Introduction
Autism is a neurodevelopmental disorder that is characterized by changes in neural communication that affect diverse sensory-motor processes such as attention and social interaction [
2,
36,
66]. Changes in frontal networks, including increased short-range and decreased long-range communication as well as changes in synchronization between cortical areas during tasks, have been described in individuals with autism [
12,
24,
31,
61,
63,
110,
124]. Anatomical studies have identified changes in the distribution and density of neurons belonging to multiple subtypes within frontal cortices [
1,
52,
130] and myelinated axons below the frontal lobes in autism [
129,
130,
133] that likely underlie these findings. However, little is known about the development of cortical pathology and the disruption of laminar-specific excitatory pathways and inhibitory circuits in the affected frontal cortical networks.
The development of cortical network pathology in the lateral prefrontal cortex (LPFC) is of particular interest because LPFC is involved in attention and the cognitive processes that are affected in autism and undergoes prolonged postnatal development and maturation [
13,
23,
52,
71,
115,
116,
129‐
131]. Layer 1 plays a significant role in the prenatal patterning of the cortex and postnatally is a chief recipient of feedback and neuromodulatory pathways in LPFC, making it an ideal candidate for the study of the development of laminar-specific pathway pathology in autism.
Layer 1 contains a distinctive set of morphologically diverse local circuit neurons along with varied populations of astrocytes, oligodendrocytes, and microglia [
11,
40,
81,
97,
102,
125,
127]. Feedback connections from cortical areas as well as the thalamus, amygdala, and neuromodulatory systems target layer 1 [
5,
7,
10,
15,
48,
60,
87,
112,
128], where they interact with local excitatory and inhibitory circuits and affect spatiotemporal characteristics of cortical activity patterns [
17,
26,
33]. In prenatal development, the intrinsic Cajal-Retzius cells of layer 1 secrete reelin to direct the development of the distinct cortical layers [
39,
53,
93]. Studies of the development of layer 1 have examined mostly the pre- and postnatal maturation of Cajal-Retzius neurons and few other cell types [
78,
82,
88,
90,
102,
114,
126,
127]. However, we know little about the postnatal changes in the diverse cellular populations of layer 1 and their relationship with the maturation of the pathways that terminate there, which serve to transition this layer from a developmental mediator to a processor of feedback input. Importantly, changes in the expression of factors that determine the maturation and activity of cortical networks have been described in layer 1 neurons in LPFC in neurodevelopmental connectivity disorders such as schizophrenia [
104] and autism [
115]. The effects of these disruptions on the cellular organization and axonal networks within layer 1 in childhood and adulthood are unknown.
To begin to address these issues we used post-mortem brain tissue from individuals with and without autism (ages 3–67 years) to systematically and quantitatively examine the postnatal development of excitatory and inhibitory circuits in layer 1 of the human LPFC. We cross-validated our findings through comparisons with neighboring anterior cingulate cortices (ACC) of neurotypical individuals and optimally-fixed non-human primate tissue. We provide evidence for significant changes in the density of neurons and maturation of pathways in typical postnatal development from childhood through late adulthood. We notably found that PV-immunoreactive neurons, which have previously only been shown in layer 1 of neonates, persist in layer 1 through late adolescence, likely influencing the signaling dynamics of the inhibitory neurons in layer 1. Comparison of the structure of layer 1 in typically developing individuals with that of individuals with autism revealed significant disruption in the developmental trajectory and structure of pathways, but not neurons, in children and adults with autism. Pathology included increased variability in axon orientation and changes in the size of myelinated axons. These findings suggest that persistent disruption of feedback pathways in LPFC in children and adults may underlie excitatory-inhibitory imbalance, cortical network disorganization, and atypical focusing and shifting of attention in autism.
Discussion
We present evidence of postnatal changes in the balance of excitation-inhibition in the maturing prefrontal cortex throughout typical development and in autism, using a large cohort of human subjects at a variety of postnatal ages. Our findings reveal specific changes in the structure of pathways and cellular populations within layer 1 of the LPFC through typical development. We also present evidence suggesting that atypical, age-associated changes in the organization and relationship between pathways and cellular populations in layer 1 of the LPFC may underlie the dysfunctional balance of excitation-inhibition in the maturing prefrontal cortex in autism.
In typical postnatal development, the density of myelinated axons in layer 1 of the prefrontal cortex increased with age, in line with previous studies on the maturation of white matter pathways [
74,
89]. Specifically, in LPFC, the relative proportion of thin myelinated axons in layer 1 of adults was significantly higher than what has been previously described in the white matter, where thin axons represented 36% of the myelinated axon population [
129,
131,
133]. This supports our understanding that axons enter the cortex and quickly branch, resulting in an overrepresentation of thin axons in superficial cortical layers. The increased density of thin axons in layer 1 was accompanied by an increase in the volume of neuropil, including an increase in unmyelinated axon density, estimated by the gray level index, and a decrease in neuron density in adults. We noted both an increase in neuron density and a decrease in the proportion of thin axons in two older adults (ages 58 and 67), suggesting that reductive changes in neuropil structure, potentially due to pruning, may be associated with normal aging.
Surprisingly, a significant proportion of neurons in layer 1 did not express the calcium-binding proteins that are typically used as markers of inhibitory interneurons in the cortex of primates, even though close to 90% of neurons in layer 1 are GABAergic [
41,
42,
54‐
56]. Similar staining patterns in human and optimally-fixed rhesus macaque brain tissue corroborated this observation. These findings suggest that there are potential differences in the origin and physiological characteristics of interneurons in layer 1 when compared to other cortical layers, which may influence the regulation of GABAergic signaling in layer 1 [
28,
102]. The emerging complexity of this layer in our study is in line with recent work that has led to the identification of novel neuronal types in layer 1 of humans [
11]. The differences that we have identified in the expression of calcium-binding proteins between childhood and adulthood further suggest that the cellular composition of layer 1 changes significantly during postnatal development. We detected a significant population of PV-immunoreactive neurons in LPFC layer 1 of neurotypical children and adolescents, previously seen only pre- and peri-natally [
30,
121]. The density of layer 1 PV-immunoreactive neurons decreased with age to negligible numbers in adults, suggesting developmental changes in the calcium dynamics, and therefore in synaptic speed and strength, of inhibitory neurons in layer 1 during development. This could have an effect on the processing of incoming feedback or neuromodulatory signals in childhood and adulthood, leading to differences in the balance of excitation-inhibition in the cortex.
We found significant evidence of disorganization of axon networks within layer 1 in individuals with autism. There was increased heterogeneity in the trajectories and the proportion of myelinated axons that were thin in children with autism compared with controls. While there was an increase in the ratio of thin myelinated axons in layer 1 in neurotypical adults, thin axon ratios did not change with age in autism, leading to a significantly lower proportion of thin myelinated axons in adults with autism compared to the control group. This suggests that layer 1 thin axon networks in adults with autism remained on average at similar levels as immature, less myelinated networks in childhood, in line with previous work that has shown increased axon branching at the border of the gray and white matter below interlinked medial and lateral prefrontal cortices in these and other cases [
35,
111,
129‐
131,
133]. Increased axon branching and a decrease in the proportion of thin myelinated axons may be due to an increase in the density of unmyelinated axons, which we could not directly assess in this study. Future studies would need to assess changes in the qualities of the neuropil through development and in autism to confirm this hypothesis.
Alterations in axon structure may have broader implications for the efficacy of signal transmission in frontal cortical networks of individuals with autism. In particular, changes in axon caliber and the relative ratios of myelinated axon sizes directly influence the strength and persistence of the action potential generated by projection neurons, as well as their firing rate [
58,
96], which play a key role in sustained activity during working memory and attentional tasks in LPFC [
22,
65,
95,
99]. These, in turn, can underlie changes in the spread of activity and oscillations [
12,
59,
67,
68,
73], and may be linked with changes in the expression of neuronal ion channels and regulatory proteins in individuals with autism [
25,
57,
94,
98]. Genes associated with autism, such as UBE3B and ZNF18, are transcriptionally regulated by membrane depolarization and are involved in experience-dependent learning and synaptic plasticity [
34]. These processes are also regulated by reelin in the cortex, and especially layer 1, throughout the lifespan in primates [
72,
83]. Importantly, changes in axons in children and adults in autism were not accompanied by alterations in the overall density of neurons nor in the density of labeled inhibitory interneuron subclasses in layer 1 of LPFC, suggesting that these changes are specifically isolated to axon networks.
Previous examination of LPFC has revealed moderate changes in the expression of reelin in layer 1 of some individuals with autism [
115]. Reelin, which is distributed throughout the soma and dendrites of most neurons in layer 1 [
83], is secreted by Cajal-Retzius cells during prenatal development. Reelin has also been implicated in the regulation of synaptic plasticity and learning in the postnatal brain and is involved in the pathogenesis of neuropsychiatric disorders (reviewed in [
37‐
39,
72]). Importantly, a study of the human temporal lobe found no change in the density of reelin-expressing neurons in autism: consistent with the findings reported here, this study further identified no change in neuron density in layer 1 in autism [
14]. Taken together, our findings and previous reports suggest that reduction in the expression of reelin may be due to changes in protein processing or intracellular distribution within the cell population of layer 1 that can contribute to cortical patterning defects and postnatal changes in network structure in autism.
Other protein factors with altered expression in LPFC layer 1 in autism include chemokine ligand 14 (CXCL14) and neuron-derived neurotrophic factor (NDNF). CXCL14, an inflammatory cytokine, is involved in the regulation of myelination; reduction in the expression of this cytokine results in a reduction in myelination [
4,
119], which is consistent with our findings of reduced myelination and increased branching in prefrontal cortices of adults with autism [
129,
131,
133]. NDNF is mainly expressed by Cajal-Retzius cells of layer 1, and promotes the growth and development of neuronal cell bodies and neurite outgrowth [
70,
106]. Disruption of any of these factors may adversely impact the development of the cortical column in LPFC, and could specifically underlie the atypical organization of circuits within layer 1 of LPFC in autism.
Placing our findings within the context of previous studies that have reported neuropathological and molecular changes in prefrontal cortices in autism offers additional clues about mechanisms that may underlie specific disruption of LPFC networks. Major feedback pathways that target layer 1 of LPFC and may contribute to the observed axon disorganization include pathways from high-order thalamic nuclei [
60,
128] and, to a lesser extent, the amygdala [
48], regions that are consistently affected in autism [
8,
9,
51,
76,
107‐
109,
117,
118,
120]. Based on extensive experiments on a large cohort of subjects we have proposed that disruption of networks in autism depends on the time of the insult during prenatal development of cortical pathways [
129]. A general, consistent theme that emerges from our findings is that feedback, short-range pathways from the deep layers of limbic cortices that target superficial layers of eulaminate cortices, which develop relatively early but mature late postnatally, are susceptible to disruption and are affected in autism [
45,
129‐
131,
133]. We previously reported that in the neighboring ACC there is an increase in the expression of GAP-43 [
129], a growth axon protein that is antagonistic to myelin basic protein [
64] and promotes branching and shedding of myelin of axons. The ACC in primates is a major contributor of robust feedback pathways that terminate in the superficial layers of LPFC, including layer 1 [
18,
20,
69,
85,
91,
131]. It is therefore conceivable that the observed axon pathology in layer 1 of LPFC may be due to disruption in this short-range feedback network linking ACC and LPFC [
45]. This hypothesis is also supported by the oft reported over-connectivity of local frontal networks in autism [
24,
113].
In conclusion, we systematically examined layer 1 of LPFC in individuals with and without autism at high resolution. We described the typical postnatal development and organization of axon circuits and local interneurons. Study of excitatory and inhibitory circuit components in parallel provided a novel framework that facilitated identification of pathological changes within cortical networks in autism. We found significant changes in the structure and organization of myelinated axons in LPFC layer 1 in individuals with autism, with important implications for the balance of excitation-inhibition and local cortical information processing. Our findings highlight feedback pathways in LPFC as an especially vulnerable node that underlies autism pathophysiology. Finally, our synthesis of the new findings with previous studies provide important clues that can help link the atypical development of frontal networks in autism with key molecular mechanisms and factors, whose interactions during development will need to be elucidated in future studies.