The procedures for handling and using the human brain tissue had been approved by the Biomedical Research Ethics Committee of Shanghai Institutes for Biological Sciences (license no. ER-SIBS-221004). The human brain tissues were obtained from the Shanghai Quyang Hospital. These tissues had to be removed surgically for the treatment of intractable epilepsy, cerebral hemorrhage, or brain tumor. Before the surgery all patients had signed a written consent form that described the experimental usage of their discarded tissues. The use and care of animals followed the guidelines of the Animal Advisory Committee at the Shanghai Institutes for Biological Sciences.

We rinsed the sections in 0.01 M phosphate-buffered saline (PBS, pH 7.4) and incubated them first with 0.3% Triton X-100 in PBS for 30 min and then with the blocking solution (10% BSA) for 1 h. Sections were incubated overnight at room temperature with the primary antibodies diluted in 0.1% TritonX-100 and 10% BSA. The following primary antibodies were used: mouse anti-Na_V1.1 (73-023; 1:200; NeuroMab), rabbit anti-Na_V1.2 (Rb-Na_V1.2; ASC-002; 1:400; Alomone Labs), mouse anti-Na_V1.2 (M-Na_V1.2; 73-024; 1:200; NeuroMab), rabbit anti-Na_V1.6 (ASC-009; 1:400; Alomone Labs), rabbit anti-Pan-Na_V (ASC-003; 1:400; Alomone Labs), mouse anti-PV (MAB1572; 1:1000; Millipore), rabbit anti-PV (PV-25; 1:1000; Swant), mouse anti-MBP (SMI-94R; 1:400; Covance), mouse anti-NeuN (MAB377; 1:1000; Millipore), goat anti-Ankyrin G (AnkG; sc-31778; 1:400; Santa cruz), mouse anti-AnkG (sc-12719; 1:400; Santa cruz), guinea pig anti-vesicular GABA transporter (V-GAT; 131004; 1:100; Synaptic Systems), rabbit anti-GABA transporter-1 (GAT-1; AB1570; 1:400; Millipore), mouse anti-contactin-associated membrane protein (anti-Caspr; 75-001; 1:500; NeuroMab). After complete wash, sections were incubated for 2 h in the secondary antibodies (1:1000, Invitrogen): Alexa 488 conjugated donkey anti-rabbit, Alexa 555 conjugated donkey anti-mouse, and Alexa 647 conjugated donkey anti-goat, or Alexa 488 conjugated donkey anti-guinea pig, Alexa 555 conjugated donkey anti-mouse, and Alexa 647 conjugated donkey anti-rabbit. Finally, the sections were washed and mounted with fluoromount-G (Electron microscopy science).

We took images using a laser scanning confocal microscope (ECLIPSEFN1, Nikon) with a 40× (NA 1.3) or 60× (NA 1.4) oil-immersion objective. The acquisition parameters were carefully adjusted to make the fluorescence signals linearly displayed and fall into the maximum dynamic range of the detector. Z-stack images were collected with a voxel interval of 0.5 or 1 μm. We then used Autoquant X2 software (Media Cybernetics) to deconvolve these images. We employed Fiji software (http://pacific.mpi-cbg.de) to obtain the maximum Z-stack projection and make appropriate adjustment of brightness and contrast.

We examined the specificity of antibodies to make sure the stainings were true positive. Omission of primary antibodies from the immunostaining protocols resulted in undetectable signals for all antibodies examined. We further tested the specificity of Na_V1.6 antibody with transgenic mice. Na_V1.6 immunosignals were absent from the AIS in the Na_V1.6^-/- mice, whereas robust staining could be observed in the control littermates (Figures S1A,B in Supplementary Material, P17, n = 2 pairs of mice). In control mice (Na_V1.6^+/+), Na_V1.2 immunosignals could be observed at proximal regions of the AIS and showed spatial segregation with Na_V1.6 signals (Figure S1A in Supplementary Material), a distribution pattern similar to previous findings in rats (Hu et al., 2009). In the Na_V1.6^-/- mice, distribution of Na_V1.2 immunosignals became relatively even along the AIS, possibly a compensatory response to Na_V1.6 loss (Figure S1B in Supplementary Material). We did observe some thin fibers positive for Na_V1.6 in the Na_V1.6^-/- mice; however, they were rarely observed and showed no co-labeling with AnkG. Despite these rare false positive staining, the lack of robust immunosignals in Na_V1.6^-/- mice and the presence of strong signals colocalized with AnkG in wild type mice indicate a high specificity of the Na_V1.6 antibody.

For Na_V1.2 antibody, it was not possible to test the antibody specificity in transgenic mice because the Na_V1.2 KO mice were lethal perinatally (Planells-Cases et al., 2000). Alternatively, we co-labeled two antibodies against different peptide sequences of rat Na_V1.2 to examine their specificity, one polyclonal antibody against the intracellular loop between domain I and II (467–485) generated from rabbit (Rb-Na_V1.2), and the other monoclonal antibody against cytoplasmic C-terminal (1882–2005) from mouse (M-Na_V1.2). Immunosignals stained with the two antibodies overlapped well with each other in rat neocortical sections (Figure S1C in Supplementary Material, n = 2 rats), indicating a high specificity of the two antibodies. For the Rb-Na_V1.2 antibody, preabsorption with its antigen for 2 h could completely block the staining, indicating that the staining with this antibody was specific (Figure S1D in Supplementary Material, n = 2 rats). Usually, M-Na_V1.2 immunosignals were more robust than those of Rb-Na_V1.2 in rodent tissue. In the human neocortex, however, only Rb-Na_V1.2 antibody showed strong stainings, whereas M-Na_V1.2 antibody exhibited no detectable signal in our experiments. This failure of M-Na_V1.2 antibody in staining the human tissue could be attributed to differences in antigen peptide sequences (1882–2005) of Na_V1.2 protein between rat and human. Therefore, we chose to use Rb-Na_V1.2 for human tissue experiments.

Although we did not examine the specificity of Na_V1.1 antibody as strictly as that of Na_V1.2 and Na_V1.6, its staining pattern in rodents was in agreement with previous findings (Yu et al., 2006; Ogiwara et al., 2007; Lorincz and Nusser, 2008). In our experiments, Na_V1.1 immunosignals could be detected only in GABAergic interneurons in the rat neocortex; in addition, blocking peptide could completely block the staining, suggesting a high antibody specificity (data not shown). The accumulation of Na⁺ channels at the AIS and nodes of Ranvier is achieved through anchoring to the actin cytoskeleton via AnkG and other proteins, and AnkG is therefore usually used as a marker of the AIS and the nodes along the axon (Srinivasan et al., 1988; Kordeli et al., 1995; Zhou et al., 1998). The Caspr is another molecule that can be used to define the paranodal regions at the nodes (Einheber et al., 1997; Menegoz et al., 1997; Peles et al., 1997). In this study, Caspr together with AnkG were used as markers to identify the nodes of Ranvier. The immunosignals of two AnkG antibodies against different antigen epitopes showed complete overlap at the AIS and at the nodes of Ranvier (Figure S2 in Supplementary Material). We normally used AnkG antibody generated from goat unless otherwise specified.

We analyzed the immunostaining signals in neurons whose AIS emitted directly from the soma (labeled by NeuN antibody) and extended for more than 40 μm in length. For each patient, we acquired images (no less than six images, 215 μm × 215 μm) from at least three cortical sections. For quantitative analysis of the fluorescence intensity along the AIS, we extracted the fluorescence signals at the AIS (for Na⁺ channels) or those surrounding the AIS (for PV signals) with the Fiji software, and calculated the fluorescence intensity in three-dimension with the Amira software (Visualization Sciences Group). Firstly, we used Autoquant X2 software to deconvolve images, and Fiji software to combine the two images of Na⁺ channel and AnkG into one 8-bit image. We then traced AIS in this image, filled out the AIS paths with different radius for Na⁺ channels and PV signals, and obtained the mask image of this fillout. We performed an AND operation of the mask image and the individual image of Na⁺ channel or PV to obtain an image with fluorescence signals only at or surrounding the AIS but without any other background noise. Next, we calculated the fluorescence intensity along the AIS with Amira. For individual AIS, we averaged the fluorescence intensity every 1 μm and normalized these values to the maximum average intensity with MATLAB (MathWorks). Finally, we obtained the distribution pattern of fluorescence intensity along the AIS by averaging the normalized values of different AIS.

To examine the percentage of nodes expressing a certain Na⁺ channel subtype in the white matter of the cortex, we performed analysis on images double stained with AnkG and channel antibodies. We counted the number of nodes positive for Na_V1.2, Na_V1.6, or Pan-Na_V in a square area (160 μm × 160 μm) and normalized the number to the total pixel number of AnkG signals with fluorescence intensity higher than a threshold value (mean + SD). For each double-staining group, six such square areas from three human patients (two from each patient) were analyzed and the normalized values were then averaged. Because Pan-Na_V antibody recognizes all Na⁺ channel subtypes, it should label all the nodes of Ranvier in the cortex. We divided the average value for Na_V1.2 or Na_V1.6 by that of Pan-Na_V to obtain the percentage of nodes containing Na_V1.2 or Na_V1.6 in the cortex.

For a particular experiment, we carried out immunostaining in different batches, but with the same protocol. For quantitative analysis of the immunosignals of PV and Na⁺ channels, we compared the normalized fluorescence intensity and examined their distribution pattern along the AIS. The absolute Na⁺ channel density could not be determined in our study.

For Na⁺ channel staining, conventional immunohistochemical reactions using 4% PFA revealed almost no labeling (Figures 1A–C, left panels), while addition of sucrose (4%) would lead to a weak staining of Na_V1.6 channels in the AIS (data not shown). Reducing the concentrations of both PFA and sucrose to 0.5% revealed a better staining. This weak fixative (0.5% PFA and 0.5% sucrose) ensured a satisfactory staining with a high signal-to-noise ratio of all antibodies against Na⁺ channels used in this study (Figures 1A–C, right panels). This result suggests that the conventional fixative is relatively strong and possibly masks the antigen epitope.

To investigate the molecular identity of Na⁺ channels distributed at the AIS of human cortical PCs, we performed immunostaining in cortical tissues obtained from patients with intractable temporal lobe epilepsy (n = 5), cerebrovascular disease (n = 1), and brain tumor (n = 3). Usage of the tissue samples were listed in Table 2. We used peri-foci or peri-tumor cortical tissues that showed clear layer structure (from L1 to L6 and the white matter) and the densities of NeuN-, Nissl-, or AnkG-positive signals were not obviously reduced (Figure 2). In our experiments, we employed Pan-Na_V antibody that could recognize all α subunits to reveal the total density of Na⁺ channels. NeuN and/or AnkG immunosignals were used to distinguish PCs from interneurons. Neurons with NeuN-positive and pyramidal-shaped somata and thick AnkG-positive AIS were considered as putative PCs. In general, the AIS of PCs uniformly projected toward deep layers. In comparison with PCs, the somata of interneurons were usually smaller and multi-shaped, and their AIS were thinner and showed unpredictable projection directions (Figure S3 in Supplementary Material).

In agreement with previous studies (Inda et al., 2006), our experiments showed a high density of Na⁺ channels at the AIS of human PCs (Figure 3). Strong fluorescence signals of Pan-Na_V were detected along the entire AIS (Figure 3A). We plotted the fluorescence distribution of Pan-Na_V in PCs of L3 (L3a and L3b) and found that the intensity remained high from the proximal to the distal AIS except for the two ends (the proximal end, 0–7 μm from the soma; and the distal end, 41–45 μm) where the intensity was relatively low (50% < intensity < 80% of the peak intensity, n = 113 from five epileptic subjects; Figures 3A,D). The distribution profile of the normalized Pan-Na_V intensity in epileptic tissues was similar to that in non-epileptic tissues (n = 62 from four non-epileptic subjects; Figure 3D), suggesting no substantial influence of epileptic seizures on the distribution pattern of total Na⁺ channels. In the following experiments, we mainly used cortical tissues from epileptic patients. No fluorescence signals could be found at the soma and proximal dendrites (stained by NeuN), suggesting an extremely low density of Na⁺ channels at these compartments. In these experiments, although the quality of the NeuN labeling was variable in tissues fixed after freezing sectioning (Figures 3A–C), the shape of the soma could still be easily distinguished.

At the AIS of cortical PCs, both Na_V1.2 and Na_V1.6 immunosignals were observed, but they accumulated at different compartments. Similar spatial segregation of channel subtypes could be found in PCs of pediatric epileptic patients, we therefore pooled the data together with adult patients. In L3b, Na_V1.2 signals concentrated at the proximal AIS region with fluorescence intensity peaked at 9–14 μm from the soma (n = 57 from five epileptic subjects, Figures 3B,E), whereas Na_V1.6 immunosignals accumulated at the distal AIS with the peak intensity at 25–29 μm from the soma (n = 62, Figures 3C,E). Although there was an overlap between Na_V1.2 and Na_V1.6 immunopositive regions, a clear segregation of these channel subtypes was found at the AIS. Interestingly, only when the distribution curves of Na_V1.2 and Na_V1.6 were respectively multiplied by a same factor, their summation curve could show a similar distribution profile with that of Pan-Na_V curve (Figure 3F), which could be reflected by a perfect overlap of the two curves after further normalization (Figure 3F). Together, these results indicate a spatial segregated distribution of proximal-Na_V1.2 and distal-Na_V1.6 at the AIS of human cortical PCs, similar to the segregated distribution of the two channel subtypes in rat PCs (Hu et al., 2009).

In addition to Na_V1.2 and Na_V1.6, Na_V1.1 is another main Na⁺ channel subtype in adult CNS. Na_V1.1 was revealed to be expressed in the axon of cortical interneurons in rodents (Ogiwara et al., 2007; Lorincz and Nusser, 2008); however, in human cortex, Na_V1.1 was reported to be distributed in PCs (Whitaker et al., 2001b). In all human tissues we examined, fluorescence signals of Na_V1.1 were only observed at the AIS of putative interneurons whose AnkG-labeled AIS were usually thin and projected unpredictably; in contrast, they were completely absent from thick AIS of putative PCs (Figure 4A). AIS with Na_V1.1 immunosignals could be found in layers from L2 to L6. Moreover, the staining intensity and distribution pattern of Na_V1.1 diversified across different interneurons (Figures 4B–E). Na_V1.1 channels could accumulate at either the proximal or the distal portion of the AIS (Figures 4B,C), and sometimes express uniformly along the AIS (Figure 4D). The intensity of Na_V1.1 immunosignals were obviously different in the two neighboring interneurons shown in Figure 4E. With triple staining of PV, Na_V1.1, and AnkG, we found that PV-positive neurons showed Na_V1.1 immunosignals at their AIS, especially concentrating at a short compartment very close to the soma (Figure 4F). In PV-negative interneurons, we also found Na_V1.1 signals at their AIS (Figure 4G), but they showed different distribution patterns. Therefore, distinct Na_V1.1 distribution profiles might be attributed to differences in interneuron cell types. However, we could not exclude the possibility that the distribution patterns of Na_V1.1 may be subject to change in response to epileptic seizures. These results indicate that Na_V1.1 channels are expressed by human inhibitory interneurons, but its distribution properties in different kind of interneurons remain to be further examined.

Together, our results reveal a high density of Na⁺ channels and a segregated distribution of channel subtypes at the AIS, with Na_V1.2 and Na_V1.6 channels accumulating at the proximal and the distal AIS of human PCs, respectively. In addition, unlike those of putative interneurons, the AIS of PCs had no detectable immunosignals for Na_V1.1. Due to the critical roles of Na⁺ channels in regulating AP generation and neuronal excitability, the presence of Na_V1.1 at the AIS of PV-positive neurons suggests a role of this channel subtype in regulating the excitability of inhibitory interneurons and thus the excitation–inhibition balance in cortical networks. Indeed, previous studies found severe spontaneous epileptic seizures in loss-of-function mutations of Na_V1.1 gene in human (Baulac et al., 1999; Escayg et al., 2000; Claes et al., 2001; Meisler and Kearney, 2005; Catterall et al., 2010).

At the myelinated axon beyond the AIS, AP is generated at the nodes of Ranvier and propagates down the axon in a saltatory mode. This saltatory conduction ensures rapid AP conduction and fast communication between neurons. In rodents, a high density of Na_V1.6 was found at the nodes of Ranvier (Caldwell et al., 2000). Here we examined the molecular identity of nodal Na⁺ channels in the human neocortex. Double staining in the white matter revealed a high density of nodes labeled with strong immunosignals of AnkG and Pan-Na_V (Figures 5A,B). Detectable AnkG signals were also observed at the paranodal regions where AnkG signals colocalized with Caspr (Figure S4A in Supplementary Material). Some nodes displayed very low AnkG signals, and the fluorescence intensity could be even lower than that in the flanking regions (Figure 5B). Moreover, we also observed gaps between labeled nodal and paranodal regions (Figure S2B in Supplementary Material). These observations may suggest a heterogeneous population of nodes of Ranvier. To disclose the molecular identity of Na⁺ channel subtypes at the nodes, we performed double staining using AnkG and Na_V1.6 antibodies. Strong Na_V1.6 signals were observed at the nodes (Figure 5C), similar to those found in rodents. The Na_V1.6-positive puncta could be also found in the gray matter (Figure 3C), especially in deep layers. To ensure that these puncta were indeed the nodes of Ranvier, we used MBP antibody to visualize the myelination of axons. Triple staining of Na_V1.6, AnkG, and MBP showed that puncta positive for both Na_V1.6 and AnkG interspersed between MBP-labeled myelin sheaths and confirmed their nodal identity (Figures S4B,C in Supplementary Material and Figure 5D).

In double staining of Na_V1.6 and AnkG, some nodes showed very weak Na_V1.6 immunosignals, or even no detectable signals (Figure 5C). Considering that this kind of nodes was rarely observed in the Pan-Na_V staining, we speculated that other channel subtypes (Na_V1.1 and Na_V1.2) may also express at the nodes. However, no Na_V1.1 signal was found at any of the AnkG-positive nodes in both the white matter (Figure 6A) and the gray matter (Figure 4A). Previous studies in the CNS of rodents showed that Na_V1.2 channels expressed at early developmental stages but were replaced by Na_V1.6 as the axon became myelinated (Boiko et al., 2001; Kaplan et al., 2001). Surprisingly, in adult human cortex, Na_V1.2 immunosignals were observed at AnkG-positive nodes in both the white matter (Figure 6B) and the gray matter (Figure 3B) of epileptic and non-epileptic tissues. The Na_V1.2-positive nodes showed diversity in terms of their length, shape and labeling patterns (Figures 6C1–C4). As shown in Figure 6C5, Na_V1.2 immunosignals were occasionally found at or around branching points of AnkG-labeled processes. The percentages of Na_V1.6- and Na_V1.2-positive nodes in the white matter were 81.6 ± 4.5% and 28.7 ± 2.3% respectively (Na_V1.6: n = 8127 nodes; Na_V1.2: n = 2978 nodes; Pan-Na_V: n = 10768 nodes from three human subjects). Because both Na_V1.2 and Na_V1.6 antibodies were generated from rabbit, we were not able to perform double staining of these channels to examine their co-localization at individual nodes in the human tissue. The summated percentage of nodes expressing the two subtypes was more than 100%, suggesting a co-expression of Na_V1.2 and Na_V1.6 at some individual nodes; however, this requires further confirmation using appropriate antibodies.

Ion channel expression could be subject to change with the development of brain disorders, for example, demyelinated axon in multiple sclerosis could regain the expression of Na_V1.2 (Craner et al., 2003, 2004). In our experiments, most of the human tissues were obtained from epileptic subjects; we therefore tried to examine the Na⁺ channel distribution in non-epileptic tissues. In addition to the similar distribution pattern of Pan-Na_V along the AIS between epileptic and non-epileptic groups (Figure 3D), the segregated distribution of proximal-Na_V1.2 and distal-Na_V1.6 was also observed in non-epileptic tissues. Similarly, we also found the presence of the two channel subtypes at the nodes of Ranvier and the absence of Na_V1.1 at the axon (data not shown). Because the tissues from different patients were not processed simultaneously, we were not able to determine whether the fluorescence intensity of Na⁺ channel staining was subject to change in response to epilepsy. However, the current experiments suggest that the distribution pattern of channel subtypes could not be influenced by epileptic seizures.

The AIS of PCs is innervated by axon terminals of GABAergic chandelier cells. This strategic innervation has been believed to play important roles in regulating AP generation at the AIS and further modulating the main neuronal output. Considering that Na_V1.6 is the channel subtype that determines AP initiation, we next sought to investigate whether axon terminals of chandelier cells selectively targeted to the Na_V1.6-enriched AIS segment.

The specialized axonal arbors of chandelier cells, usually called cartridge, are formed by arrays of vertically oriented presynaptic terminals. In primates, chandelier cells were immunoreactive to PV, and immunostaining of PV was usually employed to visualize the cartridge structure (DeFelipe et al., 1989; Lewis and Lund, 1990), though a recent study suggested that not all chandelier cells expressed PV in rodents (Taniguchi et al., 2013). Indeed, vertically distributed PV-positive cartridges could be observed and they closely surrounded the AIS of putative PCs in the human neocortex (Figure 7A). Triple staining of GAT-1, PV, and VGAT indicated co-localization of these three molecules at the cartridge structure, suggesting the PV-containing cartridges are indeed GABAergic presynaptic terminals (Figure 7B). In comparison with VGAT, much less GAT-1-positive puncta were observed in structures outside cartridges, indicating that GAT-1 specifically locates at cartridges in the human neocortex. This result is also consistent with previous findings (DeFelipe and Gonzalez-Albo, 1998; Inda et al., 2006).

Because there was a fluorescence gap between PV-positive cartridge and the soma, we speculated that the cartridges selectively innervated the Na_V1.6-enriched AIS domain. We performed triple staining of AnkG, PV, and Na_V1.6, and measured the fluorescence intensity of cartridge PV and Na_V1.6 along the entire AIS of PCs (n = 367 from three epileptic subjects) across cortical layers. Interestingly, PV-labeled cartridges in L4 showed few innervations to the proximal AIS; however, they covered the distal AIS region where Na_V1.6 immunosignals were much stronger (Figures 7C,D). PV cartridges in L5 also innervated the Na_V1.6-enriched distal AIS but with much less PV positive puncta than those in L4 (Figure 7E). Many Na_V1.6-positive nodes could be easily found in these two layers. In L4, the location of the peak fluorescence intensity for cartridge PV (16 μm away from the soma, fitting with double Gaussian function, n = 142 from three epileptic subjects) was similar to that of Na_V1.6 (15 μm; Figure 7F). At the proximal AIS, the normalized PV fluorescence intensity was relatively low but increased steeply until reaching its peak at distal AIS regions, indicating a more confined distribution of cartridge PV than that of Na_V1.6. Distribution pattern of cartridge PV and Na_V1.6 at and beyond the peak location were similar. Although the fluorescence intensity of Na_V1.6 peaked differently at the AIS across cortical layers, their peak location matched well with that of PV cartridges (Figure 7G; L2: n = 40 from two patients; L3: n = 100, L5: n = 44, L6: n = 40 from three patients). These results indicate that chandelier cells mainly target the Na_V1.6-enriched AIS region and possibly exert a strong modulatory effect on AP initiation, and thus on neuronal excitability and network activity in the human neocortex.