Introduction
The systemic lupus erythematosus (SLE) genetic component has been partially elucidated thanks to large studies that have uncovered more than 30 loci reaching very convincing disease association
Gene-gene interaction could be behind the observation that SLE loci show variability in effect sizes in function of the genetic background. For example, the Amerindian genetic background is associated with a higher effect size of several SLE loci in Hispanics
Our study included 25 top-associated SNPs in the better known SLE loci studied in 1,742 patients with SLE and in 2,101 controls from 17 collections recruited in 10 European countries, each of them with homogeneous local ancestry. The results showed a bias to larger effect sizes of the risk alleles in the Southern Europeans relative to the Central Europeans. We also found clines of risk allele frequencies.
Materials and methods
Patient data
We used DNA samples from European SLE patients and ethnically matched healthy controls recruited at 17 centres from 10 different countries (Table
Sample collections, proportion of women, frequencies of the three more informative AIMs and North/South score
DNA sample
Women (%)
AIM frequency (%)
N/S score
Collectiona
Control
SLE
Control
SLE
NL
180
104
59.4
86.5
83.0
77.6
72.5
100.0
BEb
106
147
93.5
90.5
DE
95
90
n.a.
90
77.9
68.1
56.9
80.5
CZb
100
101
32
85.1
SK
93
94
93.5
91.5
72.8
63.8
39.4
64.4
HU
95
95
48.4
90.4
63.4
59.6
39.0
54.9
ES, LCG
145
88
82.9
92
35.4
67.4
45.3
47.6
ES, OVD
200
147
69
91.8
30.5
70.4
45.4
47.2
ES, SCQ
95
109
48.4
96
35.1
68.2
41.3
46.0
ES, BCN
97
90
52.6
91.1
32.6
63.2
41.6
40.5
ES, MAD
281
92
68.8
91.3
33.1
61.9
42.3
40.0
PT
97
100
91.3
95.3
28.7
60.7
39.1
34.6
IT, MXP
106
129
42.5
86.8
47.6
49.1
19.8
26.2
IT, ROM
102
84
55.9
89.7
38.2
51.9
15.3
20.3
GR, AUMS
100
95
92
91.6
36.0
44.0
13.4
11.3
IT, NAP
109
79
100
90.1
38.6
42.2
10.0
9.6
GR, EH
100
98
67
86.7
36.7
38.0
13.5
6.8
Collections are in descending order of N/S score. aCollections from: University Medical Center Groningen, The Netherlands (NL); Université Catholique de Louvain, Belgium (BE); Hannover Medical School, Germany (DE); Institute of Rheumatology, The Czech Republic (CZ); Martin Faculty Hospital, Slovakia (SK); Albert Szent-Györgyi Medical and Pharmaceutical Centre, Hungary (HU); Complexo Hospitalario Universitario A Coruña, Spain (ES, LCG); Hospital Universitario Central de Asturias, Spain (ES, OVD); Hospital Clinico Universitario de Santiago, Spain (ES, SCQ); Hospital Val d'Hebron of Barcelona, Spain (ES, BCN); Hospital 12 de Octubre, Madrid, Spain (ES, MAD); Hospital Garcia de Orta, Almada, Portugal (PT); University of Milan and Fondazione IRCCS Ospedale Maggiore Policlinico from Milan, Italy (IT, MXP), Ospedale S. Camillo - Forlanini from Roma, Italy (IT, ROM); Athens University Medical School, Athens, Greece (GR, AUMS); Second University of Naples, Italy (IT, NAP); and Evangelismos Hospital, Athens, Greece (GR, EH). bThese two collections were not analysed further because of differences in AIM frequencies between cases and controls.
Genotyping
DNA samples were amplified in a multiplex PCR with the KAPA2G fast HotStart (Kapa Biosystems, Woburn, MA, USA) in a final volume of 10 μl (20 ng genomic DNA), using 3 mM MgCl2 and 0.2 μM each primer. Products were purified by Exo-SAP digestion with exonuclease I (Epicentre, Madison, WI, USA) and shrimp alkaline phosphatase (GE Healthcare, Barcelona, Spain). Subsequently, single-base extension reactions were performed with the SNaPshot Multiplex kit (Applied Biosystems, Foster City, CA, USA). The genotyping call rate success of the newly studied SNPs was 99.12%. Sequences of primers and probes are available from the authors upon request.
Selection of SNPs
Six ancestry informative markers (AIMs) were selected (see Table S1 in Additional file
Table S1 presenting a detailed description of all the SNPs included in the study, and Figures S1 and S2 with random effect meta-analysis results corresponding to the same data as Figure
Click here for file
Statistical analysis
Analysis of results was based on R and Statistica 7.0 (StatSoft, Tulsa, OK, USA). Conformity with Hardy-Weinberg equilibrium was tested in control samples. Allele frequencies of the AIMs were compared between patients and controls from each collection with 2 × 2 contingency tables. We created a global score by sample collection for the North-South axis of European population differentiation (N/S score) with the allele frequencies of the AIMs. First, we defined AIM allele frequencies as a function of the allele more common in Northern European populations. The most informative, nonredundant three AIMs were then selected. Finally, as a normalisation step we rescaled the frequencies from each of these three AIMs to 0 to 100%. This was done by considering 0% the frequency in the sample collection where it was less abundant and 100% the frequency where it was most abundant. The rescaled values of the three AIMs were averaged to obtain a combined normalised unique score for each collection.
Case-control allele frequencies were compared with fixed-effects and random-effects models stratifying by sample collection. For the fixed-effect model, the Mantel-Haenszel approach was used. For the random-effect model, an inverse variance meta-analysis approach was followed. Heterogeneity of effect sizes was evaluated with the inconsistency parameter
Distributions of ORs in Central European and Southern European populations were compared with the binomial distribution. Geometric mean (
Results
Analysis of population differentiation
Our study included samples from 1,742 patients with SLE and 2,101 healthy controls recruited in 17 centres (Table
Next, we divided the remaining 15 collections following the major axis of European population differentiation, the North-South axis. According to previous studies we expected two groups
Nine SLE susceptibility loci with a Southern bias
We have already replicated association of top SNPs in nine SLE susceptibility loci in the samples included in this study (Table
Association results of the 25 systemic lupus erythematosus SNPs
Mantel-Haenszel analysis
SNP (locus)
OR (95% CI)
Reference
1.70 (1.50 to 1.93)
5.16 × 10-17
1.52 (1.36 to 1.70)
6.11 × 10-14
1.34 (1.20 to 1.50)
1.33 × 10-7
1.32 (1.18 to 1.47)
1.17 × 10-6
1.27 (1.11 to 1.45)
5.57 × 10-4
1.25 (1.11 to 1.39)
6.12 × 10-5
1.23 (1.10 to 1.37)
1.92 × 10-4
1.19 (1.08 to 1.32)
1.31 × 10-3
1.15 (1.03 to 1.27)
0.02
2.25 (1.89 to 2.68)
6.41 × 10-20
-
2.17 (1.88 to 2.51)
1.09 × 10-25
-
2.00 (1.73 to 2.32)
8.38 × 10-21
1.99 (1.60 to 2.47)
2.54 × 10-10
-
1.35 (1.20 to 1.49)
1.70 × 10-7
1.34 (1.13 to 1.60)
9.15 × 10-4
-
1.26 (1.13 to 1.42)
7.30 × 10-5
-
1.25 (1.11 to 1.40)
1.32 × 10-4
-
1.21 (1.07 to 1.37)
2.25 × 10-3
-
1.19 (1.08 to 1.32)
7.22 × 10-4
-
1.18 (1.07 to 1.30)
1.27 × 10-3
-
1.17 (1.05 to 1.31)
4.35 × 10-3
-
1.09 (0.93 to 1.28)
0.3
-
1.09 (0.98 to 1.19)
0.1
-
1.06 (0.94 to 1.19)
0.3
-
1.02 (0.91 to 1.14)
0.8
-
The first nine SNPs were analysed in the first phase of the study and the remaining 16 in the second. Results of Mantel-Haenszel analysis are shown. All odds ratios (ORs) are expressed for the risk allele. Analyses of some of the SNPs in these samples have already been reported
Bias for a stronger association in Southern Europeans than in Central Europeans
Bias for a stronger association in Southern Europeans than in Central Europeans. (A) Nine systemic lupus erythematosus (SLE) loci from Suarez-Gestal and colleagues
Association analysis of additional SLE loci
We wanted to assess whether the bias found was a general phenomenon of SLE loci. We therefore selected 16 SNPs identifying other SLE genetic loci in Europeans
Stratification of the SLE patients and controls in Central Europeans and Southern Europeans showed that two of the 11 SNPs (
Southern bias for all of the SLE-associated SNPs together
When the 20 SNPs (nine from the first phase and 11 from the second) that have shown independent association in our samples were considered together, a bias towards a stronger association in the Southern subgroup was observed both as a significant deviation of the OR from a random binomial distribution (
We also analysed our data in a different way by counting the SLE risk alleles carried by each subject. Although it was possible to have from zero to 40 risk alleles, none of the subjects had less than five or more than 24 risk alleles. The distribution of frequencies stratified by disease status and by Central versus Southern subpopulations showed a gradient of values (Figure
Systemic lupus erythematosus risk alleles in Central and Southern European patients and controls
Systemic lupus erythematosus risk alleles in Central and Southern European patients and controls. Distribution of the number of systemic lupus erythematosus (SLE) risk alleles in Central European and Southern European SLE patients and controls.
The differences in number of risk alleles were not due to confounding by deviations from Hardy-Weinberg equilibrium in any of the four sample groups for any of the SNPs or by differences in call rate between the four groups (data not shown). We also checked that the differences and order hold when only women were analysed, with the lowest number in Southern controls (12.0 ± 2.4) followed by Central controls (13.0 ± 2.6), Southern SLE patients (14.1 ± 2.5) and Central European patients (14.4 ± 2.8) - showing significant differences between each of these groups except the last two (
These results showed from a different perspective the same bias in effect sizes that has been described in the previous paragraphs, because the difference in number of risk alleles between SLE patients and controls of Southern origin (difference = 2.06, 95% confidence interval = 1.85 to 2.26) was significantly larger than for patients and controls of Central European origin (difference = 1.63, 95% confidence interval = 1.25 to 2.01;
We also compared the cumulative GRS between Southern and Central European SLE patients. This parameter includes information from the sum of risk alleles and from the OR, and therefore is not independent of previous comparisons (Figure
Sum of genotypic risk score in Central and Southern European patients with systemic lupus erythematosus
Sum of genotypic risk score in Central and Southern European patients with systemic lupus erythematosus.
Finally, we wanted to analyse the mean number of SLE risk alleles for each sample collection as a function of its position along the North-South axis of population differentiation (Figure
Correlation between North-South population differentiation axis and number of systemic lupus erythematosus risk alleles
Correlation between North-South population differentiation axis and number of systemic lupus erythematosus risk alleles. Mean number of SLE risk alleles per sample collection is represented against the North/South score (N/S score) obtained in the same samples. Collection letter codes are as in Table 1.
Frequency clines of the SLE risk alleles
The previous results suggest the possibility of frequency gradients or clines of SLE risk alleles along the North-South axis of population differentiation. We therefore analysed this possibility for each of the SNPs. For this analysis we only used data from controls because they are more representative of the general population. None of the SLE-associated SNPs showed significant linkage disequilibrium with the AIMs included in this study (all pairwise
Risk allele frequencies in Central and Southern European controls and relationship with the North/South score
RAF (%)
RAF vs. N/S score
SNP (locus)
Central European
Southern European
χ2
Slope
14.9
4.6
4.9 × 10-26
0.92
4.15
2.0 × 10-6
15.7
8.7
1.9 × 10-9
0.88
4.93
4.0 × 10-5
11.8
6.1
1.5 × 10-8
0.74
4.21
2.7 × 10-3
31.0
23.2
2.2 × 10-6
0.20
0.95
NS
29.0
22.6
7.5 × 10-5
0.66
2.67
0.01
11.9
16.7
4.9 × 10-4
-0.59
-4.79
0.03
25.2
20.7
4.9 × 10-3
0.15
1.08
NS
44.5
48.6
0.03
-0.40
-1.25
NS
69.1
65.3
0.03
0.58
2.83
0.03
11.6
9.3
0.046
0.46
3.64
0.10
20.9
22.8
NS
-0.57
-2.84
0.03
28.2
26.3
NS
0.51
4.10
0.06
23.3
23.8
NS
-0.51
-4.63
0.06
19.3
17.8
NS
0.49
2.74
0.07
66.2
69.0
NS
-0.17
-1.16
NS
22.8
20.8
NS
0.16
1.28
NS
73.1
71.0
NS
0.06
0.23
NS
3.5
4.3
NS
-0.21
-3.42
NS
67.8
69.0
NS
-0.00
-0.03
NS
71.2
70.9
NS
0.32
1.99
NS
Comparison of the risk allele frequencies (RAFs) between Central and Southern European controls, and linear regression of RAFs to the North/South score (N/S score) of population differentiation (from Table 1). NS, not significant. aAnalysis performed only in women because the locus is in the X chromosome.
Discussion
Two main aspects of our results should be highlighted: a significant bias to larger effect sizes of the SLE susceptibility loci in subjects from Southern Europe than those from Central Europe, and a lower frequency of the SLE risk alleles at these loci in subjects from Southern Europe.
The bias to stronger association among Southern Europeans was shown with four types of analysis. The first, comparing the number of SLE loci showing a trend to stronger associations beyond the expected at random, is independent of the loci characteristics. In contrast, comparison of the mean OR reflects the magnitude of the differences in effect size. An advantage of these two analyses is that they were done with meta-analysis approaches to account for sample collection factors. Their limitation is that they convey little information about the nature of the differences. The next analysis, comparison of the number of SLE risk alleles carried by each subject, is more informative but does not account for sample collection effects. The last analysis, comparison of sum GRS, combines information from the two previous analyses and is therefore not independent. Concordance of results from the different analyses is reassuring.
Further confidence was gained from the gradual change of the number of SLE risk alleles per sample collection as a function of the N/S score, implying insensitivity of the results to the specific partition of Europeans we have used. It is also important to note that the studies which identified SLE risk loci were carried out with subjects of a dominant Northern European ancestry (full references in Table S1 in Additional file
The second aspect of our results is the differential distribution of SLE risk alleles, with a lower frequency in Southern Europeans than in Central Europeans. The difference was observed both in SLE patients and in controls, but was more marked in the latter. It was maximal between the subjects from Greece and Italy, on one side, and those for the Netherlands on the other. All of the remaining groups were in between. Seven of the SNPs showed a frequency cline correlating with the N/S score. Only one of these clines has already been described for the risk allele of
The existence of allele frequency clines in the European population and their main axis of differentiation are well established
A notable facet of the cline of SLE risk alleles in our study was that the difference between Southern Europeans and Central Europeans is less marked in SLE patients than in controls. We propose that this is due to the genetic structure of SLE that makes patients more similar at SLE loci across European subpopulations than the average member of the same subpopulations. This hypothesis is derived from the polygenic model of genetic inheritance
The validity of our results will be reinforced by replication in other sample collections. This will be particularly important in relation to the Central European subgroup - this was the smallest in our study, causing lower precision in the OR values and lower power to detect differential effects. Data from additional populations would also be interesting, in particular from the more extreme European subpopulations, because coverage of the full European spectrum would allow detecting additional clines. In addition, the AIMs we used were sufficient for group-level analyses but not for classification of individual subjects, which could have made our analyses more powerful. The studied AIMs were able to show the European population substructure along the North-South differentiation, however, as in the studies where these AIMs were selected
Our study also provides independent replication of SLE loci. Most of these loci were already strongly established and do not require comment (full references in Table S1 in Additional file
Also very solid is the record of
Conclusion
Our study has uncovered a bias for stronger effect sizes of SLE risk alleles in Southern Europeans than in Central Europeans. This bias was accompanied by a lower frequency of the risk alleles in the Southern European group. Difference in frequencies was more marked in controls than in patients with SLE. These results should be taken into account for genetic studies of SLE and for understanding the genetic structure of the disease and the possible presence of autoimmune disease risk allele clines and their causes. In addition, these results call for exploration of the assumptions and implications of the liability threshold concept - in particular, whether a constant threshold is consistent with GWAS data - as well as for exploration of environmental or other factors that could explain the effect size bias. Our findings therefore contribute to define the genetic epidemiology of SLE and suggest new lines of research for understanding the deep genetic structure of SLE.
Abbreviations
AIM: ancestry informative marker; CI: confidence interval;
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
EA-P and MS-G participated in the design of the study, genotyped the samples, and participated in the interpretation of the results and writing the manuscript. MC participated in the statistical analysis and in the interpretation of results. GDS, RP, CP, AK, FNS, MB, AS, MM, SM, PC, JO-R, TW, SR, MJS, NB, FJB, BRL and JJG-R participated in the acquisition of clinical data and collection of samples and in the analysis and interpretation of results. AG participated in the design of the study and in coordination of the acquisition of clinical data and collection of samples, and supervised laboratory work and interpretation and writing of the manuscript. All authors read and approved the final manuscript.