Experimental design
The aim of this study was to examine the balance of excitation and inhibition in human OFC and its disruption in ASD. To do this we studied key excitatory and inhibitory components of OFC circuits, including the myeloarchitecture, cytoarchitecture and neurochemistry of OFC gray matter, in adults with or without autism. First, we examined the distribution and morphology of myelinated axon fibers across cortical layers, to paint a picture of excitatory networks in OFC. Then we examined the laminar distribution and morphology of three largely non-overlapping, neurochemically- and functionally-distinct types of inhibitory neurons in the local circuit, to assess the balance of excitation and inhibition in OFC circuits. Typical variability or irregularities in the laminar distribution and density of these excitatory and inhibitory circuit components likely underlies the functional integrity of layer-specific feedforward, feedback, cortical and subcortical OFC networks and their pathology in ASD. An overview of our experiment design and approach is shown in Fig. 1.
Human post mortem brain tissue and sample size
Post-mortem brain tissue from 10 adults (6 Control, 4 ASD) was obtained from the Harvard Brain Tissue Resource Center through the Autism Tissue Program, Anatomy Gifts Registry, and the National Disease Research Interchange (NDRI). We used formalin-fixed tissue that was optimally prepared for correlated quantitative light and electron microscopy (EM) and immunohistochemical staining. To preserve the ultrastructure until processing, tissue blocks were cryoprotected in progressively increasing concentrations of buffered sucrose solutions (10-25% in 0.1M PB) and were then immersed in antifreeze solution (30% ethylene glycol, 30% glycerol, 40% 0.05 M PB, pH: 7.4 with 0.05% azide) and stored at −20 °C. Blocks of postmortem brain tissue containing areas of OFC were cut in 10 consecutive series of 50μm sections. We then selected adjacent sections for EM processing and immunostaining experiments. We additionally fixed sections used for EM with 6% glutaraldehyde.
All available cases were included in qualitative and quantitative axon tracing and immunohistochemistry analysis. Our observations and previous published work [34] show consistent patterns of neuron and axon organization and density across layers in different cortices with comparable numbers of subjects. Each sample yielded a large number of data points, because we typically examine a large volume fraction of the areas sampled, and thus increase the number of individual axons and neurons examined. The high sampling fraction used in our studies minimized the variability within each analyzed case, and further increased statistical power, as we have described [34-38]. Our previous work [5, 27, 35-37] and power analysis showed that the sampling ratios exceed the samples needed to detect differences with a greater than 90% probability, and large effect population size (0.8) with ≤ 10% error, as recommended [39]. We used IBM SPSS Statistics 24 or PS version 3.1.6 (PS Power and Sample Size Calculations Program) for a posteriori power analysis of studies with an independent design that are analyzed by t-tests or regression. Input variables for power and sample size calculations included the Type I error probability (α), difference in population means (δ), regression errors or within group standard deviation (σ), slope of linear regression line (λ), and the ratio of control to experimental subjects (m). Data from these analyses are presented in the Results separately for overall OFC cell density counts using light microscopy, and EM studies. We further minimized variability, using only the OFC from right hemispheres of adults (mean age of Control group ± SD: 46.3 ± 12.9; mean age of ASD group ± SD: 36 ± 7.1) with similar PMI (mean PMI of Control group ± SD: 23.4 ± 5.4; mean PMI of ASD group ± SD: 25 ± 8.3). Clinical characteristics and other data on human subjects can be found in Table 1. The study was approved by the Institutional Review Board of Boston University (Protocol X3408). The diagnosis of autism was based on the Autism Diagnostic Interview-Revised (ADI-R) in all cases. Diagnosis information for ASD cases used can be found in Table 2.
Immunostaining of inhibitory neurons and analysis
Histochemistry and immunostaining were performed as previously described [35, 37]. Briefly, we stained sections with Nissl (Thionin) to visualize the overall neuronal population in OFC (Fig. 2). Sections were pre-mounted on glass slides and air-dried before processing. Before staining, sections were defatted by incubation in chloroform/ethanol (1:1) mixture for two hours, followed by gradual rehydration in graded alcohols. Then sections were stained in 0.05% of thionin blue for 15 mins. Stained sections were dehydrated in increasing grades of alcohol solutions, cleared in Xylene, and coverslipped using Entellan (Merck).
To label the three types of cortical inhibitory interneurons, we used antibodies and immunostaining to identify their respective expression of calcium binding proteins, Parvalbumin (PV), Calbindin (CB), and Calretinin (CR). Briefly, adjacent sections were rinsed in 0.01M PBS, pH7.4, and then incubated in blocking buffer (10% donkey normal serum, 5% of bovine serum albumin, and 0.1% of Triton X-100 in 0.01M PBS) for 1hr, then incubated in respective mouse monoclonal or rabbit polyclonal primary antibody diluted in blocking buffer for 2 days at 4oC (anti-PV: 1:2000, #P3088, Sigma; anti-CB: 1:2000, #CB-300, Swant; anti-CR: 1:2000, #6B3, Swant ). Afterwards, the sections were rinsed in PBS and incubated for 3 hours with donkey anti-mouse or donkey anti-rabbit biotinylated secondary antibodies (1:300, Vector Labs), then thoroughly rinsed in PBS. We then used avidin–biotin–peroxidase kit (Vector Labs) and diaminobenzidine (Zymed Laboratories) to visualize CB-, PV-, or CR-expressing neurons. After staining, sections were mounted and cover-slipped, following the same process as for Nissl staining.
We first examined Nissl stained sections to determine the contour and boundaries of each layer, using an Olympus BX-60 microscopy system equipped with Neurolucida and StereoInvestigator software (Microbrightfield) (Fig. 2). We used stereological principles and unbiased systematic sampling to count neurons in each layer and estimate their density. Specifically, we analyzed sections within series with a z-interval of 500 μm that was kept constant across cases. Two to three counting sites across the OFC area on a section were selected. To ensure consistency and minimize variability due to cortical folding that influences layer structure, cell morphology, and density [40], we analyzed counting sites from straight gyral segments and excluded segments in sulcal depths or near the top of a gyrus. Each counting site was a rectangular ROI covering the entire cortical column (pial surface to white matter) with average width of 500 µm. Overall, we sampled 9% of the OFC area in each section. In each counting site, we outlined 6 layers based on cellular morphology and density, further dividing each rectangular ROI into 6 laminar-specific counting sites. We counted neurons within each layer contour using the optical fractionator protocol (sampling grid size: 300 µm, counting frame (disector) size: 100 µm, disector thickness: 5 µm) under the microscope, using the 100X objective. Density of neurons was calculated using two approaches. First, we divided estimated neuron counts by estimated volume of each layer to get the packing density, which highlights the layers with the most densely packed neurons in OFC. We additionally calculated the relative density of neurons, by dividing estimated neuron counts by the estimated volume of the sampled ROI (all layers from pial surface to border with white matter). The relative density highlights layers with the highest number of neurons. To accurately and reliably identify neurons in Nissl stained sections, we used an established algorithm that facilitates distinction of neurons from glia and endothelial cells based on several key and uniquely-identifying features [38]. Briefly, these features include the intensity of nuclear staining (light for neurons, astrocytes, and endothelial cells and dark for oligodendrocytes and microglia), the presence of lightly stained cytoplasm for neurons, the shape and size of the nucleus, and the distribution of heterochromatin, which is different among distinct types of cells. We then applied and adjusted the same counting sites and contours to the adjacent immunostained sections, and using the 40X objective we counted inhibitory neurons in each outlined layer exhaustively, by setting the same grid and counting frame size (300 μm). We then calculated the relative and packing density of CB, PV, CR inhibitory neurons in each layer as previously. Numbers are presented as Mean ± SE. We statistically analyzed estimates using ANOVA (Empowerstats) to compare Control and ASD groups. We also cross-validated the statistic comparison using a generalized linear regression (GLR) model, adjusting individual case heterogeneity and including PMI, age and sex as covariates. GLR analysis confirmed ANOVA results and did not reveal effects of PMI, age, and sex. Results from ANOVA analysis were presented in results and figures.
EM processing
EM processing was done as previously described [35, 36]. Briefly, 50 µm OFC sections adjacent to matching Nissl- and immunostained sections were processed for EM using a high contrast method [35, 36]. Sections were first rinsed in 0.1 M PB and postfixed in 6% of glutaraldehyde. Then sections were rinsed in 0.1M cacodylate buffer and then 0.1% tannic acid solution, followed by series of heavy metal solutions (1% osmium tetroxide with 1.5% potassium ferrocyanide, TCH aqueous solution ( 0.1 g of thiocarbohydrazide), and finally 2% osmium tetroxide) to induce heavy metal impregnation into lipid layers. We then washed the sections in distilled water and stained overnight in 1% uranyl acetate, followed by final stain in lead aspartate. Stained sections were then dehydrated in series of alcohols and cleared in propylene oxide. We then embedded sections in LX112 resin, sandwiched by thin sheets of Aclar film for long-term storage.
Before imaging, 1-mm wide rectangular segments containing desired OFC columns were cut out of the Aclar sandwich under a dissecting microscope, and each segment was then cut into 2 or 3 pieces (1X1 mm2), because the thickness of OFC is usually longer than the size limit of an EM imaging grid, and re-embedded in LX112 resin blocks for thin sectioning. Semi-thin sections (1 μm-thick) were cut and mounted on gelatin-coated slides and stained with Toluidine Blue Nissl solution, which stains all cells, the neuropil, and axons (Fig. 2). We used the stained sections to identify layer outlines and guide ROI selection for subsequent EM imaging. 100 nm-thick sections were cut and collected on single-slot pioloform grids for EM imaging.
EM imaging and analysis
We acquired high-resolution (30nm/pixel) images using a scanning electron microscope (Zeiss Gemini 300 with STEM detector, Atlas 5 software). We used tissue from 4 Control and 4 ASD subjects and imaged 268 square ROIs (120X120 μm), systematically sampling the entire section, with sampling ratio over 15%. In total, we analyzed 169 ROIs in the Control group and 99 ROIs in the ASD group, averaging 45 ROIs per layer (range from 19 – 74, depending on the size of the layer). Using matching Toluidine Blue stained sections as reference, we grouped images by layers for morphometric analysis (Figs. 2, 3).
We analyzed EM images with ImageJ (NIH). Cross-sections of myelinated axons in each image were identified, morphometrics were estimated, and summarized into a list. Measured morphometrics were grouped for comparison, and included axon size, profile area, and shape indicators (min and max diameters and circularity) that were used to estimate axon density (number of cross-sections per unit), surface area ratio (total surface area/image area), and minor diameter (excluding myelin) as indication of thickness. In addition, axon trajectory was estimated by the angle of the cross-section, as described previously [36]. We estimated the variation of angle values in each sampled image, as a measurement of axon trajectory variability. Numbers are presented as Mean ± SE. We used ANOVA to compare morphometrics between Control and ASD groups (Empowerstats). We also cross-validated the statistic comparison using a generalized linear regression (GLR) model, adjusting individual case heterogeneity and including PMI, age and sex as covariates. GLR analysis did not reveal effects of PMI, age, and sex and confirmed ANOVA analysis. Findings from ANOVA analysis were presented in results text and figures. In addition, we also used a repeated measures ANOVA followed by Tukey’s post-hoc tests to further statistically assess and confirm the effects of ASD on the myelinated axon population within the OFC gray matter. Because EM is a high-resolution, but low throughput approach, with relatively low N of subjects, a repeated measures design can also be typically used, wherein different sections, blocks of tissue, or regions of interest from each case can be pooled and compared. This framework formed the basis for a posteriori power analysis, performed as described above, to estimate appropriate sample size and power for the EM studies. All statistical analysis approaches used for the EM studies provided similar results, highlighting the same significant changes between Control and ASD groups.