The primary somatosensory barrel cortex processes tactile vibrissae information, allowing rodents to actively perceive spatial and textural features of their immediate surroundings. Each whisker on the snout is individually represented in the neocortex by an anatomically identifiable ‘barrel’ specified by the segregated termination zones of thalamocortical axons of the ventroposterior medial nucleus, which provide the primary sensory input to the neocortex. The sensory information is subsequently processed within local synaptically connected neocortical microcircuits, which have begun to be investigated in quantitative detail. In addition to these local synaptic microcircuits, the excitatory pyramidal neurons of the barrel cortex send and receive long-range glutamatergic axonal projections to and from a wide variety of specific brain regions. Much less is known about these long-range connections and their contribution to sensory processing. Here, we review current knowledge of the long-range axonal input and output of the mouse primary somatosensory barrel cortex. Prominent reciprocal projections are found between primary somatosensory cortex and secondary somatosensory cortex, motor cortex, perirhinal cortex and thalamus. Primary somatosensory barrel cortex also projects strongly to striatum, thalamic reticular nucleus, zona incerta, anterior pretectal nucleus, superior colliculus, pons, red nucleus and spinal trigeminal brain stem nuclei. These long-range connections of the barrel cortex with other specific cortical and subcortical brain regions are likely to play a crucial role in sensorimotor integration, sensory perception and associative learning.
A fundamental challenge for neuroscience is to understand the basic principles of how the mammalian brain processes sensory input, extracting and storing information in such a manner that the internally generated representations of the environment can be utilized to select appropriate goal-directed motor programs. A mechanistic and causal understanding must consider individual neurons and their synaptic interactions within complex highly-distributed neuronal networks. The difficulty of such analyses may be significantly aided by investigating relatively simple sensory systems in genetically tractable animals, such as the mouse.
Deflections of the mystacial whiskers are rapidly signalled to the primary somatosensory neocortex (S1) via two synapses, one in the brain stem and the other in the thalamus (Fig. 1A). Mechanosensitive sensory neurons of the trigeminal ganglion fire reliable direction-selective action potentials with different velocity thresholds in response to deflection of single whiskers (Szwed et al., 2003; Jones et al., 2004; Arabzadeh et al., 2005; Leiser & Moxon, 2007). This sensory information is signalled to neurons in the principal and spinal trigeminal nuclei via excitatory glutamatergic synapses in the brain stem. The brain stem neurons, in turn, signal across excitatory glutamatergic synapses to somatosensory thalamocortical neurons of the ventroposterior medial (VPM) and posterior medial (POM) thalamus (among other targets). Projections from these two thalamic nuclei to primary somatosensory barrel cortex of the mouse have begun to be characterized anatomically and functionally.
The primary somatosensory barrel cortex can be divided along its depth into anatomically defined layers, from superficial layer 1 to deep layer 6. Thalamocortical neurons located in so-called ‘barreloids’ of the VPM densely innervate layer 4 (with a more sparse innervation of upper layer 6), with each whisker being individually represented by a segregated termination field of somatotopically arranged thalamocortical axons defining the cortical barrel map (Fig. 1B and C; Woolsey & Van der Loos, 1970). The excitatory VPM axons are thought to provide the most important pathway for bringing sensory information relating to individual whisker deflections to S1. In the mouse, each layer 4 barrel is composed of a central region of high density neuropil, containing the clustered VPM axonal arborizations surrounded by a cell-dense wall of layer 4 neurons that orient their dendritic and axonal arborizations into one specific barrel (Woolsey et al., 1975). Optical stimulation has been used to study the functional projection from VPM thalamus to barrel cortex, revealing prominent VPM glutamatergic input onto neurons located in layer 4, layer 5B and layer 6 (Bureau et al., 2006; Petreanu et al., 2009; Cruikshank et al., 2010).
Thalamocortical POM neurons also project to the primary somatosensory barrel cortex, terminating densely in layer 1 and layer 5A. Functional characterization of this projection has revealed a prominent POM input onto layer 5A barrel cortex neurons (Bureau et al., 2006; Petreanu et al., 2009). In the rat, several subdivisions and additional parallel pathways have been characterized from the principal trigeminal and spinal trigeminal nuclei via different subdivisions of the VPM and POM thalamus (Deschenes, 2009). It has been suggested that these parallel pathways in the rat process different aspects of whisker sensorimotor information (Yu et al., 2006). However, in the mouse, little is currently known about the differential sensory information signalled by VPM and POM neurons, and further experimental work focusing on these issues will be of great importance.
Progress has also been made toward defining the synaptic circuits within mouse S1 barrel cortex through simultaneous whole-cell recordings (Lefort et al., 2009) and glutamate uncaging (Bureau et al., 2006; Xu & Callaway, 2009), providing complementary data to that obtained in rat (Lübke & Feldmeyer, 2007; Schubert et al., 2007). These studies have revealed several prominent synaptic pathways for processing sensory information within a cortical barrel column (defined as the entire thickness of the cortex from layer 1 to layer 6 and laterally bounded by the extent of the layer 4 barrel). Specific investigation of the microcircuits in the C2 barrel column revealed that excitatory neurons in layer 4 dominate synaptic connectivity within this barrel column (Lefort et al., 2009). Layer 4 neurons signal to neurons located in all other cortical layers and they are therefore able to robustly transmit to the entire barrel column the tactile information received via the dense layer 4 innervation by VPM. Other prominent neocortical signalling pathways in the mouse C2 barrel column are from supragranular to infragranular layers, with an interesting elevated reciprocal connectivity between layer 2 and layer 5A (Bureau et al., 2006; Lefort et al., 2009).
In vivo recordings from mouse barrel cortex neurons are beginning to shed light on how these neocortical microcircuits operate functionally during behavior (Crochet & Petersen, 2006; Poulet & Petersen, 2008; Gentet et al., 2010). However, whisker sensory perception and its contribution to driving goal-directed behaviors do not arise solely from the activity of neurons located in S1, but rather result from interactions of neurons across multiple brain areas. It is therefore of key importance to understand how sensory information is further processed in areas downstream of an individual barrel column.
Functional mapping of the cortical sensory response to whisker deflection
Voltage-sensitive dye imaging can be used to resolve the spatiotemporal dynamics of membrane potential changes in the supragranular layers with millisecond temporal resolution and subcolumnar spatial resolution (Grinvald & Hildesheim, 2004). The earliest cortical response to a single whisker deflection arises in the related barrel column in the contralateral hemisphere. If the right C2 whisker is deflected then cortical sensory processing begins in the C2 barrel column of the left hemisphere with a latency of ∼10 ms (Fig. 2A). The earliest response is highly localized to a single cortical column. However, depending upon stimulus strength, brain states and behavioral states (Petersen et al., 2003; Ferezou et al., 2006, 2007; Berger et al., 2007), the sensory response can spread across a large cortical region. If the stimulus is delivered during a quiet state, a single rapid whisker deflection evokes a sensory response which spreads to neighboring cortical columns of S1 barrel cortex and secondary somatosensory (S2) cortex. In addition, ∼8 ms after the first cortical response, a second localized region of activity is observed in the primary motor cortex (M1), which also spreads into neighboring regions. A single brief whisker deflection can therefore result in two localized regions of activity from which propagating waves of activity spread across the sensorimotor cortex. At later times, activity also spreads into the cortex ipsilateral to the stimulated whisker, appearing initially in frontal cortex, M1 and S2. Finally, but still within 100 ms of the initial whisker deflection, depolarization spreads with apparent bilateral symmetry to posterior parietal association cortex.
A single whisker deflection therefore evokes a sensory response, which extends across a large part of the dorsal neocortex in a complex spatiotemporal pattern. There are, however, many possible pathways for signalling sensory information to the neocortex. The contribution of primary somatosensory barrel cortex to the whisker-evoked sensorimotor response was thus examined by the specific inactivation of the cognate barrel column (in this case the C2 barrel column) by injection of ionotropic glutamate receptor antagonists (Ferezou et al., 2007). Inactivation of the C2 barrel column almost completely blocked the entire sensorimotor response, while leaving the response to deflection of another nearby whisker (the E2 whisker) nearly unaltered (Fig. 2B and C; Ferezou et al., 2007). A significant part of the widespread sensory response evoked by a single C2 whisker deflection is therefore driven by activity in the C2 barrel column. The axonal projections of excitatory pyramidal neurons in supragranular layers 2/3 and infragranular layers 5/6 of the C2 barrel column are therefore likely to reveal important aspects of how the C2 whisker sensory-evoked response is transmitted across these different brain regions. Furthermore, voltage-sensitive dye imaging only provides information related to the superficial dorsal neocortex, and it is likely that there are many additional targets of barrel cortex axons. The remainder of this review will focus on the anatomical connectivity of the mouse barrel cortex with specific reference to axonal output from the C2 barrel column.
Anatomical connectivity can be studied by directly injecting classical tracers or viral vectors (which can also be used as anatomical tracers) into the specific brain region under investigation. Intrinsic optical imaging provides a simple way to localize the functional representation of the mouse C2 whisker through the intact skull (Fig. 3A; Ferezou et al., 2006; Aronoff & Petersen, 2007; Lefort et al., 2009). By aligning the intrinsic optical signal with the surface blood vessels, a small craniotomy can be made over the C2 whisker representation in S1 barrel cortex, enabling direct injection of anatomical tracers into the C2 barrel column (Fig. 3B). Injection of a lentivirus into the functionally identified C2 whisker representation localized by intrinsic optical imaging results in labelling of neurons located in the C2 barrel column (Fig. 3C). Intrinsic optical imaging therefore provides a reliable map of S1, allowing anatomical tracers to be reliably targeted to the C2 barrel column.
If biotinylated dextran amine (BDA) is injected into the cortex, it is taken up locally by neuronal cell bodies and diffuses into their dendrites and axons (Fig. 3D). Because of the biotinylation, BDA can be readily stained, providing high contrast fluorescence images. BDA is therefore an anterograde tracer which can be used to study the axonal output of a given brain area. However, it should be noted that BDA is also to some extent taken up by axons near the injection site (especially when it is pressure-injected), meaning that there is also some labelling of axons with cells bodies (and their axonal collaterals) located far from the injection site. Such collateral labelling complicates the interpretation of BDA-labelled tissue.
Fluorogold (FG) injected into the cortex is taken up by axonal boutons and transported retrogradely to the soma. FG labelling is prominent in the cytoplasm of neuronal soma located in brain regions projecting to the injection site, and FG is therefore a useful retrograde tracer.
These ‘classical’ anatomical methods are now complemented by a variety of viral vector strategies for labelling (Fig. 3E and F), which may offer higher specificity for anatomical tracing and, in addition, provide the opportunity for genetic manipulation of the transduced cells. Vesicular stomatitis virus-glycoprotein (VSV-G) pseudotyped lentivirus encoding green fluorescent protein (GFP; Lenti-GFP; Dittgen et al., 2004) injected into the cortex transduces almost exclusively neurons locally near the injection site. The GFP is soluble and diffuses along the dendrites and axons of the transduced neurons, including long-range axonal projections. Lenti-GFP can therefore be used as an unequivocal anterograde anatomical tracer (Ferezou et al., 2007; Broser et al., 2008a).
Whereas VSV-G pseudotyped lentivirus only transduces neurons with somata within a few hundred microns of the cortical injection site, other viral vectors behave quite differently. Adeno-associated viruses (AAVs) are physically much smaller, so they can diffuse further, transducing neurons across larger brain regions. Different serotypes of AAV have different properties and, like adenovirus and rabies virus, some AAVs can be retrogradely transported after axonal uptake of vector (Taymans et al., 2007; Hollis et al., 2008). AAV serotype 6 (AAV6; Grimm et al., 2003) binds to heparin (like AAV serotype 2, but different from other serotypes) and probably because of this binding it diffuses less in the brain than many other AAV serotypes. Nonetheless, neurons transduced with AAV6 are found far from the injection site, presumably because of retrograde transport (Kaspar et al., 2003; Towne et al., 2008, 2010). Injection of AAV6 encoding a ‘humanized’ cre-recombinase (AAV6-Cre; Shimshek et al., 2002; Fig. 3F) into Rosa floxed-LacZ cre-reporter mice (Soriano, 1999), allows staining of transduced neurons with the blue XGal chromogenic substrate. If the AAV6-Cre vector is injected into the neocortex, it is taken up by axon boutons near the injection site (while also transducing neurons with somata near the injection site). The AAV6-Cre is then retrogradely transported to the nucleus of neurons with axonal projections to the injection site, and the subsequent expression of cre-recombinase can be monitored in cre-reporter mice. AAV6-Cre can therefore be used as a retrograde vector for anatomical labelling of neurons projecting to the injection site.
Both the classical anatomical tracers and the viral vectors can be injected simultaneously to allow labelling of both anterograde and retrograde connectivity from a single well-defined injection site.
Local connectivity within S1 and S2
Voltage-sensitive dye imaging reveals that activity within the C2 barrel column rapidly propagates to neighboring cortical columns (Fig. 2). This spread is likely to be mediated, at least in part, by the extensive local axonal projections of the pyramidal neurons located in the C2 barrel column. Injections into the C2 barrel column of the anterograde tracers Lenti-GFP (Fig. 4A and B; Dittgen et al., 2004) or BDA (Fig. 4C) indicate that C2 barrel cortex neurons extend axonal arborizations into layers 2/3 and layers 5/6, almost across the entire extent of S1 barrel cortex. The density of axons is highest close to the C2 barrel column and decreases across the neighboring cortical columns (Brecht et al., 2003; Petersen et al., 2003; Broser et al., 2008b). In addition, increased axonal innervation can be observed in the dysgranular zone (medial column of axons seen on the right side of Fig. 4B), a region immediately surrounding the S1 barrel field proper. The axons within S1 probably mediate the rapid spread of sensory information across the barrel map; this may be of importance during normal whisker sensation, when sensory input from different whiskers must be integrated to build up a representation of the external world.
Another region with high axonal density across all layers is observed ∼1 mm lateral of the C2 barrel column, corresponding to the location of S2 (Fig. 4A–C; White & DeAmicis, 1977; Welker et al., 1988; Fabri & Burton, 1991; Hoffer et al., 2003; Chakrabarti & Alloway, 2006). The high density of axonal innervation in S2, originating from S1, and the spatial proximity of S2 and S1 probably underlie the extremely rapid sensory signals that are observed in these regions with voltage-sensitive dye imaging. Indeed, the signal in S2 is only resolved with voltage-sensitive dye imaging as a separately activated region when the more medially represented E2 whisker is deflected (Fig. 2B and C). Furthermore, S1 and S2 regions are reciprocally connected, as revealed by retrograde labelling with FG (Fig. 4D) and AAV6-cre in floxed-LacZ cre-reporter mice (Fig. 4E).
There are interesting differences in the axonal projections from S1 to M1, when comparing the pattern of axonal output from superficial layers 2/3 pyramidal neurons to the pattern of axonal output from deep layers 5/6 pyramidal neurons. Supragranular S1 layers 2/3 pyramidal neurons showed the densest innervation of deeper layers 5/6 in M1 and stopped short of the outer layer 1 (Fig. 5C and D, left), whereas the infragranular S1 layers 5/6 pyramidal neurons preferentially innervated the superficial layers of M1, with a prominent innervation of the most superficial layer 1 (Fig. 5C and D, right). It will be of great interest in future studies to examine the functional consequences of the layer-specific projections from S1 to M1. In addition, anterograde tracers injected into M1 (Veinante & Deschênes, 2003) and retrograde tracers injected into S1 indicate that S1 and M1 are reciprocally connected (Fig. 5B).
S1 connectivity to other cortical areas
In addition to the prominent axonal projections from S1 to S2 and M1 on the same hemisphere of the brain, a number of reciprocal projections to other cortical regions are seen: bilateral projections to perirhinal cortex (temporal association cortex; Fig. 4A), projections to ipsilateral orbital cortex and weaker projections to the contralateral somatosensory cortex (Petreanu et al., 2007) and contralateral motor cortex. The bilateral projection from S1 to perirhinal cortex extends across a large part of the rostrocaudal axis and connectivity is clearly weaker to the contralateral perirhinal cortex. This projection from S1 to perirhinal cortex could underlie the signalling of sensory information towards brain regions involved in higher level object-oriented coding and might contribute to hippocampal sensory responses (Pereira et al., 2007).
S1 connectivity to thalamus
Sensory information in S1 arrives via ipsilateral thalamocortical inputs from at least two subdivisions of the thalamus (VPM and POM), which are labelled by injection into the C2 barrel column of FG or AAV6-Cre (Fig. 6A). These ipsilateral thalamic nuclei are also prominently innervated by corticothalamic axonal projections from S1 into VPM and POM (Fig. 6B and C; Chmielowska et al., 1989; Bourassa et al., 1995; Deschenes et al., 1998; Veinante et al., 2000). No S1 projections to contralateral thalamus are observed. Specific labelling of supragranular vs. infragranular neurons using Lenti-GFP indicates that corticothalamic projections from S1 are mediated by infragranular neurons.
Although the axonal density from infragranular S1 is high in both VPM and POM, the fine-scale structure of the boutons is quite different (Fig. 6B). The S1 projection to a barreloid of VPM, originating primarily from layer 6 corticothalamic neurons, has small boutons (Fig. 6B, bottom left), whereas the S1 projection to POM, originating from layer 5B corticothalamic neurons, has both small and very large boutons (Fig. 6B, bottom right; Hoogland et al., 1991; Groh et al., 2008). The large size boutons in POM derive from layer 5B pyramidal neurons and have been suggested as representing driver synapses (Sherman & Guillery, 1998), providing a strong excitation to the postsynaptic POM neurons (Diamond et al., 1992; Groh et al., 2008). On the other hand, the small size boutons terminating in VPM may have a more modulatory role.
The glutamatergic corticothalamic axons therefore directly contribute to depolarizing and exciting thalamic relay neurons, which in turn form excitatory projections back to the cortex. In addition, the corticothalamic axons also contact inhibitory thalamic GABAergic neurons (Cruikshank et al., 2010). After passing the internal capsule, the descending corticothalamic axons send off branches into the thalamic reticular nucleus, which contain GABAergic neurons that in turn project strongly to thalamic relay nuclei, including VPM and POM (Pinault et al., 1995; Cox et al., 1997; Crabtree et al., 1998). In addition, the cortex also projects to diencephalic GABAergic neurons in the zona incerta (Mitrofanis & Mikuletic, 1999; Barthóet al., 2007) and the anterior pretectal nucleus (Fig. 6D; Wise & Jones, 1977a; Foster et al., 1989). Neurons in the zona incerta and anterior pretectal nucleus also exert a strong GABAergic inhibition of thalamocortical neurons in higher order thalamic nuclei, including POM (Barthóet al., 2002; Bokor et al., 2005), with functionally different properties to that arising from the thalamic reticular nucleus (Wanaverbecq et al., 2008). There are thus multiple pathways providing negative feedback control loops for the corticothalamocortical system.
S1 connectivity to striatum
Another prominent region of profuse axonal arborization originating from neurons with soma located in the C2 barrel column of S1 is found in the dorsolateral striatum (caudate–putamen; Fig. 7; Wright et al., 1999; Alloway et al., 1999; Hoover et al., 2003; Alloway et al., 2006). Corticostriatal projections are predominantly from infragranular layers, but supragranular pyramidal neurons also provide input to the striatum (Royce, 1982; Gerfen, 1989; Cowan & Wilson, 1994). Excitatory input from S1 to the dorsal striatum forms an important pathway for cortex to influence the operation of the basal ganglia, which are thought to be important for motor control and action selection. Unlike the corticocortical and corticothalamic connections, no retrograde labelling by FG or AAV6-cre was observed in the striatum, suggesting a one-way flow of information. Neurons in the caudate–putamen interact with the more medially located neurons in the globus pallidus. The pallidal neurons in turn influence the thalamus, which of course interacts strongly with cortex, thus completing a long subcortical loop back to the neocortex.
S1 connectivity to superior colliculus, pons and brain stem
Further posteriorly, the S1 axons of infragranular pyramidal neurons make dense termination fields in the deep layers of the superior colliculus (Fig. 8A and B), pons (Fig. 8C and D), red nucleus and spinal trigeminal nuclei (Fig. 8E and F).
The superior colliculus (also known as the tectum) is thought to play a prominent role in spatial orientation, for example contributing to saccadic eye movements in the visual system. In the whisker sensorimotor system, the superior colliculus might well contribute to orienting whisker movements to palpate objects and surfaces that have attracted the animal’s attention. Corticotectal neurons projecting from S1 to the superior colliculus (Fig. 8A and B; Wise & Jones, 1977b) might therefore signal the presence of interesting sensory information (Cohen et al., 2008; Cohen & Castro-Alamancos, 2010) and might help direct attention and movements of whisker and head towards the region of interest (Benedetti, 1995; Hemelt & Keller, 2008).
The pons forms an important gateway for relaying information to the cerebellum, via the pontocerebellar projection to the contralateral hemisphere. In analogy to the basal ganglia circuit (linking cortex to striatum to thalamus and back to cortex), a corticocerebellar loop has also been described (linking cortex to pons to cerebellum to deep cerebellar nuclei to thalamus and back to cortex; Strick et al., 2009). The cerebellum is known to play important roles in motor refinement and learning. The corticopontine projection from S1 (Fig. 8C and D; Legg et al., 1989; Leergaard et al., 2000; Schwarz & Möck, 2001; Leergaard et al., 2004) might therefore be involved in fine-scale motor control in order to optimize the acquisition of sensory information. Interestingly, the cerebellum is apparently required for one well-studied somatosensory cortex-dependent and whisker-dependent task, known as gap crossing, in which the animal must identify the location of a target platform with its whiskers alone (Jenkinson & Glickstein, 2000).
In the brain stem, the S1 axons cross to the contralateral hemisphere forming extensive arborizations in the principal trigeminal nucleus and spinal trigeminal nuclei, with prominent labelling of caudalis (SP5c) and interpolaris (SP5i) subdivisions (Fig. 8E and F; Jacquin et al., 1990). The corticospinal projection from S1 to spinal trigeminal nuclei forms an interesting pathway by which primary somatosensory cortex can influence very early sensory processing in brain stem neurons, which are the immediate recipients of the primary sensory trigeminal ganglion input. Such a top-down input to the brain stem could influence important aspects of sensory processing; for example, it might enhance signalling of selected sensory information when the animal is attempting to actively acquire and process specific tactile whisker input.
Both functional and anatomical studies highlight the involvement of multiple well-defined brain regions in processing tactile whisker sensory information. The most prominent aspects of the long-range connectivity of the mouse C2 barrel column is qualitatively summarized in Fig. 9, including both anterograde and retrograde data. In future studies, it will be of enormous importance to establish quantitative maps of long-range anatomical connectivity in the mouse brain, perhaps in conjunction with brain atlases based on gene expression patterns (Lein et al., 2007). In addition, the specific functional roles that different brain areas contribute to whisker-dependent behaviors can now be examined with unprecedented precision. The recent development of optogenetic tools (Nagel et al., 2003; Boyden et al., 2005; Zhang et al., 2007) combined with new viral tools and the increasing number of cell-type-specific genetically engineered mice (Aronoff & Petersen, 2006; Gong et al., 2007; Livet et al., 2007; Wickersham et al., 2007a&,b; Luo et al., 2008; Cardin et al., 2009; O’Connor et al., 2009; Sohal et al., 2009) is likely to help significant progress be made over the coming decades into the causal analysis of which neurons in the brain serve which functions during whisker behaviors.
We thank Dr Laszlo Acsady for valuable comments on the manuscript. We are grateful to Derya Shimshek for the iCre clone and advice on combining the lacZ and GOD-DAB staining. We thank Pavel Osten for providing the CaMKII lentivector. We thank the histology core facility of the EPFL Faculty of Life Science for help with tissue processing, and the Trono and Aebischer laboratories at the EPFL for virus production and support, and for advice on immunohistochemistry. We are grateful to grants from Swiss National Science Foundation, SystemsX.ch, Human Frontiers in Science Program and EMBO.