The role of genetics in the etiology of systemic lupus erythematosus (SLE) is firmly established (1, 2). However, it has been difficult to identify the causal genetic factors. Many genes that are important for immune function have been studied as candidates. These studies have found clear evidence of an association with SLE for only a few factors: the histocompatibility antigens, some members of the complement cascade, and the low-affinity IgG receptors.
A different approach in the investigation of the genetics of SLE that has been followed by 4 research groups is to perform genome-wide linkage analyses in families of patients with SLE (1, 2). The results of these studies define a complex picture; none of the genetic factors have major effects, but it seems likely that many loci with small effects are involved (more than 50 loci have shown some degree of linkage). Significant linkage has already been found with 7 loci, and some of these have been replicated in different studies, yielding further support to their role in SLE. However, it has proved difficult, as in other complex diseases, to progress from linkage results to identification of the causative polymorphisms. A recent report by Prokunina et al (3) indicated that, although difficult, this progress will be possible. The authors had initially identified a locus on chromosome 2q37 that was strongly linked to SLE susceptibility (4, 5) in a genome-wide linkage study involving multicase families of Northern European descent. On the basis of this linkage result, Prokunina et al started to search for candidate genes in the 2q37 region. The authors found a very good candidate in the programmed cell death gene PDCD1.
PDCD1 is a member of the CD28 family of receptors, which are type I membrane proteins with an important function in modulating lymphocyte activation (6, 7). The most distinctive feature of PDCD1 is an immunoinhibitory domain in its intracellular tail, that is, an immunoreceptor tyrosine-based inhibition motif (ITIM), which is shared with a diverse group of inhibitory receptors. Upon tyrosine phosphorylation at the ITIMs, these molecules recruit SH2 domain–containing phosphatases, such as SH2-containing tyrosine phosphatase 1, and negatively regulate cell activity. Specifically, upon activation, PDCD1 is expressed by T and B lymphocytes, and when it binds to its ligands, PD-L1 and PD-L2, which are expressed in many tissues, it attenuates T and B cell responses.
This function is critical for the prevention of autoimmunity, as has been shown in 2 mouse strains made deficient in PDCD1 by homologous recombination. In the first strain, BALB/c-PDCD1−/−, the mice developed an autoimmune myocardiopathy mediated by antibodies against cardiac troponin I (8). The second strain, C57BL/6-PDCD1−/−, had a more relevant autoimmune phenotype for SLE, because this strain showed a late-onset, chronic, progressive, lupus-like glomerulonephritis with IgG3 and C3 deposition accompanied by other autoimmune features, including arthritis and splenomegaly (9). These experimental mouse strains have shown that PDCD1 is a necessary negative regulator of self-reactivity and is a good candidate gene in the predisposition to SLE.
Prokunina et al searched for polymorphisms in the PDCD1 gene that were overrepresented in SLE families. A single-nucleotide polymorphism (SNP) in intron 4, called PD1.3, was found to segregate with the disease. An association analysis of familial and sporadic cases of SLE found a consistent overrepresentation of the A allele among Swedish, European American, and Mexican patients with SLE. The functional consequence of the PD1.3 SNP would be to interfere with binding of a transcription factor, RUNX1, that would modify PDCD1 expression.
The identification of the role of the A allele of PD1.3 seems very significant because of the sparse data on the genetics of SLE, and because it can be expected that this finding will trigger a series of exciting discoveries about the pathogenesis of SLE and will generate new opportunities to manage the disease. This association has already been used to support the hypothesis of a special role of RUNX1 in regulating autoimmunity (10–12) and to propose a role for PDCD1 in SLE nephritis (13). However, in the present study, we found that the described relationship between the PD1.3 SNP and SLE susceptibility was not evident in Spanish patients. Therefore, new investigations will be needed to understand the true nature of the role of PDCD1 in SLE predisposition.
- Top of page
- PATIENTS AND METHODS
In an attempt to replicate the association study performed by Prokunina et al, we explored the possible association between the PD1.3 A allele and SLE susceptibility. We found that in contrast to the previously reported results, the PD1.3 A allele was significantly less frequent in Spanish patients with SLE than in Spanish controls (P = 0.006) (Table 2). Moreover, the frequency of the A allele in the Spanish controls (12.9%) was notably larger than that observed in any of the populations previously studied (in which it was 5% or lower).
Table 2. Frequency of the A allele of the PD1.3 single-nucleotide polymorphism in Spanish controls and Spanish patients with systemic lupus erythematosus (SLE)*
|Controls||140/1,076 (13.0)||66/522 (12.6)||206/1,600 (12.9)|
|SLE patients||84/930 (9.0)||12/102 (11.8)||97/1,036 (9.4)|
|Odds ratio (95% confidence interval)||0.67 (0.50–0.89)||0.9 (0.5–1.8)||0.70 (0.54–0.90)|
Therefore, we considered it necessary to check the accuracy of our results. We used 3 criteria to check accuracy. We sequenced every sample that had yielded ambiguous melting curves. We sent 72 samples to be cross-checked by authors of the original report on PD1.3 (L. Prokunina and M. Alarcón-Riquelme, Uppsala University, Uppsala, Sweden). Finally, we checked the genotype frequencies for concordance with Hardy-Weinberg equilibrium. When we applied all 3 criteria, our results were found to be accurate and we confirmed that the technology used, analysis of the melting curve with FRET probes, was reliable.
The decreased frequency of the A allele in Spanish patients with SLE was restricted to the female patients (9.0% versus 13.0% in the female controls; P = 0.005), whereas the male patients with SLE had a frequency of A alleles similar to that in the male controls (11.8% versus 12.6%) (Table 2). Nevertheless, a small difference between the male patients and male controls could not be excluded completely, due to the small number of male patients with SLE (only 51 among the 518 SLE patients). It was not possible to compare this sex effect with previous data, because only data on female subjects were available.
When we focused on the results in the female population, the decreased frequency of the A allele of PD1.3 in SLE patients, although moderate (as reflected by a 0.67 odds ratio), was consistently observed across the 5 sets of SLE cases and controls grouped by town of recruitment (Table 3). Town-specific odds ratios showed only moderate and expected fluctuations, ranging from 0.5 to 0.8. These results provided evidence against spurious differences due to unappreciated population stratification, because such an artifact would be very unlikely across 5 independent sample sets. There was no geographic trend in the frequency of the A allele or in the odds ratios along the North-South or the East-West axes of Spain.
Table 3. Frequency of the A allele of the PD1.3 single-nucleotide polymorphism in Spanish female subjects, grouped by town of recruitment*
|Controls||24/212 (11.3)||52/406 (12.8)||23/142 (16.2)||18/174 (10.3)||23/142 (16.2)|
|SLE patients||12/166 (7.2)||14/202 (6.9)||26/246 (10.6)||9/102 (8.8)||23/214 (10.7)|
|Odds ratio (95% confidence interval)||0.6 (0.3–1.3)||0.5 (0.3–0.9)||0.6 (0.3–1.1)||0.8 (0.4–1.9)||0.6 (0.3–1.2)|
There was some evidence for a gene-dose effect. Susceptibility to SLE among the female subjects with the AA genotype seemed much lower than that among the female subjects with the AG genotype; the odds ratio for the AA genotype was 0.3 (95% CI 0.1–1.0), whereas the odds ratio was 0.7 for the AG genotype (95% CI 0.5–1.0). However, this difference in SLE susceptibility between the AA and AG genotypes was not statistically significant.
We also explored the other 6 PDCD1 SNPs that were studied by Prokunina et al (3). The frequencies of the PD1.5 C allele and of the PD1.6 A allele were similar between the Spanish patients with SLE and the Spanish controls (56.9% versus 55.3%, odds ratio 1.1, 95% CI 0.9–1.3 and 11.6% versus 11.5%, odds ratio 1.0, 95% CI 0.8–1.3, respectively). There were no significant differences between male and female subjects for these SNPs, and they showed genotype distributions in agreement with Hardy-Weinberg equilibrium. The PD1.4 SNP was in complete linkage disequilibrium with PD1.5, and only PD1.5 was studied in all samples. The remaining 3 SNPs, PD1.1, PD1.2, and PD1.9, were in complete linkage disequilibrium. The allele frequency of their minor alleles was 1.1% in SLE patients (95% CI 0.3–3.9%; a total of 186 chromosomes studied) and 0% in controls (95% CI 0.0–1.9%; a total of 186 chromosomes studied). Due to their rarity, these latter 3 SNPs were not studied further.
We compared the allele frequencies of the PD1.3, PD1.5, and PD1.6 SNPs in Spanish female controls and in the female control populations studied by Prokunina et al (Table 4). There were significant differences in 2 SNPs: PD1.3 and PD1.6. As already mentioned, the PD1.3 A allele was significantly more frequent in Spanish controls than in any of the other control populations; its frequency was double that reported in the Swedish and European American sets, and was 6-fold higher than the frequency in Mexican controls. Regarding the PD1.6 SNP, the frequency in Spanish controls was more than half that reported in European Americans and much lower than that reported in Mexicans. The PD1.6 A allele seemed, however, marginally more frequent in the Spanish control subjects than in the Swedish control population, with the difference approaching statistical significance (P = 0.06).
Table 4. Differences in allele frequencies between the Spanish female control population and female controls from the study by Prokunina et al (3)*
|Population||PD1.3 A||PD1.5 C||PD1.6 A|
|n/N (%)||P†||OR (95% CI)||n/N (%)||P†||OR (95% CI)||n/N (%)||P†||OR (95% CI)|
|Spanish||140/1,076 (13.0)|| || ||540/984 (54.9)|| || ||103/956 (10.8)|| || |
|Swedish||41/602 (6.8)||0.9 × 10−4||2.0 (1.4–2.9)||220/404 (54.4)||0.9||1.0 (0.8–1.3)||47/598 (7.9)||0.06||1.4 (1.0–2.0)|
|European American||8/160 (5.0)||0.0035||2.8 (1.4–5.9)||84/160 (52.5)||0.6||1.1 (0.8–1.5)||30/160 (18.75)||0.004||0.5 (0.3–0.8)|
|Mexican||12/538 (2.2)||0.3 × 10−11||6.6 (3.6–11.9)||215/360 (59.7)||0.1||0.8 (0.6–1.0)||168/348 (48.3)||<1 × 10−16||0.13 (0.1–0.2)|
To complete this analysis, we estimated haplotype frequencies of PDCD1 among the female subjects, using the data on PD1.3, PD1.5, and PD1.6. We found subjects bearing 7 of the 8 possible haplotypes (Figure 1). The haplotype frequencies were different between the female patients with SLE and the female controls, as determined by nonparametric heterogeneity test (P = 0.04) or by chi-square test, after grouping haplotypes with low expected values (P = 0.03). The difference was mainly due to a lower frequency of the ACG haplotype (PD1.3 A, PD1.5 C, and PD1.6 G, respectively) and a higher frequency of the GCG haplotype in SLE patients. The ACG haplotype included almost all PD1.3 A alleles, and therefore it was equivalent to the allelic data on PD1.3. In contrast, the GCG haplotype was the most common haplotype and was the only one overrepresented in SLE patients (P = 0.03, odds ratio 1.2, 95% CI 1.0–1.5, compared with the other haplotypes). This difference suggested that an SLE susceptibility polymorphism could be linked to this haplotype in Spanish patients.
Figure 1. Haplotype frequencies of PDCD1 as defined by the PD1.3, PD1.5, and PD1.6 single-nucleotide polymorphisms (SNPs). Haplotypes are indicated according to the allele at the PD1.3, PD1.5, and PD1.6 SNPs, respectively. Haplotype frequencies for the Spanish female controls and Spanish female patients with systemic lupus erythematosus (SLE) were estimated from 918 and 772 chromosomes, respectively. Haplotype frequencies for the Swedish female controls were modified from Table 1 of reference3, after correction of some errors following indications from the authors (64 chromosomes). The Other haplotypes comprise GCA, ATG, and ACA. Bars show the 95% confidence intervals. ∗ = P < 0.05 versus Spanish controls.
Download figure to PowerPoint
We also compared the haplotype frequencies obtained in female Spanish controls with that reported previously in female Swedish controls (Figure 1). There were significant differences between the 2 populations in spite of the small size of the Swedish sample (64 chromosomes) (P = 0.04, as determined by chi-square test after grouping low-frequency haplotypes). In this case, the major components of the difference were the frequencies of the ACG and the GTA haplotypes. Again, the ACG haplotype included most PD1.3 A alleles, and the excess observed in the Spanish population has already been mentioned. On the contrary, the GTA haplotype was about twice as frequent in the Swedish sample as in the Spanish controls (P = 0.02, odds ratio 2.2, 95% CI 1.1–4.2, compared with the other haplotypes). Therefore, this analysis found significant differences in the haplotype structure of the PDCD1 locus between the Spanish and the Swedish populations.
- Top of page
- PATIENTS AND METHODS
In the study of the genetics of SLE, there is no precedent of a clear association to a susceptibility polymorphism as that reported by Prokunina et al (3). The discovery of the susceptibility PD1.3 A allele started from linkage results and led to a polymorphism in a gene that, on functional grounds, could clearly be involved in the pathogenesis of SLE and whose association was shown in 4 different sets of patients from 3 different populations with large sample sizes. Another 6 SNPs on the gene were excluded by analysis in a set of multicase Swedish families or in a large association study. However, we have obtained results in the present study that call into question a straightforward scenario: in a large Spanish cohort, the PD1.3 allele associated with SLE susceptibility was the G allele, not the A allele. Therefore, our results have confirmed the involvement of the PDCD1 gene in susceptibility to SLE because it was also associated with the disease in our patients, but our study casts doubt on the role of the PD1.3 SNP since its alleles were associated with opposing effects in different populations.
In light of our results, it is important to note 2 limitations of the study by Prokunina et al. First, the search for polymorphisms in the PDCD1 locus was done by sequencing the gene in only 10 individuals. Second, evidence of a functional role of the PD1.3 SNP was limited to the findings on an electrophoretic mobility shift assay (EMSA). The fact that other polymorphisms had remained undetected in the PDCD1 locus has been suggested by a recent report describing 9 new SNPs (2 of them with frequent minor alleles) in a fragment of the gene (19) and by the identification of 3 different SNPs in our very preliminary search (Ferreiros-Vidal I, et al: unpublished observations). Therefore, it is possible that other polymorphisms in PDCD1 could be associated with SLE susceptibility. With regard to the functional study by Prokunina et al, the EMSA showed diminished binding between a synthetic oligonucleotide containing the sequence of the PD1.3 A allele and the RUNX1 transcription factor. This result only indicates that the prediction of a RUNX1 binding site in this sequence was correct and that it would be altered by the PD1.3 SNP. However, this does not necessarily imply that the specific site in intron 4 is functional in vivo; many of the more than 170 RUNX1 binding sites predicted in PDCD1 (18 of them in the same intron as that of PD1.3) would also bind to RUNX1 in an EMSA.
Discordant results in genetic association studies, as has been shown between these studies, are common in the genetic investigation of complex diseases and they have been extensively discussed (20–23). Recently, several meta-analyses of association studies have identified potential causes for the lack of concordance (20, 21). A very important cause seems to be an inflated positive result in the first report describing a new association. This overestimation of the true effect has been attributed to chance deviation from the true population values, characterized as the “winner's curse” phenomenon (20, 21). However, this artifact is very improbable in the case of PD1.3, since the first report included a confirmation of the effect in 4 different sets of samples (with risk ratios ranging from 2.2 to 5.3).
A second, very common cause of lack of replication is insufficient sample size. This limitation causes underpowered studies that fail to find evidence of true, but not very strong, effects. This possibility can also be disregarded in the current situation, because we used a sufficiently large sample size and we found not only a lack of association, but also a significant difference compared with the previous study. Consequently, the post hoc power to detect even the weakest of the effects reported by Prokunina et al, a 2.2 risk ratio in European American families, was 100% in our study (with α = 0.05).
A third possible cause of lack of replication is population stratification that will lead to biased or spurious results. Recent analyses have shown that the danger of spurious association due to this cause had been exaggerated previously (22, 23). In the case of PD1.3, this cause can be ruled out confidently, because Prokunina et al used family controls and because, in our study, the population was ethnically homogeneous and the results were consistent across 5 groups of cases and controls recruited independently in distant towns.
Finally, we are left with genetic heterogeneity, which seems the most likely cause of the discrepancy. This cause of contradictory results has frequently been suspected in the study of genetic associations, but its importance has been hard to assess. It has eluded demonstration when it has been specifically analyzed in a meta-analysis of association studies, probably because of the low number of studies that have contained ethnic information (21). However, there is no doubt that genetic heterogeneity plays a role in many monogenic diseases, and there are, also, several well-documented examples in complex diseases.
Two scenarios involving genetic heterogeneity are possible with regard to PDCD1. First, the most likely cause of the discrepancy between our results and the results from Prokunina et al is a lack of effect of the PD1.3 alleles on SLE susceptibility, in addition to differences in the haplotype structure of the PDCD1 locus. The PD1.3 alleles and the causal polymorphism (or polymorphisms) will be in different haplotypes in each population; the PD1.3 A allele will be in the same haplotype as that of an unknown SLE susceptibility polymorphism in the populations studied by Prokunina et al, whereas in the Spanish population, the haplotype including the causal polymorphism will contain the PD1.3 G allele. This possibility is supported by the differences in the distribution of haplotype frequency between the Spanish and the Swedes that appeared in our analysis, and by the differences in the allele frequencies of 2 of the PDCD1 SNPs between the Spanish and the Mexican and European American populations.
A second scenario, which seems less likely, implies an interaction between PD1.3 and other environmental or genetic factors that would vary among the studied populations. In this second scenario, the PD1.3 A allele will affect PDCD1 expression, but the changes will result in contradictory effects on SLE susceptibility depending on the interaction with unknown factors.
Consequently, although it appears to be confirmed that genetic variation on PDCD1 determines the predisposition to SLE, it is still unclear which polymorphisms are implicated, and therefore what mechanisms could be involved. A more thorough search for polymorphisms in the locus and new association studies in other populations will be needed to solve the questions raised by our results.