Stereology of cerebral cortex after traumatic brain injury matched to the Glasgow Outcome Score

Abstract

Magnetic resonance imaging provides evidence for loss of both white and grey matter, in terms of tissue volume, from the cerebral hemispheres after traumatic brain injury. However, quantitative histopathological data are lacking. From the archive of the Department of Neuropathology at Glasgow, the cerebral cortex of 48 patients was investigated using stereology. Patients had survived 3 months after traumatic brain injury and were classified using the Glasgow Outcome Scale as follows: moderately disabled (n = 13), severely disabled (n = 12) and vegetative state (n = 12); and controls. Some patients from the archive were diagnosed with diffuse axonal injury post-mortem. Comparisons of changes in cortical neuron population across Glasgow Outcome Scale groups between diffuse axonal injury and non-diffuse axonal injury patients were undertaken using effect size analyses. The hypotheses tested were that (i) thinning of the cerebral cortex occurred after traumatic brain injury; (ii) changes in thickness of cortical layers in Brodmann areas 11, 10, 24a and 4 differed; and (iii) different changes occurred for neuronal number, their size and nearest neighbour index across Glasgow Outcome Scale groups. There was a greater loss of large pyramidal and large non-pyramidal neurons with a more severe score on the Glasgow Outcome Scale from all four cortical regions, with the greatest loss of neurons from the prefrontal cortex of patients with diffuse axonal injury. There were differences in the changes of number of medium and small pyramidal and non-pyramidal neurons between different cortical regions, and between patients with and without diffuse axonal injury. Generally, a decrease in the somatic diameter of pyramidal and non-pyramidal neurons was associated with a more severe clinical outcome. However, in the motor cortex a more severe Glasgow Outcome Scale was associated with an increased diameter of medium pyramidal neurons and small non-pyramidal cells. Pyramidal and non-pyramidal neurons did not follow a Poisson distribution within the neuropil of control patients. Pyramidal neurons were usually scattered while medium and small non-pyramidal neurons were clustered. An increased spacing between remaining neurons usually occurred across Glasgow Outcome Scale groups. It is concluded that loss of neurons resulted in reduced executive and integrative capability in patients after traumatic head injury.

cortical neurons, head injury, stereology

Introduction

A major consequence of traumatic brain injury in humans may be widespread injury to nerve fibres (axons) within the brain (Adams et al., 1980), and surviving patients frequently experience disability (Ghajar, 2000; Scheid et al., 2006); for example, approximately 40 out of 100 000 patients within the USA (Centers for Disease Control and Prevention, 2006), 40% of hospitalized traumatic brain injury patients in Australia (ABS, 2008) and 39.2% of hospitalized traumatic brain injury patients in the European Union (Maegele et al., 2007).

Of patients admitted to Accident and Emergency departments after head trauma in the UK, 54% experience behavioural, 39% intellectual and 29% locomotor disability at 5 years after injury (Evans et al., 2003). Changes in the lifestyle and care of these patients are a significant concern for society with an incidence of post-traumatic disability 1 year after traumatic brain injury in 120 out of 100 000 patients each year (Thornhill et al., 2000).

Head injury results in disorders of memory, attention, executive functions, slowing of information processing and adverse changes in personality and behaviour in >70% of traumatic brain injury patients (Evans et al., 2003; Hawley, 2003; Scheid et al., 2006; Kato et al., 2007). MRI provides evidence for a decrease in volume of grey matter at 12 months (Gale et al., 2005; Bendlin et al., 2008) and 4.6–5 years survival (Cohen et al., 2007; Fujiwara et al., 2008), and also of white matter in children and adults after severe traumatic brain injury (Tomaiuolo et al., 2004, 2005; Tasker, 2006; Yuan et al., 2007 Kennedy et al., 2009). Diffusion tensor imaging (DTI) provides evidence for a reduced anisotropy within a small number of tracts after mild traumatic brain injury (Kraus et al., 2007; Lipton et al., 2008) and a greater number after moderate or severe traumatic brain injury at 6 months (Tomaiuolo et al., 2005; Kumar et al., 2009), and between 3 and 8 years survival (Kraus et al., 2007; Lipton et al., 2008). In addition, fluid-attenuated inversion recovery weighted MRI has allowed visualization of white matter damage in the 30% of patients that experience a decline in cognitive function at 4.5–5 years after moderate to severe traumatic brain injury (Hofman et al., 2002; Ruttan et al., 2008; Till et al., 2008). Lastly, 2-[¹⁸F]Fluoro-2-deoxy-d-glucose positron emission tomography has suggested that post-traumatic glucose hypometabolism occurs in the injured brain, among other sites, at the anterior cingulate and medial frontal gyri of patients with diffuse axonal injury (DAI) (Nakayama et al., 2006; Kato et al., 2007).

In the animal experimental literature there are limited data for long-term response after traumatic brain injury. However, a progressive loss of myelinated axons from the fimbria, external capsule, thalamus and cerebral cortex of rats up to 1 year after fluid percussion injury (Bramlett and Dietrich, 2002; Rodriguez-Paez et al., 2005), and changes in brain volume, have been reported (Sutton et al., 1993; Smith et al., 1997).

Importantly, however, evidence for changes in volume of cortical grey matter and any correlation with changes in number or size of neurons in patients is lacking. The present study tested whether histopathological evidence could be obtained for differential loss of pyramidal and non-pyramidal neurons within four functionally associated regions of human cerebral cortex important in cognitive and executive functions, in emotional responses and control of voluntary motor function, in patients within and between different Glasgow Outcome Score (GOS) groups. More specifically, the hypotheses tested are that there is, firstly; thinning of the cortical ribbon in dorsolateral [Brodmann area (BA) 10] and ventromedial (BA 11) prefrontal, anterior cingulate (BA 24a) and motor (BA 4) cortices in patients with differing GOS (Jennett and Bond, 1975). Secondly, that changes in the number of small-, medium- and large-sized pyramidal and non-pyramidal neurons would occur in patients within different GOS outcome groups. Thirdly, that there is a change in the size of neurons, which varies between different cortical layers. Finally, that said changes may be correlated with a concurrent change in the nearest neighbour indices (NNI) of different cell types within cerebral cortical layers.

Materials and methods

Patients and clinical inclusion criteria

Material was obtained from the tissue archive in the Department of Neuropathology, Southern General Hospital, from cases that had undergone a full neuropathological examination, and had been assessed by the GOS (Adams et al., 2001; Jennett et al., 2001). This study was approved by the Research Ethics Committee of the Southern General Hospital.

After the gathering, within the three GOS groups [moderately disabled (n = 13); severely disabled (n = 12); and vegetative state (n = 12)], of raw data for stereology, patients were separated into those diagnosed with DAI and those without DAI (referred to in the text as non-DAI). These included patients diagnosed with either subarachnoid or intracerebral haemorrhage, raised intracranial pressure or diffuse ischaemic brain damage and the values obtained across that group as a whole compared with those of patients with DAI.

The number of patients within each GOS group, their sex, age at injury, range of ages in each outcome group, means and range of survivals, cause of injury and lucidity (Reilly et al., 1975) are summarized in Table 1. Blocks containing part of the dorsolateral prefrontal (BA 10), ventromedial prefrontal (BA 11), anterior cingulate (BA 24a) and motor cortices (BA 4) at the dorsum of the frontal precentral gyrus and anterior side of the central sulcus were available from all patients in this study.

Table 1

Summary of the number of male and female patients within each outcome group (moderately disabled, severely disabled and vegetative state)

	Controls	Moderately disabled	Severely disabled	Vegetative state
Number Male:female	11:0	10:3	10:2	11:1
Mean age at injury (years)		38.9	50.2	44.8
Range of ages (years)		8–65	20–76	18–74
Mean age at death (years)	47.09	48.6	55.42	45.6
Range of ages (years)	18–74	19–64	23–79	18–75
Mean survival (years)	na	8.04	1.5	0.89
Range of survival (years)	na	1.5–47	5 wks to 8	8 wks to 3
Cause (fall)		8	3	1
Cause (RTA)		2	3	5
Cause (assault)		1	4	5
Cause (other)		2	2	1
Lucid (talked)		5	1	2
Lucid (not rational)		2	11	9
Lucid (unknown)		6	0	1

The data provide the mean age in years at injury, at death and the range of ages for each group of patients, the mean period and range of post-trauma survival, the number of patients that were injured as a result of a fall, road traffic accident (RTA), assault or other type of injury and the number of patients that were either lucid, not-lucid or where lucidity was unrecorded in each GOS group.

Histopathological analysis

All material had been handled with care to minimize the incidence of ‘dark’ neurons (Cammermeyer, 1961) and the brains were fixed in 10% formal saline for at least 3 weeks prior to dissection. One-centimetre-thick blocks were obtained from the cerebral hemispheres following routine brain cut in the coronal plane and routine histological procedures (Adams and Graham, 1976; Adams et al., 1980, 1985; Graham et al., 1989) were followed. After dissection, blocks had been embedded by an automated tissue processor and only material already embedded was available for study. All blocks for BA 24a and BA 4 had been embedded in paraffin; and all blocks for BA 10 and BA 11 had been embedded in celloidin. Information concerning the pH of the brain and the size of the blocks at the time when they were taken for embedding had not been recorded, and although neuronal size and shape negatively correlates with brain pH (Harrison et al., 1995), a correction factor could not be applied during subsequent morphometric analyses.

One member of the analytical team coded the cases before sections were cut. All material was examined blind to clinical and pathological data by another member of the research group.

Serial sections (n = a minimum of 27) were cut and stained with Cresyl violet/Luxol fast blue (Kluver and Barrera, 1953) for stereology. Neurons within human cerebral cortex do not follow a Poisson distribution (Schmitz and Hof, 2005), and large pyramidal cells of cortical layers 3, 4 and 5 form ‘minicolumns’ (Mountcastle, 1957; Schlaug et al., 1995; Buxhoeveden and Casanova, 2002). In order to minimize the occurrence of ‘empty counting frames’, a counting volume 300 μm wide × 300 μm high × 100 μm deep was used, with counts made only at points where the lamellar structure of the cortex was discrete. The sampling was therefore de facto biased to sites with a discrete laminar cortical structure. However, an important correlate was that the variance was reduced and the precision increased (Benes and Lange, 2001; Schmitz and Hof, 2005).

The key features used for identification and discrimination between different cortical areas and cortical layers in controls during the present study are summarized in Fig. 1.

Figure 1

Summary of the key features used for identification and discrimination between different cortical areas and cortical layers in controls during the present study.

Methods to obtain data to test different hypotheses

The first hypothesis tested was that the depth or thickness of either the whole cortical ribbon or the thickness of each cortical layer differed between GOS groups. An Olympus photomicroscope was calibrated using a stage micrometer (Agar) and the shortest vertical distance from the pial surface to the outer surface of the underlying white matter was measured at three sites where the six discrete layers of cortex were visible. Each site was separated from an adjacent one by at least 1000 μm in the coronal plane. Three measurements were taken in each cortical region in each patient (n = 37 injured and 11 controls) and the mean value and standard deviation across all patients within each GOS group was calculated. The total thickness of the cerebral cortex and of each cortical layer in four cortical regions was compared across outcome groups (ANOVA and Dunnett Multiple Comparisons test).

The second hypothesis was that the number of sub-types of neuron differed both between cortical layers and between control, moderately disabled, severely disabled and vegetative state patients. Columns of tissue through the entire cortical thickness at points where the discrete cortical layers were clearly visible were photographed, montages assembled, the precise magnification determined and a 300 × 300 μm counting grid drawn on transparent overlays. Greater detail is provided on-line (Supplementary material). Neurons were separated into pyramidal and non-pyramidal cells, and then categorized (Benes et al., 1986; Ong and Garey, 1991; Rajkowska et al., 1998; Rivara et al., 2003) into big pyramidal [nuclear diameter >20 μm], big non-pyramidal (nuclear diameter >10 μm), medium pyramidal (nuclear diameter = 20–10 μm), medium non-pyramidal (nuclear diameter = 10–6 μm), small pyramidal (nuclear diameter < 10 μm) and small non-pyramidal (nuclear diameter < 6 μm) cells. Each subgroup of neurons was counted as a separate entity, and cells were counted only when the cell nucleus contained a discrete nucleolus in order to obtain the maximum nuclear and somatic diameter and to prevent counting the same neuron more than once. The number of cells within each layer of the cortex was estimated using the optical disector technique because results are not distorted by differences in cell size, shape or orientation (Williams and Rakic, 1988; West et al., 1991; Howard and Reed, 1998; Mouton, 2002). The mean number of each type of cell within each cortical layer was estimated and the standard deviation calculated. The result was then transformed to allow expression as the cell number within a cubic millimetre of tissue so that no correction for changes in the thickness of each cortical layer between GOS outcome groups was required. Results are presented within the following text as the percentage change (Δ%) from controls when statistical significance at the 5% level was obtained.

The third hypothesis tested was that the size of the cell body changed across GOS groups. In neurons within which the nucleolus was present, the diameter of the cell body was estimated from a minimum of two perpendicularly orientated axes measured using an objective lens (×40) and a calibrated micrometer. A minimum sample of 15 neurons from each size subgroup, within each cortical layer, from each patient was used. The mean and median values, and the standard deviation, of the diameter of the cell body were calculated within each GOS group for analysis.

The fourth hypothesis was that the nearest neighbour distance (NND) between types of cell changed across GOS groups. A randomly placed reference line was drawn perpendicular to the pia mater through the upper and lower boundaries of each cortical region. Cells crossed by the reference line, and within which a discrete nucleolus was visible, were grouped by size or form and nominated the registered neuron. The surrounding tissue was then scanned in all directions within a counting volume consisting of serial sections that included at least 150–200 similar cells containing a discrete nucleolus. Methodologically, the nucleolus was used as a point for the measurement of distance between neighbouring cells. The x–y–z co-ordinates of the nucleolus and the corresponding NND to the reference cells were determined and the mean NND for each calculated. The ratio of the observed NND to the theoretical Poisson point NND is a definition of complete spatial randomness (Diggle, 1983). Within the present study calculation of the NNI allowed determination as to whether neurons were clustered, randomly distributed or scattered.

Statistical analyses

ANOVA was used to test whether the age or sex of patients differed between patient groups. Multiple comparisons of analysis of variance (MANOVA) for the severity of outcome, changes in layer thickness, cell number, cell diameter or NNI and any relationship across outcome groups, were determined using the Spearman test because the populations of neurons did not follow a Gaussian distribution.

Differences of number, cell diameter and spacing of neurons were analysed using the Kruskal–Wallis test, the non-parametric analogue of one-way ANOVA because the neurons analysed did not follow a Poisson distribution. The test determines if there are ‘significant’ differences among the population medians rather than the population means, and if the value of ‘P’ is small the differences between subjects are not due to random sampling. Dunn's post-test was used because the sample sizes were not equal.

To determine the direction and magnitude of differences between experimental groups an effect size analysis, which provides an interpretable value for the direction and magnitude of a difference between two clinical or treatment groups, was undertaken. The result of the effect size and its confidence interval (CI) also provides an unambiguous indication that an outcome is worse or better in one group (Coe 2002; Faraone 2008). The effect size was calculated as

In the ‘Results’ section, the value Hedge's g for effect size is provided in parentheses. Hedge's g values were calculated rather than Cohen's d value because sample sizes were small and there was a strong correlation (Spearman test) between and across groups.

Regression analysis was used to determine whether there was correlation between variables and GOS group in each cortical area. Spearman's test was used and the value of the correlation coefficient r_s is provided.

Results

Clinical evaluation

Patients were predominantly male (87.5%) and the most common cause of death was bronchopulmonary complications (50%), with lesser proportions dying from cardiac disease (12.5%), or sudden unexplained death in epilepsy (10.4%). In the controls there were cases of pyelonephritis, alcoholism, drug overdose, gastro-intestinal haemorrhage and empyema as cause of death. In each year at least one member of a minimum of three GOS groups were obtained during the study. The mean post-mortem interval for all patients was 2.40 days and did not differ across GOS groups (ANOVA) or have any association with the age of individuals (P = 0.97; ANOVA). The mean age at injury of control patients was 47.09 ± 17.71 years (Table 1), of moderately disabled patients was 38.92 ± 15.94, of severely disabled patients was 52.67 ± 21.10 and of vegetative state patients was 44.83 ± 17.81 years and ANOVA (P = 0.57) showed there was no difference between GOS groups. Grade 1 DAI (Adams et al., 1989) was diagnosed in three patients within the moderately disabled group; grades 1, 2 and 3 DAI in five patients within the severely disabled group and grades 1, 2 and 3 in 11 patients within the vegetative state group.

Changes in weight of the brain

The mean values and standard deviation of the weight of brains of patients were 1442.7 ± 105.0 g for controls, 1329.6 ± 202.9 g for moderately disabled, 1330 ± 140.7 g for severely disabled and 1275 ± 135.5 g for vegetative state patients. There was a significant loss in weight of the brain in severely disabled non-DAI (P = 0.009), but not DAI, patients, and also in vegetative state both non-DAI (P = 0.036) and DAI (P = 0.016) patients (MANOVA as brain weight was assessed on a variable scale between zero and 2000 g). There was a strong association between GOS group and the incidence of DAI (P < 0.0001, regression analysis), elevated intracranial pressure (P = 0.011, regression analysis), the occurrence of an intracranial haematoma and a skull fracture (P = 0.001, regression analysis), death due to pneumonia (P = 0.009, regression analysis) and the period of post-traumatic survival (P < 0.0001, regression analysis) but not for either death due to heart disease or sudden unexplained death due to epileptic seizure. There was a strong association between greater severity of GOS after a road traffic accident (P < 0.0001), an assault (P < 0.0001) or a fall (P = 0.036, regression analysis).

In this type of analysis the square of the coefficient of variance should be <0.5 (Benes and Lange 2001). For the present study values for CV² in tests for different factors such as cortical thickness, cell number, cell soma diameter and cell spacing for large pyramidal neurons ranged between 0.18 and 0.32; for medium pyramidal neurons ranged between 0.21 and 0.38; for small pyramidal neurons ranged between 0.23 and 0.38; for large non-pyramidal neurons ranged between 0.11 and 0.27; for medium non-pyramidal neurons ranged between 0.19 and 0.29 and for small non-pyramidal neurons ranged between 0.16 and 0.31. The precision or reproducibility of the estimate obtained within the sampling scheme in this study was therefore greater that that required for optimal sampling.

Change in thickness of the cortical ribbon

Results for changes in the thickness of the cortical ribbon of ventromedial and dorsolateral prefrontal, anterior cingulate and motor cortices across GOS groups appear in Fig. 2. There was thinning of ventromedial prefrontal (P = 0.009), dorsolateral prefrontal (P < 0.0001), anterior cingulate (P < 0.0001) and motor (P < 0.0001) cortices (Dunn's Multiple Comparisons test.). Changes in total thickness of the cerebral cortex between DAI and non-DAI patients using effect size comparison showed a worse outcome for ventromedial cortex in moderately disabled patients with DAI, for dorsolateral prefrontal cortex in vegetative state patients with DAI, and anterior cingulate and motor cortices (Fig. 2) in severely disabled patients with DAI. A worse outcome for cortical thinning occurred in ventromedial prefrontal cortex of severely disabled and vegetative state non-DAI patients (Fig. 3). As 11 of the 12 patients within the vegetative group (91.6%) were diagnosed with DAI differences between non-DAI and DAI patients were very small although a worse thinning of the cortex was present in DAI patients (Fig. 2).

Figure 2

Histogram of differences in thickness of cortical ribbon across DAI and non-DAI outcome groups with size effect (values for Hedges g). The thickness of the cortical ribbon at the four regions of cortex studied and changes across GOS groups. Within data for each cortical area, the central column (in white) is the value obtained from control patients. To the right of centre are the values for patients diagnosed with DAI. To the left of centre are values for moderately disabled, severely disabled and vegetative state patients without diffuse axonal injury (non-DAI). At the top of a column the number is the value of Hedge's g for size effect analysis of severity of outcome between non-DAI and DAI patients. A negative value is a statistically valid measure that the outcome for thinning of the cortical ribbon is worse in DAI patients, a positive value indicates that thinning is worse in non-DAI patients. VS = vegetative state; SD = severely disabled; MD = moderately disabled.

Figure 3

This figure shows histograms of the change in thickness of each cortical layer within the full depth of cortex from (A) ventromedial prefrontal, (B), dorsolateral prefrontal, (C), anterior cingulate gyrus and (D) primary motor cortices. Data are presented for control, moderately disabled (MD), severely disabled (SD) and vegetative state (VS) patients. Values for DAI patients are shown to the right and for non-DAI patients to the left of centre. On the ordinate axis numbers refer to the thickness in micrometers (μm). Greek letters above individual columns indicate the P-values for the differences between populations or numbers of neurons from values in control patients obtained using Dunn's multiple comparisons test, α = P < 0.05, β = P < 0.01,χ = P < 0.001. The error bars indicate the standard deviation within each data group.

Changes in thickness of cortical layers in patients with and without DAI

The results are shown in Fig. 3A (ventromedial prefrontal cortex), Fig. 3B (dorsolateral prefrontal cortex), Fig. 3C (anterior cingulate cortex) and Fig. 3D (motor cortex). More detailed information for percentage changes from control values are provided in Supplementary Table 1A, and for effect size (Hedge's g) in Supplementary Table 1B.

Greater thinning of layer 1 occurred in DAI rather than non-DAI patients for all four cortical areas (Fig. 3; Supplementary Tables 1A and B). In layer 2a greater thinning occurred within dorsolateral prefrontal and anterior cingulate cortices of non-DAI patients (Fig. 3B and C and Supplementary Tables 1A and B) and was worst in severely disabled patients. The worst thinning of layer 3 occurred within anterior cingulate (range of Hedge's g = –0.17 to −0.52) and motor cortices (range –0.16 to –2.06) in severely disabled patients with DAI (Fig. 3C and D, Supplementary Tables 1A and B), whereas thinning from layer 3 was greater in ventromedial prefrontal cortex in non-DAI moderately and severely disabled patients and in DAI vegetative state patients (range of Hedge's g = –0.05 to +0.69). In dorsolateral prefrontal cortex thinning of cortex was worse in severely disabled and vegetative state patients with DAI (Hedge's g = –0.13 to +0.39) (Supplementary Table 1B).

Thinning of layer 4 differed between ventromedial and dorsolateral cortices. In the former, all values for Hedge's g were positive, indicating that thinning was more severe in non-DAI patients (Fig. 3A, Supplementary Table 1B). In the latter, however, negative values were obtained (Fig. 3B, Supplementary Table 1B) indicating that thinning of cortex was more severe in DAI patients. Layer 4 is not present in agranular cortex.

In layer 5 of ventromedial and dorsolateral prefrontal cortices thinning occurred across all GOS groups being worst in moderately disabled patients with DAI, as indicated by the larger negative values for Hedge's g, (Supplementary Table 1B). In layers 5a and 5b of anterior cingulate and motor cortices there was thinning across all GOS groups, being more severe in DAI patients and worst in anterior cingulate cortex of severely disabled patients and moderately disabled motor cortex (Supplementary Table 1B). Thinning of cortical layer 6 occurred across all GOS groups in all four areas of cortex. Thinning was greatest in dorsolateral and anterior cingulate cortices of moderately disabled DAI patients, and in motor cortex of severely disabled DAI patients. In ventromedial cortex, however, cortical thinning was worst in moderately disabled patients with DAI, while values for Hedge's g were positive in severely disabled and vegetative state patients when greater thinning occurred in non-DAI patients (Supplementary Table 1B).

Cellular parameters in control patients

Before the consideration of changes in non-DAI and DAI patients, presentation of cell data from control patients is required. Descriptors for morphological distinction between cortical layers have been characterized earlier in the ‘Materials and methods’ section. In the cortex of control patients, the percentage of each sub-type of neuron of the total number of neurons in each cortical layer is provided in Table 2. The value obtained is the mean number across all 11 control patients.

Table 2

The percentage value of the total number of all neurons within each cortical layer is shown within ventromedial and dorsolateral prefrontal, anterior cingulate and motor cortices for the 11 control patients included in this study

Cortical area	Pyramidal neurons			Non-pyramidal neurons
	Large	Medium	Small	Large	Medium	Small
Ventromedial prefrontal cortex
Layer 1	0	0	0	0	0	100
Layer 2	0	0	80.7	0	6.4	12.8
Layer 3	5.2	22.2	43.8	15.1	4.8	8.8
Layer 4	0	13.9	57.1	0	4.8	24.2
Layer 5	0	20.7	48.8	10.5	7.3	12.7
Layer 6	0	5.2	42.2	22.8	9.9	19.82
Dorsolateral prefrontal cortex
Layer 1	0	0	0	0	0	100
Layer 2	0	0	75.1	0	8.3	16.6
Layer 3	4.7	20.1	53.7	3.9	4.7	12.6
Layer 4	0	13.3	65.0	0	2.5	17.6
Layer 5	0	27.6	41.4	0	9.5	17.6
Layer 6	0	14.9	42.6	20.8	7.2	14.5
Anterior cingulate cortex
Layer 1	0	0	0	0	0	100
Layer 2	0	0	82.3	0	6.3	10.9
Layer 3	5.0	22.3	54.4	3.1	6.1	9.2
Layer 5a	4.5	9.5	52.3	6.9	11.3	15.5
Layer 5b	3.7	15.9	53.0	5.6	10.4	15.5
Layer 6	0	0	38.5	17.2	26.1	18.2
Motor cortex
Layer 1	0	0	0	0	37.7	62.3
Layer 2	0	0	0	16.5	32.9	50.6
Layer 3	8.3	33.3	31.8	4.6	10.2	11.9
Layer 5a	0	24.7	44.9	5.5	9.8	14.9
Layer 5b	10.3	38.3	29.6	3.7	7.2	10.8
Layer 6	0	6.9	39.2	15.9	22.1	15.7

The largest pyramidal cells occurred within cingulate cortex, with a mean cell soma diameter of 31.75 ± 4.3 μm, and motor cortex, mean diameter 43.15 ± 2.7 μm, neither of which is quite as large as reported for human Betz cells (Sasaki and Iwata, 2001). Pyramidal cells formed columnar groups with a radial orientation, termed ‘minicolumns’ (Buxhoeveden and Casanova, 2002) in layers 3 and 5 of ventromedial and dorsolateral prefrontal, anterior cingulate and motor cortices. Pyramidal neurons within prefrontal cortex are generally smaller than in agranular cortex and our results are directly comparable to the size ranges described by Ong and Garey (1991).

Layer 1 in the control patients contained only small non-pyramidal neurons. Layer 2 contained small pyramidal neurons, medium and small non-pyramidal neurons with the exception of motor cortex in which large non-pyramidal neurons were observed. Layer 3 contained examples of all six neuronal sub-types. Layer 4 of ventromedial and dorsolateral prefrontal cortices contained medium and small pyramidal and non-pyramidal neurons. Layers 5a and 5b of agranular cortex contained examples of all six neuronal sub-types. Within layer 5b of anterior cingulate cortex vertically orientated, spindle pyramidal cells occurred. Layer 6 of prefrontal and motor cortices contained medium and small pyramidal neurons and all three sub-types of non-pyramidal neuron, while in anterior cingulate cortex only small pyramidal neurons and three sub-types of non-pyramidal neurons were present (Table 2).

The specific findings of the present study are complex and differ between cortical regions. The findings are summarized by cell type or size. In addition, data are provided for the NNI and the following descriptors are used. The term ‘clustered’ means that cells were more closely packed than would be expected in a normal, random or Poisson distribution and ‘scattered’ means that cells were more widely distributed than would be expected in a random distribution (Diggle, 1983).

Effect size of changes in number of pyramidal and non-pyramidal cells across GOS groups

Loss of large pyramidal neurons occurred from ventromedial cortex in DAI and non-DAI patients across all GOS groups, with a worse loss in DAI patients (Table 3, value for Hedge's g, Supplementary Tables 2A and 3). The loss of medium pyramidal neurons was greater in non-DAI patients (Table 3 and Supplementary Table 3), whereas for small pyramidal neurons loss was greater in DAI patients (Table 3). Loss of large non-pyramidal neurons from layers 3, 5 and 6 was greater in DAI patients (Table 3). There was worse loss of middle-sized non-pyramidal neurons from layers 3 and 5 of DAI patients, whereas there was greater loss in layers 2, 4 and 6 of non-DAI patients (Table 3). There was greater loss of small non-pyramidal neurons in DAI patients (Table 3 and Supplementary Table 3), except in layer 6 where a greater loss occurred in non-DAI patients.

Table 3

Values of Hedge's g obtained in effect size

	Pnb	Pnm	Pns	nPb	nPm	nPs
Ventromedial prefrontal cortex
Layer 1
MD	na	na	na	na	na	0.00
SD	na	na	na	na	na	0.00
VS	na	na	na	na	na	1.48
Layer 2
MD	na	na	0.00	na	3.10	−0.36
SD	na	na	0.09	na	4.20	−0.26
VS	na	na	0.42	na	9.34	−0.26
Layer 3
MD	−5.18	0.07	0.00	−1.86	0.00	0.00
SD	0.0	2.06	0.00	−2.36	−6.58	−0.53
VS	0.0	2.75	1.09	0.00	−6.53	−0.80
Layer 4
MD	na	0.10	0.00	0.00	−0.47	−0.11
SD	na	0.28	0.00	0.00	2.10	−0.33
VS	na	0.41	−0.28	0.00	5.38	0.18
Layer 5
MD	na	3.34	−0.34	−7.28	−0.58	−0.36
SD	na	1.91	−0.45	−0.54	−0.64	−0.14
VS	na	0.27	−0.75	−0.21	−1.72	0.00
Layer 6
MD	na	1.03	0.00	−1.23	2.39	0.31
SD	na	0.98	0.00	−0.89	3.24	0.11
VS	na	0.65	0.98	1.60	4.18	0.07
Dorsolateral prefrontal cortex
Layer 1
MD	na	na	na	na	na	0.45
SD	na	na	na	na	na	1.58
VS	na	na	na	na	na	−3.74
Layer 2
MD	na	na	0.00	na	−0.82	3.99
SD	na	na	3.29	na	−1.15	4.05
VS	na	na	1.54	na	−0.48	0.00
Layer 3
MD	−3.28	0.80	0.00	−1.02	−0.97	−6.36
SD	−7.11	0.33	0.82	−0.31	2.79	−2.99
VS	0.00	0.72	0.87	1.10	−2.40	−1.01
Layer 4
MD	na	0.35	0.89	na	−0.72	0.89
SD	na	1.30	1.76	na	−0.68	−0.14
VS	na	1.64	0.55	na	2.40	2.43
Layer 5
MD	na	1.15	0.00	na	−0.96	−1.41
SD	na	0.49	0.00	3.95	−1.16	0.93
VS	na	3.49	1.31	3.21	−1.60	−1.69
Layer 6
MD	na	2.06	0.00	1.40	−0.89	1.49
SD	na	8.04	−0.68	0.99	−0.90	2.70
VS	na	5.60	0.84	0.00	−0.64	0.87
Anterior cingulated cortex
Layer 1
MD	na	na	na	na	na	2.75
SD	na	na	na	na	na	3.07
VS	na	na	na	na	na	3.25
Layer 2
MD	na	na	−1.02	na	−1.29	0.00
SD	na	na	−0.58	na	−1.81	0.27
VS	na	na	−0.77	na	−2.62	0.54
Layer 3
MD	−3.57	−7.21	−1.58	−0.78	−1.55	0.55
SD	−1.60	−6.11	−0.50	−1.20	−1.58	1.68
VS	0.00	−1.25	−0.05	−1.11	−1.01	2.71
Layer 5a
MD	−0.64	−2.21	−1.02	−2.50	−1.66	0.53
SD	−0.42	−1.80	−0.59	−2.18	−1.00	1.63
VS	−0.10	−1.83	−0.25	−2.40	−0.51	0.85
Layer 5b
MD	−2.27	−3.89	−0.42	−0.89	−1.07	0.28
SD	−0.11	−5.40	−1.23	−0.55	−1.07	2.79
VS	−0.11	0.92	−0.58	−0.04	−1.81	0.62
Layer 6
MD	na	na	−0.67	−1.90	−0.48	−0.28
SD	na	na	−0.85	1.69	−2.13	1.48
VS	na	na	−0.78	2.78	−1.43	2.38
Motor cortex
Layer 1
MD	na	na	na	na	−3.40	−0.60
SD	na	na	na	na	−3.86	−0.60
VS	na	na	na	na	−1.53	−0.29
Layer 2
MD	na	na	na	−3.11	−0.67	−0.62
SD	na	na	na	−2.18	−0.62	−0.68
VS	na	na	na	−0.76	−2.73	−0.72
Layer 3
MD	−3.56	−0.05	−0.33	−4.23	−0.27	−0.89
SD	−1.83	−0.65	−0.41	−1.73	−0.68	−0.87
VS	0.00	−1.26	−0.68	−3.18	−0.79	−0.53
Layer 5a
MD	na	−0.27	0.92	−3.73	−1.58	0.51
SD	na	−0.40	1.01	−1.99	−0.27	−0.79
VS	na	−0.78	0.55	−1.23	−2.04	0.26
Layer 5b
MD	−3.71	−0.92	0.44	−1.90	−0.53	0.58
SD	−1.89	−1.34	0.44	−3.10	−0.37	0.35
VS	0.00	−1.02	0.44	−2.28	−0.78	0.23
Layer 6
MD	na	−1.79	0.45	1.60	−0.71	−2.81
SD	na	−4.21	0.18	1.25	−0.611	−2.82
VS	na	−2.56	0.96	1.24	−0.77	−1.41

The value for Hedge's g for each sub-type of neuron within each cortical layer and between DAI and non-DAI patients are presented separately for ventromedial and dorsolateral prefrontal, anterior cingulate and motor corticies. (na = not appropriate because cells of a particular sub-type do not occur within that cortical layer). A negative number and its relative size compared to zero provide a statistically valid result that a worse GOS occurs in DAI patients; a positive number indicates a worse GOS in non-DAI patients. Pnb = large pyramidal neuron; Pnm = medium pyramidal neuron; Pns = small pyramidal neuron; nPb = large non-pyramidal neuron; nPm = medium non-pyramidal neuron; nPs = small non-pyramidal neuron; MD = moderately disabled patient; SD = severely disabled patient; VS = vegetative state patient; na = not available or not appropriate; nc = no change from control; ns = not significant.

Within dorsolateral prefrontal cortex changes in neuron number were similar to that described above with a worse loss for large pyramidal neurons in layer 3 of DAI patients (Table 3), a worse loss of medium and small pyramidal neurons in non-DAI patients (Table 3) and a worse loss for large non-pyramidal neurons in layer 3 of moderately and severely disabled DAI patients (Table 3 and Supplementary Tables 2B and 3). Changes for medium pyramidal neurons, however, differed in that there was a worse loss within layers 2 through to 6 in DAI patients (Table 3 and Supplementary Tables 2B and 3). For small non-pyramidal neurons a worse loss occurred from layer 2 of non-DAI patients, layers 3 and 5 of DAI patients and layer 6 of non-DAI patients (Table 3).

Within the agranular anterior cingulate cortex (BA 24a) loss of large pyramidal cells occurred across all GOS groups from layers 3, 5a and 5b (Table 3 and Supplementary Table 2C) with a worse loss in DAI patients (Hedge's g, Table 3) and the greatest loss in moderately disabled patients. Loss of medium and small pyramidal cells and large and medium non-pyramidal neurons was greater in DAI patients (Supplementary Table 2C; Table 3 Hedge's g) where loss of pyramidal cells was most severe in moderately disabled patients (Table 3) and for medium non-pyramidal cells in severely disabled and vegetative state patients. However, loss of small non-pyramidal neurons was generally worse in non-DAI patients (Table 3).

Loss of large pyramidal neurons occurred from layers 3 and 5b of motor cortex (Supplementary Table 2D) with a worse loss in moderately disabled DAI patients (Table 3). Loss was also worse for medium pyramidal neurons in DAI patients being most severe in layers 3 and 5a of vegetative state and layers 5b and 6 of severely disabled patients (Table 3). However, for small pyramidal neurons there was a worse loss within layer 3 of DAI patients and layers 5a, 5b and 6 of non-DAI patients (Hedge's g, Table 3). Loss of non-pyramidal neurons from motor cortex was worse in DAI patients (Supplementary Table 2D and Table 3) with two exceptions, large non-pyramidal neurons in layer 6 and small non-pyramidal neurons in layer 5b (Table 3). Loss of large non-pyramidal neurons was worst in layers 2, 3 and 5a of moderately disabled patients, and in layer 5b of severely disabled patients. Loss of medium non-pyramidal neurons increased in severity with a poorer GOS in DAI patients. Loss of small non-pyramidal neurons was most severe from cortical layers 1, 3 and 6 of moderately disabled DAI patients (Table 3), greatest in vegetative state patients within layer 2 of DAI patients and layers 5a and 5b of moderately disabled non-DAI patients (Table 3).

The changes in number of pyramidal and non-pyramidal cells also altered their relative proportion. However, the changes differed between prefrontal and agranular cortices (Fig. 4A and B). In ventromedial prefrontal cortex, the loss of pyramidal cells and the lesser change for non-pyramidal cells resulted in an increased proportion of non-pyramidal neurons across outcome groups; from 0.29 in ventromedial cortex of controls to 0.39 in non-DAI vegetative state patients and 0.35 in DAI vegetative state patients (Fig. 4B). Loss of neurons was more severe in non-DAI patients where the value for Hedge's g was 0.8 compared to 0.41 for pyramidal neurons, and 0.84 compared to 0.48 for non-pyramidal cells (Fig. 4B, Supplementary Table 4). In dorsolateral cortex the proportion of pyramidal cells increased from 0.66 in controls to 0.73 in DAI vegetative state patients but fell to 0.27 in non-DAI patients (Fig. 4B, Supplementary Table 4).

Figure 4

Histograms of (A) the averaged total number of pyramidal (Pn) and non-pyramidal (nP) neurons within a column 300 × 100 µm in cross section and extending through the whole depth of the cortex for each GOS group with patients separated into those with DAI and those without (non-DAI); abd (B) the ratio of pyramidal to non-pyramidal neurons across GOS groups. Statistical significance for differences between control and GOS (Dunn's test) within each cortical region is indicated as α = P < 0.05, β = P < 0.01, χ = P < 0.001). Control values are represented by the white central column in each set, black and white columns on the right are values for DAI patients, grey and white columns on the left are values for non-DAI patients. VMPF = ventromedial prefrontal cortex; DLPF = dorsolateral prefrontal cortex; AC = anterior cingulate cortex; MC = motor cortex; VS = vegetative state; SD = severely disabled; MD = moderately disabled.

In agranular cortex the proportion of the total number of neurons that were non-pyramidal did not change across GOS groups in motor cortex (Fig. 4B). In anterior cingulate cortex the relative proportion of non-pyramidal neurons increased. Although the number of pyramidal cells did fall in anterior cingulate cortex in both DAI and non-DAI patients (Fig. 4A) across all GOS groups the ratio of pyramidal to non-pyramidal cells changed only in DAI patients (Fig. 4B) where effect size analysis indicated that loss of pyramidal neurons was worse in non-DAI patients (Hedge's g = −2.42, Supplementary Table 4). On the other hand, a worse loss of pyramidal cells occurred in motor cortex of DAI patients (Supplementary Table 4, Hedge's g = −4.0) while, due to the greater loss of pyramidal neurons, the proportion of non-pyramidal neurons increased (Fig. 4B).

Changes in the proportion of pyramidal to non-pyramidal neurons therefore differ in different cortical regions following head injury. In granular cortex loss of pyramidal cells is greater, such that the proportion of non-pyramidal neurons within the cortex rises in patients with a more severe outcome. Within agranular cortex, on the other hand, loss of both pyramidal and non-pyramidal cells occurs and the proportion of one cell type to the other is unchanged across GOS groups.

There were parallel losses of both large pyramidal and non-pyramidal neurons as the GOS fell. The value of the correlation coefficient r_s was –0.94 (Table 8 and Supplementary Table 5), indicating that as the GOS worsened there was an increasing loss of these cells across outcome groups and the loss was almost linear. There was no difference between DAI and non-DAI patients with overlap of the 95% CIs (Table 8 and Supplementary Table 5) occurring in all four cortical fields. There was a strong negative correlation for changes in the number of medium pyramidal neurons in DAI patients within ventromedial, anterior cingulate and motor cortices (Table 8 and Supplementary Table 5). However, the correlation was weaker in dorsolateral cortex of DAI patients, and anterior cingulate and motor cortices of non-DAI patients (Table 8 and Supplementary Table 5). The values of CI overlapped within ventromedial and dorsolateral prefrontal, and anterior cingulate cortices indicating there was no difference between DAI and non-DAI patients. However, it is worth noting the wide spread of values for r_s and inclusion of more patients may clarify any relationships. However, CI values did not overlap in motor cortex and loss of medium pyramidal neurons was greater in DAI patients. There was a strong negative correlation for changes in number of small pyramidal neurons in both DAI and non-DAI patients within ventromedial, dorsolateral prefrontal and anterior cingulate cortices (Table 8 and Supplementary Table 5). However, CI did not overlap in dorsolateral prefrontal cortex indicating that loss of small pyramidal neurons in non-DAI patients differed from that in DAI patients, with the former being more closely associated with injury to the brain. A much weaker correlation occurred in motor cortex (Table 8 and Supplementary Table 5) which suggests that multiple factors, other that those considered in this study, may influence changes in the number of small pyramidal neurons in BA 4.

Table 4

NNI for sub-types of neurons within layers of the ventromedial prefrontal cortex

	nPb	nPm	nPs	Pns	Pnm	Pnb
Controls
Layer 1			0.289
Layer 2		0.303	0.135	1.61
Layer 3	0.585	0.418	0.345	1.98	2.02	2.4 hor 0.317 vert
Layer 4		0.287	0.0953	1.45	1.084
Layer 5	0.815	0.572	0.461	1.470	3.371
Layer 6	1.01	0.761	0.506	2.368	1.5
Moderately disabled
Layer 1			0.930
Layer 2		1.100	0.580	1.533
Layer 3	1.061	1.073	1.008	1.935	2.677	2.249 hor 3.084 vert
Layer 4		1.161	0.573	1.445	1.175
Layer 5	1.054	1.1308	0.854	1.270	2.275
Layer 6	1.068	1.097	0.445	2.359	1.351
Severely disabled
Layer 1			0.858
Layer 2		1.093	0.557	1.419
Layer 3	1.154	1.078	1.114	1.819	1.329
Layer 4		1.046	0.643	1.509	1.163
Layer 5	1.148	1.124	0.890	1.638	1.613
Layer 6	1.129	1.274	0.415	2.804	1.880
Vegetative state
Layer 1			0.768
Layer 2			0.673	1.615
Layer 3		1.073	0.642	1.718	0.949
Layer 4			1.137	1.552	1.158
Layer 5			0.837	1.446	3.637
Layer 6			0.562	2.558	1.661

For completeness, a value for NNI was calculated for all types of neuron within each layer of cortex. However, when a value >2.15 was obtained it shows that a Poisson distribution is an inappropriate model for that population of cells. When clustering of cells occurs the value for NNI is in bold font. When cells are scattered the value for NNI is in italic font and when cells are randomly distributed the value for NNI is printed in normal script. Pnb = large pyramidal neuron; Pnm = medium pyramidal neuron; Pns = small pyramidal neuron; nPb = large non-pyramidal neuron; nPm = medium non-pyramidal neuron; nPs = small non-pyramidal neuron.

Table 5

NNI for sub-types of neurons within layers of the dorsolateral prefrontal cortex

	nPb	nPm	nPs	Pns	Pnm	Pnb
Controls
Layer 1			0.790
Layer 2		0.790	0.273	1.307
Layer 3	0.660	0.580	0.354	1.471	2.541	2.129
Layer 4		0.350	0.313	1.597	1.081
Layer 5	0.910	0.470	0.360	0.620	0.783
Layer 6	0.850	0.500	0.325	1.130
Moderately disabled
Layer 1			1.006
Layer 2		1.076	0.436	1.338
Layer 3	1.102	1.036	0.599	1.511	2.625	2.156 hor 0.887 vert
Layer 4		1.046	0.666	1.629	1.168
Layer 5	1.105	1.122	0.406	0.591	0.838
Layer 6	0.857	0.822	0.379	1.210
Severely disabled
Layer 1			1.074
Layer 2		0.903	0.738	1.266
Layer 3	1.099	0.927	0.572	1.516	2.411	2.037
Layer 4		1.098	0.459	1.549	0.998
Layer 5	1.478	1.092	0.349	0.665	0.669
Layer 6	1.028	1.044	0.381	1.055
Vegetative state
Layer 1			0.827
Layer 2		1.289	0.496	1.419
Layer 3	0.877	1.098	0.644	1.384	1.598
Layer 4		1.082	0.840	1.695	1.003
Layer 5		1.222	0.516	0.656	0.621
Layer 6	0.852	0.858	0.412	1.227

For completeness, a value for NNI was calculated for all types of neuron within each layer of cortex. However, when a value >2.15 was obtained it shows that a Poisson distribution is an inappropriate model for that population of cells. When clustering of cells occurs the value for NNI is in bold font. When cells are scattered the value for NNI is in italic font and when cells are randomly distributed the value for NNI is printed in normal script. nb = large pyramidal neuron; Pnm = medium pyramidal neuron; Pns = small pyramidal neuron; nPb = large non-pyramidal neuron; nPm = medium non-pyramidal neuron; nPs = small non-pyramidal neuron.

Table 6

NNI for sub-types of neurons within layers of the anterior cingulate cortex (BA 24a)

	nPb	nPm	nPs	Pns	Pnm	Pnb
Controls
Layer 1			1.405
Layer 2		1.150	0.587	1.146
Layer 3	1.044	1.088	0.540	1.413	1.189	1.437
Layer 5a	1.010	1.085	0.538	1.249	1.127	1.24
Layer 5b	1.037	1.097	0.547	1.086	1.061	1.62
Layer 6	1.068	1.046	0.995	1.111	1.061
Moderately disabled
Layer 1			1.557
Layer 2		1.120	0.670	0.957
Layer 3	1.104	1.137	0.765	1.059	1.314	1.151
Layer 5a	1.049	1.164	0.936	1.172	1.152	1.300
Layer 5b	1.080	1.047	0.553	1.284	1.148	1.145
Layer 6	0.980	1.270	0.484	1.015	1.126
Severely disabled
Layer 1			1.746
Layer 2		1.082	0.977	1.126
Layer 3	1.322	1.360	1.031	0.975	1.327	1.177
Layer 5a	1.832	2.661	0.952	0.976	1.438	1.197
Layer 5b		1.139	0.907	0.977	1.339	1.156
Layer 6		1.512	0.502	0.975	1.141
Vegetative state
Layer 1			1.633
Layer 2		1.152	0.803	0.963
Layer 3	1.377	1.028	0.845	0.871	1.562
Layer 5a	1.716	2.691	0.811	1.093	1.158	1.140
Layer 5b		1.745	1.113	1.132	1.161	1.067
Layer 6		1.049	1.158	1.037

Nearest neighbour indices for sub-types of neurons within layers of the anterior cingulate in control (Con), moderately disabled, severely disabled and vegetative state patients classified by their GOS after head-injury. For completeness, a value for NNI was calculated for all types of neuron within each layer of cortex. However, when a value >2.15 was obtained it shows that a Poisson distribution is an inappropriate model for that population of cells. When clustering of cells occurs the value for NNI is in bold font. When cells are scattered the value for NNI is in italic font and when cells are randomly distributed the value for NNI is printed in normal script. nb = large pyramidal neuron; Pnm = medium pyramidal neuron; Pns = small pyramidal neuron; nPb = large non-pyramidal neuron; nPm = medium non-pyramidal neuron; nPs = small non-pyramidal neuron.

Table 7

NNI for sub-types of neurons within layers of the motor cortex

	nPb	nPm	nPs	Pns	Pnm	Pnb
Controls
Layer 1		1.040	0.768
Layer 2	1.043	1.011	0.923	0.710
Layer 3	1.042	0.986	0.640	1.300	1.207	2.121
Layer 5a	1.014	1.021	1.248	0.533	3.893/9.44
Layer 5b	1.240	1.002	1.017	1.643	2.53	3.659 vert 9.76 hor
Layer 6	1.000	1.009	0.428	0.440	1.12
Moderately disabled
Layer 1		1.043	0.765
Layer 2	1.037	1.006	0.806	0.898
Layer 3	1.017	1.016	0.712	1.331	1.192	1.936
Layer 5a	1.062	1.018	1.252	2.11	3.53/6.79
Layer 5b	1.173	1.018	1.155	1.410	2.260	2.44 vert 4.69 hor
Layer 6	1.016	1.015	0.529	0.703	0.654
Severely disabled
Layer 1		1.061	0.710
Layer 2	1.023	1.002	0.932	0.871
Layer 3	1.045	1.011	0.711	1.250	1.190	1.630
Layer 5a	1.092	1.014	1.237	2.326	3.18/6.99
Layer 5b	1.144	1.002	1.161	1.706	2.70	1.138 vert 2.936 hor
Layer 6	1.005	1.007	0.561	0.886	0.612
Vegetative state
Layer 1		1.013	0.895
Layer 2	1.138	0.996	1.028	0.946
Layer 3	1.060	0.994	0.771	1.360	1.16
Layer 5a	1.057	1.009	1.316	2.470	3.30/8.14
Layer 5b	0.989	0.993	1.188	1.730	2.530
Layer 6	1.011	0.910	0.615	0.730	0.410

For completeness, a value for NNI was calculated for all types of neuron within each layer of cortex. However, when a value >2.15 was obtained it shows that a Poisson distribution is an inappropriate model for that population of cells. When clustering of cells occurs the value for NNI is in bold font. When cells are scattered the value for NNI is in italic font and when cells are randomly distributed the value for NNI is printed in normal script. nb = large pyramidal neuron; Pnm = medium pyramidal neuron; Pns = small pyramidal neuron; nPb = large non-pyramidal neuron; nPm = medium non-pyramidal neuron; nPs = small non-pyramidal neuron.

Table 8

Results of a nonparametric (non-Gaussian) multivariant analysis for changes in the number, the mean diameter of the cell soma and nearest neighbour spacing between neurons in the six lamellae of the ventromedial and dorsolateral prefrontal, anterior cingulate and motor cortices across control (n = 11), moderately disabled (n = 13), severely disabled (n = 12) and vegetative (n = 12) patients classified using the GOS

	P-value	Value for Spearman rs	95% CIs
Ventromedial prefrontal cortex
Change of neuron number	0.0001	0.943	0.9165–0.9611
Change of diameter of neuron soma	0.0001	0.777	0.6860–0.8434
Change in spacing	0.0001	0.867	0.8015– 0.9046
Dorsolateral prefrontal cortex
Change of neuron number	0.0001	0.894	0.8464–0.9271
Change of diameter of neuron soma	0.0001	0.882	0.8294–0.9186
Change in spacing	0.0001	0.722	0.6134–0.8035
Anterior cingulate cortex
Change of neuron number	0.0001	0.976	0.9653–0.9841
Change of diameter of neuron soma	0.0001	0.782	0.6932–0.8477
Change in spacing	0.0001	0.782	0.6878–0.8448
Motor cortex
Change of neuron number	0.0001	0.913	0.8796–0.9434
Change of diameter of neuron soma	0.0001	0.859	0.7972–0.9023
Change in spacing	0.0001	0.871	0.8138–0.9108

There was a high negative value for r_s in medium non-pyramidal cells in ventromedial and dorsolateral prefrontal cortices of DAI patients (Table 8 and Supplementary Table 5). Nevertheless, a much weaker negative correlation occurred in ventromedial and dorsolateral cortices of non-DAI patients. For both ventromedial and dorsolateral cortices the range of CI did not overlap and suggests that multiple factors other than those considered herein may influence loss of these cells within non-DAI patients. In DAI patients however, there was a close to linear loss of medium non-pyramidal neurons. Within motor cortices the correlation was probably weaker due to the smaller loss of non-pyramidal neurons. There was a very weak positive correlation in the anterior cingulate cortex of non-DAI patients (Supplementary Table 5) that indicated a much greater complexity of factors influencing changes in cell number within this cortical region (Supplementary Table 5). However, results for medium non-pyramidal neurons allowed the suggestion that, in patients with DAI, neuronal loss was greater in the prefrontal cortex. For small non-pyramidal neurons the value of the correlation coefficient in ventromedial and dorsolateral prefrontal cortices of DAI patients was high (Table 8 and Supplementary Table 5) indicating that as GOS worsened the number of cells increased. Changes were more complex in non-DAI patients, however, (Table 8 and Supplementary Table 5) where r_s was positive, indicating an increased number of small non-pyramidal neurons in ventromedial cortex but a loss in dorsolateral cortex (Supplementary Table 5). In anterior cingulate and motor cortices only a weak or very weak correlation occurred for small non-pyramidal neurons (Supplementary Table 5). However, the direction of that correlation was different. In anterior cingulate cortex of non-DAI patients the value of r_s was negative (Supplementary Table 5) and positive for DAI patients indicating that the number of small non-pyramidal cells fell in non-DAI patients but rose in DAI patients as GOS worsened. However, the values for the correlation coefficients are so small that multiple factors not considered in the present analysis may influence changes in cell number. In motor cortex the value of r_s in non-DAI patients were very weakly positive, in DAI patients was negative and <0.4 indicating a lack of an association (Supplementary Table 5). Overall, it is suggested that either a lower level of loss of small non-pyramidal neurons occurred in parallel with loss of other neuronal sub-types, or tissue shrinkage as a result of the loss of larger neurons occurred and larger numbers of small non-pyramidal cells became included within a standardized volume of tissue within prefrontal cortex of DAI patients.

Changes in diameter of the cell soma, nearest-neighbour spacing and NNI across GOS groups

Results for changes in diameter of the neuronal cell soma are shown in Supplementary Table 6A–D; for NND in Supplementary Table 7A–D and for the NNI in Tables 4–7. Effect size analysis provided evidence for differences in cell number (Supplementary Table 4), and the NNI between DAI and non-DAI patients (Supplementary Table 6). There was no difference in the median cell diameter between DAI and non-DAI patients and these are therefore reported as a combined value. With loss of a proportion of large pyramidal neurons, the median diameter of the cell soma fell across all groups (P = 0.008, Dunn's multiple comparison's test) (Supplementary Table 6A–D). Cell diameter fell by 100% in layer 3 of anterior cingulate cortex and layers 3 and 5b of motor cortex in severely disabled and vegetative state patients (Supplementary Table 6A–D). In parallel, NND between large pyramidal neurons increased (Supplementary Table 7A–D). Large pyramidal neurons had a mean spacing of 153.6 ± 14.7 μm within layer 3 of anterior cingulate cortex and 164.6 ± 37.5 μm in motor cortex of control patients. An increase in spacing to infinity occurred for large pyramidal cells in layer 3 of granular cortex, and layers 3, 5a or 5b of agranular cortex with the complete loss of these cells (Supplementary Table 7A–D). Spacing between large pyramidal cells rose by 30% in cingulate cortex and 70% in motor cortex in severely disabled patients (Supplementary Table 7C and D) to become infinite in vegetative state patients because of their total loss (P < 0.0001, Dunn's test). In control patients, large pyramidal neurons in all four cortical regions were more widely spaced than would be expected for a Poisson distribution and are therefore scattered (Tables 4–7). Changes of NND between large pyramidal neurons occurred both within and between minicolumns (Supplementary Table 6). However, comparison of values for NNI between these cells showed that it fell (Tables 4–7) across survival groups indicating that large pyramidal cells became less scattered with increasing severity of cell loss excepting, of course, when these cells were lost completely.

Medium pyramidal cells occurred in layers 3, 4, 5 and 6 of prefrontal cortices, and layers 3, 5a, 5b and 6 of anterior cingulate and motor cortices of controls (Table 2 and Supplementary Table 2A–D). The diameter of these cells fell within ventromedial (P = 0.007, Dunn's), dorsolateral (P = 0.009) and anterior cingulate (P = 0.087) cortices, but increased in layer 3 of motor cortex (P = 0.008, Dunn's) in vegetative state patients (Supplementary Table 6D). The cells were either scattered or randomly distributed within the neuropil across all GOS groups (Tables 4–7). In layer 4 medium pyramidal cell diameter fell across all GOS groups in ventromedial (P < 0.001) and dorsolateral (P < 0.001) cortices (Supplementary Table 6A and B). In cingulate cortex the diameter of these cells fell in layers 5a and 5b (P < 0.01, Dunn's) (Supplementary Table 6C). However, in motor cortex the diameter of medium pyramidal cells fell only in layer 5a of vegetative state patients (Supplementary Table 6D). The values of NNI (Table 7) generally increased in moderately disabled patients then fell in severely disabled and vegetative state patients to result in a closer packing of neurons. However, in anterior cingulate cortex the NNI increased across all GOS groups (Table 6).

The diameter of small pyramidal neurons within layer 3 did not change across GOS groups in ventromedial, anterior cingulate and motor cortices (Supplementary Table 6A–D). However, cell diameter fell within layer 3 of dorsolateral cortex (P < 0.05) in vegetative state patients (Supplementary Table 6B). The diameter of small pyramidal cells fell within layer 4 of ventromedial and dorsolateral cortices (P < 0.05, Dunn's) in vegetative state patients (Supplementary Table 6A and B). The diameter of small pyramidal cells fell across GOS groups (P = 0.001, Dunn's) in layer 5a of anterior cingulate cortex. Cell diameter increased in layer 5a of motor cortex of moderately disabled (P = 0.001) patients but was unchanged in severely disabled and vegetative state patients (Supplementary Table 6D). The diameter of small pyramidal cells did not change in layer 5b of cingulate or motor cortices (Supplementary Table 6C). Within layer 6 of ventromedial and dorsolateral prefrontal cortices, cell diameter of small pyramidal cells fell in severely disabled and vegetative state patients (P = 0.01, Dunn's) (Supplementary Table 6A and B) and fell within motor cortex in moderately and severely disabled patients (Supplementary Table 6D). However, in vegetative state patients cell diameter did not differ from control values. In anterior cingulate cortex the diameter of small pyramidal cells increased in severely disabled patients (P < 0.05, Supplementary Table 6C) while not differing from controls in vegetative state patients. The value of the NNI in the prefrontal cortices indicated that with falling GOS a Poisson distribution was an inappropriate model for the spacing of small pyramidal neurons across all GOS groups where cells were very widely spaced (Tables 4 and 5). In anterior cingulate cortex small pyramidal cells remained randomly distributed across GOS groups, while in motor cortex within layers 2, 3, 5a and 5b the NNI increased with increasing severity of the GOS and small pyramidal cells became more widely scattered (Table 7).

The diameter of large non-pyramidal cells within layers 3, 5 and 6 of ventromedial prefrontal cortex fell in severely disabled and vegetative state patients (P = 0.006, Dunn's) and in layers 2 through to 6 of cingulate cortex (P < 0.001, Dunn's) (Supplementary Table 6C). Cell diameter within dorsolateral and anterior cingulate cortices fell sharply in vegetative state patients with complete loss of these cells (P = 0.008, Dunn's) (Supplementary Table 6B and C). The diameter of large non-pyramidal cells in motor cortex did not change across all GOS groups, except in layer 5b of vegetative state patients (P = 0.036) (Supplementary Table 6D). There was no change in the diameter of large non-pyramidal cells in layer 6 of dorsolateral prefrontal and motor cortices; but the diameter did fall within ventromedial and anterior cingulate cortices across all GOS groups (Supplementary Table 6A, P < 0.001, Dunn's; Supplementary Table 6C, P < 0.001, Dunn's). Large non-pyramidal cells were clustered in layer 3 of control ventromedial and dorsolateral cortices, were randomly distributed in layers 5 and 6 of the same and randomly distributed in anterior cingulate and motor cortices (Tables 4–7).Overall the values of the NNI did not change across GOS groups with cells remaining randomly distributed (Tables 4–7).

The diameter of medium non-pyramidal cells fell across all GOS groups within layer 2 of ventromedial cortex (P = 0.0007), in motor cortex (P = 0.0081, Dunn's) of severely disabled and vegetative state patients and in anterior cingulate cortex (P = 0.029) (Supplementary Tables 6A and D) of vegetative state patients. However, cell diameter did not change in layer 2 of dorsolateral prefrontal cortex. In layer 3, the diameter of medium non-pyramidal cells fell within ventromedial and motor cortices of moderately disabled patients (P = 0.048) but increased in vegetative state patients (P = 0.093, Dunn's) (Supplementary Tables 6A and D). Within dorsolateral cortex of severely disabled and vegetative state (P = 0.074, Dunn's) patients cell diameter fell from control values (Supplementary Tables 6B and D). However, within anterior cingulate cortex, the diameter of medium non-pyramidal neurons increased within layer 3 (P = 0.009) across all GOS groups (Supplementary Table 6C). In layer 4 of ventromedial and dorsolateral cortices, the cell diameter of medium non-pyramidal cells was unchanged across GOS groups (Supplementary Tables 6A and 6B). In layer 5a of anterior cingulate cortex cell diameter increased across all GOS groups (P = 0.0069) (Supplementary Table 6C) and spacing between these cells also increased (P = 0.0072) to the extent that medium non-pyramidal cells became so widely spaced that a Poisson or normal model of spatial distribution was no longer appropriate (Supplementary Table 7C and Table 6). There was no change in spacing between medium non-pyramidal neurons in layer 5b of cingulate cortex where cells remained either randomly spaced or scattered. Within layer 6 of ventromedial prefrontal cortex the mean diameter of the cell soma in medium non-pyramidal cells fell (P = 0.004, Dunn's) in severely disabled but increased above control values in vegetative state patients (P = 0.008) (Supplementary Table 6A), increased in cingulate cortex of vegetative state patients (P = 0.009) (Supplementary Table 6C) and fell in severely disabled and vegetative state patients in motor cortex (P = 0.008, Dunn's) (Supplementary Table 6D). Values for NND between medium non-pyramidal neurons within layer 6 of ventromedial cortex increased in moderate and severely disabled GOS groups and became infinite in vegetative state patients reflecting the loss of these cells (Supplementary Table 7A). Within dorsolateral cortex, values for NND rose in severely disabled and vegetative state (P < 0.001) patients (Supplementary Table 7B). Values of the NNI for medium non-pyramidal cells within layer 6 of agranular cortex in control patients showed that cells were randomly distributed (Tables 6 and 7). However, in all outcome groups their spacing increased so that the remaining cells assumed a random distribution (Table 4).

The diameter of small non-pyramidal neurons in most layers of all four cortical regions did not change across GOS groups (Supplementary Table 6A–D). However, cell diameter increased in layer 4 (P < 0.01, Dunn's) of ventromedial, layer 2 (P < 0.05) of dorsolateral, layer 3 (P < 0.05) of anterior cingulate and layers 1, 3 and 5a of motor cortices (P < 0.05) in severely disabled patients (Supplementary Table 6A–D). In layer 6 cell diameter fell in ventromedial cortex of severely disabled patients (P < 0.05), across all GOS groups in dorsolateral cortex, and in severely disabled and vegetative state patients within motor cortex (Supplementary Table 6A–D). Values for NND increased across all patient groups in most cortical layers within all four regions of cortex (Supplementary Table 7A–D) and were notable in that a greater increase occurred within DAI patients (Hedge's g = −2.81/–2.82) except in cingulate cortex where the greater change occurred in non-DAI (Hedge's g = 0.28) patients. Within cortical layers of ventromedial and dorsolateral cortices small non-pyramidal cells remained clustered across all GOS groups, whilst within agranular cortex a random distribution prevailed (Table 4).

Discussion

This study provides the first quantitative analysis of changes in the thickness of the cortical ribbon, and the number, size and nearest-neighbour spacing of neurons within four regions of human cerebral cortex in patients classified using the Glasgow Outcome Scale (Jennett and Bond, 1975) traumatic brain injury. In addition, novel evidence is provided, that changes in the population of cortical neurons differs both with severity of GOS, and between patients diagnosed at post-mortem either with traumatic DAI or other types of traumatic brain injury.

This study allows a definite statement to be made that neuronal cell loss is a major contributor to pathology in the frontal cerebral cortex of surviving traumatic brain injury patients, that the degree of neuronal loss correlates with the GOS outcome group of a patient and that neuronal loss is greater in patients with DAI. The present study extends the reported damage to central white matter tracts 6 months after injury (Kraus et al., 2007; Lipton et al., 2008) by providing novel data for loss of cortical neurons which, at least in part contribute to the same tracts, and that the extent of neuronal loss differs between moderately disabled, severely disabled and vegetative state patients.

Methods

Of necessity, the patient sample represents a population held on record and does not reflect the incidence of traumatic brain injury in the general population, with a more severe GOS being over-represented. A major distinction between other studies and the material examined in this study is that patients in the former were all living and data suggesting damage after traumatic brain injury were gathered using DTI. When brains of patients in the present study were referred DTI was not available and assessment of outcome was only retrospectively deduced from clinical records in some cases, and DAI diagnosed histopathologically (Adams et al., 1976, 1980, 1982, 1985, 1989, 2001; Graham et al., 1989). Furthermore, the data are probably skewed as a result of the small number of patients available within each subgroup.

The literature indicates that the mean age of patients at injury (45.87 ± 3.08) is higher that in other studies; for example 35.5 (Deb and Burns, 2007); 35.3 (Kraus et al., 2007), 20 (Ding et al., 2008), 30.5 (Fujiwara et al., 2008), 32.4 (Niogi et al 2008) and 31.37 years (Ponsford et al., 2008). Mushkadiani et al., (2007) reported that increasing age at injury was strongly linked with poorer outcome. But that premise has also been questioned by Deb and Burns (2007) who reported a greater risk for psychiatric morbidity in patients between the ages of 18 and 65 years following traumatic brain injury. Nonetheless, it is possible that the greater age at injury of patients in the present study exacerbated the pathology described, with the mean age of patients at injury being some 15–17 years older than patients in other studies.

Some controversy exists within the literature relating to thinning of the cerebral cortex with ageing (reviewed by Preul et al., 2006). The rate of cortical thinning has been reported as between 0.07 and 0.01 mm/decade (Salat et al., 2004). Across the age range of patients in the present study a thinning of cortex by 0.35 mm in motor cortex and 0.05 mm in other cortical regions may be predicted between the youngest and oldest patients in controls. The thinning of motor cortex ranged from 0.5 mm in moderately disabled, 0.8 mm in severely disabled to 1.03 mm in vegetative state patients or up to three times greater than the maximum predicted change with ageing alone. Thinning of cortex after traumatic brain injury in the patients from the current study is therefore greater than might be expected without injury and provides confidence that a change resulting from traumatic brain injury had occurred.

In the present study, patients were predominantly male (92%) and did not die from head injury per se, as the majority died of bronchopulmonary complications (59.6%), cardiac disease (22.4%), or sudden unexplained death in epilepsy (10.2%), with only two patients in this study dying as a result of either alcoholism or drug overdose. It is suggested therefore that the observed changes in the brain reflect the sum total of pathology resulting from the trauma per se and any additional insults to the injured brain after trauma.

A second feature that may compromise the results obtained is the effect of tissue shrinkage upon processing for light microscopy. The tissue blocks were used for diagnosis using routine brain cut procedures in the coronal plane (Adams et al., 1980) some 10–15 years before the present study was proposed. The blocks therefore provided only a representative slice of cortex rather than extending through the full extent of a cortical field and therefore random sampling for stereology through the entire cortical field as advocated by Schmitz and Hof (2005) could not be undertaken. Nonetheless, use of a standardized method for obtaining cortical slices (Adams et al., 1980) meant that tissue was obtained in an unbiased manner. The volume of wet tissue taken at the brain cut is unknown. However, all tissue blocks were processed in the same way in a tissue processor through cycles for embedding in either paraffin wax or celloidin so that all blocks were processed by a replicate, mechanically controlled schedule that minimized variation between different blocks as to shrinkage artefact. After cutting, all sections were mounted and stained. Celloidin sections provided slightly better preservation and contrast after staining but because the post-mortem interval was comparable across all outcome groups and fixation was achieved by immersion there was no morphometric difference of cell sizes between different parts of control brains embedded in paraffin wax or celloidin.

It is recognized that histological sections undergo a considerable collapse along the z-axis by some ∼35–60% even when tissue is embedded in celloidin (Benes and Lange, 2001). Collapse is thought to result in considerable distortion of optical depth. However, a greater distortion will occur in thicker sections (Benes and Lange, 2001). In the present study, the use of serial wax and celloidin sections reduced the percentage collapse along the z-axis, allowing a rapid and straight forward determination of separation of points along the z-axis, as well as rapid identification of cells through a volume. This has particular importance for estimation of mean NNDs as the packing density of neurons in human cortex is low compared with non-human species (Blinkov and Glezer, 1968). Thus, in the present study cognisance that stereological techniques are not ‘assumption free’ has been included and appropriate measures taken to minimize their effect.

Use of the 3D optical disector in the present study has been widely validated (Williams and Rakic, 1988; West et al., 1991; Howard and Reed 1998; Mouton 2002) in many studies. Nonetheless, the recognized wide variation between patients, both in relation to their age and level of education, results in problems when dealing with relatively small numbers of case material. Clearly, this is a major difficulty in assessing human material. For distinction between patients diagnosed with or without DAI post-mortem the conditions of a power calculation were not met in the present study. Preliminary power calculations indicate that detailed investigation of changes in neuronal number within cortical layers in moderately disabled, severely disabled and vegetative state patients will require access to at least ten DAI patients and ten non-DAI patients for each decade of life, for each GOS group, or about 420 patients. It is unlikely that such numbers might be available without large-scale international collaboration between research groups.

In the present study, calculation of the values for coefficient of variance given in the Results section for each subtype of neuron and parameter provide confidence that our findings are reproducible. A number of statistically significant differences between GOS groups were obtained in the present study using novel application of effect size analyses. Although inferential statistics provide important data concerning the reliability of a result, it is not possible to compare two scenarios or treatments when each is compared with control and/or placebo because the level of a statistical significance test depends on two things: the size of the effect and the size of the sample. In studies with less than about 50 participants there is usually a lack of statistical power (a function of sample size, effect size and the level of P) when detecting small, medium or possibly even large effects. Under such conditions the results of a significance test may be misleading as a result of Type II errors which incorrectly fail to reject the null hypothesis. Instead, use of effect sizes, including CI, serve the same function as classic inferential statistical tests such as t-tests, but are more informative because they indicate the spread of data around the mean effect size. Furthermore, if the CI does not include zero there is a reasonable certainty that a real change has been identified. Analysis of differences in thickness of layers of the cerebral cortex, and of differences in number of neurons in cortical layers between control and outcome groups, or patients diagnosed with DAI and non-DAI is appropriate in the present report because the same analytical criteria have been used across all blocks and patients. There is therefore confidence that the types of damage reported in this study may be representative of the pathology in the majority of patients within a larger sample within similar outcome groups.

Changes in the number and size of neurons

Novel quantitative evidence has been obtained that, at 3 months and longer after traumatic brain injury, there is differential loss of different sized, pyramidal and non-pyramidal neurons in different cortical areas, between different GOS groups and between DAI and non-DAI patients. These changes occur within four functionally associated regions of human cortex that are important in cognitive and executive functions, in emotional responses, in control of voluntary motor function and persistent working memory. There was differential loss of large and medium pyramidal neurons from layers 3, 4, 5 and 6 of ventromedial prefrontal; layers 2, 3, 4, 5 and 6 of dorsolateral prefrontal; layers 2, 3, 5a and 5b of anterior cingulate; and layers 3, 5a and 6 of motor cortices. There was also a differential change of the ratio of pyramidal to non-pyramidal neurons in patients with and without DAI, most notably in dorsolateral prefrontal and cingulate cortices.

The number of large pyramidal and non-pyramidal cells falls to zero in severely disabled and vegetative state patients within all four cortical areas and with a worse outcome in DAI patients (Supplementary Table 3). In the same patients, however, the number of medium-sized pyramidal cells increases within layers 3, 5 and 6 of ventromedial and dorsolateral cortices of moderately disabled non-DAI patients; in layers 5a, 5b and 6 of anterior cingulate cortex being more severe in DAI patients (Supplementary Table 3); and layers 5a, 5b and 6 of motor cortex in vegetative state non-DAI patients (Supplementary Table 3). Furthermore, the diameter of medium pyramidal cells increases in layer 3 of motor cortex across GOS groups, and of medium non-pyramidal cells in layer 3 of ventromedial cortex in vegetative state patients, layers 3 and 5a of anterior cingulate across GOS groups, and layers 3 and 5a of motor cortex across GOS groups. The data generated for the change in number and mean diameter of the neuronal soma of subsets of neurons lead to the suggestion that loss of neurons may not be the only change occurring within human cerebral cortex after traumatic brain injury. This study indicates that not all affected or injured cells die and disappear but that some cells may shrink, after de-afferentation or loss of the cues that serve to maintain normal activity and function, to become included within the next smaller size group in which the median diameter increases. Woods et al. (1999) reported a reduction in cell size or atrophy within the dorsal columns and ventral posterior lateral thalamic nuclei following dorsal rhizotomy in adult macaque monkeys. In that report, however, the type or size of neuron which underwent atrophy was not reported so that comparison with data for medium pyramidal and non-pyramidal cells obtained within the cerebral cortex in the present study is not possible. Our study does, however, extend the number of regions of the brain within which cell loss after traumatic brain injury have been reported. Moreover, it provides good evidence that changes may occur over a considerable time period after injury as has been suggested in several earlier MRI studies (Gale et al.2005; Cohen et al., 2007; Bendlin et al., 2008; Fujiwara et al., 2008).

Three-dimensional spacing of neuronal sub-types

This study provides data concerning the three-dimensional spacing of subgroups of pyramidal and non-pyramidal neurons within the cortical layers of four different regions of human cortex. To the authors’ knowledge the present study is the first analysis in head-injured patients using changes of NNDs and indices. It has long been recognized that cortical neurons have a non-Gaussian distribution within neuropil. However, an appreciation of the modes of packing of neurons between cortical layers, whether packing differs between neurons of different sizes or an appreciation of whether differences in packing density occur between different areas or regions of cerebral cortex, has been lacking. The present study reports novel information that the spatial distribution of pyramidal cells differs from that of non-pyramidal cells. Small, medium and large pyramidal cells are scattered, that is they are more widely spaced than expected for a normal or Poisson distribution. However, when pyramidal neurons form ‘minicolumns’ the model of a Poisson distribution has no relation whatsoever to the distribution of these cells. It is shown that neurons are very widely spaced in the horizontal plane and widely spaced in the vertical plane. But the precise mode of packing of these cells requires further investigation. Pyramidal neurons are scattered in layers 2 to 6 of ventromedial, dorsolateral and cingulate cortices. However, in motor cortex, small pyramidal neurons are clustered within layer 3 and layer 6. The finding that some non-pyramidal cells are clustered, that is are closer to neighbouring non-pyramidal cells than would be expected for a Gaussian or normal distribution, is novel. Both small and medium non-pyramidal cells are clustered within layers 1-3 and 5-6 of ventromedial, dorsolateral, motor and anterior cingulate cortices. But, in layer 4, small non-pyramidal cells are clustered within dorsolateral while being scattered in ventromedial cortex. Large non-pyramidal cells, on the other hand, are either randomly distributed or scattered. The three-dimensional spacing, rather than the density, of neurons within mammalian cerebral cortex has been investigated in only one other study (Schmitz et al., 2002) which related only to alterations in development of cortex in mice. There is therefore no comparable literature with which to compare the novel findings reported in the present study.

Outline of frontal lobe cortical connections and changes in function and behaviour after traumatic brain injury

Axons of pyramidal neurons in layers 2 and 3 extend largely to other cortical regions in the ipsi- and contralateral hemispheres (Thomson and Bannister, 2003; Lewis, 2004) and either form part of the cortico–cortico association inputs (Mori et al., 2002; Schmahmann et al., 2007) via long horizontal axon collaterals in layers 3 to 6 or the trans-callosal inputs between more distant regions with axons running via the underlying white matter (Thomson and Bannister, 2003). In addition, axons of pyramidal neurons in cortical layer 5 project to the striatum, brainstem and spinal cord. Axons of the largest of these cells in the primary motor cortex, the Betz cells, form part of the corticospinal tract even though Betz cells specifically form only about 10% of the total number of pyramidal cells in layer 5b of BA 4 (Rivara et al., 2003). Lastly, axons of pyramidal neurons in cortical layer 6 of the dorsal and lateral prefrontal cortices form bundles with a 34 μm centre-to-centre spacing within the cortical neuropil (Buxhoeveden and Casanova, 2002), and contribute to the corticothalamic radiation within the prefrontal white matter, the anterior corona radiata and the anterior limb of the internal capsule (Kraus et al, 2007).

Afferent fibres from the thalamus form the thalamocortical radiation and terminate on dendrites of pyramidal neurons within the deep part of layer 3 and layer 4 of the prefrontal cortex. Ventromedial and dorsolateral prefrontal cortices are linked through association fibres to each other and, among other cortical regions, to cingulate cortex. Studies of patients with lesions in the region of the anterior cingulum have reported psychological symptoms including apathy, inattention, change of personality, lack of distress, labile emotions, and changes in cognitive and emotional processing (Bush et al., 2000). The parallels with descriptors of patients with traumatic brain injury are striking. MRI and DTI confirm that several bundles or fascicles of myelinated fibres link the frontal cortex to other cortical regions: for example the superior longitudinal fasciculus, the uncinate fasciculus, the extreme capsule and the cingulate bundle (Mori et al., 2002; Schmahmann et al., 2007). These fibre bundles, among others, form a major volumetric component of the prefrontal white matter. In the earlier histopathological literature, lesions of DAI were reported to be most numerous and widespread in the frontotemporal white matter and at the grey-white matter boundary (Adams et al., 1989). More recently, a reduced anisotropy in frontal and temporal white matter has been reported both in patients diagnosed with post-concussional disorder (Hofman et al., 2002), and with deficiencies of attention and memory (Niogi et al., 2008). The present study extends the literature in providing quantitative evidence for loss of neurons from the overlying grey matter and provides confidence that their axons will have been lost from the frontotemporal white matter.

Within the traumatic brain injury literature, MRI and DTI have provided evidence that white matter density falls within the superior longitudinal fasciculus, uncinate fasciculus, inferior longitudinal fasciculus, sagittal stratum, anterior corona radiata and corticospinal tract after mild, and more severe head injury (Kraus et al., 2007; Niogi et al., 2008). Furthermore, DTI provides evidence that a more widespread loss of white matter occurs from the corpus callosum, fornix, internal and external capsules, corona radiata, superior frontal gyrus, superior longitudinal fasciculus, cingulum, sagittal strata and parahippocampal gyrus in moderately and severely head injured patients who survive either 90 days (Tomaiuolo et al., 2005), 6 months (Kraus et al., 2007; Kumar et al., 2009), 1–3 (Niogi et al., 2008) or 4.6–5 years (Cohen et al., 2007; Fujiwara et al., 2008) after traumatic brain injury. Also reported is a disproportionate loss of white matter from the temporal lobe after head-injury in young to middle-aged adults (Bigler et al., 2002), a selective vulnerability to loss of frontotemporal white matter in children (Wilde et al., 2005, 2006) and adults who experienced traumatic brain injury in childhood (Lipton et al., 2008), and a reduced fractional anisotropy within the internal capsule (Yuan et al.2007; Lipton et al., 2008). The present study extends that literature through the provision of rigorous quantitative evidence that the number of large and medium pyramidal neurons, whose axons are likely contribute to callosal and projection tracts, falls in ventromedial, dorsolateral, anterior cingulate and motor cortices after traumatic brain injury. Our quantitative evidence for loss of larger pyramidal and non-pyramidal neurons from anterior cingulate cortex in long term survival non-DAI and DAI patients also extends reports of changes in fractional anisotropy in the cingulum, and in particular, the differing reduced anisotropy between mild traumatic brain injury and moderate and severe head injured patients reported by Kraus et al. (2007). It is suggested that the neuronal losses described in the present study also reflect disruption or loss of white matter in the context of accepted thinking (Maxwell et al., 1997) about traumatic axonal injury and provide support for the hypothesis that disconnection of neuronal circuits within the prefrontal—anterior cingulate axis and the prefrontal—thalamic axis occurs in these patients. The finding also provides support to an earlier hypothesis (Maxwell et al., 2004, 2006) in GOS assessed patients that loss of neurons within some thalamic nuclei contributes to changes within the thalamocortical radiation. The data within the present study extend the hypothesis that neurons probably contributing to the corticothalamic radiation are lost. The quantitative data also provide support for the hypothesis that injury to, and loss of, frontotemporal white matter occurs in humans after traumatic brain injury and provides indirect evidence that loss of axons occurs from, for example, the cortico–cortico association networks, the superior longitudinal fasciculus linking dorsal motor and prefrontal cortices to medial and dorsal parietal cortices, the extreme capsule linking the ventral and lateral prefrontal cortices to the mid-portion of the superior temporal region and the cingulum. Although definitive quantitative evidence for changes in the number of axons within any of the above human tracts is lacking, it is suggested that the evidence for loss of neurons from appropriate cortical areas obtained in the present study provides strong support for such a hypothesis.

It is widely recognized that head injury results in disorders of memory, working memory, attention, executive functions, slowing of information processing and adverse changes in personality and behaviour in a majority of patients (Evans et al., 2003; Hawley, 2003; Scheid et al., 2006; Kato et al., 2007; Niogi et al., 2008, Kumar et al., 2009). In the present study novel information concerning loss of non-pyramidal local circuit and or integrative neurons is reported. Loss of large non-pyramidal cells occurred across all GOS groups in all four regions of cortex. Widespread changes in the number of medium non-pyramidal neurons occurred, with loss from ventromedial and motor cortices across all patient groups. However, there was an increased number of these cells in dorsolateral cortex of moderately disabled and cingulate cortex of severely disabled patients. Loss of small non-pyramidal cells occurred from cortical layers in ventromedial cortex while increased numbers of cells occurred within dorsolateral cortex of non-DAI and DAI patients, in anterior cingulate cortex in severely disabled and vegetative state DAI patients and in layer 6 of motor cortex of severely disabled and vegetative state non-DAI patients. Lewis (2004) reported that non-pyramidal cells are local circuit, GABAergic, inhibitory, interneurons that form some 20–25% of the total number of cortical neurons. In very simple terms, GABAergic non-pyramidal cells are thought to regulate cortical activity. Loss of interneurons, together with the concurrent loss of pyramidal neurons reported in the present study will alter cortical function and activity. This change, together with regional cortical disconnection following white matter injury will hypothetically result in widespread cortical dysfunction, which may be manifested as alteration of working memory or cognitive activity. It is suggested that the quantitative data provided in the present study provide strong support for the above hypothesis.

A widely reported psychofunctional effect of head injury is impairment of attention (Cossa and Fabiani, 1999; Gale et al., 2005; Niogi et al.2008) and this has been correlated with foci of reduced anisotropy in the dorsolateral prefrontal cortex (McAllister et al., 1999) frontal white matter (Hofman et al., 2002), white matter of the cerebral hemispheres (Tomaiuolo et al.2005), and anterior corona radiata and uncinate fasciculus (Niogi et al., 2008). There is also a consensus that ventromedial and dorsolateral prefrontal cortices are involved in executive functions while anterior cingulate cortex is the largest and most recently evolved of the limbic structures important in integration of emotion and cognition (Devinsky et al., 1995; Yount et al., 2002). However, Yount et al. (2002), using MRI, were unable to demonstrate atrophy of the anterior cingulate gyrus at 22.8 months after traumatic brain injury in patients with a mean quoted GOS of 8.33 (patients falling in the moderate to severe range). However, more recently, use of Freesurfer® software with MRI volumetric methods has provided good evidence of cortical thinning in children about 3 years after moderate to severe traumatic brain injury (Merkley et al., 2008). The present study extends the above by providing novel morphometric evidence for a reduction in cortical thickness within four regions of human cerebral cortex and changes in the number and size of pyramidal and non-pyramidal neurons within that thinned cortex over a range of severities of GOS outcome groups. In addition, the quantitative evidence for loss of cortical neurons provides strong support for the suggestion (Cohen et al., 2007) that atrophy of grey and white matter visualized by MRI-volumetry after mild to severe head injury may be correlated with loss of cortical neurons. The data obtained in the present study also provide a mechanistic explanation for the recently reported progressive cross-sectional global atrophy in both mildly (Cohen et al., 2007) and mildly to severely head-injured patients (Trivedi et al., 2007).

In conclusion, although the patient material used in the present study may show some bias toward a poorer outcome because this type of analysis can only be undertaken using post-mortem material and the mean ages of the patients at injury or death are higher that in some other studies, the novel quantitative data obtained provide strong evidence that loss of cortical neurons occurs in patients who survive traumatic brain injury. This study also provides strong evidence that greater loss of pyramidal neurons occurs from prefrontal cortex of patients diagnosed with DAI, and that the extent of loss of pyramidal and non-pyramidal neurons differs both between cortical region and with the GOS of a patient. There are changes in the somatic diameter of subgroups of neurons with the greatest change occurring for the largest cells where these are completely lost in severely disabled and vegetative state patients. The latter data also allow the suggestion that with changes in cell size a proportion of neurons may fall from a subgroup containing larger neurons into a subgroup containing smaller neurons. Generation of values of the NNI for neuronal subtypes provide the first, statistically validated data that pyramidal neurons within the cerebral cortex have a scattered distribution and that spacing between cells increases with a greater severity of outcome. Lastly, changes in the NNI, most notably in vegetative state patients, in neuronal number and neuronal diameter all strengthen the hypothesis that there is a structural explanation for thinning of frontal cortex suggested (Cohen et al., 2007; Kennedy et al., 2009) following quantitative analysis of MRI images and data.

Acknowledgements

The authors would like to acknowledge helpful discussions with the late Professor Emeritus William B. Jennett and Professor Emeritus J. Hume Adams, University of Glasgow. The authors would also like to thank Dr Ben Torsney, Department of Statistics, University of Glasgow for helpful discussions about statistical techniques. The authors would like to thank especially the referees and editors of Brain for their great help, patience and valuable suggestions during the processes of revision of the manuscript before acceptance for publication.

Supplementary material

Supplementary material is available at Brain online.

Abbreviations

Abbreviations
 
• BA

  Brodmann area

 
• CI

  confidence interval

 
• DAI

  diffuse axonal injury

 
• DTI

  diffusion tensor imaging

 
• GOS

  Glasgow Outcome Score

 
• NNI

  nearest neighbour index

 
• NND

  nearest neighbour distance

 
• non-DAI

  patient without diffuse axonal injury