Postnatal development and maturation of layer 1 in the lateral prefrontal cortex and its disruption in autism

Light microscopy

Layer 1 was identified in stained sections using reference maps [6, 77, 100, 101, 123, 124]. Neurons were identified and quantified in Nissl-stained sections using a validated algorithm [46]. We quantified neuron density in multiple ROIs on one series of coronal sections per case. Sequential sections in children were on average 500 μm apart, while those in adults were on average 400 μm apart. We used an unbiased statistical sampling method to sample the areas of interest using commercially available software (StereoInvestigator, MicroBrightfield). The counting frame for these analyses was set to 150–180 μm with a height of 8 μm and grid spacing of 150–560 μm. The thickness of the section was measured at each counting site and a guard zone was set at the top of each section (2 μm). We calculated the density of neurons in layer 1 (neuron density, also referred to as the packing density) by dividing the estimated number of cells in layer 1 in each case by the measured volume of layer 1 in mm³. Quantification of immunolabeled tissue was performed at 400x, while Nissl-stained tissue was quantified at 1000x. Quantification of immunohistochemistry-labeled tissue sections was exhaustive, meaning that all counting frames were analyzed (counting grid and frame size were equal).

We measured the area fraction of neurons in high-resolution images of Nissl-stained tissue (thionin and toluidine blue) and the area fraction of axons in toluidine blue- and osmium-stained sections. Area fraction is a measure of the percent of the overall tissue surface area which contains axons or neurons. We used a consistent protocol to modify brightness and contrast of each image through adjustment of the levels of the histogram to maximize the black and white tonal range, while keeping the midtones unchanged (value 1.00). We thresholded images to increase the contrast between neurons, myelinated axons, and neuropil and to ensure reproducibility. Thresholding was performed using a standardized function within ImageJ that maintains the image moment, or weighted average of the intensity of the image pixels, in the whole image: this separates the image into meaningful gray-level classes. All background correction occurred automatically at image acquisition, and was maintained through the use of identical levels of luminescence for all image acquisitions. Standardized functions within ImageJ performed all measurements.

To place our findings within the context of postnatal maturation and myelination of frontal cortical layer 1 in the human brain we studied myelination patterns throughout the neurotypical lifespan using the classical atlas of cortical myelination by Kaes [62]. Kaes processed tissue using Weigart’s myelin stain (Max Wolter’s variant) in thinly cut tissue (under 50 μm) and examined 12 regions of the cortex from both hemispheres across 45 human brains, ranging from 3 months to 97 years old. His study focused on the thickness of the layers, defined by their myelin level, during postnatal development [92]. Comparison of the density of myelin in adults in Kaes’ material and our processed sections allowed us to match analyzed ROIs between datasets based on their similarity. Among Kaes’ ROIs, the anterior and posterior frontal lobes included the anterior and lateral prefrontal cortices, while Kaes’ Gyrus Fornicatus ROI is the cingulate cortex, including the ACC, in line with Kaes’ descriptions [62] and previous assessments [92].

We scanned the illustrations of columns from 31 brains (3 months - 65 years old, mean age = 27 years old; 7 female) from the atlas (original publication) at high resolution and used the scans to estimate the plot profiles of acquired cortical columns in ImageJ. Kaes’ published atlas used in this study is out of copyright and the book is in the public domain. We estimated the average gray level (optical density) of layer 1 in these columns using the acquired plot profile.

Electron microscopy

We acquired high-resolution images using a scanning electron microscope (Zeiss Gemini 300 with STEM detector, Atlas 5 software) at magnifications ranging from 2000x to 50,000x as previously described [133].

Maps of processed tissue labeled with toluidine blue were used to identify layer 1 in EM-processed tissue. We sampled grids in an unbiased way, similar to that described above and as shown in Fig. 1. We acquired all images at 30 nm resolution and performed all analyses of these images in Image J. We directly circled all myelinated axon profiles and obtained the measurement of the angle, major and minor diameter, surface area, and perimeter of each axon.

Axon segment trajectory measurements were acquired directly from the outlined profiles of axons analyzed in individual electron micrographs. For each outlined axon profile we estimated the major and minor diameter, aspect ratio, and the angle trajectory of the major diameter. Previous studies have successfully used this method to describe changes in axon trajectory in autism [133], and these measurements have been shown to correlate with the local trajectory of myelinated axons using 3D-reconstruction [129]. We processed the raw measurements of the axon segment angles following a previously reported protocol [133]. Briefly, we aligned the peak angle for each section to 90 degrees to account for bias from image acquisition, and we calculated the standard deviation of the angles for each case. We used the proportion of elongated axons (axons with aspect ratio > 3) as a weighting factor to produce the final reported standard deviation measurement. Axons with aspect ratio < 3 appeared more circular, and the angles of these axons were not a reliable indicator of the axon trajectory. Axons with aspect ratios > 3 were more elongated, and produced reliable measurements.

Statistics and cross-validation

All values are reported as the mean ± standard deviation (SD). To ensure that the results of these studies had statistical power, we selected a sampling fraction and studied a tissue volume that resulted in an error of under 10%, as described [44, 47, 129, 131, 133]. Data was statistically analyzed using ANCOVAs in SPSS (IBM) in order to determine correlations with age, cause of death, PMI, gender, and other identifying characteristics of our sample. The Levene statistical test, which assesses the homogeneity of variances between studied groups, was used to assess differences in variability between groups. Results were considered statistically significant with a p-value under 0.05.

Many independent experimental and analytical methods, ranging from light and electron microscopy, toluidine blue labeling, and myelin staining were used to cross-validate the results and minimize the effects of any potential experimental bias. We used archival monkey tissue to further validate our findings in human tissue, ensuring that factors in human tissue analysis and acquisition, such as PMI or quality of fixation, did not bias our conclusions. Additionally, multiple researchers analyzed subsets of the same data.

Normative development of myelinated axons in layer 1

We examined functionally relevant subpopulations of myelinated axons in layer 1 in neurotypically developing children and adults and assessed the increase in myelination using two complementary approaches. First, we measured the surface area occupied by myelinated axons using area fraction measures in osmicated and Nissl toluidine blue-stained sections. Myelinated axons occupied little of the surface area of layer 1 in childhood (mean ± SD: 0.89 ± 0.29%); this increased significantly (p = 0.000) to 5.4 ± 0.44% in adults (Fig. 2a). We also calculated the angle trajectories of all myelinated axon profiles, and we evaluated the organization of the network within layer 1 by determining the variability in the trajectories of the axons in all cases. Variability in myelinated axon trajectory deviation was small in childhood, ranging from 3.6^o - 5.0^o (4.20 ± 0.56^o), and increased in adulthood (6.08 ± 2.60^o), potentially reflecting the increased myelination of more diverse pathways with age (Fig. 2b).

Analysis of the laminar density of myelin in layer 1 of the anterior and posterior prefrontal cortex and the cingulate cortex (gyrus fornicatus) from cases used in Kaes’ atlas was in line with our quantitative analysis at the light and electron microscope. We observed a gradual and significant increase in myelination of layer 1 with age: anterior and posterior frontal regions had a steeper rate of increase in myelination during development than the cingulate cortex and higher levels of myelin in adulthood [p(Frontal Lobes) = 0.005; p(Gyrus Fornicatus) = 0.041] (Fig. 2d).

Axon caliber is correlated both with the length of the pathway and with efficiency of signal transduction: thick axons participate mainly in long-range connections and transmit signals faster than thin axons, which are mainly found in short-range connections [58, 131]. Therefore, we assessed the sizes of myelinated axons in layer 1 in order to identify potential changes in the relative proportion of different pathways and overall changes in the efficiency of signaling at different developmental stages. We divided axons into three categories: thin axons (diameter < 0.84 μm), medium-caliber axons (0.84 μm < diameter < 1.18 μm) and thick axons (diameter > 1.18 μm), as described [129, 131, 133]. In neurotypically developing children, 62.3 ± 2.6% of myelinated axons in layer 1 were thin while 33.9 ± 3.9% were of medium caliber; few thick myelinated axons were present (Fig. 2f, h, j). There were no significant differences in these proportions between children and adults. In adults there was an age-related decline in the relative proportion of thin myelinated axons and increase in the relative proportion of medium-caliber myelinated axons in layer 1 (p = 0.041), potentially relating to changes in pruning or branching characteristics in the neuropil that may occur with normal aging.

Changes in the myelinated axons of layer 1 in autism

We found significant changes in the organization and structure of myelinated axons in layer 1 of both children and adults with autism when compared with neurotypical individuals. In order to ensure appropriate age-matching of subjects, we excluded unmatched older control adults from comparisons.

The density of myelinated axons increased with age in individuals with autism and controls (Fig. 2a, c) (p = 0.000; no difference between autism and control, p = 0.349). The variability in myelinated axon trajectory deviation was significantly larger in children with autism compared to neurotypical children (p = 0.033): this difference did not persist in adults, although there was a significant increase in myelinated axon trajectory deviation with age (p = 0.023) (Fig. 2e). This finding suggests that there is significant local circuit disorganization in individuals with autism beginning early in postnatal development. In children with autism, we found significantly increased variability in the proportion of myelinated thin axons (SD Autism = 12.5%, SD Control = 2.6%, p = 0.048) (Fig. 2g). In adults with autism, the proportion of myelinated thin axons in layer 1 was significantly lower than that in controls (mean Control = 73.6 ± 7.7%, mean Autism = 57.9 ± 4.1%, p = 0.034) (Fig. 2g), accompanied by a significant increase in the proportion of myelinated medium-caliber axons (mean Control = 22.5 ± 5.1%, mean Autism = 36.2 ± 1.8%, p = 0.011) (Fig. 2i). These changes were not associated with parallel alterations in the density of axons in adults with autism (p = 0.390).

Normative development of neurons in layer 1

We measured neuron density in layer 1 in neurotypically developing children and adults in Nissl-stained tissue. In order to ensure accuracy, tissue was oversampled to account for the low density and anisotropic distribution of cells in layer 1. Overall neuron density in layer 1 in adults was 7059 ± 1242 cells/mm³. Neuron density declined from childhood to adulthood (Fig. 4a, b), likely due to the increase in neuropil volume associated with branching and myelination during development. There was a trend towards an increase in neuron density in layer 1 with age in older adults (Fig. 4e), potentially relating to the changes in axon pruning with age described above. In children, we saw significant variability in the neuron density in layer 1, in line with reported heterogeneity in the maturation of the cortex [127]. Cells occupied 10.6 ± 3.8% of the area of layer 1 in children; this increased moderately to 12.1 ± 5.4% in adults (Fig. 4f).

Inhibitory neurons in layer 1 during neurotypical development

In order to further study this population of inhibitory neurons, we labeled tissue with the three calcium-binding proteins (PV, CB, and CR). The most prevalent inhibitory interneuron subtype in layer 1 in neurotypically developing individuals were CR+ interneurons (54–94% of labeled neurons) (Fig. 4l), which, in the upper cortical layers, primarily serve a disinhibitory role through selective targeting of CB+ interneurons [29, 86]. This is in line with previously reported findings in layer 1 in the adult human and non-human primate cortex, which has been shown to predominately contain CR neurons, including remnants of the Cajal-Retzius cell population [40, 81, 82]. There were also few calbindin-expressing neurons (mean = 194 ± 127 cells/mm³), which did not vary significantly with age. In children, CR neurons had a mean density of 1270 ± 475 cells/mm³, which increased in adults to 1437 ± 902 cells/mm³ (Fig. 4h). We separately analyzed younger (37–47 years old) and older adults (58–67 years old) to identify possible age-related changes in CR neuron density. In younger adults CR neuron density in layer 1 was 1009 ± 488 cells/mm³, while preliminary data from two older adults suggests an increase in CR neuron density (mean = 2508 ± 847 cells/mm³, p = 0.006). Quantification of additional cases with ages over 60 years would be necessary to confirm this change. Of note, the majority of neurons in layer 1 in adults were not labeled by calcium-binding proteins (69–94%) (Fig. 4m).

PV-immunolabeled neurons in layer 1 of children and adolescents

Our results show for the first time to the best of our knowledge that neurotypically developing children and adolescents have a population of PV neurons in layer 1 in LPFC (mean = 329 ± 194 neurons/mm³) (Fig. 4j). Previous studies have identified PV-expressing Cajal-Retzius and non-Cajal-Retzius neurons in layer 1 of prenatal and newborn humans [30, 121]: the neurons identified in this study may be members of these neuronal populations, which persist later in development than previously believed. Further studies are needed to identify the functional role of these PV-expressing neurons of layer 1, which may or may not serve the same strongly-inhibitory role of the PV-expressing neurons of the deeper cortical layers.

There was a significant age-related reduction in the density of PV interneurons in adults, with many cases having no detectable PV neurons in layer 1 (mean = 62 ± 22 neurons/mm³, p = 0.017) (Fig. 4j). Younger adults had a mean PV neuron density of 33 ± 39 cells/mm³, which increased to 148 ± 32 cells/mm³ in older adults, based on preliminary data.

No changes in neurons in layer 1 in autism

There was no significant difference between control and autism groups in overall neuron density in layer 1 in adults (mean Autism = 6939 ± 217 neurons/mm³; mean Control = 6602 ± 1186 cells/mm³, p = 0.453) (Fig. 4c). The decline in neuron density associated with age (Fig. 4d) and the percent surface area occupied by cells (Fig. 4g) were also similar between groups. Previous reports have identified as much as a 30% increase in neuronal density in the prefrontal cortex in autism [3, 16]; our findings suggest that these increases are not due to changes in layer 1.

We assessed the densities of the three interneuron subtypes in layer 1 of LPFC in individuals with autism and found no differences from controls [p(CR, Autism vs Control) = 0.470, p(PV, Autism vs Control) = 0.828, p(CB, Autism vs Control) = 0.711] (Fig. 4i, k, n). This suggests that the changes reported above in myelinated axon structure in autism are not due to changes in the cellular populations intrinsic to layer 1, but instead are due to changes in branching and maturation of local networks or incoming pathways.