Sample

Formalin-fixed brain tissue was used post-mortem from subjects who were selected on the basis of the assessment of clinical notes by a consultant psychiatrist (T. J. C. or Dr S. J. Cooper, Belfast). The DSM—IV (American Psychiatric Association, 1994) criteria for schizophrenia and schizo-affective disorder were used in the selection of cases of schizophrenia. Comparison group brains were collected prospectively from individuals who had died without a history of neuropsychiatric disorder, and whose next of kin had given consent for their tissues to be used in medical research. Pathological assessment of tissue samples was carried out by a consultant neuropathologist (M. M. E. or Dr B. McDonald, Oxford), and cases with significant neurological disorders, such as Alzheimer's disease, were excluded using the criteria of the Consortium to Establish a Registry for Alzheimer's Disease (Reference Mirra, Heyman and McKeelMirra et al, 1991).

Subsequent to sectioning and staining, the series consisted of 36 cases, comprising 10 men and 8 women with schizophrenia, and a comparison group of 9 men and 9 women.

The degree of neuroleptic medication received during life was assessed from case notes, and categorised as ‘little’, ‘average’ or ‘much’. This was based on a clinician's (T. J. C. or S. A. C.) judgement on the basis of the available clinical records, the variability of information available precluding a more quantitative estimation. A three-section categorisation of causes of death indicates that 14 subjects died of cardiac causes, 9 of respiratory causes and 12 of ‘other’ causes. Of these, 22 deaths were prolonged and 13 were sudden. One subject died of uncertain causes. Other demographic details and potentially confounding variables, including age of symptom onset, duration of illness, age at death, post-mortem interval and fixation time, were noted for statistical analysis (Table 1).

Table 1 Demographic variables and details of the brain collection for the amygdala study

Variable	Men		Women
	Comparison group (n=9)	Schizophrenia group (n=10)	Comparison group (n=9)	Schizophrenia group (n=8)
Age at onset (years)
Mean	NA	28.5	NA	34.41
s.d. (range)		7.7 (21-42)		11.3 (17-47)
Duration of illness (years)
Mean	NA	26.5	NA	40.31
s.d. (range)		14.3 (1-44)		13 (26-64)
Age at death (years)
Mean	68.9	55.0	70.8	75.8
s.d. (range)	11.0 (40-80)	13.2 (29-69)	14.3 (48-90)	13.8 (48-91)
Post-mortem interval (hours)
Mean	42.9	42.01	46.7	39.8
s.d. (range)	31.8 (10-98)	34.0 (6-113)	26.4 (19-104)	39.2 (10-128)
Fixation period (months)
Mean	25.3	46.3	28.0	67.6
s.d. (range)	15.9 (1-53)	22.4 (20-91)	17.7 (3-61)	13.6 (44-85)
Brain volume (cm3)
Mean	1144	1191	982	893
s.d. (range)	115 (995-1290)	104 (994-1378)	144 (846-1296)	144 (641-1087)

Tissue preparation

Following removal of the leptomeninges and brain-stem, and bisection of the brain, the temporal lobes were removed from the rest of the hemispheres at the posterior end of the sylvian fissure. The temporal lobe was cut into slices 5 mm thick along its anteroposterior axis. Embedding the slices in paraffin resulted in blocks of tissue approximately 3 mm thick owing to shrinkage. Blocks containing the amygdala were exhaustively sectioned from a position posterior to, and random with respect to, the posterior border of the amygdala. For a structure with a regular shape such as the amygdala, a sample of approximately 10 sections should be adequate for the estimation of volume (Reference Gundersen and JensenGundersen & Jensen, 1987). Selecting sections 25 μm thick with an interslice distance of 650 μm provided about 8-12 sections per amygdala. The sections were counterstained with Luxol fast blue, a myelin stain useful for picking out fibre tracts, and cresyl violet, which stains Nissl's substance in cell bodies (Fig. 1) (reagents obtained from Raymond A. Lamb Ltd, Eastbourne, UK).

Fig. 1 Cross-section of the amygdala and hippocampus. Staining with Luxol fast blue and cresyl violet results in dark staining of areas containing predominantly white matter fibre tracts and lighter staining of nuclei-containing cell bodies. The amygdala lies superior to and overlaps the hippocampus.

Measurement

A Cavalieri estimate of volume for each amygdala was obtained by stereological point-counting within the bounds of the amygdala on all sections in which it appeared. A 4-mm² point density grid printed on acetate was laid over each cross-section and viewed under a dissecting microscope × 7 magnification (Fig. 2). Each point represented a volume of 10.8 mm³ (4 mm × 4 mm × 0.675 mm). Three placements of the grid, each time in a random orientation, were used to obtain a mean count for each section. The total count was multiplied by the volume represented by each point for an estimate of total volume. The most anterior section containing the hippocampus and the most anterior section containing the temporal horn were also noted. This enabled a calculation to be made of the percentage of amygdala volume overlapping with each of those structures along the axis of the temporal lobe.

Fig. 2 The black crosses of the acetate counting grid appear large when superimposed over a section under a × 7 magnification. Points falling within the structure of interest such as the amygdala grey matter (circled) are counted.

All measures were performed by the same researcher (S. A. C.) who was blind to the diagnosis and gender of the cases. Left, right and mean amygdala volumes were obtained.

Pilot study

A pilot study was used to determine the grid size of points and the number of counts per section required for a satisfactory estimate of amygdala volume. Point counts were made for each of three different grid sizes, varying the number of recounts. Trial grid sizes had interpoint distances of 2 mm, 3 mm and 4 mm, respectively. The 4-mm² grid yielded a total point count of up to about 75 points per amygdala. Three counts of this grid gave a coefficient of error of 0.04, which was deemed satisfactory.

Reliability

The volume was estimated twice for 10 randomly selected amygdalas to test intrarater reliability. This provided an intraclass correlation coefficient of 0.94, suggesting that the reliability of this measure was satisfactory.

Statistics

Statistical analyses were carried out using the software Statistical Package for the Social Sciences, version 9.0 (SPSS, 1998). Unfortunately, for some amygdalas, insufficient tissue was available to provide a systematically random sample of total volume. This was due to an idiosyncratic tissue response to embedding procedures in some tissue blocks which resulted in unsatisfactory sections in a few amygdalas and prohibited their inclusion. Consequently, three cases provided volumetric data only for the right side and four cases only for the left side. Since a repeated-measures analysis of variance (ANOVA) would exclude all seven cases with one side missing, an alternative univariate ANOVA was used to test the mean of both sides for each case, replaced by the value for just one side in cases that did not have both. This approach was used to test for the between-subject effects of diagnosis and gender. A repeated-measures ANOVA was still applied to test the within-subject effects of side using those cases that had data for both sides. In addition, an asymmetry statistic was calculated for these cases: (R—L)/(R+L) × 100, where R andL are the values for the right and left sides respectively.

The influence of potentially confounding variables — age at death, post-mortem interval and fixation time — for all cases were examined as variables in separate ANOVAs of groups selected by gender and/or diagnosis. Lifetime medication of the patients with schizophrenia was examined as a factor in ANOVA of the measured variables. Clinical information was felt to be unsatisfactory for correlating symptoms or sub-syndromes with data. All variables, including covariates, fitted with a normal distribution as tested by a Kolmogorov—Smirnov goodness-of-fit test (Reference ZarZar, 1984) for each gender and diagnosis group. Homogeneity of variance of the measured variables between groups was accepted after testing with a Box's M test (Reference Glantz and SlinkerGlantz & Slinker, 2000) before each ANOVA.

Accuracy

The observed coefficients of error for the estimates of volume were less than or equal to 0.1 for both left and right amygdalas in each diagnosis × gender subgroup, indicating that their means offer a satisfactory estimate of the true population means.

Results of analysis of variance

A univariate ANOVA of mean amygdala volume found no significant difference between schizophrenia cases and controls (F=0.5, d.f. 1,30,P=0.47), no difference between genders (F=0.1, d.f. 1,30,P=0.73) and no interaction between diagnosis and gender. Analysis of covariance (ANCOVA) included fixation time, which was not a significant covariate, and age, which showed a significant negative relationship with volume (F=5.4, d.f. 1,30, P=0.03). Inclusion of brain volume in a secondary ANCOVA revealed that it was a significant covariate (F=10.8, d.f. 1,29, P<0.01) which had a positive relationship with amygdala volume, and resulted in a reduced P value in the effect of age (F=0.9, d.f. 1,29, P=0.37), suggesting a relationship with age described below. When controlling for brain volume all other results remained, including the lack of difference between patients and controls, and between the genders.

A repeated-measures ANOVA revealed no significant volume asymmetry of the amygdala and no difference in amygdala symmetry between groups. An ANCOVA found that age and fixation time were not significant covariates (although there was an interaction between fixation time and side: left amygdala smaller with longer fixation time, F=8.8, d.f. 1,22, P<0.01). Brain volume also had no effect on amygdala asymmetry.

The percentage of mean amygdala volume overlapping with the hippocampus, along the temporal lobe axis, was 56% (s.d. 15) in control brains and 63% (s.d. 17) in schizophrenia. The volume overlap with the temporal horn of the ventricles was 76% (s.d. 12) in controls and 84% (s.d. 12) in schizophrenia. No significant correlations were found between amygdala volumes and temporal horn volumes.

Artefacts and covariates

As shown in Table 1, the women (F=7.0, d.f. 1,32, P=0.01), particularly those with schizophrenia (F=3.7, d.f. 1,32, P=0.06), were older at death than the men. Consequently, age at death was included as a covariate in all tests of the measured variables. As expected, the total brain volumes were larger in the men than in the women in this series. Amygdala volume was positively correlated with brain volume, which was included as a covariate in secondary ANOVAs of the measured variables to observe the effect of controlling for brain size that might mask changes in the amygdala.

MRI limitations

Although consistent with other post-mortem findings (Reference Heckers, Heinsen and HeinsenHeckers et al, 1990; Reference PakkenbergPakkenberg, 1990) the absence of a reduction of amygdala size apparently runs counter to the weight of published neuro-imaging research. Recent MRI reviews and meta-analyses include studies from the early 1990s, since which time imaging technology has improved. The most obvious limitations of older studies is low scan resolution. Several studies provide only one or two slices through the amygdala (Table 3). Such limited sampling exacerbates the problem of delineating the amygdala, separating it from the hippocampus and the temporal horn of the ventricle. Several authors concluded that the delineation of the amygdala was unreliable and excluded it from their analysis (Reference Flaum, Swayze and O'LearyFlaumet al, 1995), or excluded the region of overlap with the hippocampus (Reference Breier, Buchanan and ElkashefBreier et al, 1992). Many studies (e.g. Reference DeLisi, Hoff and SchwartzDeLisi et al, 1991) simply included it with the hippocampus. Some studies (e.g. Bogerts et al, Reference Bogerts, Ashtari and Degreef1990, Reference Bogerts, Lieberman and Ashtari1993; Reference Seidman, Faraone and GoldsteinSeidman et al, 1999) that divided a segmentation including both amygdala and hippocampus into a posterior portion and an anterior portion (deemed to represent the amygdala) found a volume reduction only when the posterior portion was included.

Table 3 surveys the MRI studies from the period 1988-2000 that have measured the volume of the amygdala in schizophrenia. Studies that provided only one or two slices through the amygdala have been noted. The outcome of the study is reported for the measurement deemed to be closest to an amygdala volume. In some cases this is the anterior part of a hippocampus—amygdala complex. In studies in which the amygdala was defined as a separate entity, the landmark used for the posterior boundary is important. An interslice spacing has been reported for studies that measured more than one slice through the amygdala. The interslice spacing is composed of both slice thickness and, when present, interslice gap, and is split into categories of less than or greater than 3 mm, since a distance greater than 3 mm has been reported to reduce significantly the accuracy of MRI measurements (Reference Luft, Skalej and WelteLuft et al, 1996).

Limitations of meta-analysis

Although apparently consistent evidence of amygdala volume reduction emerges from compilations of studies, such meta-analyses conceal a potential source of bias. In most studies a landmark external to the temporal lobe was used either to mark the division into anterior and posterior portions, or to begin segmentation of the amygdala. Few MRI studies that provide more than two slices through the amygdala delineate it without the use of an external landmark (Fig. 4). The landmarks include the mamillary bodies (Bogerts et al, Reference Bogerts, Ashtari and Degreef1990, Reference Bogerts, Lieberman and Ashtari1993; Reference Shenton, Kikinis and JoleszShenton et al, 1992; Reference Rossi, Stratta and ManciniRossi et al, 1994; Reference Becker, Elmer and SchneiderBecker et al, 1996; Reference Hirayasu, Shenton and SalisburyHirayasu et al, 1998), the optic tract (Reference Staal, Hulshoff Pol and SchnackStaalet al, 2000), the pons (Reference Hoff, Riordan and O'DonnellHoff et al, 1992), and the anterior and posterior commissures (Reference Seidman, Faraone and GoldsteinSeidman et al, 1999). However, the temporal lobes have been reported to be preferentially shortened in schizophrenia (Reference Bartzokis, Nuechterlein and MarderBartzokis et al, 1996). For example, in the present series of brains, measurement from the pole to the posterior sylvian fissure (Reference Highley, Esiri and McDonaldHighley et al, 1998), and to the ventricular trigone (further details available from the author upon request), have found shorter temporal lobes despite controlling for brain size. In this situation, landmarks that lie outside the temporal lobe in cases of schizophrenia will be further forwards relative to structures within the temporal lobe. Consequently, the use of external landmarks in MRI studies could constitute one systematic source of error, yielding smaller estimates of amygdala volume such as are identified in the meta-analyses.

Fig. 4 The mamillary body can be used to delineate the boundary between hippocampus and amygdala. The approximate positions of the amygdala (horizontal stripes) and hippocampus (vertical stripes) are shown. The line B runs through the mamillary body orthogonal to the line (A) between the anterior and posterior commissures.

Furthermore, this study found that at least half of the volume of the amygdala overlaps the hippocampus along the axis of the temporal lobe. This emphasises the necessity of clearly differentiating the two structures to obtain accurate volumes. Although the mean amygdala volume was not changed in this study, the percentage of its volume overlapped by the neighbouring hippocampus and temporal horn was slightly more in the brains from subjects who had schizophrenia. The greater surface of amygdala bordered by the high contrast of an enlarged temporal horn in schizophrenia could lead to a more conservative segmentation of the amygdala on MRI scans.

Hemisphere and gender

No significant asymmetry of amygdala volume was found in these brains — a finding similar to that of Heckers et al (Reference Heckers, Heinsen and Heinsen1990). No difference of asymmetry was noted between patients with schizophrenia and comparison subjects. In contrast to the finding of Fukuzako et al (Reference Fukuzako, Yamada and Kodama1997), who showed that a relatively larger right hippocampus was correlated with later age of illness onset, no correlation between amygdala asymmetry and age of onset was found. This was accompanied by an absence of significant difference between the genders, similar to the findings of Bryant et al (Reference Bryant, Buchanan and Vladar1999), although the mean amygdala volume in the women seemed to be a little less than that in the men (see Table 2) — presumably associated with the normal human sexual dimorphism of total brain size.

Interpreting normal volume

The negative findings of this study (particularly in relation to imaging studies) raise the possibility of a type II error (i.e. a failure to identify a difference between groups because of small sample size). A power analysis using the Statmate program, version 1.01 (GraphPad Software, San Diego, USA) indicates that this study had about 15% power to detect a change of 5-7% of amygdala volume, as suggested by MRI meta-analyses. This rises to about 30% power to detect a change of 10% volume, and 80% power to detect a change of 20% (power=1 — p (type II error); the model was a two-samples t-test with α=0.05). The observed coefficients of error, which take into account sample size to determine how good an estimate of the true population mean is provided by the sample mean, were all satisfactory, at less than or equal to 0.1.

The absence of a gross volumetric reduction in this study does not discount a cytoarchitectural or neurochemical disturbance. Increased dopamine innervations on the left (Reference ReynoldsReynolds, 1983) and reduced binding of γ-aminobutyric acid (Reference Simpson, Slater and DeakinSimpson et al, 1989) in the amygdala in schizophrenia, have been interpreted as a loss of inhibition which might induce positive symptoms (Reference ReynoldsReynolds, 1995). Such changes may reflect a change in connectivity, for example of the dopaminergic afferent fibres, rather than an alteration in gross volume.

The study was limited by the use of elderly, medicated subjects, and Bogerts et al (Reference Bogerts, Falkal and Greve1991) have suggested that age-related brain atrophy may obscure reductions in limbic structures. In our study the negative association of age with volume could act to reduce any apparent diminution in volume in the subjects with schizophrenia, since the control subjects were on average 5 years older (means: controls 69 years, patients 64 years). However, age was controlled for as a covariate in the analyses of volume. The use of MRI can avoid these complications.

Magnetic resonance imaging is most appropriate for assessment of macroscopic measurements, while post-mortem examination is still the only option at a microscopic scale. Currently, the amygdala stands at the limit of structures that can be satisfactorily determined using MRI. Recent and improving methods of assessment, which make use of visual tracing of the amygdala—hippocampus boundary, using the alveus and local anatomical features rather than other external landmarks, should provide a more reliable source of measurements (Reference Kates, Abrams and KaufmannKates et al, 1997; Reference Niemann, Hammers and CoenenNiemann et al, 2000; Reference Pruessner, Li and SerlesPruessneret al, 2000). Further studies are required to clarify which, if any, components of the limbic system are affected in schizophrenia. The current study supports the conclusion that volume reductions of the amygdala in schizophrenia are not large, and that small reductions reported in MRI may be due to coarse delineation methods that could introduce bias to subsequent meta-analyses.

CLINICAL IMPLICATIONS

1. • The volume of the amygdala is unchanged in schizophrenia.

2. • Schizophrenia does not affect all components of the limbic system equally.

3. • Structural change elsewhere in the temporal lobes may introduce bias in magnetic resonance imaging studies.

LIMITATIONS

1. • Although satisfactory for a post-mortem study, the sample size does not exclude small differences of amygdala volume in schizophrenia.

2. • There may be alterations in other morphological aspects of the amygdala such as cellular composition.

3. • Age-related brain changes may obscure size reductions in limbic structures.