Distinctive Structural and Molecular Features of Myelinated Inhibitory Axons in Human Neocortex

Visual Abstract


Introduction
Myelinated axons, which account for approximately half the volume of the human brain (Filley and Fields, 2016;Wandell, 2016), enable the dense, rapid, and efficient signal transmission central to human cognitive capacities. The insulating myelin sheath, created by a complex interaction between neuronal and glial cells, keeps axonal impulse propagation velocity high and energy consumption low, while allowing very small overall fiber diameters.
Myelinated axons convey information from a variety of neuronal types, and are likely to exhibit specialized features; yet, their properties have rarely been studied in relation to cell type (Gärtner et al., 2001;Jinno et al., 2007;Tomassy et al., 2014;Micheva et al., 2016;Stedehouder et al., 2017). Thus, very little is known about the diversity of CNS myelinated axons. Most of the available evidence comes from studies in rodents. Recently, it was shown that a large fraction of myelin in the mouse neocortex ensheathes axons of inhibitory neurons, specifically of parvalbumin-positive basket cells (Micheva et al., 2016). These inhibitory myelinated axons differ significantly from the excitatory axons in structural organization (e.g., length of nodes of Ranvier and internodes), and in molecular organization (e.g., cytoskeletal composition and protein content of myelin). Parvalbumin-positive myelinated axons have also been observed in human neocortex (Chung et al., 2013), and a recent study showed that the majority of cortical fast-spiking basket cells form myelinated axons (Stedehouder et al., 2017).
Do inhibitory myelinated axons in human cortex share the distinctive features reported in mouse cortex? There are no available data regarding their relative abundance, laminar distribution, or possible differences from excitatory myelinated axons. In the present study, we address these questions, using array tomography on surgically excised human neocortical tissue.

Human surgical specimens
Human neocortical tissue specimens were obtained during neurosurgeries for epilepsy treatment or tumor removal (Table 1). It was necessary to remove the overlying neocortical tissue to gain access to the diseased tissue. Before surgery, informed consent was obtained for the use of neurosurgical tissue for research purposes under protocols approved by the institutional review board of the University of California, San Francisco, and Swedish Medical Center, Seattle.
Resected cortical samples from neurosurgeries for epilepsy treatment were rinsed in saline and placed in room temperature fixative for 1 h. The tissue was further fixed for 23 h at 4°C, for a total time of 24 h in fixative. The average time between resection and placement in fixative was 8 min, with a range of 4 -18 min. The tissue was then transferred to PBS with 0.01% sodium azide and stored at 4°C before further processing.

Array tomography
The tissue was dehydrated and embedded using standard array tomography protocols (Micheva et al., 2010). Small tissue chunks from neocortex were dissected out and rinsed in PBS with 50 mM glycine at 4°C (three changes for up to 30 min total). The tissue was then dehydrated in graded series of ethanol (50%, 70%, 95%, 100%, 100%), followed by a mixture of equal amounts of LRWhite and 100% ethanol, and finally three changes in 100% LRWhite. Each step was done for 10 min at 4°C. After that, the samples were left for 24 h at 4°C in LRWhite for complete infiltration, then transferred to gelatin capsules filled with LRWhite, and polymerized for 24 h in an oven set at 55°C. The polymerized blocks with tissue were stored at room temperature. Additionally, separate tissue chunks from three of the samples (Q1010, Q1011, Q1014) were prepared by freeze-substitution and embedding in Lowicryl HM20, following the protocol reported in Micheva et al. (2016). This tissue was used for the experiments with mitochondrial markers.
To prepare ribbons of serial sections, the blocks were trimmed around the tissue to the shape of a trapezoid, and glue (Weldwood Contact Cement diluted with xylene) was applied with a thin paint brush to the leading and trailing edges of the block pyramid. The embedded plastic block was cut on an ultramicrotome (Leica Ultracut EM UC6) into 70-nm-thick serial sections, which were mounted on gelatin-coated coverslips.

Mouse tissue
Six adult mice (three to seven months old) were used. All animal procedures were performed according to National Institutes of Health and University of North Carolina guidelines. Three of these mice were used for a previous study (Micheva et al., 2016). The mice were perfusionfixed with 2% glutaraldehyde and 2% formaldehyde in phosphate buffer, freeze substituted, and embedded in Lowicryl HM20, as described in Micheva et al. (2016). Array tomography was performed as described above for the human tissue. Somatosensory cortex (the three mice from Micheva et al., 2016) and visual cortex (the additional three mice) were analyzed.

Immunofluorescence
Sections were processed for standard indirect immunofluorescence, as described in Micheva et al. (2010). Antibodies were obtained from commercial sources and are listed in Table 2. Array tomography-specific controls are presented in Extended Data Table 2-1, Figure 2-2. The antibodies against GABA, parvalbumin, ␣-tubulin, and neurofilament heavy chain have previously been characterized for array tomography (Bennett and Brody, 2015;Micheva et al., 2016) and gave the expected characteristic tissue staining patterns; therefore, they are not included in the analysis in Extended Data Table 2-1. The sections were pretreated with sodium borohydride [1% in Tris-buffered saline (TBS), pH 7.6 for 3 min] to reduce non-specific staining and autofluorescence. After a 20min wash with TBS, the sections were incubated in 50 mM glycine in TBS for 5 min, followed by blocking solution (0.05% Tween 20 and 0.1% BSA in TBS) for 5 min. The primary antibodies were diluted in blocking solution as specified in Table 2 and were applied for 2 h at room temperature or overnight at 4°C. After a 15-min wash in TBS, the sections were incubated with Alexa Fluor dyeconjugated secondary antibodies, highly cross-adsorbed (Life Technologies), diluted 1:150 in blocking solution for 30 min at room temperature. Finally, sections were washed with TBS for 15 min, rinsed with distilled water, and mounted on glass slides using SlowFade Gold Antifade Mountant with DAPI (Invitrogen). After the sections were imaged, the antibodies were eluted using a solution of 0.2 M NaOH and 0.02% SDS, and new antibodies were re- applied. Several rounds of elution and restaining were applied to create a high-dimensional immunofluorescent image.
The immunostained ribbons of sections were imaged on an automated epifluorescence microscope (Zeiss Ax-ioImager Z1) using a 63ϫ Plan-Apochromat 1.4 NA oil objective. To define the position list for the automated imaging, a custom Python-based graphical user interface, MosaicPlanner (obtained from https://code.google.com/ archive/p/smithlabsoftware/), was used to automatically find corresponding locations across the serial sections. Images from different imaging sessions were registered using a DAPI stain present in the mounting medium. The images from the serial sections were also aligned using the DAPI signal. Both image registration and alignment were performed with the MultiStackReg plugin in FIJI (Schindelin et al., 2012).

Immunofluorescent image analysis and statistics
For each sample, we used volumes spanning at least four cortical layers. In five of the samples, all layers were included. For each layer, a field of view of ϳ135 ϫ 100 m was imaged and analyzed. The volumes comprised of 33 serial sections on average (range of 12-67 sections). Immunofluorescence measurements were performed on raw images using FIJI. Myelin basic protein (MBP) or proteolipid protein (PLP) immunofluorescence were used to define regions of interest (ROIs), which were either the myelin sheath for measurements of MBP, PLP, and 2',3'cyclic nucleotide 3'-phosphodiesterase (CNP) signals, or the axon under the myelin sheath for measurement of axonal immunofluorescence for GABA, parvalbumin (PV), cytoskeletal and mitochondrial proteins. The mean gray value of immunolabels was compared between GABA and nonGABA, or PV and nonPV axons from the same coverslip, using the nonparametric Mann-Whitney U test, with statistical significance set to p Յ 0.01. To prepare the boxplots, the web application BoxPlotR was used (http:// boxplot.tyerslab.com; Spitzer et al., 2014).

Results
The presence and distribution of myelinated axons was quantified in samples from human temporal neocortex, using array tomography. The tissue samples were obtained from surgical resections for epilepsy, immersion fixed with a mixture of paraformaldehyde and glutaraldehyde, and embedded in resin (Table 1). Serial ultrathin (70 nm) sections were cut from the tissue blocks, immunostained for the inhibitory neurotransmitter GABA, MBP, and other relevant markers (Table 2), and imaged with a fluorescence microscope.
MBP immunofluorescence clearly outlines myelinated axons, some of which contain the inhibitory neurotransmitter GABA (Fig. 1). Myelinated GABA axonal profiles are observed in all human cortical layers at a density of 0.001-0.002 per m 2 , with the exception of layer 2, where they are significantly sparser (p Ͻ 0.01 compared to layers 1, 3b, 4, and 5, and p Ͻ 0.05 compared to layers 3a and 6). Myelinated GABA axons are rarely encountered in subcortical white matter. The proportion of myelinated axons that contain GABA decreases in cortical layers 4, 5, and 6, as the density of nonGABA myelinated axons substantially increases with cortical depth. Overall, the proportion of myelinated axons that contain GABA is significantly lower in the human temporal cortex compared to mouse somatosensory and visual cortex (Fig.  1E,F). For example, the highest percentage of GABA myelinated axons in human cortex is in layer 3a, 10 Ϯ 2%, whereas GABA myelinated axons constitute 48 Ϯ 3% of axons in mouse cortical layer 2/3 (Micheva et al., 2016). This is mostly due to the lower density of GABA myelinated axons in human cortex, while the density of nonGABA myelinated axons in not significantly different between human and mouse cortex in all layers, except layer 1 (p Ͻ 0.01). Layer 1 is also the only cortical layer, where the density of GABA myelinated axons is higher in human cortex compared to mouse (p ϭ 0.012).
A previous study of mouse cortex reported a number of differences between myelinated inhibitory and excitatory axons, which were also seen in human cortex (Fig. 2). Myelinated GABA axons in human cortex have significantly higher neurofilament content (average immunofluorescence of 810 Ϯ 49 average units (a.u.) vs 248 Ϯ 9 a.u. for nonGABA myelinated axons, p Ͻ 0.00001) and lower tubulin content compared to myelinated nonGABA axons (immunofluorescence of 222 Ϯ 17 vs 413 Ϯ 17 a.u., p Ͻ 0.00001; n ϭ 164 GABA and n ϭ 216 nonGABA myelinated axons from layers 3a through 5 from three different samples; Fig. 2A-D). The nodes of Ranvier (Rasband and Antibody controls are presented in Extended Data Table 2 nofluorescence of 1026 Ϯ 41 vs 884 Ϯ 29 a.u., p ϭ 0.007; n ϭ 164 GABA and n ϭ 216 nonGABA myelinated axons from layers 3a through 5 from three different samples), although their PLP content is essentially the same (1708 Ϯ 62 vs 1643 Ϯ 48 a.u., p ϭ 0.68). Thus, the distinctive characteristics of myelinated inhibitory axons seen in mouse are also present in human cortex. In mouse, the overwhelming majority of myelinated GABA axons come from parvalbumin-containing basket cells (Micheva et al., 2016). To determine the source of myelinated GABA axons in human cortex, we immunostained the samples with an antibody against parvalbumin. More than half of the myelinated GABA axons are clearly immunoreactive for parvalbumin, with PVimmunofluorescent signal present on consecutive sections (Fig. 3). On average, the PV immunofluorescence within GABA axons is significantly higher than within nonGABA axons (68 Ϯ 2 vs 42 Ϯ 1 a.u., p Ͻ 0.0001, n ϭ  To confirm our findings, we also used cortical tissue from a non-epileptic patient who had overlying cortical tissue removed during a biopsy to diagnose an inflammatory lesion. In this cortical specimen, we observe numerous myelinated axons immmunoreactive for parvalbumin (Fig. 3C). The proportion of the myelinated PV axons is very similar to the GABA axons in the tissue from epilepsy surgeries. For example, in layer 3a, 15% of the myelinated axonal profiles are PV-immunopositive, and in layer 3b, 14% are PV-immunopositive (693 myelinated axons from layer 3a and 2019 from layer 3b). This is within the range observed for GABA myelinated axons from epilepsy surgeries (3-21% in layer 3a, and 3-15% in layer 3b). The biopsy tissue, however, was fixed only with paraformaldehyde, which prevented simultaneous labeling with an anti-GABA antibody, since glutaraldehyde is required for satisfactory preservation of GABA (Somogyi et al., 1985;Schiffmann et al., 1988). However, other characteristics of the PV-immunopositive myelinated axons from the nonepilepsy biopsy tissue, such as their cytoskeletal content and length of nodes of Ranvier (Fig. 3D,E), are also very similar to the GABA axons in the epilepsy surgery tissue, suggesting that these axon populations are comparable. Thus, parvalbumin interneurons appear to be a major source of inhibitory myelinated axons in human neocortex.
Parvalbumin-positive basket cells are fast-spiking interneurons capable of generating long trains of action potentials at very high frequency (Kawaguchi et al., 1987;Kawaguchi and Kubota, 1997), which requires a constant supply of ATP. Thus, the high energy demands of parvalbumin interneurons may be reflected in a higher content of mitochondria within their myelinated axons to ensure the local production of ATP. Their cell bodies (Gulyás et al., 2006; in rodent cortex), as well as presynaptic boutons (Cserép et al., 2018; in rodent and human cortex), are known to have abundant mitochondria, but the mitochondrial content of their axons has not been assessed. We used antibodies against three mitochondrial proteins: MDH2, which is present in the mitochondrial matrix, and TOMM20 and VDAC1, which are proteins of the outer mitochondrial membrane (Fig. 4). Accordingly, immunolabels for TOMM20 and VDAC1 were seen surrounding the MDH2 label (Fig. 4B). We quantified the distribution of these three mitochondrial markers in GABA and nonGABA myelinated axons in the human epilepsy surgery samples. The immunofluorescence for all three mitochondrial markers was significantly higher in myelinated GABA compared to nonGABA myelinated axons (MDH2, p Ͻ 0.00001; TOMM20, p ϭ 0.001; VDAC1, p ϭ 0.0002), indicating that myelinated GABA axons in human cortex are indeed enriched in mitochondria (Fig. 4).
How do these axons receive the nutrients necessary for the mitochondrial production of ATP? Trophic support to myelinated axons in the CNS is likely provided through a system of cytoplasmic channels within myelin (for review, see Simons and Nave, 2015), whose maintenance requires the myelin protein CNP (Snaidero et al., 2017). Our experiments reveal that GABA myelinated axons have a significantly higher CNP content compared to nonGABA axons (321 Ϯ 17 vs 271 Ϯ 7 a.u.; p ϭ 0.002, n ϭ 133 GABA and n ϭ 489 nonGABA myelinated axons) suggesting the existence of more cytoplasmic channels within their myelin, consistent with a higher need for trophic support. Interestingly, this difference in CNP content between GABA and nonGABA axons is most pronounced among myelinated axons with a diameter larger than 0.6 m (Fig. 5).

Discussion
Using samples of surgically excised brain tissue, we show here that many of the characteristic features of myelinated axons in cortical gray matter previously described in the mouse (Micheva et al., 2016) are also present in human, despite 100 million years of divergence. As in mouse, many human myelinated inhibitory axons originate from PV-positive interneurons and have distinctive features, including high neurofilament and low microtubule content, short nodes of Ranvier, and high content of MBP in their myelin sheath. The implications of these differences will require further study, but we speculate that they have direct functional significance. For example, the cytoskeleton is directly involved in axonal transport (Conde andCáceres, 2009: Hirokawa et al., 2010); it also determines the mechanical properties of axons and therefore their susceptibility to mechanical injury (Ratzliff and Soltesz, 2000; Grevesse et al., 2015). The length of the continued samples). C, PV neuron and processes in layer 3b of human cortex from a biopsy; volume reconstruction from 49 serial sections. Yellow arrowheads point to PV-positive myelinated axons. D, Cytoskeletal composition and node length of myelinated axons in biopsy tissue. Left, Comparison of the immunofluorescence intensity (mean Ϯ SE) for neurofilament heavy chain and ␣-tubulin within myelin profiles of nonPV (blue) and PV axons (green) from the biopsy sample. The differences are statistically significant (p Ͻ 0.001 for NFH and p Ͻ 0.0001 for ␣-tubulin; 48 nonPV and 26 PV axons from layers 3a, 3b, and 4 from one sample). Right, Nodes of PV axons are shorter than nonPV axons in the biopsy sample (p Ͻ 0.01; 34 nonPV and 24 PV nodes). E, Cytoskeletal composition and node length of myelinated axons in epilepsy surgery tissue. Left, Comparison of the immunofluorescence intensity for neurofilament heavy chain and ␣-tubulin within myelin profiles of nonGABA (blue) and GABA axons (red) from epilepsy surgery samples. The differences are statistically significant (p Ͻ 0.0001, 196 nonGABA and 177 GABA axons from layers 3a, 3b, and 4 from three different samples). Right, Nodes of GABA axons are shorter than nonGABA axons in epilepsy surgery samples (p Ͻ 0.0001; 100 nonGABA and 39 GABA nodes from eight different samples). ‫‪p‬ءء‬ Ͻ0.01. Figure 4. GABA myelinated axons have more mitochondria. A, A single 70-nm section from layer 4 of human cortex immunolabeled with GABA (red), PLP (white), MDH2 (magenta), TOMM20 (green), and VDAC1 (cyan). B, The boxed area in A is enlarged to show two immunolabeled mitochondria, with the MDH2 mitochondrial matrix protein surrounded by the outer mitochondrial membrane proteins TOMM20 and VDAC1. C, Five serial sections through axons 1 and 2, marked on A, top section. Axon 1 is nonGABA and axon 2 is GABA. Adjacent to axon 2, there is also a nonGABA axon. D, Boxplots of the immunofluorescence for MDH2, TOMM20, and VDAC1 within nonGABA and GABA myelinated axons. Center lines show the medians, box limits indicate the 25th and 75th percentiles as determined by R software; whiskers extend 1.5 times the interquartile range from the 25th to the 75th percentiles, outliers are represented by dots, n ϭ 88 and n ϭ 41 axons for MDH2, n ϭ 95 and n ϭ 43 axons for TOMM 20, and n ϭ 97 and n ϭ 47 axons for VDAC1, from layers 3a through 5 from three different samples.
nodes of Ranvier affects the speed and timing of actionpotential propagation (Arancibia-Cárcamo et al., 2017). MBP is a target in multiple sclerosis (Allegretta et al., 1990) and differences in its content may influence the fate of axons in brain pathologies. We further show that the inhibitory myelinated axons have more mitochondria, as well as more CNP, a protein enriched in the myelin cytoplasmic channels that is thought to provide access for trophic support from ensheathing oligodendrocytes. This is consistent with the high energy demands of fast-spiking PV interneurons and suggests that, in addition to influencing conduction velocity, the myelination of inhibitory axons is likely beneficial for managing their energy consumption by increasing the efficiency of action potential propagation (Hartline and Colman, 2007)   . GABA myelinated axons have more CNP in their myelin. A, Volume reconstruction of 21 serial sections from human cortical layer 5, immunostained with CNP (magenta) and MBP. B, Three consecutive sections through a GABA myelinated axon immunolabeled for CNP (magenta) and MBP (white). The first section is also shown immunolabeled for GABA. C, Thicker GABA axons (0.6 -1.2 m in diameter) have significantly more CNP in their myelin sheath compared to nonGABA axons of similar thickness. Center lines in boxplots show the medians, box limits indicate the 25th and 75th percentiles as determined by R software; whiskers extend 1.5 times the interquartile range from the 25th to the 75th percentiles, outliers are represented by dots. For the boxplot on the left (axon diameter 0.1-0.6 m), n ϭ 341 nonGABA and n ϭ 98 GABA myelinated axons from two samples), and for the boxplot on the right (axon diameter 0.6 -1.2 m), n ϭ 142 nonGABA and n ϭ 34 GABA axons from two samples).
There are also important differences between human and mouse neocortical myelinated axons. The density of GABA myelinated axons in human cortex is substantially less than in mouse cortex, while the density of nonGABA myelinated axons is rather similar, with the exception of layers 1 and 2. This may appear puzzling at first glance, because of the association of interneurons with high-level cortical functions (for review, see Tremblay et al., 2016). However, this is consistent with the lower densities of other neuronal structures in human cortex. For example, synapse density is ϳ2.5-3 times less in human cortex compared to mouse, and neuron density is five times less (DeFelipe et al., 2002). This is likely due, at least in part, to the increased proportion and size of nonneuronal cells in human cortex (for review, see Herculano-Houzel, 2014). If the same scaling down of density applies to myelinated axons, then it would appear that GABA myelinated axons are following the general trend, while nonGABA myelinated axons have a disproportionately high density, possibly because of the much longer distances many of these axons have to cover in the human brain and therefore the advantages offered by myelination to speed up action potential propagation. One notable exception to the lower density of GABA myelinated axons in human cortex is layer 1, which has more GABA myelinated axons compared to mouse layer 1. Layer 1 is the site where input from a variety of cortical and subcortical neurons is integrated (Douglas and Martin, 2004;Harris and Shepherd, 2015) and which appears to have undergone large evolutionary changes. For example, recent studies revealed two new interneuron types in human cortex that are not present in mouse cortex, and both of these interneurons are enriched in layer 1 (Boldog et al., 2018;Hodge et al., 2018). The source of the GABA myelinated axons in human layer 1 remains to be determined and will provide further insight into the functional significance of this evolutionary change.
The characteristic features of myelinated inhibitory axons in human cortical gray matter are likely to have important implications for neurologic disorders that involve pathologies of myelinated axons. The distinct molecular and structural organization of inhibitory and excitatory myelinated axons may underlie differences in their vulnerability in neurologic disorders or to injuries. Furthermore, disturbances of myelin around inhibitory axons are expected to have very different functional implications compared to disturbances of myelin on excitatory axons and may require different strategies for prevention and treatment.