|Home | About | Journals | Submit | Contact Us | Français|
Neural communication is disrupted in autism by unknown mechanisms. Here we examined whether in autism there are changes in axons, which are the conduit for neural communication. We investigated single axons and their ultrastructure in the white matter of post-mortem human brain tissue below the anterior cingulate cortex (ACC), orbitofrontal (OFC), and lateral (LPFC) prefrontal cortices, which are associated with attention, social interactions, and emotions and have been consistently implicated in the pathology of autism. Area-specific changes below ACC (area 32) included a decrease in the largest axons that communicate over long distances. In addition, below ACC there was over-expression of the Growth Associated Protein 43 accompanied by excessive number of thin axons that link neighboring areas. In OFC (area 11) axons had decreased myelin thickness. Axon features below LPFC (area 46) appeared to be unaffected, but the altered white matter composition below ACC and OFC changed the relationship between all prefrontal areas examined, and could indirectly affect LPFC function. These findings provide a mechanism for disconnection of long distance pathways, excessive connections between neighboring areas, and inefficiency in pathways for emotions, and may help explain why individuals with autism do not adequately shift attention, engage in repetitive behavior, and avoid social interactions. These changes below specific prefrontal areas appear to be linked through a cascade of developmental events affecting axon growth and guidance, and suggest targeting the associated signaling pathways for therapeutic interventions in autism.
Communication problems are at the core of the entire spectrum of autism disorders, disrupting particularly the social interactions of affected individuals. Genetic studies in autism have implicated changes in expression of genes that affect connectivity and neuronal excitability [(Morrow et al., 2008; Glessner et al., 2009; Weiss et al., 2009); reviewed in (Rubenstein and Merzenich, 2003; Walsh et al., 2008)]. At the brain level studies have identified functional abnormalities in neural networks in autism that prominently involve the frontal cortex (Herbert et al., 2004; Barnea-Goraly et al., 2004; Casanova, 2004; Courchesne and Pierce, 2005; Just et al., 2007), and functionally related distant association areas (Just et al., 2007; Koshino et al., 2008). Interestingly, the white matter below the frontal lobe is enlarged in young children with autism but not in adults, as assessed by structural imaging (Herbert et al., 2004). However, adults with autism continue to exhibit deficits associated with the disorder, and imaging studies show decreased functional connectivity between brain areas, desynchronization of cortical activity, and changes in the fractional anisotropy of the white matter (Barnea-Goraly et al., 2004; Kana et al., 2006; Just et al., 2007; Keller et al., 2007; Minshew and Williams, 2007; Koshino et al., 2008; Thakkar et al., 2008; Minshew and Keller, 2010). These findings suggest compromise of the structural integrity of the white matter that may be below the resolution of magnetic resonance imaging.
In spite of evidence indicating disruption of cortical pathways in autism, there is no information as to whether there are structural defects in single axons, which are the conduit for neural communication. To address this issue we investigated the fine structure of myelinated axons in post-mortem brain tissue of adults with autism and matched controls (Table 1 lists cases and clinical characteristics). We investigated exclusively myelinated axons because they make up the large majority (~90%) of axons (LaMantia and Rakic, 1990b), and focused on the white matter below three prefrontal regions: the anterior cingulate cortex (ACC), the orbitofrontal cortex (OFC), and lateral prefrontal cortex (LPFC). These functionally distinct regions are associated with attention, emotions and executive function, in processes that are severely affected in autism (Luna et al., 2002; Courchesne and Pierce, 2005; Bachevalier and Loveland, 2006; Hardan et al., 2006; Girgis et al., 2007; Loveland et al., 2008; Thakkar et al., 2008; Griebling et al., 2010).
The objective was to investigate whether or not abnormalities of the white matter below frontal areas in autism observed with structural imaging in children persist in the brains of adults with autism. We used unbiased quantitative stereology to study myelinated axons at high resolution at the light microscope (LM) and their fine structure at the electron microscope (EM) below the ACC (A32), OFC (A11), and LPFC (A46) areas (Figure 1A–C) in the brains of autistic (n=5, 1 female) and age-matched, typically developed controls (n=4, 2 females). We investigated the density of axons and thickness of axons and myelin sheaths. We examined only myelinated axons because they constitute the vast majority of axons in the frontal cortical white matter (~90%), the corpus callosum, anterior and hippocampal commissures in primates (LaMantia and Rakic, 1990b). Further, myelinated axons can be labeled using immunohistochemical methods, which we used for an independent evaluation at the light microscope.
We investigated axons in the superficial and deep white matter separately for two reasons. First, structural imaging studies suggested possible differences in pathology in autism (Herbert et al., 2004). And second, the deep white matter contains axons that communicate over long distances, whereas the superficial white matter contains axons that communicate mostly over short or medium distances (Schmahmann and Pandya, 2006). We thus divided the white matter into superficial (outer or radiate) and deep (inner or sagittal) compartments, based on axon orientation and distance from the cortical grey matter (Meyer et al., 1999). We determined axon alignment at the LM and at the EM in serial coronal ultrathin sections at gradually increasing distances from the grey-white matter border. The superficial compartment included axons that were mostly aligned radially and were immediately adjacent to layer VI of the overlying cortical areas (at a distance up to 2 mm from layer VI). The deep compartment included axons that run mainly sagittally and more or less parallel to the cerebral surface (Figure 1D–H).
Post-mortem prefrontal brain tissue was obtained from the Harvard Brain Tissue Resource Center through the Autism Tissue Program from five autistic adults (one female) and four typically developed, age-matched, controls (two females). The selection of matched cases used was based on tissue availability. The study was approved by the Institutional Review Board of Boston University. The diagnosis of autism was based on the Autism Diagnostic Interview–Revised (ADI-R) in all cases (Supplemental Table 1). Clinical characteristics are summarized in Table 1 and Supplemental Table 1. Some autistic cases were diagnosed with seizure disorder (case 5173), depression (case 4871) and schizophrenia (case 4541). Results from the analysis of the features of axons in these and the three female cases did not differ from other cases within each group, in this and other studies that used tissue from the same cases (Buxhoeveden et al., 2006; Schumann and Amaral, 2006; Yip et al., 2007).
We excised small blocks (~2 × 3 cm) of matched ACC (A32, A24), OFC (A11) and LPFC (A46) cortices containing grey and white matter (Figure 1A–C) based on the human brain atlas from the Autism Tissue Portal (www.atpportal.org) and [(von Economo, 2009), re-issued], and additional cytoarchitectonic studies of human prefrontal cortex (Selemon et al., 1998; Stark et al., 2004; Miguel-Hidalgo et al., 2006).
We postfixed tissue blocks in 2% paraformaldehyde and 2.5% glutaraldehyde, in 0.1M phosphate buffer (PB, pH: 7.4) for 2 days at 4°C. To preserve the ultrastructure until processing, tissue blocks were immersed in anti-freeze solution (30% ethylene glycol, 30% glycerol, 40% 0.05M PB, pH: 7.4 with 0.05% azide) and stored at −20°C. The blocks were then rinsed in 0.1M PB and cut coronally at 50µm thick sections on a vibratome (Pelco, series 1000). In all cases, tissue blocks through the grey matter of the areas of the associated white matter sampled were frozen in −70°C isopentane, cut in a cryostat (CM 1500, Leica) in the coronal plane at 20 µm in 10 series and mounted on chrome-alum coated slides.
We conducted several immuno-assays to label specific axon features and white matter oligodendrocytes. At the light microscope we labeled oligodendrocytes with an antibody against Myelin and Oligodendrocyte Specific Protein (MOSP). We also labeled myelinated axons with an antibody against NeuroFilament Protein 200 KDa (NFP-200), and examined branching axons with an antibody for Growth Associated Protein 43 KDa (GAP-43). To sort out axons from the thalamus we used antibodies against calbindin (CB) and parvalbumin (PV), which label excitatory thalamic projections to the cortex, and examined labeling at the confocal microscope and EM.
Series of free floating coronal tissue sections (50 µm thick) or cryosections mounted on slides (20 µm thick) were used in all immunohistochemical procedures. Sections were rinsed in 0.01 M PBS, pH 7.4, followed by 10% normal goat serum, 5% bovine serum albumin, and 0.1% Triton X-100 in 0.01 M PBS blocking solution for 1 h and incubated for 1–2 days in primary antibody.
We labeled axons in the white matter with antibodies against calbindin (CB; mouse monoclonal; dilution 1:1,000; Swant and/or Sigma), parvalbumin (PV; rabbit polyclonal; dilution 1:1000; Swant and/or Sigma), NeuroFilament Protein 200 KDa (NFP-200; rabbit polyclonal; dilution 1:200; Chemicon), and Growth Associated Protein 43 KDa (GAP-43; mouse monoclonal; dilution 1:2,000; Chemicon). We labeled oligodendrocytes with a monoclonal antibody against Myelin and Oligodendrocyte Specific Protein (MOSP; mouse monoclonal; dilution 1:1,000; Chemicon). The sections were rinsed in PBS, incubated for 4 h with goat anti-mouse or anti-rabbit secondary antibodies conjugated with the fluorescent probes Alexa Fluor 488 (green) or 568 (red; 1:100; Invitrogen) and thoroughly rinsed with PBS. In some cases, a biotinylated secondary antibody and an avidin–biotin–peroxidase kit was used to label CB-positive or PV-positive axons with diaminobenzidine (DAB, Zymed laboratories, San Francisco, CA), which were further processed for EM (see below). To test for non-specific labeling we performed control experiments with sections adjacent to those used in the experiments. These included omission of the primary antibodies and incubation with secondary antisera. Control experiments showed no immunohistochemical labeling.
Tissue processing and pre-embedding immunohistochemical labeling for serial EM is an especially challenging technique for post-mortem human brain tissue because of limited control over tissue extraction protocols, post-mortem interval and post-fixation. In addition, processing and labeling of the tissue can degrade the ultrastructure and preclude quantitative analyses. To address these issues we have developed several novel protocols that maximize tissue quality and specificity of labeling (Zikopoulos and Barbas, 2006; Zikopoulos and Barbas, 2007). We preserve tissue blocks or sections at −20°C in anti-freeze buffer solution for long periods of time, fix and process tissue using a variable microwave, and label tissue before embedding, all of which enhance and accelerate penetration of reagents in brain sections during processing, reduce non-specific background staining, minimize the need for detergents that degrade fine structure, and decrease potential damage of a series. These protocols have markedly increased tissue quality and made it possible to conduct 3D quantitative reconstruction of identified structures.
Sections were rinsed briefly in 0.1M PB and postfixed in a variable wattage microwave oven (Biowave, Pelco) with 6% glutaraldehyde at 150W. Small blocks of sections containing the outer (superficial) or inner (deep) parts of the white matter below prefrontal cortices were cut under a dissecting microscope, postfixed in 1% osmium tetroxide with 1.5% potassium ferrocyanide in PB, washed in buffer (PB) and water and dehydrated in an ascending series of alcohols. While in 70% alcohol they were stained with 1% uranyl acetate for 30 min. Tissue sections were then cleared in propylene oxide and embedded in araldite at 60°C. Serial ultrathin sections (50 nm) were cut in the horizontal plane with a diamond knife (Diatome) using an ultramicrotome (Ultracut; Leica) and collected on single slot grids to view with a transmission electron microscope (100CX; Jeol), as described (Zikopoulos and Barbas, 2006; Zikopoulos and Barbas, 2007). Myelinated axons were easily identified at the EM by the darkly stained electron dense myelin sheath (Peters et al., 1991).
One series of sections was stained for Nissl using thionin to view neurons and glia and examine the cytoarchitecture of each area, as described (Barbas and Pandya, 1989; Dombrowski et al., 2001). Sections were dried, defatted in a 1:1 solution of chloroform and 100% ethanol for 1 h, rehydrated through a series of graded alcohols and dH2O, stained with 0.05% thionin (pH 4.5) for 15 min, differentiated through graded alcohols and xylenes and coverslipped with Entellan (Merck, Whitehouse, NJ).
To determine adequate sample size we performed a priori power analysis, using repeated measures from a pilot study, and a posteriori power analysis, using the actual data. The a posteriori power analysis took into consideration all known and estimated variables, including age, sex, post-mortem interval (PMI), and other diagnoses, and was used to test the validity of the pilot study and the a priori power analysis, which always rely on fewer data points and make more assumptions. These analyses, which had an estimated large effect size in the population, 0.80, showed that the sampling ratios used exceeded the samples needed to detect differences with a greater than 90% probability. We used several additional computational and statistical methods to establish adequate sample size, including progressive means analysis with exhaustive sampling, and the formula of West et. al., (West et al., 1991). We used higher sampling fractions in all analyses than the minimum of 3 cases and 3 sections required by the power analyses, and in most cases we expanded these numbers to 5 brains from autistic individuals, 4 brains from control cases, and more than 5 sections per case. In all analyses the sample size included number of cases, volume fraction of areas sampled, and number of individual axons examined, which were not only adequate but exceeded the estimated minimum requirements. Moreover, for each case we examined thousands of axons at very high resolution, totaling nearly 50,000 for the study. In one analysis (A32) tissue was available only for 3 control cases (1 female and 2 male).
We estimated the overall and laminar density of neurons in A32, A11, and A46 overlying the sites of white matter analysis, and the density of oligodendroglial cells in the superficial part of the white matter below OFC (A11) using the unbiased stereological method of the optical fractionator (Gundersen, 1986; Howard and Reed, 1998) with the aid of commercial software (StereoInvestigator; Microbrightfield, Williston, VT, USA), as described (Zikopoulos and Barbas, 2006). For LM quantitative analyses we used a minimum of three sections from one series of coronal sections (20 µm thick) from each case. To ensure unbiased estimate of the number of neurons we first measured the thickness of each section, and used StereoInvestigator to set a guard zone at the bottom and top of each section to correct for objects plucked during sectioning; the disector thickness was thus smaller than the thickness of the section (Gundersen, 1986; West et al., 1991; Howard and Reed, 1998). The sampling fraction was 1/50 of the total volume of the area examined, and was determined in pilot studies using exhaustive sampling and progressive means analysis so that final estimates had a standard error ≤ 10%. The use of uniform random sampling ensured that every part of the area examined had the same chance of being included in the sample. The numbers of neurons and volumes of the corresponding area and layers estimated with the Cavalieri method were divided to assess the density of neurons in each case. We normalized data by expressing the density of neurons as a percentage of the total density of all labeled neurons in each area in each case.
For the analysis of the number of myelinated axons with branches, and to estimate the number of axons that express GAP-43, we double labeled a minimum of three sections from one series per case with NFP-200 and GAP-43 and used systematic random sampling (sampling fraction 1:50) to capture stacks of confocal images at high magnification (×1,000). For analysis of axons with GAP-43 we examined another anterior cingulate area (A24), in addition to the neighboring A32, in order to increase the power of the analysis for the control cases in the anterior cingulate (3 control cases for A32 and 4 control cases for A24).
To reduce the fluorescent glare, we applied three-dimensional deconvolution algorithms to images prior to analysis with the aid of Autodeblur (Media Cybernetics). We used these image stacks to create three-dimensional projections in ImageJ, which we viewed and re-sliced in the X, Y or Z axis, in order to decipher axon branches from crosses. Profile counts of axons with branches, or axons that expressed GAP-43, were obtained using ImageJ and normalized by dividing with the total number of sampled axons in each case. We also assessed GAP-43 expression using an independent method by estimating the ratio of the surface area of GAP-43/NFP-200-positive axons to the total surface area of all axons labeled with NFP-200. Both high resolution quantitative methods allowed accurate quantification of axons expressing GAP-43, whilst excluding unrelated signal in glia or non-neural tissue that might have concealed potential differences.
We estimated the thickness of the cortical grey matter of ACC (A32) in a series of 20 µm thick coronal sections per case, including its divisions at the bottom of a sulcus, where the cortex is compressed, at the top of a gyrus, where the cortex is thick, and at relatively straight parts of the cortex, using ImageJ as described (Hilgetag and Barbas, 2006).
To determine the density of axons and the thickness of axons and myelin in the white matter, we sampled a volume of approximately 1cm3 below each prefrontal cortical area, with a systematic random sampling fraction of 1:1,000 that yielded more than 2,000 axons, per case, per area. We divided the white matter (as described above) into a superficial part (closer to the grey matter) and a deep part.
High resolution images of areas of interest were captured with a digital camera attached to the electron microscope (ES1000W, Gatan), imported in ImageJ and calibrated. We estimated the overall density of axons at low magnification (×3,300) by dividing the number or the surface area of the axon profiles by the total surface area of the sampled region. We estimated the maximum inner and outer diameter as well as the thickness of the surrounding myelin sheath, at high magnification (×10,000). To minimize variability and test for errors due to sectioning we measured axons that were perpendicular to the cutting plane and appeared cylindrical. We then repeated the analysis to include all axons by measuring the diameter perpendicular to the center of the maximum diameter of the axon profile. The two analyses yielded similar results and were combined.
We followed axons in the superficial part of the white matter of ACC in three autistic and three control cases in long, uninterrupted series of ~400 ultrathin sections (thickness: 150 nm each). The volume examined was ~600,000 µm3. We viewed at least 200 axons per case at high magnification (×10,000) and photographed them throughout their extent in the series using exhaustive sampling. High resolution digital images were imported as a series in Reconstruct [http://www.bu.edu/neural (Fiala, 2005)] and aligned, as described (Zikopoulos and Barbas, 2006). Axons were traced, reconstructed in three dimensions, and their average diameters calculated. Traces of small, medium, large and extra-large axons were color-coded for easy visualization. We estimated the number of all branching points and the number of axons (by size) with branches in each series. Branches were reliably identified and associated with parent axons based on the continuity of the axoplasm, and the thinning or disappearance of myelin at the branching points.
Data were gathered blind to condition and cortical region. Random codes for cases and images were broken after completion of each part of the study. In all cases data collection was performed by at least two investigators. Values obtained from the two independent measures were highly correlated (Pearson R = 0.97, p = 0.001). The samples were obtained from widely spaced sections (one every ten) and fields of view through systematic random sampling to minimize the likelihood of sampling axons from the same parent branch. This sampling scheme and the fact that most axons branch very close to or after they enter the grey matter minimized the likelihood of counting segments of the same axon more than once.
Data distributions for continuous variables were not significantly different from normal as determined by the Kolmogorov–Smirnov test, and thus allowed the use of parametric statistics. We initially used x2 and Kolmogorov–Smirnov tests to examine axon size distributions and multiple linear regression analysis to examine correlations. Data were evaluated with Statistica (StatSoft, Tulsa, OK), through scatter and frequency distribution plots and K-means cluster analysis with parameters set to maximize initial between-cluster distances. We used MANOVA to test for differences among axon and neuron populations and densities. We then used post hoc analyses using Bonferroni’s/Dunn’s (all means) to identify possible differences between groups. For the axon branching analysis we used a two-tailed t-test. For all analyses p values <0.05 were taken as statistically significant.
We also employed three different multivariate analyses to assess global similarities and dissimilarities of the white matter below prefrontal cortices based on all the ultrastructural features examined. We performed discriminant analysis to identify experimental measures that minimize the overlap and clearly separate the distributions of individual data points belonging to different cortical areas for each case. We performed hierarchical cluster analysis (HCA) to group areas based on (dis)similarities in their parameter profiles. In this test, the relative similarity of areas is expressed as the distance between two branching points in a cluster tree diagram. Finally, we used nonmetric multidimensional scaling (NMDS), to arrange prefrontal areas in control and autistic cases in a low-dimensional (2D) space based on the pairwise correlation (dis)similarities between areas. The relative proximity among items in an NMDS diagram represents their relative similarity. We performed NMDS using both the mean values for each area for the autistic and control cases to maximize their separation, as well as using the entire range of values for each case to take into account sample variability. HCA and NMDS analyses employed squared area (dis)similarity matrices derived from the normalized areal profiles by Pearson's correlation.
We also examined potential effects of sex, PMI, age at death, and other diagnoses (i.e. seizures) on all estimates for axon density, size, branching, expression of GAP-43, as well as neuronal and glial cell densities, using correlation analysis (Supplemental Figure 1A–F). In addition we compared all estimated variables between and within control and autistic cases using MANCOVA with sex, PMI, age at death, and other diagnoses as the covariate and compared the results from this analysis with the MANOVA outcome.
The PMI for the control cases averaged 17.5 hours (±1.5 SEM) and for the autistic cases 39.8±16.9 hours (Table 1). This number was significantly higher for the autistic cases because of case 4871, which had a PMI of 99 hours. Without this case the PMI for the autistic cases was comparable to the control cases (25±4.8 hours). Examination of the structural integrity of the tissue and the quality of labeling revealed no differences between case 4871 and the other autistic cases at the light or confocal microscope. At the EM the density of axons in case 4871 was not affected but the membranes of some glial cells and small parts of the myelin surrounding some axons were compromised. As a result we did not use this case for glial cell density estimates at the EM and we sampled a much larger area to estimate axon and myelin diameter. The results obtained from case 4871 correlated well with measurements from the other autistic cases and were thus included in the analyses.
We performed additional statistical analyses to assess the generalizability of the results and to estimate whether the data could be used to accurately predict relationships between the estimated variables in independent samples. To this effect we used cross-validation techniques specifically designed to test the validity of the results and groupings derived from ANOVA, cluster and discriminant analyses, and NMDS. This method involved partitioning the sample of cases into N=30 complementary subsets, performing the analyses on N-1=29 subsets, validating it on the other subset, and then repeating the process N times. The N=30 sample subsets was determined so as to include all possible combinations of control and autistic cases in groups of 3, which was a minimum requirement to perform all analyses (e.g., autistic group 1: cases 4541, 4871, 5173; control group 1: cases 4786, 4981, 5353; autistic group 2: cases 6232, 6677, 5173; control group 2: cases 6004, 5353, 4981; … etc). This method yields fits of predicted and actual data, which are acceptable if the root mean squared errors remain low. Finally, we used two complete datasets collected independently by two investigators and repeated all analyses using a repeated measures design.
Myelinated axons made up approximately 40% of the white matter in brain tissue from both autistic and control cases, resulting in an average density of 0.36 axons / µm2, in agreement with previous studies in non-human primates (LaMantia and Rakic, 1990a; LaMantia and Rakic, 1990b; LaMantia and Rakic, 1994). The rest of the white matter was occupied by glia, mainly oligodendrocytes.
Overall axonal density between normal and autistic groups was similar below all prefrontal areas and parts of the white matter examined (Supplemental Table 2). This finding is in agreement with previous reports indicating that the enlargement of the frontal white matter observed in children with autism is transient (Herbert et al., 2004).
However, axons vary in thickness, which affects their physiologic properties (Rushton, 1951; Wang et al., 2008), but their key features in prefrontal white matter and potential disruption in autism are unknown. We addressed this issue in brain tissue from both autistic (n = 5, 1 female; 2,000 axons/case) and control (n = 4, 2 female; 2,000 axons/case) individuals using electron microscopy (EM). Myelinated axons varied in diameter, ranging from 0.1–7 µm (axon thickness, inner diameter). The average diameter of axons in the superficial white matter was 0.8 ± 0.02 µm, and in the deep white matter axons were slightly thicker (0.9 ± 0.02 µm). Similar axon sizes have been reported in other white matter regions of the primate brain, such as the corpus callosum, anterior commissure and hippocampal commissure (LaMantia and Rakic, 1990b).
Cluster analysis of all axons segregated them into four groups based on inner diameter (without myelin), which were used for further comparisons (small < 0.35 µm; medium, 0.35–0.69 µm; large, 0.7–1.4 µm; and extra-large > 1.4 µm; p = 0.01). In all cases, most axons were small and medium in size (small: 36%; medium: 46%; large: 15% and; extra-large: 3%).
The relative position of axons within the white matter is an indicator of their termination in nearby or distant brain areas. The deep white matter includes mostly long range pathways (Herbert et al., 2004; Hilgetag and Barbas, 2006; Schmahmann and Pandya, 2006; Petrides and Pandya, 2006; Petrides and Pandya, 2007; Sundaram et al., 2008). Previous studies have suggested that long range cortico-cortical pathways that link frontal areas with other cortices are weak and disorganized in autism (Just et al., 2004; Courchesne and Pierce, 2005; Just et al., 2007) but the cause is unknown. To address this issue we measured the inner diameter of axons in the deep white matter. We found that the autistic group had significantly fewer extra-large axons only in area 32 of ACC (hereafter called ACC) compared to controls (p = 0.03; Figure 2A–I; Supplemental Table 2). This group constituted a small proportion of all axons, but is within the range of densities of long distance pathways, which are sparse in comparison with short range pathways (Barbas, 1988). Nevertheless, long distance pathways have considerable influence on the cortex. The prefrontal cortex, in particular, relies on sparse cortico-cortical pathways for all its sensory input.
To narrow down the list of possible long range pathways affected in autism, we labeled axons with the calcium binding proteins calbindin or parvalbumin which are expressed by distinct classes of excitatory thalamic neurons that project to the cortex (Jones, 1998; Zikopoulos and Barbas, 2007; Jones, 2007). Quantitative analysis with EM showed no significant differences in the proportion of thalamo-cortical axons in autistic and control cases (p = 0.23; Supplemental Figure 2). This finding suggests that in autism there is reduction in other long distance pathways. Among long range pathways, the cortico-cortical are likely affected, in view of their functional disruption in autism (Just et al., 2004; Courchesne and Pierce, 2005; Just et al., 2007). However, the involvement of other cortico-subcortical pathways, including cortical projections to the striatum, or connections with the basal forebrain and brainstem cannot be excluded.
The superficial white matter contains mostly short and medium range axons that connect nearby areas, but also includes long distance axons as they pass through to reach or exit the cortex. However, short and medium range pathways make up the bulk (>80%) of all cortico-cortical connections (Barbas, 1988). We compared axons in the superficial white matter in autistic and control cases. Figure 2J–S shows the results from this analysis and provides evidence that the density of small axons was significantly higher in the autistic than in the control group, specifically in the superficial white matter of ACC (p=0.01), but not below the other areas (Supplemental Table 2).
We tested whether the higher density of small axons could be explained by differences in the density of the overlying neurons in ACC, or the thickness of the cortex. There were no significant differences in the overall or laminar density of neurons in ACC in the brains of control (overall density = 32,536/mm3 ± 5,450 SEM) and autistic (32,388/mm3 ± 3,145 SEM) cases (Figure 3A–D), or in overall cortical thickness (average thickness, control = 2.9 mm ± 0.2 SEM; autistic = 3.0 ± 0.3, p = 0.7), or in the segments of ACC separated into sulcal, straight cortex, or the top of the gyrus (Figure 3E).
We reasoned that the higher density of small axons in the autistic cases must be due to increased branching of axons in the superficial white matter. To address this hypothesis we followed and reconstructed axons (> 800) in three dimensions from large uninterrupted series of EM images. We found that the white matter below ACC had a significantly higher percentage of axons with branches in autistic compared to control cases only for medium axons (Figure 4; p = 0.03), which give rise to small axons at bifurcations. The average number of branches per axon in the superficial white matter below ACC was 2.6 ± 1.4 (branches ± SEM; control cases: 1 ± 0.5 for small axons; 1.8 ± 0.1 for medium axons; 4.2 ± 3.1 for large axons; 1 ± 0.7 for extra-large axons; autistic cases: 2.2 ± 0.9 for small axons; 3.2 ± 1.8 for medium axons; 3.2 ± 1.6 for large axons; 0.5 ± 0.2 for extra-large axons). Most points of bifurcation were unmyelinated or arose after thinning of the myelin [typically seen near the nodes of Ranvier (Peters et al., 1991)]. Axon branches were in most cases thinner than their parent axons, in agreement with previous reports (Ramon y Cajal, 1911; Peters et al., 1991; Schmitt et al., 2004).
Using an alternative method, we labeled myelinated axons with an antibody against a neurofilament protein (NFP-200) and determined the proportion of axons that branched, using image stacks at the confocal microscope, which confirmed the EM results (Figure 4C, D). Taken together these data show an excess of axons that most likely course over short or medium distances. This finding is consistent with the hypothesis that prefrontal areas are over-connected in autism (Casanova, 2004; Courchesne and Pierce, 2005).
Supernumerary branching may be associated with increased axon production or decreased axon pruning occurring in the post-natal period (LaMantia and Rakic, 1990a). GAP-43 is expressed at high levels during rapid axon growth, and is subsequently markedly reduced (Benowitz and Routtenberg, 1997). In the adult brain GAP-43 is found in significant amounts only in association cortices, including the ACC, and at focal sites after brain injury (Benowitz and Routtenberg, 1997). Based on this evidence we hypothesized that the high proportion of axons with branches immediately below ACC (A32) may reflect a high level of GAP-43 in autism. Immunohistochemical analysis of the superficial white matter showed over a two-fold increase in the proportion of axons that express GAP-43 only below ACC (A32) in autistic (22% ± 5 SEM) compared to control (9% ± 2) cases (p = 0.02; Figure 5), but not below the other areas (Supplemental Figure 3).
We then investigated whether other factors might have affected GAP-43 levels, such as medication, or co-morbidity with schizophrenia, epilepsy, or depression, reported for three of the autistic cases. A correlation analysis did not reveal significant associations at the 95% confidence level, however, more autistic cases will need to be examined in future studies to fully address this issue. Moreover, analysis of a nearby cingulate area (A24) showed no difference in GAP-43 between autistic and control cases (n = 4 control and 5 autistic cases; Supplemental Figure 3). This evidence suggests that the increased proportion of axons that express GAP-43 in ACC A32 may be specific, though more areas must be studied to address this issue. It is not clear why neighboring cingulate areas 32 and 24 differ in the expression of GAP-43 in autism. The difference may be related to their developmental patterns, and in particular myelination, which is completed much earlier in A24 than in A32 (Flechsig, 1901).
We next investigated the thickness of myelin, which insulates axons and affects conduction velocity. We found a positive linear correlation between the thickness of axons and the thickness of their myelin sheath (Pearson, R=0.7, p<0.05; Supplemental Figure 4). This finding is consistent with the classic relationship of the inner to the outer diameter of axons, known as the g-ratio (Rushton, 1951). In all cases and areas the g-ratio increased significantly with axon size (Figure 6A), in agreement with recent studies (Paus and Toro, 2009).
Below ACC overall myelin thickness was lower in autistic than in control cases, but the difference could be explained by the higher prevalence of thin axons, as predicted by the g-ratio (Supplemental Table 2). In contrast, in the superficial white matter of OFC area 11 (hereafter called OFC) the myelin was significantly thinner in autistic than in control cases regardless of axon diameter (Figure 6B–D). Moreover, below OFC there were no differences in the relative proportions of small, medium, large and extra-large axons between autistic and control cases that could account for the overall thinner myelin. This evidence suggests that myelination per se is affected only below OFC among the areas studied. Consequently, the overall g-ratio in OFC increased to 0.63 (± 0.01 SEM) from 0.58 (± 0.01) in all other cases and areas (Figure 6A).
To determine whether the thinner myelin could be due to reduction in the number of oligodendrocytes, which myelinate central axons, we estimated their density in the white matter below OFC. We conducted independent stereologic analyses at the light microscope after immunohistochemical labeling for myelin and oligodendrocyte specific protein, and at the EM, based on the distinctive morphology of oligodendrocytes (Supplemental Figure 5). The two independent methods yielded similar data, and showed no significant differences (p = 0.53). The respective densities (±SEM) were: EM (oligodendrocyte profile counts/mm2), control cases: 703 ± 19; autistic cases: 784 ± 135; and LM (stereology, oligodendrocytes/mm3), control cases: 137,203 ± 24,397; autistic cases: 141,220 ± 71,877; p = 0.53 (Supplemental Figure 5).
We then performed a discriminant analysis to identify the experimental measures that were most informative in distinguishing the white matter below specific prefrontal areas in control and autistic cases and constructed detailed fingerprint diagrams (Figure 7). All parameters used were highly characteristic for identifying individual areas in all cases, with the exception of overall axon density. Hierarchical cluster analysis showed a clear separation of ACC, OFC, and LPFC areas in control cases, which was less apparent in the autistic group. Independent nonmetric multidimensional scaling analysis using both the mean values for each area for the autistic and control cases to maximize their separation, and using the entire range of values for each condition to take into account sample variability (Figure 8; Supplemental Figure 1G–K), corroborated these results and showed that the axons below lateral prefrontal A46 (LPFC) were not affected in autism. However, the white matter below ACC and OFC had altered characteristics in autism and were more similar to each other than with their respective controls.
The significant increase in the number of small axons below ACC in the autistic cases changed the relationship between the prefrontal areas examined and revealed additional differences (Figure 9). In control cases OFC had a higher proportion of small axons than ACC (Figure 9A), and a lower proportion of large axons than LPFC (Figure 9B). In turn, ACC had a higher proportion of medium axons than LPFC (Figure 9C). These inter-areal differences were not apparent in the autistic group, which had more small axons in the deep white matter of LPFC compared to OFC. This evidence indicates that the relationships among these prefrontal areas are altered in autism, suggesting widespread repercussions on neural communication (Bauman and Kemper, 2005; Loveland et al., 2008).
We performed a series of additional statistical analyses to assess the robustness and generalizability of the findings. Correlation analysis showed no potential effects of sex, PMI, age at death, and other diagnoses (i.e. seizures) on all estimates for axon density, size, branching, expression of GAP-43, as well as neuronal and glial cell densities at the 95% confidence level (Supplemental Figure 1A–F). The only significant correlation was between sex and the density of thalamocortical axons, where the 3 female cases (one autistic and two controls) had higher values than the male cases. In addition, we compared all estimated variables between and within control and autistic cases using MANCOVA with sex, PMI, age at death, and other diagnoses as the covariate and detected the same changes as with MANOVA with no significant correlations with the covariant variables.
In view of evidence for significant changes in the structure, neurochemistry, and function of the cortex with age, both in autistic and in typically developing individuals, we selected post-mortem tissue from cases that were closely matched regarding age at death, and ranged from 30–44 years. In contrast to the major neural changes that take place during childhood, teenage years and late adulthood, there is no evidence for significant changes in the density of neurons, axons, and synapses as well as in the myelination of axons within this age range in autistic or typically developing individuals [e.g., (Yakovlev and Lecours, 1967; Bauman and Kemper, 2005; Redcay and Courchesne, 2005; Amaral et al., 2008)]. In agreement with these studies we found no significant correlation of age at death with any of the estimated parameters (Supplemental Figure 1A–F).
The cross-validation tests conducted to estimate whether the data could be used to accurately predict relationships between the estimated variables in independent samples, were specifically designed to test the validity of results and groupings derived from ANOVA, cluster and discriminant analyses, and NMDS. These tests, which were performed on all possible combinations of control and autistic cases in groups of 3, yielded highly accurate fits of predicted and actual data with low root mean squared errors (≤ 0.01), and concurred with the previous analyses. Finally, we used two complete datasets collected independently by two investigators and performed all analyses in a repeated measures design, which also yielded similar results.
The robustness of the results was strengthened by the fact that variability in all estimated parameters was low and the observed differences were found in all cases with no exception (Supplemental Figures 6, 7).
Previous studies have suggested that the fundamental defect in autism is at the synapse (Sudhof, 2008; Bourgeron, 2009). Our findings show physical changes in single axons below prefrontal areas in autism that likely affect synaptic function. The changes in axons were found in all autistic cases regardless of the presence or absence of epilepsy or mental retardation (Table 1, Supplemental Figures 6, 7). These findings suggest a fundamental autism phenotype in axons that make up the brain’s communication system.
The high density of thin axons below ACC is consistent with studies suggesting excessive short range connectivity in autism (Courchesne and Pierce, 2005). The presence of supernumerary axons likely accounts for the increased cortical folding in the frontal lobe in autism, consistent with the hypothesis that tension exerted by cortico-cortical connections is a significant factor in shaping the gyrencephalic cerebral cortex (Van Essen, 1997; Hilgetag and Barbas, 2006). Diseases of developmental origin lead to atypical folds (Levitt et al., 2003; Nordahl et al., 2007) suggesting abnormal connectivity.
The areas studied are robustly interconnected, and have a key role in emotions, attentional mechanisms and executive control (Barbas, 2000), in processes that are severely affected in autism. In particular, the OFC has an overview of the sensory environment, and through robust connections with the amygdala participates in the process of assessing the emotional significance of events (Barbas and Zikopoulos, 2006). The change in the optimal relationship between axon diameter and myelin in OFC, and the reduction of neurons in the amygdala in autism (Schumann and Amaral, 2006), provide the anatomic basis for disrupted transmission of signals for emotions (Bachevalier and Loveland, 2006; Loveland et al., 2008). Specifically, the increase in the g-ratio of axons above the optimal average value (~0.6), suggests suboptimal conduction velocity (Rushton, 1951; Paus and Toro, 2009), and these factors have been linked to cytoskeletal defects that affect cell metabolism and neurotransmission (Paus and Toro, 2009).
On the other hand, the ACC has a role in attentional processes, and has the most widespread connections within the prefrontal cortex (Barbas et al., 1999). Through these robust connections the ACC may affect function in LPFC, which has a key role in cognition. For example, when excitatory axons from ACC form synapses with inhibitory neurons in LPFC they target preferentially calbindin inhibitory neurons (Medalla and Barbas, 2009), which have modulatory effects on pyramidal neurons, increasing the signal-to-noise ratio (Wang et al., 2004). These synaptic specializations suggest that ACC can reduce noise in LPFC, and facilitate holding attention on a task. The exuberance of thin axons in ACC in autism suggests a potential exaggeration of this mechanism, consistent with atypical LPFC activation in autism reported in functional imaging studies (Luna et al., 2002).
The ACC develops early in ontogeny in non-human primates (Rakic, 2002), suggesting early engagement of synaptic sites by its axons. Exuberance of thin axons that course over short or medium distances in autism may lead to occupation of sites normally available to the considerably sparser long distance pathways. The latter are at a competitive disadvantage, not only because they develop later, but also because they need additional time to extend long axons to form synapses in the prefrontal cortex. Reduction in the strength of long distance pathways in autism may thus be secondary to the excessive short range connections of ACC. This connectivity bias may help explain why individuals with autism do not adequately shift attention when necessary, and engage in repetitive and inflexible behavior [e.g., (Luna et al., 2002; Thakkar et al., 2008; Minshew and Keller, 2010)].
In contrast to ACC, in OFC there were no differences in the composition of the four size groups of axons, but the myelin was overall thinner. What underlies these seemingly disparate changes in axons in ACC and OFC? The varied genetic mutations that confer susceptibility to autism affect, in general, aspects of development (Walsh et al., 2008; Weiss et al., 2009). The interplay of developmental events may help explain all the observed changes in axons, as summarized in Figure 10. Normal development is initiated with neurogenesis and migration followed by axon elongation in an environment enriched with GAP-43. Myelination occurs after axons grow, and extends well beyond the second year of life (Yakovlev and Lecours, 1967) and into adulthood (Paus et al., 1999). Once myelination is initiated, signals from myelin proteins help stabilize axons by inhibiting GAP-43 synthesis and halting axon growth (Kapfhammer and Schwab, 1994). Importantly, the inhibitory effects are reciprocal, so that GAP-43 exerts inhibitory effects on myelin (Kapfhammer and Schwab, 1994).
In non-human primates, neuronal migration in ACC is completed first among the areas studied (Rakic, 2002) but myelination occurs considerably later in A32 (Flechsig, 1901; Von Bonin, 1950), suggesting that the ACC has prolonged exposure to GAP-43 in an environment that is permissive for axon growth. Moreover, within the white matter, GAP-43 is highest in its outer border, which is closest to the cortex, consistent with its rapid transport to axon terminals (Benowitz and Routtenberg, 1997). The significant increase in GAP-43 in the superficial white matter of ACC in autism is consistent with the exuberance of short and medium range axons. The assumption in our model, that GAP-43 in ACC A32 is elevated during development, is based on the increased levels seen in adults with autism (this study) and findings that the frontal white matter is enlarged in children with autism (Herbert et al., 2004). Further, our finding that myelin thickness in ACC was unaffected in autism can be explained by the fact that ACC axons in A32 myelinate very late (Flechsig, 1901), when GAP-43 levels are lower. On the other hand, OFC (area 11) develops after ACC (Rakic, 2002) but its myelination begins earlier than in ACC (Flechsig, 1901). The level of GAP-43 in OFC at the time of myelination onset is unknown, since our analysis was on adult brains. However, because the interval from cell migration to myelination is shorter in OFC than in ACC, even a small elevation in GAP-43 expression may be sufficient to retard myelin growth without causing excessive branching of axons, as seen here. Finally, LPFC (area 46) develops after ACC and OFC (Rakic, 2002) and myelinates very late (Flechsig, 1901; Von Bonin, 1950), so neither axon growth nor myelination is affected in autism. This model suggests that the distinct abnormalities in ACC A32 and OFC A11 in the autistic cases are linked and traced to a common developmental disturbance that affects the onset and perhaps the duration of expression of GAP-43 and its interaction with myelin.
Neuronal migration appears to be intact in autism, since neither neuronal density nor cortical depth was affected, at least in the parts of the three areas studied (Supplemental data: Cytoarchitecture of ACC, OFC, and LPFC). Our data suggest that the insult occurs later, when axons connect with other areas in the presence of high GAP-43 expression, and possibly other growth factors, which may remain elevated in adulthood in response to inflammation (Vargas et al., 2005). The associated signaling pathways need to be investigated for therapeutic interventions in autism.
GAP-43 is up-regulated by a variety of external factors as well, including estrogenic agents that disrupt endocrine function, such as bisphenol A, used for lining plastic food and drink containers, linoleic acids found in some oils, and by immunosuppressive and psychiatric drugs used for a variety of common disorders, including psoriasis, asthma, rheumatoid arthritis, depression and anxiety (Wong et al., 1989; Jyonouchi et al., 2001; Granda et al., 2003; Croen et al., 2005; Ostensen et al., 2006; Sairanen et al., 2007; Nguyen et al., 2009; Brown, Jr., 2009). Several of these substances came into heavy use in the early 80s at a time when the prevalence of autism began to rise (Blaxill, 2004). Epidemiologic studies are necessary to investigate if these events are merely coincident or if the cumulative effects of dietary factors and drugs change the uterine and postnatal environment and perturb the expression of factors implicated in axon growth and guidance in autism.
We thank Clare Timbie, Amar Patel, Mary Louise Fowler, Seung Yeon Michelle Kim, and Sue Paul for technical assistance and Marcia Feinberg for assistance with electron microscopy. We also thank Dr Alan Peters and Dr Claus Hilgetag for useful comments. We gratefully acknowledge the Autism Tissue Program and the Harvard Brain Tissue Resource Center for providing human brain tissue. This work was supported by Autism Speaks, and NIH grants from NIMH, NINDS, and NSF CELEST grant.