Tumor necrosis factor promoter haplotype associated with schizophrenia reveals a linked locus on 1q44

Abstract

Using restriction fragment length polymorphism and pyrosequencing methods, we genotyped two TNFA gene promoter SNPs (−G308A, −G238A) and analyzed the haplotype structure in 24 Canadian families of primarily Celtic origin. Our results demonstrate that after correction for multiple testing based on simulations of 10 000 replicates of unlinked/unassociated data, there is evidence for association (P=0.026) of a specific haplotype (−308A, −238G) with schizophrenia and schizophrenia spectrum disorders with a family-based trimmed haplotype linkage disequilibrium test (Trimhap). Stratifying the 22 families with genome scan data by TNFA promoter haplotypes followed by reanalysis of linkage to schizophrenia throughout the genome, we identified few loci that exhibit a considerable increase in LOD/HLOD scores. A locus on chromosome 1q44 (D1S1609) demonstrated a significant increase (P=0.025) in LOD score from 0.15 to 3.01 with a broad definition of the schizophrenia phenotype and a dominant mode of inheritance. This result replicates a previously reported positive result of linkage of schizophrenia spectrum disorders to this area of the genome. We also illustrated that simulation studies are pivotal in evaluating the significance of results obtained with newer statistical methods, when multiple, but not independent, tests are performed, and when sample stratification is utilized to reduce the impact of heterogeneity or assess the interaction between loci.

Schizophrenia (SCZD [MIM 181500]) is a debilitating psychiatric disorder that affects approximately 1% of the human population worldwide. Twin, family and adoption studies are the principal methods that have been used to elucidate the genetic contribution to schizophrenia.¹ It is considered a complex disorder since it cannot be explained by a single gene effect or environmental influence alone.^2,3 The etiology of this common disorder remains unknown, although numerous regions in the human genome may influence susceptibility to the disease, as shown by several published linkage analysis studies.^3,4

One of the problems with linkage analysis studies of schizophrenia is that they rarely satisfy strict criteria for statistical significance.⁵ Markers from only two regions in the human genome have produced highly significant evidence for linkage when multiple pedigrees were used for the analysis: D1S1679 on chromosome 1q21–q22 and D13S779 and D13S174 on chromosome 13.^6,7 The difficulties in identifying susceptibility loci for schizophrenia may be explained by the intrinsic complex etiology of the disease. The phenotypes associated with schizophrenia may be caused by the simultaneous presence of multiple interacting genes with various morbid dose-effects in a population. Environmental and other heritable and nonheritable influences may also contribute to the etiology of schizophrenia⁸ and may be different for each subject, making genetic research more difficult due to incomplete penetrance in subjects with favorable environmental, epigenetic and genetic backgrounds. The presence of undetected genetic heterogeneity can greatly reduce the power to obtain a significant LOD score in a data set of multiple families if a true negative result at one locus in one family overrides a true positive result at the same locus in another family.

One approach to reduce the impact of heterogeneity is to make the data set more genetically homogenous. Traditionally, this has been attempted by limiting sample collection to subjects of a single ancestry, environment, or diagnostic subclass. Another approach is to divide an existing sample into hopefully more genetically homogeneous subsets based on an independent, reasonable criterion.

A good candidate for such a criterion might be the major histocompatibility complex (MHC) or a polymorphic marker that resides within the complex. The MHC has been studied extensively in schizophrenia and associations of several HLA alleles have been reported repeatedly in various population groups.⁹ Linkage studies also suggest there may be one or more loci on chromosome 6p that affect susceptibility to schizophrenia.^10–14 In addition, there is some evidence that genetic loci on chromosome 6p may influence the manifestation of the endophenotype of eye-tracking dysfunction¹⁵ and modify the severity of positive symptoms in affected subjects.¹⁶ An example of the utility of stratifying linkage data set on the basis of HLA genotypes (DR3 and DR4) was demonstrated in the successful identification of a causative locus for Type I diabetes mellitus on chromosome 11q,¹⁷ a finding that was confirmed by other linkage^18,19 and linkage disequilibrium studies.^20,21

The tumor necrosis factor alpha (TNFA) gene [MIM 191160] is located on chromosome 6p21.3, and polymorphisms within this gene have been shown to be in linkage disequilibrium with other genes in HLA I, II, and III.^22,23 Recently, at least two studies have provided positive results for association of the TNFA promoter −G308A polymorphism with schizophrenia in Caucasian populations.^24,25 Immunopathological findings are common in schizophrenia, with individuals sometimes showing signs of infectious factors and nonspecific inflammatory reactions that would be consistent with a possible role of TNFA dysfunction in the disease etiology.²⁶

Several polymorphisms have been identified in the TNFA gene, a number of which reside within the promoter region of the gene. We have analyzed the haplotype structure of two TNFA single-nucleotide polymorphisms (−G308A, −G238A) in 24 Canadian families of primarily Celtic origin. Our results demonstrate that after correction for multiple testing based on simulations of 10 000 replicates of unlinked/unassociated data there is evidence for association (P=0.026) of a specific haplotype (−308A, −238G) with schizophrenia and schizophrenia spectrum disorders. Furthermore, our results show that upon stratification of these families by TNFA promoter haplotypes, several loci throughout the genome exhibit considerable increases in LOD/HLOD scores. Among these is a locus on chromosome 1q44 (D1S1609), which demonstrates a significant increase (P=0.025) in LOD score from 0.15 to 3.01 with a broad definition of the schizophrenia phenotype and a dominant mode of inheritance. This result replicates a previously reported positive result of linkage of schizophrenia spectrum disorders to this area of the genome.²⁷

Material and methods

Subjects

The family sample used for this study is an extension²⁸ of a set of Canadian families that was used for a genome scan for schizophrenia susceptibility loci.⁶ Briefly, this is a sample of 24 families of Celtic (n=23) and German (n=1) origin, with 330 subjects with comprehensive phenotypic assessments and 332 subjects with DNA samples available for study. Written informed consent was obtained from all subjects after an explanation of possible consequences. Protocols were approved by the institutional review boards of Rutgers University, University of Toronto, and by the Centre for Addiction and Mental Health. Direct interviews were conducted using the Structured Clinical Interview for DSM-III-R (SCID-I) for major disorders and SCID-II for personality disorders. The interviews, collateral information, and medical records were used to make consensus diagnoses for 319 subjects based on DSM-III-R criteria. For 11 additional subjects—two alive, but unavailable for full phenotypic assessment, and nine deceased—only medical records and collateral information were reviewed, through the consensus diagnosis procedure. Details of the diagnostic and ascertainment procedures have been described previously. ^29,30 In total, 85 subjects were considered affected under a narrow diagnostic classification (schizophrenia and schizoaffective disorder) and 125 subjects under a broad diagnostic classification that also includes nonaffective psychotic disorders, schizotypal personality disorder, and paranoid personality disorder. Of these families, 22 were used for the original genome scan.^6,7 In addition to two new families (17 participating subjects), additional members of the previously studied families (n=25) were also included. Another 65 subjects without DNA samples or phenotype data were included for statistical analysis as pedigree founders and connecting individuals.

Genotyping

To assure the quality of the genotype data, two methods of DNA analysis were used. The first genotyping method used PCR amplification and restriction fragment length polymorphism detection by NcoI for the −G308A polymorphism and AvaII for the −G238A polymorphism. Primers and PCR conditions for −G308A were taken from a previously published study,²⁴ with SNP detection by overnight incubation with NcoI (New England BioLabs Inc.). The primers for identification of −G238A were designed using Primer3 (Whitehead Institute) using 6p chromosome sequence available from build 34 of the human genome assembly (UCSC). A 127-bp fragment of the TNFA promoter was amplified by PCR using primers 5′-AAAAGAAATGGAGGCAATAGGT-3′ and 5′-CACTCCCCATCCTCCCTGGTC-3′. This second primer contained a single-base mismatch with the genomic sequence, which introduced an AvaII site into the PCR product when the −238G allele was present. PCR reactions contained 40 ng of template DNA, 0.5U AmpliTaq Gold polymerase (Applied Biosystems), 0.01 μM of each primer, 0.1mM dNTP, 1.0mM MgCl₂, and 1 μl of GeneAmp 10× buffer II (Applied Biosystems), in a 10 μl volume. After 3 min at 95°C, 30 cycles were carried out at 94°C for 15 s, at 62°C for 15 s, at 72°C for 15 s, followed by a final extension step at 72°C for 3 min. The SNP was detected using overnight incubation with AvaII (New England BioLabs Inc.). Both alleles of −G238A had one permanent site for AvaII, which served as an internal control for the enzymatic reaction. The digest products were resolved by Higher Resolution Microplate Array Diagonal Gel Electrophoresis, ³¹ using a 7.5% polyacrylamide gel run at 120V for 45 min.

The second genotyping method was Pyrosequencing^™, performed as simplex assays on the automated PSQ HS96A platform.^32,33 PCR primers were designed using Primer3 (Whitehead Institute) and the sequencing primer used for the Pyrosequencing assay was designed using the Pyrosequencing SNP Primer Design Software v1.0 (Pyrosequencing). For both polymorphisms the following PCR primers were used: 5′-CTGTCTGGAAGTTAGAAGGAAACAG-3′ (left), 5′-GGACACACAAGCATCAAGGATAC-3′ (right). PCR reactions contained 40 ng of template DNA, 0.5U AmpliTaq Gold polymerase (Applied Biosystems), 0.01 μM of each primer, 1.5mM of MgCl₂, 1μl of GeneAmp 10× buffer II (Applied Biosystems), in a 10 μl volume. After 3 min at 95°C, 30 cycles were carried out at 94°C for 30 s, at 54°C for 30 s, at 72°C for 15 s, and then a final extension step at 72°C for 3 min. For the −G238A polymorphism the right primer was biotin labeled and 5′-AGACCCCCCTCGGAAT-3′ was used as the sequencing primer. For the −G308A polymorphism the left primer was biotin labeled and 5′-CCTGGAGGCTGAACCCCGTC-5′ was used as the sequencing primer. The initial concordance between the two genotyping methods was 97.6%. Discordant samples were re-genotyped by pyrosequencing, with 100% concordance with the original pyrosequencing read.

For evaluation of linkage in TNFA stratified family subsets, the previously published genome scan data were used.⁶ Genotypes were available from 288 subjects from the original 22 pedigrees for the 381 markers from the Weber Version 6.0 Screening Set. These genotypes were generated both in our laboratory and in the laboratories of the Center for Inherited Disease Research (CIDR), as previously described.⁶

Error-checking and statistical analysis

Genotyping errors are known to affect family-based tests of association,³⁴ and we therefore undertook several error-checking steps. Genotype data from TNFA −G308A and −G238A were first checked for Mendelian inconsistencies using the program Ped-check. ³⁵ Second, the genotyping error probabilities were estimated with five runs of Simwalk v2.82,^36–38 using different starting conditions. In total, 16 samples with a probability of mistyping ≥0.25 were regenotyped with pyrosequencing. For 13 samples, repeat genotyping resolved the apparent errors. Data for three other genotypes were removed from further analysis due to ambiguity of the Pyrosequencing reads. Five Simwalk2 runs were performed on the data after error checking, with a variety of initial analysis parameters, to identify the most likely haplotypes. Genotypes from both SNPs were used simultaneously for haplotype generation. The resulting haplotypes were compared across runs to identify and remove any haplotypes with markers that could not be definitively phased. Six haplotypes were removed because they could not be unambiguously phased. In total, 723 fully phased haplotypes were defined in this sample, including 658 haplotypes for subjects with available genotyping data and 65 inferred definitive haplotypes for subjects without genotyping data. No recombination events were detected between TNFA −G308A and −G238A in our sample. These haplotypes were arranged into an input format for the Trimhap program. The error-checking steps for the genome scan data used are as previously described.⁶

The trimmed haplotype test for linkage disequilibrium was performed using Trimhap.³⁹ The test is based on the construction of the sequence of ancestral haplotypes, analysis of identity by descent in the pedigree, and calculation of haplotype-sharing score for each haplotype in the sample. Due to the complexity of the pedigree structures and therefore the haplotype distribution in the sample, the reported P-value is based on a specified number of random rapid permutation replications. The replicates are formed by shuffling all observed haplotypes among the existing pedigrees within the sample without changing the pattern of the inheritance to avoid the interference of the linkage signal with the linkage disequilibrium value. The genetic models for the analysis were the same as those used for the genome scan.⁶ The following additional parameters were specified for the analysis: 200 generations since the ancestral mutation event, 10⁻⁵ mutation rate per marker per generation, 0.01 error rate per marker, 10 000 replicates for permutation analysis, proportion of the linked founder haplotypes is 1.0, and proportion of linked founder haplotypes descended from a given ancestor is 0.2.

For our original genome scan for linkage, branches of extended pedigrees that were connected through more than one individual without available DNA were removed from the main pedigrees and analyzed as separate pedigrees, to minimize inflations of the LOD score due to errors in pedigree structure, including undetected nonpaternity.⁶ This resulted in three small branches (total of 23 individuals) being removed from three pedigrees. However, for association analysis, consideration of these small branches as independent families could inflate the evidence for association. To avoid these possible errors, we reanalyzed our data excluding these three small branches. This did not significantly change the results of the association analysis and therefore for the remainder of analysis these branches were included as separate families.

Several factors may affect the significance of these association analysis results, including multiple testing under narrow and broad phenotype definitions and dominant and recessive modes of inheritance and the complexity of these pedigree structures, including the presence of one pedigree with an inbreeding loop. We therefore evaluated the significance of results from the Trimhap analysis with a simulation study. In total, 10 000 replicates with LD between the marker loci preserved as in the original data but without linkage or linkage disequilibrium to the affection status locus were generated. First, we calculated the conditional probabilities of observing each of the two alleles at the −G238A SNP given each of the two alleles at the −G308A SNP, using all the founder haplotypes stably defined by our Simwalk analyses. Next, the genotypes for the first SNP were simulated without regard to affection status using the program Simulate,^40,41 using the allele frequencies observed in our sample. Genotypes for the second SNP were then generated conditional on the allele present at the first SNP, using the probabilities calculated from all founders. Finally, each replicate was analyzed by Trimhap under the same four conditions as the real data to produce four nominal P-values, for each of the genetic models: narrow-dominant, narrow-recessive, broad-dominant, and broad-recessive. The distribution of the simulation study results was assembled to allow empiric evaluation of the real data.

Furthermore, we stratified the 22 pedigrees with genotypes available from the original genome scan into two groups for additional linkage analysis. The first group, H1 positive, included families that segregate a relatively rare haplotype H1 (−308A, −238G), and the second group included families that did not have this haplotype. Two-point linkage analysis was performed on each subset separately using the Vitesse 2.0 program.⁴² Heterogeneity testing was conducted using the HOMOG program.⁴³ The same genetic models originally used for the genome scan⁶ were used for this analysis.

All markers from the original genome scan were reanalyzed for linkage in the two subsets. Since a number of markers produced LOD or HLOD scores > 3.0 in our original genome scan, we wanted to develop a criterion for evaluation of results that would consider not just the absolute magnitude of the subset LOD/HLOD score, but also the significance of any increase in the subset LOD/HLOD over the LOD/HLOD of the entire sample. Therefore, at each marker, the maximum LOD and HLOD scores from each subset were compared to the results that had been obtained for the total sample. The difference between the whole sample LOD and HLOD and the higher of the two subset LOD and HLOD scores was recorded as ΔLOD and ΔHLOD, respectively, for that marker.

To assess the significance of the increase in subset LOD/HLOD scores, we conducted another simulation study. The program Simulate was used to generate 1000 replicates of full autosomal genome scans under the null hypothesis of no linkage, using actual allele numbers, allele frequency, and recombination fractions for each marker used for our actual scan. Then, each of the replicates was analyzed with Vitesse 2.0 and HOMOG, under the same four genetic models as the original genome scan.⁶ The pedigrees were then divided into two subsets, based on the presence (sixteen families) or absence (six families) of the H1 haplotype in the real pedigree data. Each subset in each simulated replicate was again analyzed under the four models as the real data. In each subset, the maximum LOD and HLOD score was recorded for each marker and, if greater or equal to 3.0, was compared to the maximum LOD and HLOD values at the same marker under the same model in the whole replicate. For each replicate, the maximum values of ΔLOD and of ΔHLOD from all subset LOD/HLOD scores >3.0 were recorded for each of the two subsets and arranged into a table to allow empiric evaluation of the results.

Results

Haplotype construction and trimmed haplotype analysis

Multiple runs of Simwalk defined three TNFA promoter haplotypes in this sample, assigned the labels H1 (−308A, −238G; frequency: 0.18), H2 (−308G, −238G; 0.78), and H3 (−308G, −238A; 0.04). The results of the Trimhap analysis indicated association of TNFA promoter haplotype H1 with both the narrow and broad schizophrenia phenotypes. All four tested models yielded nominal P-values of <0.05 and for three of them the P-values were <0.01. The most interesting result was obtained with the narrow definition of the phenotype under the dominant model of inheritance (nominal P=0.0052), which corresponded to an empiric P-value of 0.073 when calculated by considering only the probability of obtaining a single nominal P-value of equal or smaller size from among the four P-values generated for each replicate. However, we observed that for some of the simulated replicates, only one of the four models produced a nominally significant P-value. Therefore, we also calculated the significance of observing this overall pattern of three nominal P-values <0.01 within a single unlinked/unassociated replicate. This chance was equal to 0.026, supporting the association of TNFA promoter haplotype H1 with schizophrenia in these families. The output results of Trimhap test and simulations are summarized in Table 1.

Table 1.

Results of Trimhap test and simulations: association of schizophrenia phenotypes with TNFA promoter haplotypes

Models	Trimhap test, nominal P-value	10 000 simulations, empiric P-value
Narrow-dominant	0.0052	0.073
Narrow-recessive	0.0329	0.281
Broad-dominant	0.0082	0.105
Broad-recessive	0.0056	0.077
Overall probability	Three values ≤0.01	0.026

Linkage analysis in stratified samples

Examining the Trimhap output, we observed that the associated haplotype H1 was present only in a subset of the families in our sample. We therefore stratified the available genotyped pedigrees, for which the autosomal genome scan data were available, into two groups based on the presence or absence of the TNFA H1 haplotype.

The subset of H1-positive families comprised 16 families with 206 genotyped subjects, and in the smaller H1-negative subset there were six families with 75 genotyped subjects. In the H1-negative subset marker locus D1S1609 located at 1q44 reached a LOD/HLOD value of 3.01 with ΔLOD equal to 2.87 and ΔHLOD equal to 1.67 (Table 2).

Table 2.

Increase in LOD/HLOD scores in subsets of families and significance values based on 1000 simulations

Marker	Subset	Cytogenetic position	Model	LOD (ϕ) stratified sample	LOD (ϕ) unstratified sample	ΔLOD (p)	HLOD (á, ϕ) stratified group	HLOD (á, ϕ) unstratified group	ΔHLOD (p)
D1S1609	H1(−)	1q44	BD	3.01 (0.01)	0.15 (0.3)	2.86 (0.025)	3.01 (1, 0.01)	1.34 (0.25, 0)	1.67 (0.102)
D1S1679	H1(+)	1q22	NR	5.69 (0.01)	5.77 (0.05)	−0.08 (0.137)	5.82 (0.85, 0)	5.80 (0.95, 0.05)	0.02 (0.176)
D1S1677	H1(+)	1q22	NR	3.65 (0.05)	2.15 (0.1)	1.50 (0.037)	3.65 (1, 0.05)	2.26 (0.8, 0.1)	1.39 (0.035)
D3S3045	H1(+)	3q13.12	NR	3.30 (0.1)	2.11 (0.2)	1.19 (0.064)	3.30 (1, 0.1)	2.40 (0.75, 0.1)	0.90 (0.086)

P-values <0.05 are in bold.

In the H1-positive subset, two of the three LOD/HLOD scores of 3.0 or greater that were obtained upon the stratification were in the 1q22 region known to contain a schizophrenia susceptibility locus in our sample.⁶ The D1S1677 locus showed a significant increase in LOD score, while the LOD score for D1S1679, site of the peak in the original and fine-mapping studies,^6,44 remained virtually unchanged using this smaller sample (Table 2). The other locus, D3S3045, located on 3q13.12 reached a LOD/HLOD value of 3.30 with ΔLOD of 1.19 and ΔHLOD of 0.90 (Table 2).

Table 2 summarizes the results of the stratified linkage analysis and provides the significance level of the LOD/HLOD score increases as empiric P-values obtained from the simulation analysis. The results demonstrate that among the previously unreported loci from our sample that reach LOD scores of 3.0 or higher in one of the TNFA haplotype stratified subsets, there is one locus, D1S1609, which has a statistically significant increase in LOD score as compared to the whole sample.

Discussion

This study has demonstrated the association of a TNFA promoter haplotype with both the narrow and broad definition of schizophrenia-associated phenotypes using a family-based trimmed haplotype linkage disequilibrium test (Trimhap).³⁹ We further showed that stratification of our sample based on TNFA haplotypes followed by reanalysis of linkage to schizophrenia throughout the genome yielded significant evidence for a locus on chromosome 1q44 that is linked to schizophrenia phenotypes and was previously undetected in this sample as a whole. We also illustrated that simulation studies are pivotal in evaluating the significance of results obtained with newer statistical methods, particularly when the structure of the sample is complex, and when multiple, but not independent, tests are performed. Simulation studies are also necessary to evaluate linkage results after sample stratification, since the traditional interpretation of LOD scores in comparison to a fixed threshold may not be sufficient to fully evaluate the significance of LOD/HLOD increases observed within the subsets.

Many hypotheses regarding the etiology of schizophrenia have been proposed, one of which involves an immunological theory. The progress in immunological research techniques over the past two decades and the development of genetic methods for the investigation of polymorphisms in the genes related to the immune system has revived this theory. These developments has revealed the role of the components of the immune system in normal behavior and psychosis.⁴⁵ A recent review article has summarized the proven effects of key cytokines in the central nervous system, their possible function with respect to schizophrenia, and the results of original studies elucidating the changes in cytokine systems in schizophrenic individuals.⁴⁶ Clinical reports on the beneficial effects of the anti-inflammatory COX-2 inhibitors in combination with the regular antipsychotic treatment on the total schizophrenia psychopathology provide additional evidence that the immune dysfunction seen in schizophrenics may be part of the etiology and/or pathophysiology of the disorder.^47,48

In light of the immunological theory of schizophrenia, the MHC has long been proposed as a candidate locus that may influence the susceptibility to schizophrenia. The MHC spans approximately 4 million base pairs of DNA and has been mapped to the distal portion of 6p21.3. MHC genes are divided into three groups: Class I or HLA A, B and C; Class II or HLA DP, DQ and DR; and Class III (located between Classes I and II). The genes in the HLA complex exhibit extreme polymorphism and marked linkage disequilibrium. It is well known that genes in HLA I, II, and III are involved in the pathogenesis of a variety of diseases where infectious or autoimmune components are involved. There are many established associations of HLA polymorphisms with common diseases such as types I and II diabetes mellitus, rheumatoid arthritis, systemic lupus erythematosus, obesity, sepsis, septic shock, and acute graft rejection. HLA complexes have been studied extensively in schizophrenia, and associations of HLA I A9, A28, A10 and HLA II Dr1/DRB1*01, DQB1*0602, DRB1*04 and DRw6 with the disease have been reported in various population groups using immunological as well as genotyping techniques, although there are also numerous negative studies.⁹ Linkage studies have also suggested that there may be at least one locus on chromosome 6p that affects susceptibility to schizophrenia,^10–14 eye-tracking dysfunction,¹⁵ and/or the severity of positive symptoms.¹⁶

There are many genes of potential interest in MHC region that demonstrated association with schizophrenia phenotypes in previous studies— NOTCH4,^49–52 TNXB,⁵³ DRB1,^54–57 DQB1.^58–60 We wanted to select a single gene from the region to test for association to limit the multiple testing performed. The TNFA gene (OMIM 191160) is located in the HLA III complex, chromosome 6p21.3, and is in linkage disequilibrium with genes in HLA I, II, and III.^22,23. Plasma TNFA levels may be higher in patients with schizophrenia than controls,⁶¹ with levels normalizing in schizophrenia on treatment with antipsychotic medications.^62,63 Another study has reported that TNFA concentrations demonstrate a significant correlation with Brief Psychiatric Rating Scale and Scale Assessment of Positive Symptoms scores.⁶⁴

Recently, it was shown that the −308A allele is present significantly (P=0.0042) more often in Northern Italian individuals with schizophrenia than healthy controls.²⁴ An association study of −G308A TNFA in a mixed sample of Caucasian schizophrenic sib-pairs and parent–offspring trios provided evidence for significant association with the −308G allele with schizophrenia.²⁵ If the −G308A polymorphism is not the causative risk mutation but merely in linkage disequilibrium with the true risk mutation, then differences in the associated allele in these two populations could be explained by differences in the population genetics of these two samples. This is supported by the reported difference in the allele frequencies of this polymorphism, ranging from 0.11 for the minor allele in Northern Italians to 0.19 in the mixed Caucasian sample from Germany, Hungary, and Israel. Association to TNFA may be stronger in Caucasian populations, with Japanese,⁶⁵ Chinese,^66,67 Korean⁶⁸ and some Asian-Pacific populations⁶⁹ failing to show association between TNFA promoter markers and schizophrenia.

TNFA plays an important role in initiation and regulation of the inflammatory process. Changes in expression may trigger an anomalous response to infection, which could lead to a subtle brain injury at critical stages during development and contribute to the onset of schizophrenia latter in life. As recently reviewed, immunopathological findings are common in individuals with schizophrenia who sometimes show signs of infectious factors and a certain degree of nonspecific inflammatory reactions that are consistent with a possible role of TNFA dysfunction in disease etiology through an infection or inflammatory mechanism.²⁶ Recent research indicates the possible role of glial TNFA and other cytokines in synaptic strength control at excitatory synapses⁷⁰ and apoptosis of oligodendrocytes through glutamate excitotoxicity. ⁷¹ This provides a different kind of mechanistic link to schizophrenia that would be compatible with the increasing genetic evidence for the glutamate hypothesis.^72,73

To date, no other association study of TNFA promoter polymorphisms with schizophrenia has analyzed haplotypes in extended pedigrees. This approach provides the opportunity to increase the information content of the biallelic markers and elevate the quality control of the genotypes through robust methods of error detecting. In addition, this is the first study of TNFA that investigated a polymorphism other than −G308A.

To decrease the number of tests we chose to evaluate haplotypes in a single test rather than each SNP separately. We chose a trimmed haplotype method of association analysis³⁹ since it can be used in extended pedigrees and it evaluates haplotype data as a whole without a need to convert the haplotypes into a multiallelic pseudomarker, which would then be analyzed as unique independent alleles. Despite the strengths of this approach and the built-in permutation test of Trimhap, we have also demonstrated the utility of simulations to accurately evaluate the significance of test results when multiple tests are performed on complex pedigree structures.

The reanalysis of our genome scan data in subsets of families stratified by TNFA haplotypes produced only a four loci that exceeded a LOD score threshold of 3.0. Only two of these, D1S1609 and D1S1677, exhibited a statistically significant increase in LOD score over the results from the entire sample. D1S1609, located on 1q44, was found to be linked to the broad phenotype under a dominant mode of inheritance in the small subset not segregating haplotype H1. Linkage of schizophrenia to the telomeric portion of chromosome 1 has been reported in a large family with a balanced translocation (1;11) (q42;q14.3).⁷⁴ The family was followed for 20 years and showed an increase incidence of DSM-IV defined schizophrenia, schizoaffective disorder, and recurrent major depression in family members with the translocation. Linkage between these three phenotypes and the translocation was detected with a LOD score of 7.1. Linkage with seven cases of schizophrenia alone (other disorders coded unknown) was also significant (LOD score 3.6). In the isolated population from the northeastern part of Finland a maximum LOD score of 3.82 was detected in a three-stage genome scan at locus D1S2891, located approximately 35cM centromeric from locus D1S1609.⁷⁵ Linkage was obtained to schizophrenia/schizoaffective disorder using a dominant affected-only model. Another study performed on 134 affected sib-pairs from Finland detected linkage with a LOD score of 2.62 to the region between markers D1S439 and D1S1656.⁷⁶ Interestingly, this result was obtained using a broad phenotype definition and dominant model of inheritance—the same conditions under which we obtained significant results for linkage to D1S1609, approximately 20cM more distal on chromosome 1. More recently, the same authors conducted a chromosome 1 screening for schizophrenia loci in samples consisting of subjects from a northeastern isolate of Finland and from the rest of the country.²⁷ They found strong evidence of linkage at marker D1S2709 (LOD=3.21) located at 1q42.2, about 17cM centromeric from D1S1609. These results were obtained from the sample that did not include the northeastern isolate. In the combined sample, the authors obtained a maximum LOD score of 2.71 at the same locus. In both cases, linkage was obtained with a schizophrenia spectrum phenotype and a dominant model of inheritance.

Another interesting observation in our study relates to the locus on chromosome 1q22, previously identified during the genome scan analysis of these family samples.^6,44 The maximum heterogeneity LOD score of 6.5 with a Z-max support interval of <3cM was obtained approximately 80cM away from locus D1S1609. Upon stratification on the basis of TNFA promoter haplotypes, virtually the entire signal at 1q22 was found to fall within the subgroup that does segregate the H1 haplotype, without a significant change in LOD/HLOD at D1S1679 and a significant strengthening of the signal at the adjacent D1S1677 marker. Review of the literature and the observation that the two chromosome 1 regions (1q22 and 1q44) appear to be linked to different subgroups of families strongly suggests that there are two distinct schizophrenia susceptibility loci segregating in our sample, one associated more strongly with a narrow phenotype (1q22) and the other with a broad phenotype (1q44). The relationship of these loci to TNFA remains unclear. It is not know if there may be an interaction with TNFA with one or both of these loci, or if the different TNFA promoter haplotypes are merely serving as a proxy for another MHC locus or other population differences. Additional studies on these population subgroups will clearly be needed.