Brain autopsies

Autopsies from 110 individuals were obtained from three brain banks and each collection contained an equal number of tissue samples from individuals diagnosed with schizophrenia and individuals without psychiatric diagnosis that were used as controls. The same sample set was previously used for the analysis of myelination-related genes [8, 15]. The Stanley Foundation Brain Consortium (Bethesda, USA) contributed with brain autopsies from 30 individuals, 48 samples were provided by Maudsley Brain Bank (Institute of Psychiatry, Dept. of Neuropathology, London, UK) and 32 came from Harvard Brain Tissue Resource Centre (Massachusetts General Hospital, Massachusetts, USA). The diagnostic criterion used by the Stanley foundation was DSM-IV, whereas the Maudsley brain bank used ICD-10 and the Harvard Resource Centre used the Feighner criteria. Forty-six samples were from females and 64 were from males. In total 110 tissue samples from frontal cortex, Brodmann area 8 and 9 (from Harvard and Stanley brain banks) and the left side of the superior frontal gyrus (from Maudsley brain bank), were included in the study. The mean time post mortem was 34.3 hours for controls (standard deviation 22.6) and 34.6 hours for schizophrenic subjects (standard deviation 20.5), where as the average age of controls and affected subjects were 58.5 (18.0) and 54.7 (17.3), respectively. Fifty-five tissue samples came from controls (CTRL) and 55 from patients. Seven of the schizophrenic (SCH) individuals were treated with atypical neuroleptics (ATYP), nineteen were treated with typical neuroleptics (TYP) and eleven schizophrenic individuals were not treated with neuroleptics prior to death (NTD). For the remaining eighteen schizophrenic individuals no information regarding medication was available (UNKNW). The typical neuroleptics group included patients treated with chlorpromazine, promazine, haloperidol, thioridazine, stelazine, trifluroperazine, thiothixene and sulpiride. The atypical neuroleptics group included patients treated with clozapine and risperidone. Each autopsy sample consisted of 300 mg to 500 mg of prefrontal cortex tissue with similar amounts of grey and white matter in each sample. Moreover, duplicate autopsies were collected from each individual to control for differences in the proportion of white matter in each sample.

Demographic information about each individual has been loaded to the miame express database (ArrayExpress accession number E-MEXP-857) and a detailed description of each subject can also be found in Supplementary Table 1. Ethical approval for the use of these samples was received according to Swedish regulations (Ups dnr 99277).

Microarray hybridization

To identify genes that differed in mRNA expression levels between all schizophrenic patients and controls or between a patient subgroup and controls we carried out several different microarray experiments. Messenger RNA from all individuals was combined into five sample pools, with each individual in a pool contributing with an equal amount of mRNA. The mRNA samples were pooled as follows: patients treated with typical neuroleptics (TYP, n = 19), patients treated with atypical neuroleptics (ATYP, n = 7), patients that did not receive any neuroleptics at the time of death (NTD, n = 11), all control individuals (CTRL, n = 55), and all schizophrenic patients (n = 55). One hundred ng from each of the four schizophrenic mRNA pool was hybridized together with 100 ng mRNA from the control pool on slides printed with 29 750 human cDNA clones [16]. Detailed information on the array designs is available on ArrayExpress (accession number A-MEXP-114). Each hybridization was repeated four times, with the dye assignment reversed in the two of the hybridizations (dye swap pairs), resulting in a total of 16 microarrays.

The MICROMAX TSA™ Labeling and Detection kit (NEN^® Life Science Products, Inc.) was used to label the sample pools with Cyanine-3 (Cy-3) and Cyanine-5 (Cy-5) respectively. The TSA™ procedure was performed according to manufacturer's protocols.

Microarray analysis

The microarrays were scanned at 10 μm resolution using a GenePix 4100A scanner (Axon Instruments, Inc.). Spots on the resulting images were quantified with the software package GenePix Pro 5.0 (Axon Instruments, Inc.). The mean intensity of the two samples (Cy5-labelled sample = R, the Cy3-labelled sampled = G) were used to calculate the log-transformed ratio between the two samples for each spot: M = log₂ (R/G). Microarray data was submitted to ArrayExpress.

where β ₀ referrers to the intercept, x _i refers to the four design variables NTD, TYP, ATYP and UNKWN, and β _i to the corresponding regression coefficient estimating the mRNA difference between subgroup i and the control. The arrays containing samples from all the schizophrenic patients was represented by the sum of all four design variables, each multiplied by the fraction of individual contributing to the specific subgroups. To identify common mRNA responses in all the three subgroups with known medication status, we also tested whether the mean of the three estimated responses (β _i ) differed from zero (n = 3). We penalized all F-ratio by adding a constant (a₀) to the denominator and choose a₀ to be the 90^th percentile of the mean square errors of all analyzed clones. We considered penalized F-ratios above 7 to be indicative of a differential expression for a schizophrenic subgroup, and a penalized F-ratio above 2.2 to indicate a common response in the three subgroups as determined by a 10,000 permutations. The cut-offs values corresponded to four expected false positives for each clone list.

Overrepresentation of a biological function within a list of selected genes was assessed with GOstat [17, 18]. That is, to assess statistically overrepresentation we compared the observed frequencies of genes within Gene Ontology (GO) classes for a list of selected clones with the frequencies of all genes represented by all analyzed clones, using the FDR option to correcting for multiple testing. We only considered the biological function hierarchy, and used a minimal GO path length of five.

Real-time RT PCR

All samples, without pooling, were analysed by real-time RT PCR. Tissue sample preparation, total RNA preparation, poly-A RNA purification as well as the reverse transcription reactions were performed as previously described [19]. Briefly, total RNA was prepared using Trizol reagent (Life Technologies, Sweden), mRNA was extracted using the PolyATtract mRNA isolation system (Promega SDS, Sweden) and reverse transcription employed Superscript II (Life Technologies, Sweden). The 110 individuals, covering 55 patients and 55 controls, were distributed in duplicates on three plates according to a balanced incomplete block design, with respect to diagnosis, sex and brain bank. Real-time RT-PCR was performed with an ABI PRISM 7000 Sequence Detection System (Applied Biosystems, Foster City, USA) as follows: 2 minutes at 50°C and 10 minutes at 95°C followed by 40 cycles of 15 seconds at 95°C and 1 minutes at 60°C. Each reaction was carried out in a total volume of 25 μl, consisting of 9.8 μl TaqMan universal PCR master mix, 0.125 μM probe (Applied Biosystems, Foster City, USA), 0.25 μM of each primer (Thermo Electron Cooperation, Germany) and ~ 10–100 ng of cDNA.

The primer/probe-set for the reference genes (ACTB and GAPD) as well as for the IFITM2, IFITM3 and MAG were uniquely designed using Primer Express (Applied Biosystems, Foster City, USA). Primers for TF, SCD, SERPINA3, GBP1 and HLA-A were ordered as "Assay on Demand" (Applied Biosystems, Foster City, USA). The expression data was collected with the ABI PRISM 7000 SDS software (Applied Biosystems, Foster City, USA).

qPCR analysis

To test if the expression of the target genes were disturbed in patients suffering from schizophrenia and/or affected by patient sub-group with respect to neuroleptic treatment, we analyzed the mRNA levels with an ANCOVA model (Proc GLM, SAS 8.2) as we previously described [8]. The model included the main factors disease (SCH/CTRL), and disease subgroup (ATYP, TYP, NTD) nested within disease status. Thus, the 18 patients with unknown medication status were included in the comparisons of all patients versus controls, but they were excluded in the test of patient subgroups. In addition, we included the categorical factors brain bank, gender, and replicate PCR-plate, and the covariates, age, post-mortem time and reference gene expression (the geometric mean of ACTB and GAPD) to account for the effects of these variables on mRNA levels of the target genes. We used logarithmic transformed expression data and averaged the mRNA levels obtained from the duplicate samples from each individual prior to the statistical analysis.

The co-expression patterns of genes that were differentially expressed in schizophrenic subjects and/or varied between schizophrenic subgroups (IFITM2, IFITM3, SERPINA3, GBP1, SCD, MAG and TF) were explored for schizophrenic subjects and controls respectively. Target gene expression was normalized, prior to the analysis, by fitting the linear model described above to the observed data (Proc GLM, SAS 8.2). The difference between the observed and the fitted values (i.e. the residuals) were considered the normalized expression level, and all pair-wise Pearson correlation coefficients were calculated.

Cell cultures and interferon treatment

Human telomerase-immortalized dermal microvascular endothelial cells (TIME) [20], were cultured in EBM MV2 growth medium (Promocell). The human oligodendrocyte-derived cell line (HOG)[21] was kept in Dulbecco's modified Eagle's medium (DMEM) with low glucose and GlutaMAX (Invitrogen), supplemented with 5% fetal bovine serum and 1% penicillin/streptavidin. TIME and HOG cells were treated for 3, 8, 24 or 48 hours with either 10 ng/ml IFN-γ (Peprotech), 10 ng/ml IFN-α (Peprotech) or 1 ng/ml TNF-α (R&D), and thereafter harvested for further analysis. All cell culture experiments were performed as independent triplicates. To test if the expression levels of inflammatory genes were induced by cytokines in human cell lines, we analyzed the mRNA levels of target genes in each experiment with an ANCOVA model (Proc GLM, SAS 8.2). The model included the main factors, treatment (Cytokine addition vs CTRL), and time (3, 8, 24, 48 h) nested within treatment. In addition, we included reference gene expression (ACTB) to account for the systematic variation that was due to sample mRNA concentration. Expression levels of target and reference genes were log-transformed prior to the statistical analysis.

Analysis of expression differences that are independent of medication received

Analysis of expression differences for each of the three patient subgroups separately

After the screening for genes that were differentially expressed in schizophrenia independently of received neuroleptic treatment, we identified clones that showed a large average expression difference between schizophrenic patients and controls, as compared to random technical variation, for each of the three patient subgroups separately. (The cut-off value for the ranking statistics was adjusted so that four false positives were expected in each list due to random technical variation). We note that this analysis served a primarily exploratory purpose, as pooling does not allow for estimation of population variance. Consequently the results from any individual genes (Supplementary tables 2–5) should be interpreted with extreme caution, until mRNA levels have been quantified on an individual level. Nevertheless, results based on multiple genes are less likely to be affected by a few extreme values and our results suggest that both the number and direction of differentially expressed genes may vary between the three subgroups (Figure 1). In particular we note that the group treated with atypical neuroleptics had approximately four times as many genes showing an average expression deviations from the control group than other schizophrenic subjects, and that the distribution of clones showing increased or decreased mRNA levels strongly depended on schizophrenic subgroups (p-value < 0.0001, Fishers exact test). Although these preliminary findings need to be confirmed on an individual level similar trends have been reported from experiments with haloperidol and risperidone on rats [22].

Real-time RT-PCR

To rule out the possibility that increased mRNA levels of immune response genes in schizophrenic subjects were due to infectious disease at the time of death, we performed two additional analyses. First, a senior physician classified the death cause of all 110 subjects into two categories, "likely infectious" (bronchopneumonia, intra-abdominal sepsis, multiple organ failure, peritonitis, pneumonia, pulmonary tuberculosis) or "non-likely infectious". The data was then re-analyzed, taking account of the effect of infectious/non-infectious death cause as a factor in the statistical analysis. In a second analysis we excluded all subjects that died from a "likely infectious" cause and we re-analysed the remaining patients and controls. Both these analyses confirmed that mRNA levels of SERPINA3, IFITM2, IFITM3 and GBP1 were increased in schizophrenia and were not simply the consequence of an infection immediately prior to death. P-values for the four genes were 0.003, 0.004, 0.004, 0.023 for the analysis accounting for infectious/non-infectious death cause, and 0.0006, 0.004, 0.003, 0.035 for the analysis excluding subjects that died due to infectious causes.

Expression of immune-related genes in oligodendroglial and endothelial cell lines

As it has been suggested that schizophrenia may be the consequence of a vascular-inflammation [14], we studied the mRNA levels of SERPINA3, IFITM2, IFITM3, and GBP1 in a human endothelial cell line (TIME), before and after stimulation with TNF-α (Figure 2A). All four genes were expressed and stimulated in the endothelial cell line. The mRNA levels of SERPINA3 and GBP1 increased 80-fold and 9.4-fold in response to TNF-α addition, and the corresponding response of IFITM2 and IFITM3 was 3.8 and 3.4-fold respectively (p-values for the four genes were 0.0215, <0.0001, 0.0200 and 0.0118).

Since the expression of myelination genes is disturbed in schizophrenia, we also decided to examine whether one or more of the inflammation-related genes were expressed in a human oligodendrocyte-cell line (HOG). Unexpectedly, all four genes were expressed in HOG cells and their expression was stimulated by inflammatory cytokines (Figure 2B, C, D). In particular, the mRNA levels of GBP1 increased 24-fold (p < 0.001) with IFN-γ stimulation. GBP1 expression was also stimulated by IFN-α and TNF-α (p-values were, <0.0001, and 0.0022), but the response was less pronounced in magnitude and time. IFITM2 expression was only induced by IFN-α (p = 0.0069) and SERPINA3 was only significantly induced by TNF-α (p = 0.05).