Morphometric methods were applied to brains of 14 individuals with autism from 4 to 60 years of age, and 14 control subjects from 4 to 64 years of age. In the examined autistic cohort, there were 2.5 × more males than females (10 males and four females). The proportion between males and females was similar in the control group (nine males and five females).
Tissue preservation and morphometric methods
The postmortem interval (PMI), corresponding to the period between death and autopsy, ranged from 6 to 28 h in the control group (16.7 h on average; standard deviation [SD] 6.6 h) and from 8 to 50 h in the autistic group (21.9 h on average; SD 11.4 h). The difference in the PMI between the two groups was not significant (Table
2). Selected structures were examined in 11 right hemispheres and 3 left hemispheres in the autistic and control groups.
Table 2
Tissue samples, demographics of autistic and control subjects, brain weight, and changes during processing
A | 1 | M | 4 | 30 | 1,280 | R | 4,560 | 28 | 49 |
A | 2 | F | 5 | 13.2 | 1,390 | R | 1,568 | 28 | 52 |
A | 3 | M | 7 | 25 | 1,610 | R | 330 | 37 | 55 |
A | 4 | M | 8 | 22.2 | 1,570 | R | 196 | 36 | 45 |
A | 5 | F | 11 | 12.9 | 1,460 | L | 308 | 34 | 52 |
A | 6 | M | 13 | 8 | 1,470 | L | 75 | 33 | 39 |
A | 7 | F | 17 | 25 | 1,580 | L | 470 | 36 | 51 |
A | 8 | F | 21 | 50 | 1,108 | R | 136 | 35 | 43 |
A | 9 | M | 22 | 25 | 1,375 | R | 1,034 | 39 | 38 |
A | 10 | M | 23 | 14 | 1,610 | R | 533 | 45 | 60 |
A | 11 | M | 36 | 24 | 1,480 | R | 721 | 37 | 44 |
A | 12 | M | 52 | 27.6 | 1,625 | R | 1,650 | 39 | 40 |
A | 13 | M | 56 | 3.3 | 1,570 | R | 692 | 38 | 52 |
A | 14 | M | 60 | 26.5 | 1,210 | R | 398 | 39 | 38 |
Average |
21.9
|
1,453
| |
905
|
36
|
47
|
SD |
11.4
|
162
| |
1,160
|
4
|
7
|
p< (difference between autistic and control groups) |
ns
|
ns
| |
ns
|
ns
|
ns
|
C | 1 | F | 4 | 17 | 1,530 | R | 126 | 41 | 49 |
C | 2 | F | 4 | 21 | 1,222 | R | 233 | 30 | 43 |
C | 3 | M | 7 | 12 | 1,240 | R | 130 | 41 | 51 |
C | 4 | F | 8 | 20 | 1,222 | R | 650 | 36 | 51 |
C | 5 | M | 14 | 20 | 1,464 | R | 1,067 | 38 | 44 |
C | 6 | F | 15 | 9 | 1,250 | R | 372 | 41 | 49 |
C | 7 | F | 20 | 9 | 1,340 | R | 245 | 37 | 52 |
C | 8 | M | 23 | 6 | 1,520 | R | 95 | 45 | 41 |
C | 9 | M | 29 | 13 | 1,514 | R | 89 | 41 | 49 |
C | 10 | M | 32 | 24 | 1,364 | R | 460 | 37 | 42 |
C | 11 | M | 48 | 24 | 1,412 | L | 215 | 38 | 39 |
C | 12 | M | 51 | 18 | 1,450 | L | 1,819 | 20 | 26 |
C | 13 | M | 52 | 13 | 1,430 | L | 158 | 47 | 48 |
C | 14 | M | 64 | 28 | 1,250 | R | 52 | 37 | 51 |
Average |
16.7
|
1,372
| |
408
|
38
|
45
|
SD |
6.6
|
118
| |
491
|
7
|
7
|
The average weight of the brains in the autistic cohort (1,453 g) was not significantly different from that of the control group (1,372 g). The brain hemisphere with cerebellum and brainstem was fixed with 10% buffered formalin for an average of 408 days in the control group (range 52–1,819 days; SD 491 days). The average time of fixation in the autistic group was 905 days (range from 75 to 4,560 days, SD 1,160 days). The brains were dehydrated in a series of ascending concentrations of ethyl alcohol. The average time of dehydration was 36 and 38 days in the autistic and control group, respectively. Dehydration was associated with reduction of brain hemisphere weight by 47% (SD 7%) on average in the autistic group and by 45% (SD 7%) in the control group. Brain hemispheres were embedded in 8% celloidin [
41]. Serial 200-μm-thick sections were stained with cresyl violet and mounted with Acrytol.
Autistic and control subjects were very precisely age matched. There was no significant difference between the autistic and the control group in the PMI, autopsy brain weight, fixation time, dehydration time, or brain weight loss during dehydration. Therefore tissue preservation and processing impact on age-related changes is expected to be comparable in both cohorts and to affect estimates of neuron volume in a similar way. Moreover, robust statistical methods were applied to minimize any distortion of results due to outlying cases.
The neuropathological examination was carried out (a) to identify the type, distribution, and severity of qualitative developmental changes and (b) to eliminate brain samples affected by either pathology not related to autism, changes associated with mechanisms of death, or postmortem autolytic tissue degradation. On average, 120 cresyl violet–stained serial hemispheric sections were examined per case in blinded fashion with regard to diagnosis. The results of our study of the developmental abnormalities in these cohorts were previously reported [
42]. In summary, the neuropathological evaluation revealed a broad spectrum of focal developmental alterations in the 13 examined autistic brains, including (a) subependymal nodular dysplasia, (b) subcortical and periventricular heterotopias, and (c) dysplastic changes in the neocortex, archicortex, dentate gyrus, Ammon’s horn, and cerebellum. The pathology detected in 92% of the autistic brains reflects focal modifications of neurogenesis, migration, and alterations of the cytoarchitecture. The absence of these changes in the control brains indicated that the detected alterations were autism-associated.
The morphometric measurements of the material being analyzed were performed without knowledge of the subject’s age, gender, clinical diagnosis, or neuropathological status. The volume of neurons was estimated in 16 brain structures or neuronal populations, including the amygdala, entorhinal cortex, Ammon’s horn, claustrum, caudate nucleus, putamen, globus pallidus, nucleus accumbens, thalamus and two parts (magnocellular and parvocellular) of the lateral geniculate body (LGB), substantia nigra, the magnocellular basal complex (MBC), including the acetylcholinergic system; Purkinje cells and dentate nucleus in the cerebellum, and the inferior olive in the brainstem.
Neuronal morphometry was performed using a workstation consisting of an Axiophot II (Carl Zeiss, Goettingen, Germany) light microscope with Plan Apo objectives 1.25× (numerical aperture, N.A., 0.15); 2.5× (N.A. 0.075), 40× (N.A. 0.75), and 63× (N.A. 0.9), a specimen stage with a three-axis, computer-controlled stepping motor system (Ludl Electronics; Hawthorne, NY, USA), a CCD color video camera (CX9000 MicroBrightfield Bioscience, Inc., Williston, VT, USA), and stereology software (Stereo Investigator and Nucleator, MicroBrightField Bioscience, Inc.). The volume of neurons was estimated with the nucleator method [
43] using the MicroBrightfield software (Nucleator).
Sampling scheme
The parameters and procedures that were applied to estimate the volume of the neuronal soma in the 16 brain structures are presented in (Additional file
1: Table S1). An optical fractionator systemic random sampling scheme was applied (Stereo Investigator, MicroBrightField Inc.). At least four equidistant serial sections [
44] were used for neuronal measures in the anatomical subdivisions of the amygdala (total 12 sections), substantia nigra (total 9 sections), Ammon’s horn sectors CA1–CA4) (total 14 sections), claustrum (total 9 sections), and magnocellular basal complex (total 9 sections). In the large striatal subdivisions (caudate nucleus, putamen, and globus pallidus) with a relatively uniform cytoarchitecture, four sections were examined. Examination of the relatively large number of neurons (213/ROI/case, on average) resulted in a less than 0.01 Scheaffer coefficient of error (CE). The coefficient of error is essentially a measure of the uncertainty of the population estimate. The number of neurons measured in individual region has been controlled to reach a CE close to the required 0.05 (MicroBrightField, Inc.). Scheaffer coefficient of error decreases with an increasing number of measured neurons, counting frames per section and number of sections. The grid size and the virtual counting space were designed for each brain structure individually to adjust to the size and shape of the region of interest and to reduce the variation, as reflected in the SD and the CE. Very heterogeneous structures were oversampled (more sections, denser grid, more counting frames and more neurons measured) to achieve acceptable precision. The mean number of virtual counting spaces ranged from 59 for very small magnocellular basal complex (CE, 0.005), to 664 for Purkinje cells, with irregular distribution and low numerical density (CE, 0.002).
The results of the examination of the cytoarchitectonic subdivisions of the amygdala, entorhinal cortex, Ammon’s horn, and substantia nigra are presented as a mean value for the entire structure.
Evaluation of measurement reliability
As random differences in the axes upon which measurements are taken are introduced by the Stereo Investigator protocol to avoid a measurement bias, per-neuron reliability cannot be meaningfully computed for each observer (or among observers). Consistency of each pathologist’s measurements was assessed by comparing repeated measurements of structures by the same rater. Interrater reliability was assessed as an intraclass correlation coefficient using data from a sample of five structures, each of which was examined by all four observers (raters). The raters were considered to be a random sample of all possible raters, so a random effects model was employed. This corresponds to ICC(2,1) proposed by Shrout and Fleiss [
45], though the computed ICC is the same if the raters are assumed to comprise all possible raters, that is, if a mixed model assumption is employed, equivalent to ICC(3,1). As observers’ measurements are not in practice pooled, the reliability of their absolute agreement was assessed (rather than the more liberal standard of their consistency).
The evaluation of intrarater consistency revealed that differences in observers’ repeated measurements of a single structure ranged from 0.1% to 6.8%. The mean absolute value of the difference between the two ratings was 2.3%. Random differences in the axes upon which measurements were taken, introduced by the Stereo Investigator protocol to avoid a measurement bias, constitute a small component of these differences. Interrater reliability for five structures, each measured by all four observers, was very strong, with an intraclass correlation coefficient of .961 (95% C.I. .835, .995).
Anatomical boundaries of examined brain structures and their cytoarchitectonic subdivisions
In this study, five components of the basal ganglia, including the telencephalic striatum (the caudate nucleus, putamen, nucleus accumbens), the diencephalic pallidum (external and internal globus pallidus), the mesencephalic substantia nigra, and the magnocellular basal complex, including the nucleus basalis of Meynert, were examined [
46]. They are an integral part of the cortico-subcortical circuits involved in the programming and execution of movements, and they provide the necessary timing (ramp signals) for smooth guided movements. Abnormal movements and changes in muscle tone (hyper- or hypotonia) are prominent signs of basal ganglia disorders [
46,
47].
The caudate nucleus and putamen represent one cellular mass (striatum) partially separated by a thick internal capsule. The putamen extends from the external capsule to the external medullary lamina of the pallidum. In the ventral striatum, the nucleus accumbens was examined. The n. accumbens is located underneath the rostroventral aspect of the internal capsule. The border between the nucleus accumbens and dorsal striatum (caudate nucleus and putamen) was arbitrarily defined by a line perpendicular to the midline of the internal capsule at its lower end [
48]. The study concentrated on the striatal small neurons, whereas a few sparse large neurons [
49] were not included in this neuronal size analysis. The globus pallidus is separated from the putamen by the external medullary lamina, whereas the medial medullary lamina divides the globus pallidus into the external and internal pallidum [
46]. Neurons in both parts of the pallidum were examined.
In the substantia nigra, the pars compacta was distinguished from the underlying pars reticulata by the accumulation of melanin in neuronal soma, and the nucleator examination was limited to melanin-positive neurons [
50]. To identify the three major subdivisions of the pars compacta, Fearnly and Lees’ [
51] classification of neuronal nuclei into the dorsal and ventral tier was used. The bundle of the putaminonigral pathway was used as a marker separating the pars lateralis from the ventral tier.
In a series of CV-stained sections, four subdivisions of the magnocellular basal complex, including the most prominent nucleus basalis of Meynert [
52], were identified. Four subdivisions (Ch1-4) were delineated using Vogels [
53] mapping. The small Ch1 magnocellular group is located in the medial septum, whereas a substantial population of magnocellular neurons is found in the Ch2 nucleus of the vertical limb of the diagonal band of Broca. Ch2 neurons are continuous with a sparser population of large neurons that are located along the medio-ventral surface of the hemisphere in the nucleus of the horizontal limb of the diagonal band (Ch3). At the preoptic area, the magnocellular group of the basal nucleus of Meynert (Ch4 group) [
54] was examined. In the amygdala, the boundaries of the lateral, basal, accessory basal and central nuclei were identified on CV-stained sections using cytoarchitectonic criteria described in detail by Schumann and Amaral [
22].
The mediodorsal surface of the thalamus is limited by the wall of the lateral ventricle, whereas the lateral and dorsal border of the thalamus is accentuated by the reticular thalamic nucleus that separates the thalamus from the posterior limb of the internal capsule. To obtain global characteristics of the thalamus, which consists of more than 50 nuclei with often poorly defined borders, the entire thalamus was randomly sampled using equidistant sections.
A thin ribbon of claustrum gray matter is separated from the insula by the extreme capsule and from the putamen by the external capsule [
49]. Two major anatomical subdivisions including the claustrum (compact insular claustrum and ventral claustrum), and the prepiriform claustrum were examined. Poorly defined and diffuse subregions such as preamygdalar claustrum [
55] were excluded from this study.
Purkinje cells were identified on the basis of their distribution along the border between the molecular and granular layer, their large soma size, and their apical dendrite spatial orientation. The study of deep cerebellar nuclei was limited to neurons within the dentate nucleus to eliminate random contributions of neurons of the small and medially located fastigial, globose, and emboliform nuclei.
The study of the inferior olive, divided into a principal nucleus, dorsal and medial accessory olives, dorsal cap of Kooy, ventrolateral outgrowth, and nucleus
β[
56], was limited to the largest principal nucleus.
The entorhinal cortex (EC, Brodmann area 28) spreads over the gyrus ambiens and the anterior portion of the hippocampal gyrus. The anterior limit of the EC starts about 5 mm rostrally to the amygdaloid complex. The EC ends anterior to the rostral pole of the lateral geniculate body. Rostrally, the medial portion of the EC borders with the preamygdaloid cortex in the area marked by the sulcus semiannularis. Caudally, the median portion of the EC borders with the subicular complex. The lateral border is distinguished as the border with Brodmann area 35 that lacks a distinct layer IV [
57]. The transentorhinal zone [
58] was not included in this study.
Neurons were measured within four layers of the EC. Layer II consists of islands of relatively large modified pyramidal and stellate neurons separated by an acellular gap from layer III [
58]. A broad layer III is built up of medium-sized pyramidal neurons separated from layer V by an acellular lamina dissecans labeled as layer IV. Layer V consists of the sublayer Va with large pyramidal neurons, Vb with smaller and more loosely arranged neurons, and sublayer Vc with relatively few dispersed neurons. Layer VI is built up of the smallest neurons, and the border of this layer in the rostral portion is characterized by a decreasing gradient of neuronal density, whereas in the caudal portion the border is sharp [
57].
Ammon’s horn neurons, in the pyramidal layer sectors CA1–4, were examined. The CA1 sector pyramidal neuronal layer extends from the subiculum to the CA2 sector as a thick band of relatively small pyramidal neurons. The border between the CA1/CA2 sector and CA2/CA3 sector is accentuated by a compact arrangement of large pyramidal neurons in CA2. Moreover, the pyramidal layer is narrowest in CA2. The size of neurons in the CA3 and 4 is similar, but parallel arrangement of neurons in the CA3 sector distinguishes CA3 from the hilus neurons (identified also as CA4 sector) [
59,
60]. The mapping of sectors on serial sections representing the entire rostro-caudal extension was based on the Amaral and Insausti [
57] classification, but with the CA4 identification as a separate subdivision.
A six-layered lateral geniculate nucleus (LGN) borders at the base (hilus) [
61] with the transverse fissure, whereas the posterior limb of the internal capsule separates the LGN from the lateral wall of the thalamus. The major portion of the LGN is present on the level of the border between the head and body of the hippocampus. Neurons were measured separately in two major ventral laminae (1 and 2; magnocellular LGB) and in the parvocellular laminae (3 to 6; parvocellular LGB), with neurons approximately half the size of neurons in the magnocellular laminae.
Tissue and medical records were handled in accordance with the NIH Guide for the use of human tissue. The research project and protocols were approved by the Institutional Review Board of the New York State Institute for Basic Research in Developmental Disabilities (IBR). The tissue was obtained from the Harvard Brain Tissue Resource Center, the Brain and Tissue Bank for Developmental Disorders of the National Institute of Child Health and Human Development at the University of Maryland, the Brain Bank at IBR, Staten Island, NY, and the Mount Sinai Medical School, NY. Tissue samples were coded with the number assigned by the brain banks, and this ID was the only identifier of the tissue, MRI scans, and clinical records. The Autism Tissue Program (Autism Speaks, Princeton, NJ) provided access to brain tissue samples and to the database with coded and anonymous characteristics of the autistic subjects, including the subject’s age at time of death, clinical diagnosis, results of application of the ADI-R, cause of death, PMI, and fixation time.
Statistical analysis
This study analyzed the effect that autism has on neuronal soma, with age-matched controls as a comparison group. Data have been post-stratified to adjust for the differing numbers of neurons sampled per individual so as to weight each individual equally. Any data points lying more than 1.5 times the interquartile range below the 25th percentile or an equal amount above the 75th percentile for each structure or subdivision were considered outliers and omitted from analyses. Removed defective records accounted for 0.3% of all records.
Analyses were carried out separately for three groups of cases: 4 to 8 years, 11 to 23 years, and 36 to 60 years of age. Three groups of controls were selected to provide age-matching at the group level. The age ranges for control groups were 4 to 8 years, 14 to 23 years, and 29 to 64 years of age. Comparisons were carried out across 16 regions. Adjustment for multiple comparisons was made using the Benjamini-Hochberg method [
62] to maintain a False Discovery Rate (FDR) of 0.05. Accordingly, p values of 0.027 or less were considered statistically significant. Analyses were conducted using versions 11.1 and 12.1 of the Stata statistical package [
63,
64].
Adjusted mean volumes and standard errors were computed using the multilevel sampling survey data procedures of the Stata package. Preliminary analyses regressed neuronal size, normalized by expressing each neuron’s size as a proportion of the size of the mean control neuronal size for that structure, on PMI (in hours), fixation time (in days), brain weight (in grams), brain weight loss during processing and dehydration (as a percentage), duration of dehydration (in days), autism status and log age (in years). This analysis resulted in the detection of significant effects for each of these factors. Age and these potential confounders of autism’s effect on neuronal size accounted collectively for 1.09% of the variance in size. Autism status, when entered into the model, raised the explained variance to 1.91%. As autism status univariately explained 1.45% of variance, all potential confounders considered together diminished the variance explained by the autistic status by approximately 43%, while autism remained by a large margin the strongest predictor in the model.
Further analyses of the potential effects of confounders, including analyses of the effect of a history of seizures and of sudden and unexplained death in patients with known epilepsy, were performed using the Stata svy:regress command, with potential confounders and autistic status (or age group) entered as predictors of neuronal or soma volume.
The statistical significance of the differences in mean neuronal volumes between autistic and control brains and between pairs of age ranges was computed in general linear mixed models adjusted for autocorrelation of neurons within each brain using the xtmixed procedure in version 12 of the Stata statistical package. This method takes into account the variance between neurons while assessing the differences between brains of autistic and control individuals. Exploratory analyses computing Generalized Estimating Equations (GEEs) using the Stata xtgee procedure yielded nearly identical results. Analyses were controlled for post-mortem interval (PMI, log-transformed), days of dehydration, and weight loss. Fixation time was also examined as a potential confounder but proved both redundant and highly correlated with the other control variables and often resulted in non-converging models. In a small number of models PMI was also removed to allow stable estimation. With the sole exception of the dentate nucleus in older subjects, controlling for seizures or sudden unexplained death in epilepsy (SUDEP) in analyses did not affect the significance of the difference in cell soma volume between autistic and control subjects.