Introduction

The human hippocampus is involved in memory and regulation of affect, both of which are compromised in schizophrenia [34]. Structural magnetic resonance imaging and postmortem studies in schizophrenia have shown volume loss in the medial temporal region, especially in the hippocampus, as one of the most consistent structural abnormalities [32]. Additionally, postmortem studies showed volume loss in the hippocampal subregions in schizophrenia, which may be related to positive symptoms [12–14]. The small tissue reduction of approximately 5% suggests that some cellular subfractions within this structure might be reduced in number. However, a recent postmortem study of hippocampal gray matter in schizophrenia using the gray-level-index (GLI) method revealed no cytoarchitectonic alterations, suggesting no changes in neuronal perikarya [40]. Earlier quantitative studies of the hippocampus described reduced density of interneurons, mainly of parvalbumin-immunoreactive cells, without alterations of pyramidal neurons [9, 84]. While most of these studies reported no alterations in cell density [3, 20, 29], the density of pyramidal neurons has been shown to be decreased in schizophrenia [23]. Additionally, there are findings of both smaller [7, 83] or unchanged [35] pyramidal neuron volumes in schizophrenia. Although some authors have noted a reduction of glial cells in the hippocampus of male patients with schizophrenia [23], astrogliosis has not been detected in the hippocampus of postmortem brains from patients with schizophrenia [17, 24, 65]. However, two-dimensional cell counting in only a few sections without considering the volume of the region carries methodological limitations through the influence of volume differences (e.g., mediated by neuropil degeneration) and tissue shrinkage by fixation procedures. Additionally, problems may be caused by irregular cell shape and size, non-random orientation and cutting of cells during sectioning [79]. To date, two design-based stereologic estimates of total cell numbers in the hippocampus reported no differences between patients with schizophrenia and matched controls. It should be noted, however, that one study investigated only the total number of neurons [33], while the second study failed to differentiate between different cell types and included both patients with schizophrenia and schizoaffective patients [75]. Thus, the differentiation of neuronal and glial cell types has not actually been investigated using stereologic methods.

A large number of structural MRI studies has shown a reduced volume of the posterior part of the hippocampus [6, 14, 36, 51, 57, 73, 82], while only some others have found a pronounced reduction in the anterior part [54, 67]. In functional MRI studies, decreased activation of the posterior part of the hippocampus has been related to smooth pursuit eye movement deficits [69] and impaired verbal learning in schizophrenia [22]. Additionally, in the posterior part of the hippocampus, cornu ammonis (CA) subregions can be identified based on histological criteria using Nissl-stained sections [23]. To shed light on alterations in this region, we examined hippocampal subregions (CA1, CA2/3, CA4 and subiculum) in serial postmortem sections from the posterior part of the hippocampus using design-based stereology, posting the hypothesis that the numbers of neurons and oligodendrocytes are decreased in schizophrenia.

Material and methods

Human postmortem brains

Postmortem brains were obtained from the Düsseldorf brain collection [13]. Patients fulfilled ICD-9 criteria for schizophrenia. They had been treated with neuroleptic drugs for most of their illness. Exclusion criteria were alcohol or drug abuse and other neuropsychiatric disorders. Brains from ten patients with schizophrenia [mean age 55.1 ± 7.7 years (mean ± SEM), five males, five females, postmortem interval (PMI) 40.0 ± 17.0 h, mean disease duration 24.8 ± 7.5 years] and ten age and gender matched healthy controls without a history of neuropsychiatric disorders, alcohol or drug abuse, dementia, neurological illness, trauma or chronic terminal diseases (mean age 50.2 ± 10.1 years, 5 males, 5 females, PMI 41.4 ± 23.4 h) (Table 1) were in toto uniformly fixed in 10% phosphate-buffered paraformaldehyde for about 7 months (pH 7.0, T = 15–20°C). Then, the frontal and occipital lobes were separated from the middle part of the brain by coronal sections anterior to the genu of the corpus callosum and posterior to the splenium. These sections contain both hemispheres. The frontal and middle block including the entire hippocampus were embedded in paraffin and cut into 20-μm whole brain coronal sections on a Polycut S Leica microtome. Every 50th section (intersectional distance, 1 mm) was Nissl (cresyl violet) and myelin (luxol fast-blue) stained. Afterwards, the actual thickness of each section after histological processing was controlled by focussing the upper and lower surface with a light microscope at a magnification of 1,000×. The mean section thickness after histological processing was 18.9 ± 1.2 μm.

Table 1 List of normal controls and patients with schizophrenia investigated in the present study
Full size table

Within the hippocampal formation we analyzed the posterior part, spanning from the lateral geniculate nucleus to the level of the splenium of the corpus callosum. Subregions to be investigated separately were CA (cornu ammonis) 1, 2–3 and 4 (deep polymorph layer of the dentate gyrus) and subiculum [55, 76]. According to the literature [21], CA2 and CA3 were lumped together because these small regions are difficult to separate on the microscopic level due to histological criteria. (Note that the anatomical delineation of CA3 and CA4 is a traditional way to segment the hippocampus, but has no functional consequence, since the CA4 neurons originate in CA3 with extensions into the hilus of the dentate gyrus [1]. Thus, according to a new functional view, CA3 and CA4 neurons should be considered together as one hippocampal subregion.) We differentiated between neurons, oligodendroglia and astroglia based on histological and morphological criteria in our Nissl-stained sections (Fig. 1). Neurons were distinguished from other cell types on the basis of cytological characteristics. Neurons showed a large cytoplasm, less distinct nuclear membrane, clearly visible nucleolus within a pale nucleus and a less heterogeneous distribution of chromatin material in the nucleus. Oligodendroglial cells were identified by the absence of cytoplasmic staining, intense staining of the nucleus with dispersed chromatin and lack of a nucleolus [72]. Astrocytes were less densely stained and showed a clear rim of cytoplasm [38, 62, 64]. We determined densities and numbers of these specific cell types along with the volumes of the entire subregions within the posterior part of the hippocampus.

Fig. 1
figure 1

Representative high-power photomicrograph of a 20 μm-thick coronal section from a control. The arrows point to neurons, the arrowheads to astrocytes and the asterisks to oligodendrocytes. Scale bar = 25 μm

Full size image

Stereologic analyses

All stereological analyses were performed with the rater (C.St.) blind to diagnosis using a stereologic workstation, consisting of a modified light microscope (BX50; Olympus, Tokyo, Japan), Olympus Uplan Apo objectives (1.5 × , 20 × , 50 × oil, 100 × oil), motorized specimen stage for automatic sampling, electronic microcator, CCD color video camera, PC with frame grabber board, stereology software (Stereoinvestigator, MBF Bioscience, Williston, USA) and a 17-in. monitor. Boundaries of hippocampal subfields were traced on video images displayed on the computer screen (Fig. 2) and total volumes were calculated according to Cavalieri’s principle [59] using on average 10.1 ± 2.3 sections in controls and 10.0 ± 1.6 sections in patients with schizophrenia. Total numbers of cells were estimated using the optical fractionator [58–60, 77]. We performed a count of the different cells (neurons, oligodendroglial cells, astrocytes) that came into focus within unbiased virtual counting spaces [26, 66] distributed in a systematic-random fashion throughout the different regions of interest. Estimated cell numbers were calculated from the numbers of counted cells and the sampling probability. The predicted coefficient of error (CE) of the total numbers of neurons, astrocytes and oligodendrocytes was evaluated using the prediction methods described by Schmitz [58] and Schmitz and Hof [60]. Details of the stereologic counting procedure are outlined in Table 2.

Fig. 2
figure 2

Representative photomicrograph of a 20 μm-thick frontal section from a control, showing the subregions of the posterior part of the hippocampus investigated in the present study. Scale bar = 1 mm

Full size image
Table 2 Details of the stereologic counting procedures
Full size table

The rater was blind to the diagnosis and performed repeated measurements of both the left and right CA2/3 in eight cases. For the intra-rater reliability the intraclass correlation coefficients (ICC) were 0.982 for astrocytes, 0.992 for oligodendrocytes, 0.984 for neurons and 1.00 for volumes of the subregions.

Statistical analysis

For both, patients with schizophrenia and controls, mean and standard error of the mean were calculated for all investigated variables, separately for the left and the right hemispheres. Comparisons between patients with schizophrenia and controls were performed using generalized linear model multivariate analysis (MANOVA), with diagnosis and hemisphere as fixed factors and the following variables as covariates: (i) the patients gender, (ii) the postmortem interval and (iii) the adjusted illness duration of the patients with schizophrenia (calculated as individual age at death minus age at onset plus the mean age at onset of all schizophrenic patients) or the age of the controls, respectively [Note that use of the actual individual illness duration of the patients with schizophrenia instead of the adjusted ones as covariate would have caused invalid results of the MANOVA model because there was no illness duration of the controls, and the mean illness duration of the schizophrenic patients was significantly different from the mean age of the controls (Student’s two-tailed t test; P < 0.001), whereas the mean adjusted illness duration was not (P = 0.974)] (see also [46]). For each investigated variable, all investigated brain regions were tested simultaneously. Post hoc tests in the analyses of covariance were performed with linear regression analysis (in case of the adjusted illness duration of the patients with schizophrenia or the age of the controls, respectively), or with two-way repeated measures (RM) ANOVA (in case of the patients gender). In all analyses, an effect was considered statistically significant if its associated P value was smaller than 0.05. Calculations were performed using SPSS (Version 12.0.1 for Windows, SPSS, Chicago, IL).

Results

No significant differences between the patients with schizophrenia and the controls were found with respect to the mean volumes, the mean densities of neurons and the mean numbers of neurons in the investigated subregions of the posterior part of the hippocampus (Figs. 3, 4, 5; Fig. S1 in Electronic Supplement Material; Table 3; Table S1 in Electronic Supplement Material show all P values of the MANOVAs performed).

Fig. 3
figure 3

Representative high-power photomicrographs of 20 μm-thick coronal sections of the dorsal part of the hippocampus of a patients with schizophrenia (a, c, e) and a control (b, d, f), showing details of the subregions CA4 (a, b), CA2/3 (c, d) and CA1 (e, f). Note that the reduced number of oligodendrocytes in CA4 in schizophrenia was not necessarily obvious at simple observation of the tissue but necessitate the type of rigorous quantitative analysis undertaken in this study to be revealed. Scale bar = 50 μm

Full size image
Fig. 4
figure 4

Volumes of CA1 (a–c), CA2/3 (d–f), CA4 (g–i) and the subiculum (k–m) in both hemispheres of the brains from ten patients with schizophrenia (S open bars in a, d, g, k, and dots and squares in c, f, i, m) and ten age-matched controls (C; closed bars in a, d, g, k, and dots and squares in b, e, h, l). In a, d, g, k, data are shown as mean and standard error of the mean for the left (l) and right (r) hemispheres from patients with schizophrenia (S-l and S-r) and controls (C-l and C-r). In b, c, e, f, h, i, l, m, individual data for the left hemispheres (closed dots males, open dots females) and right hemispheres (closed squares males, open squares females) from controls (b, e, h, l) and patients with schizophrenia (c, f, i, m) are shown as function of the controls age (or the illness duration of the patients with schizophrenia, respectively)

Full size image
Fig. 5
figure 5

Number of neurons in CA1 (a–c), CA2/3 (d–f), CA4 (g–i) and the subiculum (k–m) in both hemispheres of the brains from ten patients with schizophrenia (S; open bars in a, d, g, k, and dots and squares in c, f, i, m) and ten age-matched controls (C; closed bars in a, d, g, k, and dots and squares in b, e, h, l). In a, d, g, k, data are shown as mean and standard error of the mean for the left (l) and right (r) hemispheres from patients with schizophrenia (S-l and S-r) and controls (C-l and C-r). In b, c, e, f, h, i, l, m, individual data for the left hemispheres (closed dots males, open dots females) and right hemispheres (closed squares males, open squares females) from controls (b, e, h, l) and patients with schizophrenia (c, f, i, m) are shown as function of the controls age (or the illness duration of the patients with schizophrenia, respectively

Full size image
Table 3 Results of statistical analysis (P values) with generalized linear model multivariate analysis of variance (MANOVA)
Full size table

Neither the mean density nor the mean number of astrocytes was significantly different between the patients with schizophrenia and the controls in any of the investigated subregions of the posterior part of the hippocampus (Fig. 6; Fig. S2 in Electronic Supplement Material, respectively). However, multivariate analysis revealed an effect of gender (F = 5.088; degree of freedom = 4.0 [this was the case in all MANOVAs applied]; P = 0.03) and age (controls) or duration of the disease (patients) on the number of astrocytes (F = 13.621; P < 0.001) as well an effect of age (controls) or duration of the disease (patients) on the density of astrocytes (F = 9.014; P < 0.001). Additionally, we found a significant, positive relation between (i) the density of astrocytes and the controls age in CA4 and the subiculum, (ii) the density of astrocytes and the age of the patients with schizophrenia in CA1 and the subiculum, (iii) the number of astrocytes and the controls age in CA2/3, CA4 and the subiculum, and (iv) the number of astrocytes and the age of the patients with schizophrenia in CA1 and the subiculum, respectively (Fig. 6; Fig. S2 in Electronic Supplement Material, respectively).

Fig. 6
figure 6

Number of astrocytes in CA1 (a–c), CA2/3 (d–f), CA4 (g–i) and the subiculum (k–m) in both hemispheres of the brains from ten patients with schizophrenia (S; open bars in a, d, g, k, and dots and squares in c, f, i, m) and ten age-matched controls (C; closed bars in a, d, g, k, and dots and squares in b, e, h, l). In a, d, g, k, data are shown as mean and standard error of the mean for the left (l) and right (r) hemispheres from patients with schizophrenia (S-l and S-r) and controls (C-l and C-r). In b, c, e, f, h, i, l, m, individual data for the left hemispheres (closed dots males, open dots females) and right hemispheres (closed squares males, open squares females) from controls (b, e, h, l) and patients with schizophrenia (c, f, i, m) are shown as function of the controls age (or the illness duration of the patients with schizophrenia, respectively). In c, e, h, l, m, results of linear regression analysis (regression lines, 95% confidence intervals of the regression lines, regression coefficients and corresponding P values, respectively) are shown, indicating statistically significant relations between the number of the astrocytes and the controls age (or the illness duration of the patients with schizophrenia, respectively)

Full size image

MANOVA also showed significant effects of diagnosis (F = 3.587; P = 0.017) and gender (F = 3.841; P = 0.012) on the mean numbers of oligodendrocytes. Subsequent univariate tests demonstrated then that, compared to the controls, the patients with schizophrenia showed a significantly reduced mean number of oligodendrocytes only in CA4 [−27.6% on the left side (l) and −37.8% on the right side (r); F (1) = 6.767 (one degree of freedom); P = 0.014 for the fixed factor Diagnosis in the MANOVA] (Fig. 7; Fig. S3 in Electronic Supplement Material). For CA1, CA2/3 and the subiculum, no significant differences in the mean number of oligodendrocytes were found between the patients with schizophrenia and the controls. Note that the reduced number of oligodendrocytes in CA4 in schizophrenia was not necessarily obvious at simple observation of the tissue (as shown in Fig. 3), but necessitate the type of rigorous quantitative analysis undertaken in this study to be revealed.

Fig. 7
figure 7

Number of oligodendrocytes in CA1 (a–c), CA2/3 (d–f), CA4 (g–i) and the subiculum (k–m) in both hemispheres of the brains from ten patients with schizophrenia (S; open bars in a, d, g, k, and dots and squares in c, f, i, m) and ten age-matched controls (C; closed bars in a, d, g, k, and dots and squares in b, e, h, l). In a, d, g, k, data are shown as mean and standard error of the mean for the left (l) and right (r) hemispheres from patients with schizophrenia (S-l and S-r) and controls (C-l and C-r). In b, c, e, f, h, i, l, m, individual data for the left hemispheres (closed dots males, open dots females) and right hemispheres (closed squares males, open squares females) from controls (b, e, h, l) and patients with schizophrenia (c, f, i, m) are shown as function of the controls age (or the illness duration of the patients with schizophrenia, respectively). *P < 0.05 for the fixed factor Diagnosis in general linear model multivariate analysis of variance (MANOVA)

Full size image

The postmortem interval and the hemispheres had not any influence on the investigated variables. In contrast, MANOVA showed a significant effect of gender on the mean volumes of the subregions of the posterior part of the hippocampus (F = 2.930; P = 0.037), as well as on the mean numbers of neurons within these subregions (F = 3.295; P = 0.024). Subsequent univariate analysis showed that the gender of the controls and the patients with schizophrenia had a significant influence on the volume of CA4 as well as the number of neurons in CA2/3, CA4 and the subiculum. By pooling the data from the controls and the patients with schizophrenia, we found that on average the females had a significantly smaller volume of CA4 than the males [−28.9% (l) and −23.8% (r); P hemisphere = 0.097; P gender = 0.019; P interaction = 0.148] as well as significantly smaller numbers of neurons in CA2/3 [−30.9% (l) and −16.8% (r); P hemisphere = 0.389; P gender = 0.044; P interaction = 0.019] and CA4 [−29.5% (l) and −23.7% (r); P hemisphere = 0.132; P gender = 0.035; P interaction = 0.154] but not in the subiculum (Hemisphere = 0.892; P gender = 0.068; P interaction = 0.654) (Fig. S4 in Electronic Supplement Material).

Discussion

The present postmortem study revealed no differences in mean neuron numbers or neuron densities in the subfields of the posterior part of the hippocampus between patients with schizophrenia and controls. In line with our results, Heckers et al. [33] and Walker et al. [75] also did not detect differences in neuronal numbers in all cornu ammonis subdivisions of the total hippocampus in schizophrenia using a stereologic approach. Thus, we could not confirm earlier findings of a decreased density of pyramidal cells in the hippocampus in schizophrenia [23]. In contrast, Benes et al. [10] found reduced numbers of pyramidal neurons in CA1 of the posterior part of the hippocampus in patients with schizophrenia which also suffered from mood disturbances. Since we did not investigate a group of patients with affective disorders, it cannot be excluded that neuron number is altered in patients with mood disturbances. Additionally, we did not distinguish between pyramidal neurons and interneurons, because these cell types cannot be separated in Nissl-stained sections due to overlapping morphological criteria.

We detected effects of age on density and number of astrocytes in several subregions, but no effects of diagnosis on astrocyte number. Our findings confirm animal studies showing increased mean astrocyte numbers in brains of aged mice [50], and the well-known lack of astrocytosis in schizophrenia [17, 24]. Thus, the increasing number of astrocytes in the ageing brain shown in our study might serve as internal validity control for the results from the stereologic investigations. Furthermore, we did not find effects of age on neuronal number in our study. Another study showed that age had only little influence on this parameter in the human brain [53]. However, we did not confirm a decline in the number of neurons in the hilus and subiculum of the hippocampus, which has been shown by West and colleagues [32]. Altogether, age-related effects on neurons may reflect neuronal dysfunction rather than cell loss [49].

According to our results of fewer neurons in CA4 and CA2/3 in the brains from females compared to males, male gender is known to be related to larger neuron numbers in the subiculum of the rat hippocampus [2], although interspecies differences may render comparisons between rats and humans difficult. While a stereologic study in human cortical regions did show a higher neuronal number in men [63], these results may not be related to changes in the hippocampus. Since the size of our samples of males and females was small, our results should be taken with caution in this regard and confirmed in a larger cohort.

We found a decrease in the mean number of oligodendrocytes in the left and right CA4 of the posterior part of the hippocampus in schizophrenia compared to controls. Among hippocampal subfields, CA4 is the deep polymorph layer of the dentate gyrus and receives collateral mossy fibers from the granule cells [1]. CA4 is among the regions especially suggested to be involved in disturbances of connectivity in schizophrenia [30]. To the best of our knowledge, these are the first findings of reduced oligodendrocyte numbers in any subfield of the posterior part of the hippocampus in schizophrenia. These results point in the same direction as the noted decrease in the mean number of oligodendrocytes in the prefrontal cortex and thalamus of patients with schizophrenia, and, interestingly, in patients with major depression and bipolar disorder [16, 37, 38, 64, 68, 70, 74]. Thus, specifity for diagnosis remains unclear in our sample lacking other psychiatric disorders.

A low number of oligodendrocytes might be associated with a defect in proliferation, disturbed maturation of oligodendrocytes and lack of normal myelination [5]. In the hippocampus, ultrastructural investigations in schizophrenia revealed atrophy of axons and swelling of periaxonal processes [71]. Oligodendrocytes are involved in the myelination process of axonal projections during development. Myelination of projections in the frontal and temporal lobes, including the posterior part of the hippocampus, in turn peaks in late adolescence and early adulthood, occurring in close temporal proximity with the onset of schizophrenia [8, 11, 81]. Interestingly, polymorphisms in the neuregulin-1 gene, which is involved in myelination and proliferation of oligodendrocytes, have been reported to be associated with schizophrenia [18, 31, 61].

One of the most pronounced decreases in expression of oligodendrocyte-related proteins have been reported in the hippocampus of patients with schizophrenia [19, 42]. These microarray data and our light microscopic findings in the posterior part of the hippocampus are supported by in vivo diffusion tensor imaging (DTI) studies of white matter tracts of the fornix body and posterior part of the hippocampus in schizophrenia, showing decreased fractional anisotropy and supporting the hypothesis of functional dysconnectivity in schizophrenia [41, 47, 78, 85]. Moreover, cognitive function such as verbal declarative memory is correlated with the fractional anisotropy of the hippocampus in schizophrenia [48]. This disruption of connectivity in this region could result in cognitive as well as clinical negative symptoms [28, 47, 56]. This is due to the fact that neuronal fibers traversing the limbic pathways from the posterior part of the hippocampus are connected to prefrontal regions and pathways involved in higher cognition [25]. The prefrontal cortex and posterior part of the hippocampus belong to a neuronal network showing disturbances in schizophrenia, most likely of neurodevelopmental origin [15].

One major limitation of our study is the partial sampling strategy investigating only the posterior part of the hippocampus, while the very posterior tail of the hippocampus was not always completely included. Additionally, the posterior part of the hippocampus has been defined using external landmarks. The definition, therefore, may vary slightly depending on slight differences in the angle of the cut surfaces as well as differences in the volumes of the landmarks. However, meta-analyses of regional brain volumes did not find reductions in the volume of the lateral geniculate nucleus in schizophrenia. The corpus callosum was reduced in volume mainly in first-episode patients, while medicated patients with chronic schizophrenia showed larger volumes [4, 21, 39, 80]. Another limitation is the treatment of patients with neuroleptics, which may influence cell numbers and volumes of hippocampal subregions. In animal studies, antipsychotics reduced astrocytes and, to a non-significant degree, also oligodendrocytes, while increasing neuronal density [44, 45]. In the hippocampus, cell proliferation has been reported to be increased after treatment with atypical antipsychotics [27, 43, 52]. In our study, however, the decreased number of oligodendrocytes in CA4 in the posterior part of the hippocampus of patients with schizophrenia was not related to either the duration of the disease and, thus, the probable duration of medication. Further studies on animal models are needed to investigate the influence of long-term antipsychotic medication on cell number in the posterior part of the hippocampus.

While it should be noted that the postmortem intervals in our study sample were rather long, we did not find any influence of the postmortem interval on any of the variables investigated. Accordingly, we opted not to exclude cases with longer postmortem intervals.

In summary, we found decreased mean numbers of oligodendrocytes in the left and right posterior hippocampal subfield CA4 in schizophrenia using a design-based stereologic approach. Our results point to disturbances in connectivity of subregions of the posterior part of the hippocampus and confirm earlier findings in the literature of hippocampal pathology in schizophrenia. Further studies investigating the total hippocampus in larger samples and immunohistochemical approaches are called for to replicate these results. Furthermore, animal studies should be conducted to investigate effects of antipsychotic treatment in more detail.