μm3). These sampling volumes were not significantly different across the treatment groups (ANOVA, p=0.58, F=0.67, total df=27). The total number of spines and synapses counted in subsequent groups of treatment is summarized in table 1, where also the total number of ultra-thin sections/photographs is provided.
Conditioning did not change total synapse density in the hollow of the stimulated B2 barrel, but the density of synapses in the pseudoconditioned group (2.66 ± 0.69 per μm3 ) was 82 % higher than the density of synapses in the control group (1.46 ± 0.19 per μm3 ; ANOVA, p<0.001, F=11.34, total df=27), 46% higher than in the conditioned animal group (1.82 ± 0.21 per μm3; ANOVA, p<0.01) and 37% higher than in the stimulated group (1.94 ± 0.28 per μm3 ; ANOVA p<0.05) (Fig.1B). These data indicate that only pseudoconditioning is capable of inducing changes in total synapse density. Our finding does not preclude the possibility that whisker stimulation and conditioning could evoke synaptic changes in specific sub-groups of synapses that are less numerous and therefore obscured by the variance in total synapse density. We tested for this possibility by measuring symmetrical and asymmetrical synapse densities independently and also recognising that some of them are located on dendritic shafts and some on dendritic spines.
The asymmetrical synapse density in both pseudoconditioned (2.34 ± 0.62 per μm3) and stimulated (1.69 ± 0.24 per μm3) animals was about 97% and 42% higher than in controls respectively (1.19 ± 0.15 per μm3), which was statistically significant (ANOVA, p<0.001 and p<0.05 respectively with F=14.46 and total df=27). The mean of the synaptic density of conditioned animals (1.34 ± 0.21 per μm3) was found not to be different from controls (Fig.1C).
The mean density of symmetrical synapses in the conditioned group (0.47 ± 0.09 per μm3) was 74% higher than the mean of the control (0.27 ± 0.05 per μm3; ANOVA, p<0.001, F= 14.94, total df=27), 88% larger than stimulated (0.25 ± 0.06 per μm3; ANOVA, p<0.001) and 52% larger than in the pseudoconditioned group (0.31 ± 0.08 per μm3; ANOVA, p<0.01) (Fig.1D).
In this study, only synapses located on the dendritic shafts and spines were counted in the barrels’ hollows. The density of synapses located on dendritic shafts was several-fold lower than the density of synapses found on dendritic spines, independent of treatment (synapses located on dendritic shafts: control, 0.29 ± 0.07 per μm3; conditioned, 0.39 ± 0.10 per μm3; stimulated, 0.32 ± 0.05 per μm3 and pseudoconditioned, 0.42 ± 0.14 per μm3) (Fig.2A). A two-fold increase in the number of synapses on dendritic spines was found in the pseudoconditioned animals when compared to control animals (2.24 ± 0.64 per μm3 and 1.17 ± 0.19 per μm3 respectively; ANOVA, p<0.001, F=11.66, total df=27) with no significant differences found between conditioned, stimulated or control animals (Fig.2B). The large increase in asymmetrical synapse density due to pseudoconditioning was therefore localised to the synaptic spines.
Asymmetrical synapses located on dendritic spines made up about 73.3% of all synapses found in the area (1.07 ± 0.17 per μm3) while 8.2% of them were located on the dendritic shafts (0.12 ± 0.06 per μm3). Among the remaining synapses, 11.6% were recognised as symmetrical synapses connecting to the dendritic shafts (0.17 ± 0.03 per μm3) and 6.9% were located on dendritic spines (0.10 ± 0.03 per μm3).
The density of dendritic spines increased about two-fold following pseudoconditioning (control, 1.07 ± 0.17 per μm3 and pseudoconditioned, 2.08 ± 0.60 per μm3; ANOVA, p<0.001, F=12.30, total df=27), but not after conditioning (1.13 ± 0.20 per μm3) or whisker stimulation (1.49 ± 0.25 per μm3) (Fig.3A).
Most dendritic spines have only one synapse located on them and this is usually an asymmetrical synapse (single-synapse spines), but a small percentage of dendritic spines host two synapses (double-synapse spines) and in the majority of cases one is asymmetrical, located on the head of a dendritic spine and one is symmetrical, located on its neck (Fig.1A) (Knott et al., 2002; Jasinska et al., 2006; Kubota et al., 2007). In the present study, all synapses located on single-synapse spines were recognised as being asymmetrical, and on double-synapse spines, one was always symmetrical and the other asymmetrical. Spines with more than two synapses were not observed in this region of the barrel cortex. In the control animals, the density of single-synapse spines was calculated as 0.96 ± 0.16 per μm3 and constituted about 90% of the total number of dendritic spines in the area (1.07 ± 0.17 per μm3). The density of double-synapse spines was estimated as 0.10 ± 0.03 per μm3. Pseudoconditioning increased the density of single-synapse spines by about two-fold (from 0.96 ± 0.16 per μm3 in control animals to 1.92 ± 0.57 per μm3; ANOVA, p<0.001, F=14.80, total df=27) while conditioning and stimulation alone did not induce significant increases (conditioned group, 0.84 ± 0.22 per μm3 and stimulated group, 1.36 ± 0.24 per μm3) (Fig.3B). Thus the large increase in the density of asymmetrical synapses due to pseudoconditioning (Fig.1C) could be interpreted as an increase in the density of asymmetrical synapses connecting to the single-synapse dendritic spines (Fig.3B).
Whisker conditioning increased the density of the double-synapse spines almost 3-fold, from 0.10 ± 0.03 per μm in control animals to 0.29 ± 0.07 per μm3 (ANOVA, p<0.001, F=18.95, total df=27) with no change in the case of stimulated (0.13 ± 0.04 per μm3) and pseudoconditioned animals (0.16 ± 0.05 per μm3) (Fig.3C). Thus the large increase in the density of conditioning evoked symmetrical synapses is correlated with the increase in double-synapse spines (Figs.(Figs.1D1D and and3C3C).
The conditioning paradigm led to an increase in the density of inhibitory synapses on double-synapse spines. To answer the question of whether this increase is accompanied by an increase in the concentration of GABA in the presynaptic terminals of inhibitory synapses, we immunolabelled ultra-thin sections cut through layer IV of conditioned and untreated animals with anti-GABA antibodies, visualised with gold-conjugated secondary antibodies (Fig.4). We counted the number of gold particles in the symmetrical and asymmetrical terminals in the same sections cut from the barrels that were involved in the process of conditioning (153 symmetrical and 308 asymmetrical terminals counted) and in matching controls (148 symmetrical and 309 asymmetrical terminals counted) and measured the areas of all terminals. While the presence of gold particles in the asymmetrical synapse terminals is evident (Fig. 4) the presence of large amounts of GABA in them is unlikely (Martin and Rimvall, 1993) so we took the density of gold particles in the asymmetrical synapse terminals as the background. Having estimated the densities of gold particles in the individual synaptic terminals we plotted the distributions of these densities for symmetrical and asymmetrical “trained” and “untrained” data pooled from all experimental animals as depicted in Fig.5. Only the histogram representing the symmetrical synaptic terminals is shifted to the right due to the conditioning, suggesting an increase in the concentration of GABA. To compensate for the background, in each animal we subtracted the 95th percentile value obtained from the distribution of density of gold labelling in the asymmetrical terminals from the density of gold particles measured in all symmetrical terminals individually and then averaged the results. From this average for each animal we then calculated an average value of the three animals within the treatment groups, that is control and conditioned and compared them using the Mann-Whitney U test (mean ± SEM; control group: 12.31 ± 1.89 gold particles per μm2, conditioned group: 27.06 ± 1.06 gold particles per μm2, p< 0.0001). This analysis unequivocally shows that whisker conditioning increases the concentration of GABA in the terminals of the symmetrical synapses.
During the initial trials of whisker conditioning (CS+UCS), the mice often reacted to vibrissal stimulation by turning the head towards the stimulus, but during the subsequent trials the frequency of this reaction decreased. The number of head turnings during CS application, which possibly indicates the amount of learning by the mouse, was counted from video recordings. In the course of CS-UCS the number of head turnings decreased from 10.40 ± 0.93 in the first session to 4.80 ± 0.83 in the third session, p<0.05; in the case of CS from 20.00 ± 0.95 in the first session to 11.40 ± 0.93 in the third session, p<0.05, but the decrease has not been observed during pseudoconditioning (session 1; 22.60 ± 1.95 and session 2; 27.00 ± 4.24) (Fig.6). This shows that the animals learn not only during CS-UCS sessions, but also during CS sessions albeit in a different manner than in the CS-UCS case and far slower.
These results demonstrate that brief classical conditioning, where stimulation of whiskers is paired with tail shock, leads to a substantial up-regulation of the density of GABA-ergic synapses in the hollow of ‘trained barrels’. The supernumerary GABA-ergic synapses were found on double-synapse spines which showed an increased number following conditioning. Moreover, GABA content increased in the presynaptic terminals of symmetrical synapses (not only in double-spine synapses) in the ‘trained’ barrel hollows. Pseudoconditioning led to a specific, high variance, increase in the density of the asymmetrical synapses on the single-synapse spines, while whisker stimulation alone increased the density of excitatory synapses in an unspecified pool.
One interpretation of the data would be that whisker stimulation in the absence of a meaningful context (i.e., pseudoconditioning or whisker stimulation alone) results in an increased density of excitatory synapses, and whisker stimulation paired with reinforcement (i.e., conditioning) leads to an increase in the density of GABA-ergic synapses. However, at first glance, this reasoning is at odds with the results of other experiments where whisker stimulation alone (Knott et al., 2002) led to an increase in the density of GABA-ergic synapses or whisker deprivation (Micheva and Beaulieu 1995) led to a decrease, specifically on double-synapse spines. This would suggest that the length, intensity, frequency or pattern of stimulation and not the behavioural paradigm is the differentiating factor leading to a specific change in synapse density. If so, the reinforcement in our experiments would eventually only modulate the demand for an appropriate stimulation needed for a particular synaptic change. Alternatively, the aforementioned ‘stimulation/deprivation’ experimental paradigms may contain elements of learning.
Whisker conditioning produces an upregulation of GABA-ergic synapses that is consistent with an elevation of several GABA-ergic markers in the barrel hollows of ‘trained’ vibrissae, like the density of small GABA-immunoreactive, non-parvalbumin containing neurones (Siucinska et al., 1999; Siucinska and Kossut, 2006), GAD-67 mRNA (Gierdalski et al., 2001), GABA-ergic puncta (Siucinska, 2006) and GAD-67 protein level (Gierdalski et al., 2001). Moreover, in the barrels’ excitatory neurones representing vibrissae involved in training, a selective increase in frequency, but not amplitude of spontaneous inhibitory, but not excitatory postsynaptic currents was observed (Tokarski et al., 2007). Also, the amplitude of field potentials evoked by the stimulation of layer VI and recorded in layer IV was significantly reduced (Tokarski et al., 2007).
As the increase in GABA-ergic synapse density appears on double-synapse spines with no apparent change in the density on single-synapse spines, a concomitant increase in excitatory synapse markers would be expected and these indeed have been found again exclusively in barrels involved in learning (Jablonska et al., 1996; Skibinska et al., 2001, 2005).
Although synapses were counted only in ‘conditioned’ barrels, there is evidence that their density in neighbouring barrels either does not change, or the change does not have an impact on neuronal transmission. Firstly, the 2-DG response to stimulation of the principal whisker of the barrel immediately neighbouring the ‘conditioned’ barrel is indistinguishable from the control (Kossut and Siucinska, 1998). Secondly, neuronal transmission is affected by the learning procedure only in the output from, not input to, the ‘conditioned’ barrel (Urban-Ciecko et al., 2005). The changes in GABA-ergic markers, GABA-ergic synapse density and in vitro physiology appear to be well orchestrated. A similar coherence between physiological, histological and anatomical response was observed in the barrel cortex in response to sustained whisker stimulation (Welker et al., 1989; Welker et al., 1992; Knott et al., 2002; Quairiaux et al., 2007). Taken together these data unequivocally show that the response of barrels’ neurones to vibrissae conditioning is net inhibitory in nature at least at the time when measurements have been taken, and confined to ‘conditioned’ barrels.
On the other hand, the “trained” whisker-evoked 2-DG uptake spreads well beyond the borders of the “trained” barrels (Siucinska and Kossut, 1996). This result is intuitively at odds with an inhibitory response to learning. It should be expected that rising inhibition inside the barrel diminishes the uptake; intensity and spread of the 2-DG response (Fox et al., 2003; Tokarski et al., 2007). With regard to the 2-DG uptake inside the barrel, the unchanged response could be a consequence of both the additive nature of the 2-DG signal, where inhibition and excitation have the same metabolic meaning (Sokoloff et al., 1977) and high metabolic activity of inhibitory neurones (McCasland and Hibbard, 1997). In such circumstances, a slight increase in inhibition leading to a diminished excitatory response could be metabolically more pronounced than a loss of label due to decreased excitation (Melzer et al., 1985). 2-DG spread could be similarly explained as an effect of a learning-induced shift in lateral excitation and inhibition. The main substrates for such changes, that is excitatory and inhibitory connections between barrels (Aroniadou-Anderjaska and Keller, 1996; Salin and Prince, 1996; Fox et al., 2003; Schubert et al., 2003), including the inhibitory network (Gibson et al., 1999; Sun et al., 2006; Cruikshank et al., 2007) are known, but details of the mechanism are missing. Alternatively, as our data were obtained a day after the cessation of training, we may be observing up-regulation of the GABA-ergic system in response to training-enhanced excitation. We do not know at present how long the changes in the inhibitory system last after cessation of the conditioning procedure. The metabolic change fades within 5 days of the last training session (Siucinska and Kossut, 1996), which corresponds well with the time course of the disappearance of electrophysiological changes due to passive whisker stimulation, the procedure that may contain elements of learning (Knott et al., 2002).
The source of excitatory and inhibitory terminals engaged in synaptic contacts on double-synapse spines in the barrel cortex is not precisely known. The excitatory terminals could originate from the thalamus, as was recently shown in the rat frontal cortex (Kubota et al., 2007) or be intracortical, most likely belonging to excitatory cells of the same barrel (Feldmeyer et al., 1999; Lubke et al., 2000). The inhibitory terminals are intracortical (Fox, 2008), but it is not known which cells they belong to from the variety of dendrite-targeting inhibitory cell types (Markram et al., 2004). Initially, double bouquet cells were thought to be likely candidates, as they are involved in the cat visual cortex (Tamas et al., 1997; Knott et al., 2002), but later it appeared that they are missing in rodent brains (Yanez et al., 2005). As non-parvalbumin GABA-ergic cells were found to be up-regulated by conditioning (Siucinska and Kossut, 2006), the “donors” of GABA-ergic innervation to double-synapse spines possibly belong to this class of neurons with the Martinotti cells as possible and testable candidates (Markram et al., 2004). Double-synapse spines are most likely located on the dendrites of stellate and/or star pyramidal cells of the same barrels from which their afferentation originates (Woolsey et al., 1975; White and Peters, 1993; Salin and Prince, 1996; Aroniadou-Anderjaska and Keller, 1996; Lubke et al., 2000; Fox et al., 2003; Schubert et al., 2003). Spiny inhibitory interneurones are another possible, but less likely, target (Kawaguchi et al., 2006; Sun et al., 2006).
The increase in density of symmetrical synapses occurs on double synapse-spines and may be obtained in two ways that are not mutually exclusive. The first possibility relies on the addition of symmetrical synapses to the pre-existing single-asymmetrical synapse spines with concomitant replenishment of their own density. The second possibility is that the double-synapse spines are constructed de novo. New synapses could be formed using an existing pattern of neuronal connectivity or alternatively, orchestrated with changes in the density of neuropil inside the ‘trained’ barrel. Retractions and additions of small axonal elements over a timescale of days and the distances comparable with the dimensions of barrel hollows have recently been observed during development of the barrel cortex (De Paola et al., 2006). The dendrites, excluding dendritic spines, appear to be more stable, at least over a relatively short timescale (Trachtenberg et al., 2002).
We showed that animals reduce head movements in response to whisker stimulation coupled with conditioning, signalling an inescapable fearful stimulus, significantly more than mice that received only whisker stimulation. This possibly indicates that the animal is undergoing learning (see also Siucinska and Kossut, 1996, Cybulska-Klosowicz et al., 2009). Additionally, the animals clearly remember the conditioning for at least 24 hours as it is indicated in Fig.6. The mechanism underlying this may reside in the barrels. Such a mechanism should include a specific memory trace, but also regulated activity environment necessary for its formation. In our case the increased inhibitory interactions in layer IV, triggered by sensory training could decrease the effects of learning-associated release of neuromodulators and in consequence general arousal in order to maintain activity homeostasis (Froemke et al., 2007). With time, conditioning-induced changes observed in barrels could propagate elsewhere (Berry et al., 2008).
This work was supported by the Wellcome Trust Grant to SG and MK, Royal Society and Physiological Society grants to S.G. and grants from the Erasmus, Sieć and BW/IZ/72/2007 to MJ. We thank Dr. Alison Barth, Dr. Kevin, Fox, Dr. Egbert Welker and Dr. Gerald Finnerty for their advice on the manuscript, Dr. Graham Knott, Dr. Egbert Welker and Karen Walker for the training of MJ and Dr. Polwart and Dr. Preater for advice on the statististical analysis.