Cyclooxygenase-2 (COX-2) is a rate-limiting enzyme for prostanoid synthesis that is present in cortical pyramidal neurons and highly implicated in control of cerebral blood flow during neural activity. We examined the electron microscopic localization of COX-2 and neuronal nitric oxide synthase (nNOS), a functionally related enzyme, in the somatosensory cortex of rat brain to determine the relevant functional sites. COX-2 immunoreactivity was detected in significantly more somatodendritic than axonal profiles, while nNOS was more often seen in axon terminals. The dendritic COX-2 was localized to endomembranes near synaptic inputs from axon terminals, some of which contained nNOS. Conversely, COX-2 terminals formed asymmetric, excitatory-type synapses with dendrites containing nNOS. The dendritic and axonal profiles containing COX-2 as well as those containing nNOS were minimally separated from penetrating arterioles and capillaries by filamentous glial processes. The perivascular COX-2 labeled terminals were among those that also formed axo-dendritic synapses, suggesting that the release of prostanoids and/or excitatory transmitters from a single terminal may simultaneously affect neuronal activity and cerebral blood flow. Thus, COX-2 has a compartmental distribution in somatosensory cortical neurons consistent with the local neuronal synthesis of prostanoids that are involved in neurovascular coupling and whose actions are modulated by nitric oxide.
The prostaglandin synthesizing enzyme, cyclooxygenase-2 (COX-2) is present in glutamatergic neurons of the cerebral cortex and many other regions of adult brain (Yamagata et al., 1993; Kaufmann et al., 1997). Unlike the COX-1 isoform of this enzyme, COX-2 is induced by synaptic activity in adult and developing brain (Worley et al., 1991; Kaufmann et al., 1996). In the limbic cortex and hippocampal formation, COX-2 also has been implicated in memory and learning (Shimizu and Wolfe, 1990; Yamagata et al., 1993). In addition, COX-2 plays an important role in mediating the changes in neocortical blood flow evoked by neural activity (Niwa et al., 2000), a critical homeostatic mechanism that matches the delivery of nutrients with the energy needs of the active brain (Iadecola, 2004).
Like COX-2, neuronal nitric oxide synthase (nNOS) is present in the cerebral cortex and other brain regions (Aoki et al., 1998; Calabrese et al., 2000; Contestabile, 2000), and has been implicated in several aspects of synaptic function including the regulation of cerebral blood flow during synaptic activity (Iadecola and Niwa, 2002). Furthermore, in some organs including the brain, nitric oxide can enhance the catalytic activity of COX-2 (Salvemini, 1997; Nogawa et al., 1998). However, the structural relationships between nNOS and COX-2 neurons have not been fully elucidated. In the cerebral cortex, nNOS has a regional distribution comparable to COX-2 in that both enzymes are prevalent in layer I as well as in layers III–V (Bredt et al., 1991; Degi et al., 1998). Light microscopic studies suggest, however, that COX-2 and nNOS are differentially enriched in cortical pyramidal cells and interneurons, respectively (Bidmon et al., 2001). Moreover, electron microscopy shows that nNOS and GABA are present in perivascular neuronal networks in the cerebral cortex (Vaucher et al., 2000). Synaptic interactions are known to occur between pyramidal neurons and subsets of interneurons in the cerebral cortex (Blatow et al., 2003), suggesting that synapses occur between neurons differentially containing COX-2 and nNOS. The detection of these synapses depends, however, on whether each of these enzymes is selectively targeted to distal dendritic and axonal processes. While nNOS has been localized by electron microscopic immunocytochemistry in both dendrites and axon terminals of the cerebral cortex (Aoki et al., 1998), neither the neuronal compartmental nor subcellular distribution of COX-2 has been established. We used electron microscopic immunocytochemistry to determine the subcellular distribution of COX-2 and to determine whether this enzyme is located within either dendritic or axonal profiles having synaptic associations with neurons containing nNOS. We focused the analysis on layers II–V of the somatosensory cortex, where vibrissae stimulation produces activation of pyramidal neurons (Filipkowski et al., 2000) and enhancement of cerebral blood flow (Kleinfeld et al., 1998; Niwa et al., 2000). The subcellular distribution of COX-2 in these layers was qualitatively compared with the COX-2 labeling near penetrating arterioles in somatosensory cortical layer I.
The results provide the first evidence that COX-2 has an endomembrane dendritic distribution near synaptic inputs from terminals, some of which contain nNOS, and is also present in excitatory-type terminals presynaptic to nNOS dendrites. Each of these enzymes also is shown to be present in neuronal processes that are minimally separated from the walls of penetrating arterioles and capillaries by thin glial processes.
Materials and Methods
Adult male Sprague–Dawley rats were obtained from Taconic Farms (Germantown, NY). All animals used were cared for in accordance with the National Institutes of Health Guidelines for the Care and Use of Laboratory Animals. The animal protocol used was approved by the Research Animal Resource Center at Weill Medical College of Cornell University. A total of eight rats were used for COX-2 localization: four for immunoperoxidase and two for immunogold single labeling, and two for COX-2 and nNOS dual labeling in which the immunoperoxidase and immunogold markers were used reversible for detection of each of the enzymes. All efforts were made to minimize the number of animals used and their suffering.
A rabbit affinity polyclonal antiserum (catalog no. 160106, Cayman Chemical Company, Ann Arbor, MI) was raised against a synthetic peptide (amino acids 584–598) mapping at the C-terminal of murine COX-2 (Fletcher et al., 1992; Zhang et al., 1996). The C-terminal region of COX-2 is unique for this COX isoform, and recognizes a 72 kDa protein corresponding to COX-2 cloned from rat and several other species including human (Hla and Neilson, 1992; Yamagata et al., 1993). This antiserum has been previously tested for specificity (Iadecola et al., 1999), and shown to produce immunolabeling that is selectively removed by prior adsorption of the antiserum with blocking peptide (catalogue no. 360106, Cayman Chemical Company). A mouse monoclonal antibody (catalog no. 610309, Becton Dickinson BD Biosciences/Pharmingen) was generated against a sequence at the C-terminal of human nNOS (amino acids 1095–1289). The antiserum was tested in Western blot and in immunolabeling studies (Togashi et al., 1997; Sasaki et al., 2000; Yu et al., 2000).
Animals were anesthetized with sodium pentobarbital (100 mg/kg, i.p.) and their brains were fixed by rapid sequential vascular perfusion with 10 ml of heparin saline (1000 U/ml), 50 ml of 3.8% acrolein and 2% paraformaldehyde solution, and 200 ml of 2% paraformaldehyde. All fixation solutions were made in 0.1 M phosphate buffer (PB), pH 7.4, at room temperature. The aldehyde-fixed brains were removed from the cranium and cut rostrocaudally in a coronal plane into thick slices including the region of the somatosensory cortex. These slices were postfixed in 2% paraformaldehyde for 30 min and then sectioned at 40 μm thickness using a Leica Vibratome VT1000 S (Leica Instruments GmbH, Nussloch, Germany) in chilled 0.1 M PB. Sections through the somatosensory cortex were collected at levels corresponding to adult rat brain atlas between 2.20 mm anterior and 0.30 mm posterior to Bregma (Paxinos and Watson, 1986).
Immunoperoxidase Labeling for COX-2
Sections were first incubated in a 1% sodium borohydride solution for 30 min and rinsed in 0.1 M PB until no bubbles emerged from the tissue. To enhance the penetration of immunoreagents while maximally preserving cellular membranes, 0.5% Triton X-100 was included in the primary antisera solutions only for light microscopy; whereas electron microscopic immunolabeling was preceded by rapid freeze–thawing of tissue sections that had been incubated for 15 min in a cryoprotectant (25% sucrose and 2.5% glycerol in 0.05 M PB) solution. For this, the cryoprotected sections were immersed in liquid freon followed by liquid nitrogen, and thawed in PB at room temperature. These sections were washed (2 × 5 min) in 0.1 M Tris buffer, pH 7.6, containing 0.9% saline (TBS), and placed in a blocking solution of 0.5% bovine serum albumin (BSA) in TBS for 30 min. This was followed by TBS rinse and antisera incubation. The sections were incubated for 1 day at room temperature and a consecutive day at 4°C in the rabbit COX-2 antiserum, diluted at 0.5 μg/ml in 0.1% BSA in TBS. The sections were then incubated for 30 min in donkey anti-rabbit biotinylated IgG (1:400 in 0.1% BSA and 0.1 M TBS; Jackson ImmunoResearch, West Grove, PA), followed by another 30 min incubation in avidin–biotin–peroxidase complex (1:100 in 0.05% BSA and 0.1 M TBS; Vectastain Elite kit, Vector Laboratories, Burlingame, CA). The peroxidase bound to the sections was visualized by a 6 min incubation in 0.022% 3,3′-diaminobenzidine and 0.003% hydrogen peroxide. Sections were rinsed in 0.1 M TBS between incubations.
Immunogold-silver Labeling for COX-2
Sections were incubated in rabbit COX-2 antiserum diluted at 2 μg/ml in 0.1% BSA and 0.1 M TBS for 1 day at room temperature followed by a consecutive day at 4°C. These sections were next rinsed in 0.01 M phosphate-buffered saline (PBS), pH 7.4, blocked for 10 min in 0.8% BSA and 0.1% gelatin in 0.01 M PBS, and then placed for 2 h in goat anti-rabbit IgG conjugated with 1 nm gold particles (AuroProbeOne; Amersham, Arlington Heights, IL; 1:50 in the BSA-gelatin blocking solution). Sections were rinsed in 0.01 M PBS and post-fixed in 2% glutaraldehyde in 0.01 M PBS for 10 min. The gold particles were silver enhanced by using the IntenSE-M kit (Amersham) for 6–7 min at room temperature.
Dual labeling for COX-2 and nNOS
Sections were incubated in a mixture of a rabbit anti-COX-2 antiserum and a mouse anti-nNOS antibody. The concentrations used were 0.5 μg/ml (immunoperoxidase) and 2 μg/ml (immunogold) for COX-2, and 1.25 μg/ml (immunoperoxidase) and 5 μg/ml (immunogold labeling) for nNOS. After the 2 day primary antisera incubation, sections were first processed for peroxidase labeling as described above, using donkey anti-rabbit (for COX-2) or horse anti-mouse (for nNOS) biotinylated IgG (Jackson), respectively. For gold labeling, sections were also processed as described above using goat anti-rabbit (for COX-2) or goat anti-mouse (for nNOS) IgG conjugated with 1 nm gold particles (AuroProbeOne; Amersham; 1:50 in the BSA-gelatin blocking solution).
Immunolabeled sections for electron microscopy were postfixed for 1 h with 2% osmium tetroxide in 0.1 M PB, dehydrated through increasing concentrations of ethanol and propylene oxide, and incubated overnight in a 1:1 mixture of propylene oxide and epon (EMbed-812 kit, Electron Microscopy Sciences). The sections were transferred to 100% epon for 2–3 h and then flat-embedded between two sheets of Aclar plastic film (Allied Signal, Pottsville, PA). The flat-embedded tissue was viewed with a light microscope in order to select regions for ultrastructural analysis of layer I or layers II–V of the somatosensory cortex. In each of these regions, ultrathin sections (70 nm) were cut with a diamond knife (DiATOME US, Fort Washington, PA) on an ultramicrotome (Leica Ultracut UCT; Leica, Wien, Austria) and collected on 400 mesh copper grids (Electron Microscopy Sciences). These ultrathin sections were counterstained with 5% uranyl acetate followed by Reynold's lead citrate (Reynolds, 1963) prior to electron microscopic analysis.
Light and Electron Microscopic Data Analyses
The immunoperoxidase-labeled sections were processed either as described above for electron microscopy, or were mounted on glass slides and examined with a Nikon Microphot-FX light microscope (Nikon, Garden City, NY) equipped with a digital CoolSNAP camera (Photometrics, Huntington Beach, CA). The ultrathin sections from the plastic-embedded tissues prepared for single or dual labeling were examined at 60 kV with a Tecnai transmission electron microscope (FEI Company, Hillsboro, OR). The electron microscopic images were captured by using AMT Advantage HR/HR-B CCD Camera System (Advanced Microscopy Techniques, Danvers, MA). The acquired light and electron microscopic images were adjusted for contrast and brightness using Photoshop 6.0 software, and imported into PowerPoint 2001, to add lettering and prepare the composite figures.
The images acquired for electron microscopy were obtained exclusively at the outer (∼1 μm) surface of the tissue to minimize artificial differences in labeling that might be attributed to antibody penetration. In tissue processed for COX-2 immunoperoxidase labeling, the ultrathin sections were collected from layers I or II–V of eight vibratome sections from four rats. In layers II–V, we also examined ultrathin sections from tissue processed by immunogold labeling for COX-2 (four vibratome sections from two rats) or dual immunogold labeling for COX-2 and immunoperoxidase labeling for nNOS (eight vibratome sections from two rats). The total examined electron micrographs were 484 representing 16 383 square microns. The numbers of labeled neuronal processes were grouped according to their dendritic or axonal features, and compared by using a chi-squared test to determine statistically significant differences in numbers of profiles in each group. Neuronal (somata, dendrites, dendritic spines, axons and axon terminals) as well as glial profiles and symmetric or asymmetric synapses and vascular structures were classified according to Peters et al. (1991). Arterioles were distinguished from capillaries both by their larger diameter and presence of layers of smooth muscle cells that are usually not seen in either capillaries or veins (Peters et al., 1991).
A profile was considered as selectively labeled when either the number of gold particles or the electron microscopic density of the peroxidase reaction product was greater than that of other morphologically similar profiles in the neuropil.
Light microscopy showed COX-2 immunoreactivity within neuronal perikarya and processes that were prevalent throughout the somatosensory cortical layers (Fig. 1A,B). COX-2 labeled processes resembling apical dendrites of pyramidal neurons were oriented perpendicularly to the cortical surface. These processes passed near several blood vessels within the neuropil (Fig. 1B). The peroxidase labeling for COX-2 was considerably less dense than that observed for nNOS, which was seen in isolated cells having the features of cortical interneurons (Fig. 1C,D). The nNOS-immunoreactive neurons were present in small numbers in most layers of the somatosensory cortex, and typically showed two or more thick straight processes presumed to be dendrites. The nNOS immunoreactivity also was prevalent in thinner varicose processes resembling axon collaterals arising from the interneurons (Fig. 1D). Electron microscopy confirmed the prominent somatodendritic and axonal distributions of COX-2 and nNOS in layers III–V as well as in many neuronal processes in layer I. In addition, the ultrastructural analysis established a compartmental distribution of COX-2 within somatosensory cortical neurons and identified reciprocal synaptic interactions between these neurons and those containing nNOS. These results are described in detail in the following sections, where we also demonstrate that COX-2 and nNOS-containing neurons are separated from arterioles and capillaries by only thin glial processes.
Immunolabeling for COX-2 was distributed within the cytoplasm and along the outer nuclear membranes of neuronal somata within the pyramidal cell layer. The COX-2 immunoreactivity was particularly abundant in large dendritic profiles (Fig. 2A), which were distinct from those that were intensely labeled for nNOS (Fig. 2B). While COX-2 was detected in dendrites that were lightly immunolabeled for nNOS (Fig. 2C), the differential distributions of these enzymes in many neuronal profiles ensures that the dual labeling is not due to a non-specific attachment of silver to the peroxidase reaction product. In dendritic profiles, the COX-2 immunogold was located near endomembranes resembling smooth endoplasm reticulum beneath synaptic inputs and also seen in the cytoplasm of dendritic shafts (Figs 2A, 3). The COX-2 immunogold labeling also was detected near endomembranes in dendritic spines, the majority of which were without nNOS immunoreactivity (Fig. 4A). In sections processed for single immunoperoxidase labeling of COX-2, the endomembrane distribution of the enzyme also could be seen in associations with endomembranes of dendritic spines (Fig. 4B). In comparison with COX-2, nNOS somatodendritic immunoreactivity was more variable in density. Intense nNOS labeling (Fig. 2B) was seen in only a few dendrites, which probably originate from the interneurons that were also seen by light microscopy to be intensely nNOS immunoreactive (Fig. 1D). Within these somatodendritic profiles, nNOS was distributed throughout the cytoplasm, which contained the usual cytoplasmic organelles including smooth and rough endoplasmic reticulum, Golgi and mitochondria. Some of these profiles also contained a few large dense core vesicles (Fig. 2B), which is consistent with the known presence of neuropeptide Y in similar storage vesicles within nNOS-containing cortical interneurons (Aoki and Pickel, 1990). In addition to the few intensely nNOS-immunoreactive dendrites, many other dendrites showed nNOS labeling that was barely above background, and only these dendrites were among those that also contained COX-2 (Fig. 2C). The numbers of dendrites containing nNOS and/or COX-2 are quantitatively compared in Figure 5, which includes the nNOS profiles that were either densely or lightly labeled.
Axonal Distributions and Synaptic Associations
Many small axons and axon terminals within the cortical neuropil were separately labeled for COX-2 or nNOS. In the axon terminals, the COX-2 immunogold was seen on or near the plasma membrane and in association with membranes of small synaptic vesicles (Fig. 4A), whereas the immunoperoxidase labeling was diffusely distributed throughout the axoplasm (Fig. 4C,D). Almost all COX-2 labeled terminals with recognizable membrane specializations formed asymmetric and sometimes perforated synapses with dendrites or dendritic spines (Fig. 4D). The postsynaptic structures included many dendrites and dendritic spines that expressed low levels of nNOS (Fig. 4C). Of 113 COX-2 labeled terminals, ∼30% contacted dendrites that were lightly immunolabeled for nNOS, while <1% were presynaptic to dendritic profiles that were intensely nNOS labeled or which contained both nNOS and COX-2. The remaining target dendrites were without detectable immunoreactivity. In contrast with COX-2, which was detected in significantly more dendrites than terminals (chi-squared test, P < 0.0001), nNOS terminals significantly outnumbered the nNOS-labeled dendrites (P = 0.0026; Fig. 5). In terminals, nNOS immunoreactivity was densely distributed throughout the cytoplasm around synaptic vesicles (Fig. 3A,B). These terminals either apposed without clearly defined junctions (Fig. 3A), or showed symmetric, inhibitory-type contacts (Fig. 3B). While nNOS was also detected in several axon terminals forming asymmetric-synapses with larger dendrites, these terminals were far less intensely immunolabeled for nNOS (Fig. 3C). The nNOS-labeled terminals of each type were observed presynaptic to the large dendrites immunolabeled for COX-2 (Fig. 3). A few of the more lightly nNOS immunoreactive terminals also contained detectable levels of COX-2 (not shown).
Vascular and Glial Associations
COX-2 immunolabeled dendrites were apposed to arterioles and capillaries, but separated from their basal laminae by thin glial processes (Fig. 6A). Filamentous glial processes also separated nNOS labeled somatodendritic profiles (Fig. 6B) and COX-2 immunoreactive terminals (Fig. 6C,D) from the basal laminae of the blood vessels. In some cases, the glial membranes separating the COX-2 immunoreactive profiles from the vascular basal lamina were in continuity with glial profiles containing bundles of intermediate filaments characteristic of astrocytic processes (Fig. 6C). In layer I, apposed vessels were identified as arterioles by the prominent layer of smooth muscle (Peters et al., 1991) (Fig. 6C). With few exceptions (Fig. 6D), the vascular glial processes, endothelial and muscle cells were without immunolabeling for either COX-2 or nNOS. The COX-2 labeled terminals in contact with perivascular astrocytes sometimes formed recognizable asymmetric axo-spinous synapses (Fig. 6D), suggesting that the release of prostanoids and/or excitatory transmitters from a single terminal may simultaneously affect neuronal activity and cerebral blood flow.
We have shown that in rat somatosensory cortex COX-2 is preferentially targeted to dendritic compartments near inputs from axon terminals, some of which contain nNOS. Conversely, COX-2 is localized within excitatory-type terminals presynaptic to dendrites that are either intensively or lightly labeled for nNOS, the latter of which also sometimes contain both enzymes. These results are consistent with light microscopic evidence that COX-2 and nNOS are largely differentially expressed in cortical pyramidal neurons and interneurons identified in the present and earlier studies (Bidmon et al., 2001). These results provide the first ultrastructural evidence for COX-2 distributions within both dendritic and axonal compartments where synaptic associations with nNOS-containing neurons may facilitate nitric oxide-dependent activation of COX-2 (Salvemini, 1997). In addition, we show that the somatosensory dendrites as well as terminals containing COX-2 or nNOS are often separated from the vascular walls by only thin layers of unlabeled astrocytic processes (Fig. 7). Together, our results have important implications for involvement of prostanoids in nitric oxide signaling and in the homeostatic regulation of cerebral blood flow with synaptic activity in the somatosensory cortex.
The COX-2 and nNOS antisera were shown to be selective for the C-terminal peptide sequences used for generation of the primary antisera (see Materials and Methods), but may also recognize homologous sequences in other proteins. In addition, the levels of basal expression of COX-2 in cortical neurons is relatively low and widely distributed (Madrigal et al., 2003). This is reflected in the modest light and electron microscopic immunolabeling for COX-2 in the somatosensory cortex of the present study. Immunogold–silver labeling is also less sensitive than the avidin–biotin–peroxidase method (Chan et al., 1990) resulting in our detection of few gold–silver particles for COX-2 in dendritic spines and other small profiles. The morphological features of the COX-2 immunoreactive dendrites and terminals were the same, however, irrespective of the labeling method used for detection of the enzyme. This suggests that the immunogold and immunoperoxidase methods are each capable of identifying COX-2 in cortical neurons. The more limited detection of COX-2 by immunogold–silver may have contributed, however, to an underestimation of the number of COX-2 labeled profiles as well as the frequencies with which these profiles are associated with either nNOS-containing neurons or blood vessels.
Somatodendritic COX-2 Distribution
COX-2 was intensely localized to radially oriented, often spiny, dendrites comparable to those described for pyramidal neurons (Nieuwenhuys, 1994). In these neurons, COX-2 was detected principally near endomembranes located beneath synapses on the plasma membrane or within the shaft of dendrites. This subcellular distribution is consistent with the known association of COX-2 with caveolea-like organelles in non-neuronal cells (Liou et al., 2001; Spisni et al., 2003). The observed endomembrane distribution of COX-2 in dendrites and dendritic spines of cortical pyramidal neurons provides the first ultrastructural evidence for a previously hypothesized dendritic compartmentation of COX-2 in brain (Kaufmann et al., 1997).
The compartmental distribution of COX-2 in postsynaptic dendrites could facilitate the activation of COX-2 by incoming signals resulting in the generation of retrograde signaling molecules affecting transmitter release (Bazan, 2003). COX-2 is crucial for the oxygenation of arachidonic acid to form prostaglandin H2, but also can generate glyceryl prostaglandins from 2-arachidonylglycerol (2-AG), one of two known endocannabinoids (Di Marzo et al., 1999; Beltramo and Piomelli, 2000; Kozak et al., 2000; Murakami et al., 2000; Nirodi et al., 2004). Increased synthesis of prostaglandins as well as the degradation of endocannabinoids could dramatically affect either the presynaptic release or postsynaptic responsiveness of cortical neurons that express COX-2.
Axonal COX-2 Distribution
The axonal location of COX-2 in the somatosensory cortex provides the first ultrastructural evidence that this enzyme is locally available in axon terminals for the generation of prostanoids that may affect glutamate-induced ion currents, as has been shown by patch-clamp recording in rat periaqueductal gray (Shin et al., 2003). Consistent with this view, we frequently observed COX-2 in axon terminals forming asymmetric synapses, which are typical of those that contain excitatory-amino acids (Nieuwenhuys, 1994). The morphology of the COX-2 labeled terminals is also consistent with their likely origin from glutamatergic COX-2 containing pyramidal neurons within the same, or functionally interrelated, cortical layers (Schubert et al., 2001, 2003). COX-2 is expressed, however, in glutamatergic neurons of many brain regions (Kaufmann et al., 1997). Thus, at least some of the excitatory- type terminals containing COX-2 may be derived from the glutamatergic neurons of the thalamus that are activated by tactile stimuli and extensively project to the somatosensory cortex (Broman, 1994; Pinto et al., 2000). Collectively, these observations suggest that prostanoids locally generated by COX-2 in axon terminals may dynamically affect synaptic transmission between cortical neurons or between these neurons and thalamic inputs within the somatosensory cortex.
The COX-2 terminals forming asymmetric synapses with dendritic spines included those with notable perforations of their postsynaptic membrane densities. Perforated axo-spinous synapses are those most commonly associated with activity-dependent plasticity (Jones and Harris, 1995; Neuhoff et al., 1999). Such plasticity can be evoked by repetitive activation of excitatory pathways from the thalamus to the somatosensory cortex (Schlaggar et al., 1993; Cahusac, 1995; Fox, 1994; Fox et al., 1996). Both prostanoids and cannabinoids are implicated in glutamatergic synaptic transmission and plasticity (Auclair et al., 2000). Cortical pyramidal neurons also provide major input to the striatum (Bolam et al., 2000), where cannabinoid-1 (CB1) receptors are often localized to excitatory-type terminals forming perforated synapses on dendritic spines (Pickel et al., 2004). Thus, the detection of COX-2 in terminals forming this type of synapse in the somatosensory cortex may be a further indication of the involvement of prostaglandins or cannabinoids in synaptic plasticity in this brain region.
nNOS Interneuronal Distribution and Input to Neurons Containing COX-2
The intense labeling for nNOS within the cytoplasm of cells resembling inhibitory cortical interneurons confirms and extends earlier studies on the cortical distribution of this enzyme (Valtschanoff et al., 1993; Tong and Hamel, 2000). The symmetry of the synaptic junctions formed by many of the nNOS-labeled terminals is consistent with their origin from inhibitory interneurons that provide major inputs to cortical pyramidal neurons (Nicoll et al., 1996). Together, these observations support the conclusion that inhibitory interneurons are the only cells that express high levels of nNOS in the somatosensory cortex.
The nNOS immunoreactivity was lightly distributed in spiny dendrites as well as axon terminals forming asymmetric, excitatory-type synapses in the somatosensory cortex. The lightly nNOS-labeled neuronal profiles included some of those that also contained COX-2. These observations suggest that glutamatergic pyramidal neurons, including those containing COX-2, have at least some capacity to synthesize nitric oxide in the somatosensory cortex. Others have shown that nNOS is detectable by electron microscopic immunocytochemistry in dendritic spines of the visual cortex, where the expression is correlated with synaptic activity (Aoki et al., 1997, 1998). Conceivably, the constitutively expressed nNOS may also be induced by synaptic activity in somatosensory cortical pyramidal neurons containing COX-2.
The observed major input from nNOS terminals to COX-2 dendrites is consistent with the prominent differential distributions of these enzymes in interneurons and projection neurons in the somatosensory cortex. Cortical interneurons show considerable variation in the shape and span of their axonal arbors, but most provide significant input to the pyramidal neurons that are highly implicated in lateral inhibition affecting functional interactions within and between cortical columns (Krimer and Goldman-Rakic, 2001). Our observation that some of the more lightly nNOS labeled terminals form excitatory-type synapses with COX-2 dendrites suggests that nitric oxide co-released with either GABA or glutamate can produce postsynaptic activation of COX-2 in the somatosensory cortex.
The observed microvascular association of nNOS-containing neurons in the somatosensory cortex confirms and extends prior studies implicating nitric oxide in cortical vasodilation (Tong and Hamel, 2000). These nNOS-containing neurons, as well as those containing NPY (Aoki and Pickel, 1990) or vasoactive intestinal peptide (VIP; Wang et al., 2002), belong to subpopulations of GABAergic interneurons that send numerous projections to local microvessels (Wang et al., 2002; Hamel, 2004). The activity of these vascular-associated interneurons is potently regulated by release of acetylcholine from axon terminals that are mainly extrinsic in origin (Wang et al., 2002; Hamel, 2004). Calcium entry through glutamate NMDA receptors or other Ca2+-permeable ion channels also elicits nNOS activation (Bredt, 1996), which is consistent with our detection on nNOS in dendrites postsynaptic to excitatory-type terminals including those that contain COX-2. Moreover, nitric oxide release from the target dendrite may induce activation of COX-2 (Salvemini, 1997). Thus, the observed synapses between COX-2 excitatory-type terminals and nNOS dendrites may provide an important retrograde feedback circuit for controlling the vascular release of nitric oxide or co-stored neuropeptides.
The synaptic input from COX-2 terminals to nNOS containing perivascular interneurons is consistent with the fact that COX-2 derived prostanoids, most likely PGE2, and nNOS derived nitric oxide are involved in the neocortical vasodilation evoked by activation of the whisker barrel cortex (Niwa et al., 2000; Iadecola and Niwa, 2002). In addition, however, the release of the prostanoid products of COX-2 from neurons in the somatosensory cortex also may affect cerebral blood flow through mechanisms that are independent of nitric oxide. We observed COX-2 within postsynaptic dendrites and excitatory-type terminals showing dual neuronal and vascular associations. This arrangement provides an ideal substrate for COX-2 dependent generation of prostaglandins that might regulate cerebral blood flow to match synaptic activity.
The COX-2 immunoreactive neuronal processes were separated from the vascular basal lamina by thin layers of astrocytic processes that were almost exclusively without COX-2 labeling. This observation is consistent with reports indicating that normal astrocytes infrequently contain COX-2 (Hirst et al., 1998). The findings differ, however, from the proposal that vascular astrocytes are the major sources of prostaglandins evoking an increase in cerebral blood flow to match synaptic activity (Zonta et al., 2003). While cortical neurons and/or astrocytes may express COX-1 or other isoforms, the data from our laboratory suggest that COX-2 is the enzyme involved in the synthesis of the prostanoids that mediate the increases in blood flow produced by neural activity. The data show that while COX-2 null mice have a reduced cerebral blood flow response to whisker stimulation (Niwa et al., 2000), COX-1 null mice have a normal response (Niwa et al., 2001). Furthermore, the cerebral blood flow response to whisker stimulation is attenuated by COX-2 inhibitors, but not by COX-1 inhibitors (Niwa et al., 2000, 2001). Together with our present results, these observations suggest that the arrival of action potentials may result in COX-2-dependent release of prostaglandins along with glutamate from somatosensory cortical pyramidal neurons. These agents could influence blood flow either by acting directly on local blood vessels or by inducing the release of vasoactive factors from perivascular glia (Iadecola, 2004). Vascular astrocytes express both glutamate (Zonta et al., 2003) and prostanoid receptors (Kitanaka et al., 1996) (Fig. 7). Conceivably, the activation of glutamate or prostaglandin receptors in astrocytes could evoke the release of other vasodilators including nitric oxide, p450 metabolites, ATP and adenosine (Ko et al., 1990; Ito et al., 1992; Murphy et al., 1992; Peng et al., 2002; Simard et al., 2003). Alternatively, the prostaglandins could reach contractile cells within the vascular walls (Chakravarthy and Gardiner, 1999; Rucker et al., 2000) by diffusion through astrocytic gap junctions (Martinez and Saez, 1999). Although these mechanisms require further investigation, our results provide ultrastructural evidence that neuro-glial signaling underlies the COX-2 dependent component of functional hyperemia (Niwa et al., 2000).
This work was supported by grants from the NIH: DA14214 to H.W.; NS35806 and HL18974 to C.I.; MH40342 and DA04600 to V.M.P.