Identification of previously unrecognized predisposing factors for ankylosing spondylitis from analysis of HLA–B27 extended haplotypes in sardinia




To define the contribution of HLA genes other than HLA–B27 in conferring susceptibility to ankylosing spondylitis (AS), through analysis of HLA–B27 haplotypes in Sardinian subjects.


Ninety-eight patients with AS, 133 HLA–B27–positive controls (of whom 33 were positive for HLA–B*2709), and 190 randomly selected controls were genotyped for microsatellites and single-nucleotide polymorphisms (SNPs) spanning the HLA region.


Haplotypes carrying either the B*2705 or the B*2709 allele were found to share a conserved region downstream of the HLA–B gene and a functional polymorphism in the HLA–E gene (R128G), while differing in all other markers. Notably, the presence of an A at SNP rs1264457, encoding for Arg-128, was significantly increased in the cohort of patients (P = 6 × 10−6, corrected P = 3 × 10−5) but not in B*2705- or B*2709-positive controls. Comparing the alleles co-occurring at each HLA marker, we identified a region differentiating patients with AS and B*2705-matched controls. In particular, there was a markedly increased prevalence of heterozygosity at rs1264457 among B27-positive controls (74%, versus 47% in patients and 54% in random controls), suggesting a protective role of G128 in AS. Moreover, other markers around the HLA–B gene were also differentially represented.


These results demonstrate a significant difference in the frequency of some HLA markers between AS patients and B*2705-positive controls, which could be attributed to the opposite chromosome. In particular, the differential distribution of a functional polymorphism in the HLA–E gene suggests a possible role of natural killer function in AS pathogenesis.

Ankylosing spondylitis (AS) is a rheumatic disorder that is strongly associated with HLA–B27 (1, 2). However, although its involvement in the disease has been demonstrated in animal models (3, 4), a clear role of this molecule in AS pathogenesis has not been elucidated as yet. A further complication comes from the demonstration that HLA–B27 is a family of alleles, which are not all equally associated with AS (5). In particular, HLA–B*2709, an allele that occurs in high frequency among the population of Sardinia, is not associated with the disease. This allele differs from the worldwide-distributed and AS-associated HLA–B*2705 by a single amino acid at position 116 (H116D), located in the peptide-binding groove (6). The presence of either amino acid confers to the 2 B27 molecules different binding specificities (7, 8) and structural/functional antigen-presenting properties (9) that could account for the differential association with AS (10).

In addition, the B*2709 allele in Sardinia has been found within a haplotype different from that harboring the B*2705 allele (11). This leaves open the possibility that other genes mapping in the same haplotype could contribute to the lack of association. In fact, findings of several studies suggest the presence of additional susceptibility genes within the HLA region (12, 13). However, the identification of such genes is hampered by the strong linkage disequilibrium (LD) present in the HLA region, which makes it difficult to single out genes with possibly weak influence in the presence of the major susceptibility factor HLA–B27 (which confers a relative risk of >80). In this context, fine mapping of the HLA region, comparing the haplotypes harboring B*2705 and B*2709, may provide insight.

The Sardinian population is an ancient genetic isolate with a unique distribution of alleles at multiple loci and is therefore particularly appropriate for association studies (14). Two additional features make this population especially informative for the study of AS: 1) HLA–B*2709 is relatively frequent in Sardinia (occurring in approximately one-fifth of all B27-positive individuals) (15) and 2) B*2705 occurs within an extended haplotype (HLA–A2;Cw2;B*2705;DR2) (11, 16). A previous study in which the allelic frequencies of the classic HLA class I and II genes were analyzed showed no major differences between AS patients and B*2705-matched controls, while allowing identification of an extended B*2709 haplotype (HLA–A32;Cw1;B*2709;DR5) (11).

In the present study, we further characterized the 2 haplotypes and undertook a more detailed comparison of B*2705-positive patients and controls. We found significant differences in the frequency of several markers between patients and B*2705-matched controls, which could be attributed to the opposite chromosome. This strongly suggests that genes mapping in the HLA region have an HLA–B27–independent role in disease susceptibility.



A total of 190 randomly selected controls, 98 HLA–B27–positive patients with AS, and 133 HLA–B27–positive controls of whom 33 are B*2709 positive have been identified for study during the last 5 years. All have been typed for HLA–A, B, C, and DR using commercial kits (HLA SSP kits; Biotest, Dreieich, Germany), for MICA transmembrane microsatellite (short tandem repeat [STR] MICA) by heteroduplex analysis as previously described (17), and for polymorphisms at −857 of the tumor necrosis factor (TNF) promoter (rs1799724) by restriction enzyme digestion. A subset of these subjects have also been typed for other markers, depending on DNA availability.The markers, besides the classic HLA–A, B, C, and DR genes, included the following: microsatellite D6S248, mapping upstream of the HLA–A gene and reported to be associated with AS in Basque patients (18); a functional single-nucleotide polymorphism (SNP) (rs1264457), mapping in the third exon of the HLA–E gene; microsatellite D6S2674, mapping between HLA–C and HLA–B; a microsatellite mapping in the transmembrane region of MICA (STR_MICA); SNP rs2284178, mapping in HLA complex P5-1 (HCP5-1), an open-reading frame related to human endogenous retroviruses and possibly transcribed in lymphoid cells (19); microsatellite D6S2793 in the MICB gene, encoding (together with MICA) for ligands of natural killer (NK) receptors (20); rs2071591 in NFKBIL1, a gene encoding for a potential regulator of NF-κB function (21) and found to confer susceptibility to rheumatoid arthritis (22); and finally, rs1799724 at position –857 of the TNFα promoter. Primers are shown in Table 1.

Table 1. Oligonucleotide sequences
Marker (gene)*ForwardReverseMinisequencing oligonucleotides or restriction enzyme
  • *

    D6S248, D6S2674, STR_MICA, and D6S2793 are microsatellites; rs1264457, rs2284178, rs2071591, and rs1799724 are single-nucleotide polymorphisms.


All patients with AS were diagnosed according to the modified New York criteria (23). All subjects provided informed consent prior to participation in the study.

Polymerase chain reaction (PCR) and SNP typing.

For PCR, the following reagents were used: 20 ng genomic DNA in a volume of 15 μl, containing 20 mM Tris HCl (pH 8.4), 50 mM KCl, 2.5 mM MgCl2, 0.2 mM dNTPs, 5 pmoles forward and reverse primers, and 0.75 units of Taq polymerase (Biotaq DNA Polymerase; Bioline, London, UK). The mixture was incubated at 94°C for 3 minutes, followed by 25–30 cycles of denaturation at 94°C for 30 seconds, annealing at 55–62°C (depending on the marker being analyzed) for 30 seconds, extension at 72°C for 1 minute, and a final step of 72°C for 10 minutes.

SNPs were typed by minisequencing or restriction enzyme digestion. For minisequencing, the amplified products were treated with 0.5 units of shrimp alkaline phosphatase (Roche, Indianapolis, IN) and 2.5 units of exonuclease I (New England Biolabs, Ipswich, MA) at 37°C for 120 minutes, and at 75°C for 15 minutes for inactivation. For minisequencing reactions, the commercial fluorescence-based minisequencing kit SNaPshot multiplex (Applied Biosystems, Foster City, CA) was used, with 2 pmoles of primer. Single base extension reactions were performed as follows: initial denaturation at 94°C for 3 minutes, followed by 50 cycles of primer extension at 96°C for 15 seconds, annealing at 55–59°C (depending on the oligonucleotide annealing temperature) for 15 seconds, and extension at 60°C for 1 minute. The sequences of the SNaPshot primer are shown in Table 1. After extension and labeling, unincorporated dNTPs were removed by enzymatic treatment with 0.5 units of shrimp alkaline phosphatase at 37°C for 120 minutes, and at 75°C for 15 minutes for inactivation. The products were analyzed on an ABI Prism 310 Genetic Analyzer. Genotyping was performed using 310 ABI Prism GeneScan 2.1 software (Applied Biosystems). For the TNF promoter SNP rs1799724, the amplified product was digested overnight at 37°C with HpyCH4IV (Bioline). Fragments were separated on 10% polyacrylamide gels, stained with ethidium bromide, and photographed under ultraviolet light. The underlined base (Table 1) was introduced to create the specific HpyCH4V restriction site.

Typing of microsatellites.

PCR was performed using forward primers labeled at the 5′ end with fluorescent dye. The diluted (1:200) amplified products were run on ABI Prism 310 Genetic Analyzers, by capillary electrophoresis. Genotyping was performed using GeneScan and Genotyper (Applied Biosystems) software.

Statistical analysis.

Fisher's 2-tailed exact test was used to assess the differences in the proportions of polymorphic alleles and associations in randomly selected controls versus HLA–B27–positive AS patients, HLA–B*2705–positive controls, and HLA–B*2709–positive controls. Two-tailed P values were calculated using the Open-Epi Epidemiological calculators at All alleles with frequency of >10% in at least 1 category were tested, for a total of 120 statistical tests. In order to control for multiple-comparison error, the false discovery rate (FDR) principle (24) was applied. The FDR criterion controls the “expected fraction of false rejections” when multiple tests are simultaneously carried out; hence, it is less affected by the lack of power typical of correction methods based on the familywise error rate principle, such as the Bonferroni correction. Under the assumption of independence, P values less than 0.05 were considered significant. Significance was demonstrated in 58 of the 120 statistical tests after application of the FDR correction (note that 61 statistical tests showed a P value of less than 0.05 in a single, uncorrected test, and 32 were also significant after application of the more conservative Bonferroni procedure).

The Cochran-Armitage trend test was performed using StatXact7 software. Fisher's 2-tailed exact test was again used to assess the difference in allelic proportions between B*2705-positive patients, B*2705-positive controls, and pooled B27 (B*2705 and B*2709)–positive controls. A total of 31 statistical tests were performed. Benjamini and Hochberg FDR (24) was used under the assumption of independence to assess the significance of the test results. P values less than 0.05 were considered significant. Significance was demonstrated in 3 of the 31 statistical tests after application of the FDR correction (note that 5 statistical tests showed a P value of less than 0.05 in a single, uncorrected test, and 1 was also significant after application of the more conservative Bonferroni procedure).

Linkage disequilibrium.

Haplotype frequencies (and their distributions) were estimated, and their significance for LD calculated, using PyPOP software. Hardy-Weinberg equilibrium testing was performed using Pearson's goodness-of-fit chi-square test with 1 df. All markers were in Hardy-Weinberg equilibrium except TNF rs1799724 (χ2 = 5.3, P = 0.02).


Analysis of interassociation among markers mapping in the HLA region in Sardinian subjects.

HLA–A, B, C, and DR loci in the Sardinian population are clustered in a number of defined haplotypes occurring with relatively high frequency (16). To better characterize the extended haplotypes, we analyzed additional markers spanning the HLA region in a subset of randomly selected controls, HLA–B*2705–positive AS patients, HLA–B*2705–positive controls, and HLA–B*2709–positive subjects. Figure 1 shows their approximate positions on chromosome 6, together with the D′ values of the interlocus associations, as established in a set of 100 randomly selected controls, using PyPOP software. As expected, there was a central block of higher LD, which was weaker when the SNPs were considered (Figure 1).

Figure 1.

Significance of linkage disequilibrium (LD) among the HLA markers studied. Haplotype frequencies (and their distributions) were estimated, and significance for LD calculated, using PyPOP software. Approximate locations on chromosome 6 are also indicated (

Definition of the markers defining HLA–B*2705 and B*2709 haplotypes.

We previously reported that the chromosomes harboring the B*2705 and B*2709 alleles were characterized by different HLA–A, HLA–C, and HLA–DR alleles (11). In the present study, we investigated whether the same was true for the other markers under study. Their frequencies are shown in Table 2. Because patients with AS were selected for HLA–B*2705 positivity, a cohort of B*2705-positive controls was also studied, in order to compare patients and controls sharing the same risk factor (HLA–B*2705) and, presumably, the other markers in LD with it. For the loci demonstrating the highest degree of polymorphism, only those reaching a frequency of >10% in at least 1 cohort were considered.

Table 2. Distribution of HLA markers in patients with AS and in randomly selected controls, HLA–B*2705–positive controls, and HLA–B*2709–positive controls*
Marker (no. of random controls/ B*2705+ controls/ AS patients/ B*2709+ controls)Random controls, no. (%)B*2705+ controls, no. (%)P/PcorrB27+ AS patients, no. (%)P/Pcorr (OR [95% CI]), AS patients vs. random controlsP/Pcorr (OR [95% CI]), AS patients vs. B* 2705+ controlsB*2709+ controls, no. (%)P/Pcorr
  • a

    For multiallelic markers, in the absence of any trend of association, the 2 most frequent alleles are shown. Significant P values are reported. AS = ankylosing spondylitis; Pcorr = corrected P; OR = odds ratio; 95% CI = 95% confidence interval; NS = not significant.

  • B*2705-positive or B*2709-positive controls versus random controls.

D6S248 (171/52/84/30)
 1299 (29)50 (48)5 × 10−4/0.00266 (39)0.020/0.044 (1.6 [1.0–2.3]) 17 (28) 
 2116 (5)1 (1) 5 (3)  10 (17)0.002/0.006
 2392 (27)24 (23) 44 (26)  13 (22) 
 Others135 (39)29 (28)0.03/NS53 (32)  20 (33) 
HLA–A (162/98/64/33)
 286 (27)78 (40)0.002/0.00660 (47)5 × 10−5/0.002 (2.4 [1.5–3.8]) 11 (17) 
 3230 (9)25 (13) 12 (9)  28 (42)8 × 10−10/8 × 10−8
 Others208 (64)93 (47)2 × 10−4/8 × 10−456 (44)8 × 10−5/3 × 10−4 (2.0 [1.4–3.5]) 27 (41)5 × 10−4/0.002
HLA-E rs1264457 (116/53/79/24)
 A125 (54)64 (60) 121 (77)6 × 10−6/3 × 10−5 (2.8 [1.7–4.5])0.006/0.015 (2.1 [1.2–3.8])29 (60) 
 G107 (46)42 (40) 37 (23)  19 (40) 
HLA–C (154/82/82/28)
 112 (4)23 (14)1 × 10−4/3 × 10−416 (10)0.01/0.025 (2.7 [1.1–6.3]) 30 (54)4 × 10−19/4 × 10−17
 225 (8)73 (45)1 × 10−19/ 1 × 10−1790 (55)1 × 10−28/1 × 10−26 (13.8 [8.0–24.0]) 1 (2) 
 Others271 (88)68 (41)4 × 10−26/ 4 × 10−2458 (35)5 × 10−32/5 × 10−30 (13.4 [8.1–22.0]) 25 (46)8 × 10−12/8 × 10−10
D6S2674 (183/55/85/31)
 127 (7)47 (43)2 × 10−16/ 2 × 10−1483 (49)9 × 10−27/9 × 10−27 (12 [7.1–20.3]) 5 (8) 
 673 (20)17 (15) 9 (5)4 × 10−6/2 × 10−4 (0.22 [0.1–0.46])0.006/0.015 (0.3 [0.13–0.7])6 (10)0.044/NS
 1728 (8)12 (11) 21 (12)  27 (44)2 × 10−11/2 × 10−11
 Others238 (65)34 (31)3 × 10−10/ 3 × 10−857 (34)1 × 10−11/1 × 10−9 (3.6 [2.4–5.5]) 24 (39)1 × 10−4/4 × 10−4
MICA (142/80/99/23)
 A4101 (35)112 (70)2 × 10−12/ 2 × 10−10125 (63)2 × 10−9/1 × 10−7 (3.1 [2.1–4.6]) 31 (67)7 × 10−8/7 × 10−6
 A519 (7)6 (4) 21 (11) 0.015/0.032 (3.0 [1.1–9.4])1 (2) 
 A5.136 (13)9 (6)0.021/0.04410 (5)0.004/0.010 (0.37 [0.18–0.76]) 5 (11) 
 A672 (25)21 (13)0.002/0.00625 (13)5 × 10−4/0.002 (0.43 [0.26–0.7]) 4 (9)0.013/0.028
 A956 (20)12 (8)5 × 10−4/0.00217 (8)7 × 10−4/0.002 (0.38 [0.21–0.68]) 5 (11) 
HCP5 rs2284178 (169/51/50/28)
 C201 (59)33 (32)2 × 10−6/1 × 10−527 (27)1 × 10−8/1 × 10−6 (0.25 [0.15–0.41]) 21 (37)0.003/0.008
 T137 (41)69 (68) 73 (73)  35 (63) 
MICB D6S2793 (153/46/80/21)
 267 (22)46 (50)6 × 10−7/2 × 10−592 (58)4 × 10−14/4 × 10−14 (4.8 [3.1–7.4]) 16 (38)0.03/NS
 346 (15)10 (11) 21 (13)  9 (21) 
 Others193 (63)36 (39)6 × 10−5/2 × 10−447 (29)5 × 10−12/5 × 10−10 (0.24 [0.16–0.37]) 17 (40)0.007/0.017
NFKBIL rs2071591 (148/48/70/23)
 C251 (85)84 (87) 112 (80)  26 (57)3 × 10−5/2 × 10−4
 T45 (15)12 (13) 28 (20)  20 (43) 
TNF rs1799724 (190/82/87/28)
 C311 (82)96 (59)3 × 10−8/1 × 10−690 (52)1 × 10−12/1 × 10−10 (0.24 [0.16–0.35]) 53 (95)0.012/0.027
 T69 (18)68 (41) 84 (48)  3 (11) 
HLA–DR (109/76/69/18)
 2 (15, 16)54 (25)59 (39)0.004 0.01054 (39)0.005/0.013 (1.9 [1.2–3.2]) 5 (14) 
 3 (17)53 (24)20 (13)0.008/0.01810 (7)3 × 10−5/2 × 10−4 (0.24 [0.12–0.5]) 11 (31) 
 5 (11, 12)29 (13)35 (23)0.017/0.03517 (12)  13 (36)0.002/0.042
 Others82 (38)38 (25)0.013/0.02857 (41)  7 (19)0.04/NS

Starting from the telomere, microsatellite D6S248 was represented by 21 alleles, among which alleles 12 and 23 were the most frequent. Allele 12 was found in high prevalence within the B*2705 haplotype (P = 0.0005, corrected P [Pcorr] = 0.002, B*2705-positive controls versus randomly selected controls). There was also a high frequency of allele 21 in the B*2709 haplotype (P = 0.002, Pcorr = 0.006 versus randomly selected controls). With regard to the HLA–A locus, the current analysis confirmed previous reports showing the association of HLA–A2 with HLA–B*2705 (11). In fact, A2 was more frequent in patients with AS (P = 5 × 10−5, Pcorr = 0.002 versus randomly selected controls) than in B*2705-positive controls (P = 0.002, Pcorr = 0.006 versus randomly selected controls), whereas HLA–A32 was associated with HLA–B*2709 (P = 8 × 10−10, Pcorr = 8 × 10−8 versus random controls).

Next, we analyzed a functional polymorphism (G128R) in the third exon of the HLA–E gene. Of note, patients with AS showed a higher frequency of allele A encoding for R (77% of cases) compared with random controls (54%; P = 6 × 10−6, Pcorr = 3 × 10−5, odds ratio [OR] 2.8, 95% confidence interval [95% CI] 1.7–4.5), compared with B*2705-positive controls (60%; P = 0.006, Pcorr = 0.015, OR 2.1, 95% CI 1.2–3.8), and compared with B*2709-positive controls (60%). With regard to HLA–C, the strong association of Cw2 with HLA–B*2705 and of Cw1 with HLA–B*2709 (11) was confirmed.

The microsatellite D6S2674, mapping between HLA–C and HLA–B and possessing 22 alleles, also showed a skewed distribution, with allele 1 associated with B*2705 (49% and 43% in B*2705-positive cases and controls, respectively, versus 7% in random controls) and allele 17 with HLA–B*2709 (44%, versus 8% in random controls). Moreover, the presence of allele 6 was significantly decreased in patients with AS compared with random and B27-matched controls (P = 4 × 10−6, Pcorr = 2 × 10−4, OR 0.22, 95% CI 0.1–0.46 and P = 0.006, Pcorr = 0.015, OR 0.3, 95% CI 0.13–0.7, respectively)

As previously reported (25), the MICA-A4 microsatellite was strongly associated with HLA–B27, independent of the subtype. HCP5-1 maps centromeric to the MICA gene. Allele T rs2284178 was associated with HLA–B27 (either B*2705 or B*2709), independent of the disease. Allele 2 of the MICB microsatellite also showed an association with HLA–B*2705 (allelic frequencies 58% in patients, 50% in B*2705-positive controls), a trend shared by the B*2709 haplotype.

The distribution of NFKBIL rs2071591 did not differ between the HLA–B*2705 cohorts and random controls. However, B*2709 was significantly associated with the presence of allele T. TNF-857 showed a strikingly different distribution, with a strong association between B*2705 and allele T independent of the disease (76% and 73% of B*2705-positive patients and controls respectively) (data not shown). B*2709 was associated with C, which also occurred with high frequency in the random controls. As previously reported (11), HLA–DR was also skewed, with HLA–DR2 associated with B*2705 and HLA–DR5 with B*2709.

These data allowed us to further dissect the 2 haplotypes harboring either the B*2705 or the B*2709 allele, which differed in most of the markers except a central block from MICA to MICB. In addition, allele A encoding for R128 in HLA–E was shared by the B*2705 and B*2709 haplotypes. Figure 2 depicts the 2 haplotypes as derived using PyPOP software. The B*2705 haplotype was highly characterized, with the only exception represented by the MICB microsatellite, which was present within the same haplotype either with allele 2 (34%) or with allele 3 (18%). B*2709 was more ambiguous in a few markers (D6S248, MICB, and NFKBIL1), probably also due to the small panel analyzed. However, the haplotype presented represents >70% of all possible B*2709 haplotypes.

Figure 2.

Schematic representation of the B*2705 and B*2709 haplotypes. Markers were attributed as derived using PyPOP software. The haplotypes shown represent >50% and >70% of all possible B*2705 and B*2709 haplotypes, respectively. Double indication of alleles at any marker indicates uncertain attribution.

The data shown in Table 2 also reveal that there were at least 2 markers, HLA–E and D6S2674, for which AS patients, but not B*2705-positive controls, exhibited a significantly different allelic distribution compared with random controls. Allele A rs1264457 in exon 3 of the HLA–E gene was more frequent in patients, while the opposite was the case for the allele 6 of the D6S2674 microsatellite.

Genotype distributions of the 4 SNPs (rs1264457, rs2284178, rs2071591, and rs1799724) are presented in Table 3, along with the results of the Cochran-Armitage test for trend. When the 2 cohorts of controls (B*2705- or B*2709-positive) shared the same marker, such as for HLA–E 128 and HCP5, the data were pooled. The P values for trend were calculated comparing patients with AS both with random controls and with B27-positive controls.

Table 3. Genotype distribution of 4 single-nucleotide polymorphisms in patients with AS, in randomly selected controls, and in HLA–B27–positive controls*
Marker (no. of random controls/B27+ controls/AS patients)Random controls, no. (%)B27+ controls, no. (%)B27+ AS patients, no. (%)P (OR [95% CI]), AS patients vs. random controlsP (OR [95% CI]), AS patients vs. B27+ controls
  • *

    See Table 2 for definitions.

  • For AS patients versus random controls, P for trend = 1.3 × 10−6 (Pcorr for trend = 1 × 10−5); for AS patients versus B27-positive controls, P for trend = 7.6 × 10−5 (Pcorr for trend = 6 × 10−4), by Cochran-Armitage trend test.

  • For AS patients versus random controls, P for trend = 8.6 × 10−9 (Pcorr for trend = 6.8 × 10−8); for AS patients versus B27-positive controls, P for trend not significant, by Cochran-Armitage trend test.

  • §

    For AS patients versus random controls and versus B27-positive controls, P for trend not significant, by Cochran-Armitage trend test.

  • For AS patients versus random controls, P for trend = 7.5 × 10−12 (Pcorr for trend = 6 × 10−11); for AS patients versus B27-positive controls, P for trend not significant, by Cochran-Armitage trend test.

HLA–E rs1264457 (116/77/79)
 AA31 (27)18 (23)42 (53)ReferentReferent
 AG63 (54)57 (74)37 (47)0.009 (0.43 [0.23–0.8])0.0003 (0.28 [0.14–0.55])
 GG22 (19)2 (3)0 (0)2.4 × 10−7 (0.016 [0.001–0.28])NS
HCP5 rs2284178 (169/79/50)
 CC60 (35)2 (3)1 (2)ReferentReferent
 CT81 (48)50 (63)25 (50)0.0002 (18.5 [2.8–772])NS
 TT28 (17)27 (34)24 (48)3.4 × 10−9 (51.4 [7.3–2,149])NS
NFKBIL rs2071591§ (148/48/70)
 CC108 (73)36 (75)46 (66)ReferentReferent
 CT35 (24)12 (25)20 (29)NSNS
 TT5 (3)0 (0)4 (6)NSNS
TNF rs1799724 (190/82/87)
 CC132 (69)22 (27)21 (24)ReferentReferent
 CT47 (25)52 (63)48 (55)7.2 × 10−10 (6.4 [3.5–11.8])NS
 TT11 (6)8 (10)18 (21)1.5 × 10−7 (10.2 [3.9–27.3])NS

All SNPs were in Hardy-Weinberg equilibrium except rs1799724 (TNF-857), for which a significant deviation from Hardy-Weinberg equilibrium was found (χ2 = 5.3, P = 0.02). Genotyping for this SNP was repeated in 1 of every 5 DNA samples (n = 38) and found to be identical. It is known that ∼1 in 20 SNPs is, by chance, not in Hardy-Weinberg equilibrium (26), and rs1799724 might be one of them.

Of note, 3 SNPs showed a significant P value for trend when AS patients were compared with random controls, but only HLA–E rs1264457 showed a significant trend when patients were compared with HLA–B*2705–positive controls (P for trend = 7.6 × 10−5, Pcorr = 6 × 10−4), consistent with the association data reported in Table 2. The frequency of AG heterozygosity was significantly higher (74%) in the B27-positive controls than in patients (OR 0.28), suggesting that the presence of HLA–E G128 on the opposite chromosome from that carrying HLA–E A128-B*2705 might play some protective role in the disease.

Comparison of markers characterizing the chromosome opposite that harboring HLA–B*2705 in patients and controls.

The above findings prompted us to investigate whether other markers surrounding the HLA–B genes might also have a different frequency in AS patients and controls matched for HLA–B*2705 status. B*2705-positive patients and controls possessing the markers characterizing the B*2705 haplotype as shown in Figure 2 were identified and the co-occurring alleles were counted, with the assumption that their frequencies should not be significantly different from those in random controls, as would be expected in the absence of any selection. Conversely, any significant difference would suggest involvement of the marker, or of markers in strong LD with it, in disease susceptibility and this involvement could be interpreted as B27 independent, the marker being located on the opposite chromosome from that harboring the markers characterizing the B*2705 haplotype.

Frequencies of the alleles co-occurring with the HLA–B*2705 markers are reported in Table 4. For those markers clearly shared by the B*2705 and B*2709 haplotypes, i.e., HLA–E, MICA, and HCP5, the comparison between patients and B*2705- and B*2709-positive controls pooled together is also reported. Reviewing each single marker and starting from the telomere, we found that allele 12 of the D6S248 microsatellite was expressed by 34 (65%) and 52 (62%) of B*2705-positive controls and patients, respectively. When the frequency of the co-occurring allele was determined, we found no significant difference in the expression of the 2 major alleles or any others, although there was a trend toward a higher frequency of homozygosity for allele 12 in controls (47%, versus 27% in patients).

Table 4. Comparison of HLA markers present in the opposite chromosome from that harboring the B*2705 allele*
Marker (allelic frequency in random controls) (no. of B*2705+ controls/B27+ controls/AS patients)B*2705+ controls, no. (%)B27+ controls (B*2705+ B*2709), no. (%)B*2705+ AS patients, no. (%)P/Pcorr (OR [95% CI]), patients vs. B*2705+ controlsP/Pcorr (OR [95% CI]), patients vs. B27+ controls
  • *

    As represented in Figure 2. For multiallelic markers, in the absence of any trend of association, the 2 most frequent alleles are shown. Significant P values are reported. ND = not determined (see Table 2 for other definitions).

  • Alleles shared by B*2705 plus B*2709 haplotypes.

  • Determined in 100 B*2705-positive controls and 63 AS patients.

 12 (29%) (34/ND/52)16 (47) 14 (27)  
 23 (27%)7 (21) 18 (35)  
 Others11 (32) 20 (38)  
 2 (27%) (66/ND/51)12 (18) 9 (18)  
 30 (18%)12 (18) 13 (25)  
 Others42 (64) 29 (57)  
HLA–E Rs1264457
 A (54%) (51/75/79)13 (25)18 (24)42 (53)  
 G (46%)38 (75)57 (76)37 (47)0.002/0.03 (0.3 [0.14–0.65])0.0002/0.008 (0.28 [0.14–0.55])
 2 (8%) (60/ND/65)13 (22) 25 (38)0.05/NS 
 Others47 (79) 40 (62)  
 1 (7%) (42/ND/66)5 (12) 17 (26)  
 6 (20%)15 (36) 7 (11)0.003/0.03 (0.21 [0.08–0.57]) 
 Others22 (52) 42 (63)  
 18 (29%)26 (26) 18 (28)  
 58 (10%)9 (9) 1 (2)  
 Others65 (65) 44 (70)  
 A4 (34%) (80/103/96)32 (40)40 (39)25 (26)0.05/NS 
 A5 (6%)7 (9)8 (8)20 (21)0.03/NS0.01/NS
 Others41 (51)55 (53)51 (53)  
HCP5§ Rs2284178
 C (59%)29 (60)50 (67)25 (50)  
 T (41%) (49/75/49)20 (40)25 (33)24 (50)  
MICB D6S2793
 2 (22%) (34/ND/68)12 (35) 24 (35)  
 3 (15%)2 (6) 11 (16)  
 Others20 (59) 33 (49)  
NFKBIL Rs 2071591
 C (84%) (48/ND/66)36 (75) 46 (70)  
 T (16%)12 (25) 20 (30)  
TNF Rs1799724
 C (82%)51 (86) 48 (73)  
 T (18%) (59/ND/66)8 (14) 18 (27)  
 2(15,16) (25%) (43/ND/50)10 (23) 9 (18)  
 5(11,12) (13%)13 (30) 10 (20)  
 Others20 (47) 31 (67)  

The most striking difference was found in the HLA–E genotypic distribution: allele G co-occurred with allele A in 75% of HLA–B*2705–positive controls compared with 47% of patients (P = 0.002, OR 0.3, 95% CI 0.14–0.65) and 46% of random controls. In the absence of any selective pressure, the frequency of co-occurrence of G with A, which is considered a distinctive marker for both the B*2705 and B*2709 haplotypes, being carried by all B27-positive individuals, would be expected to be approximately the same as in random controls. The presence of G in 76% of B27-positive controls strongly suggests a protective effect of this allele against AS (P = 0.0002, OR 0.28, 95% CI 0.14–0.55, patients versus B*2705 and B*2709–positive healthy controls).

Other significant differences between B*2705-positive patients and controls were found in the HLA–C locus, and differences were even more evident in the D6S2674 marker, where allele 6 co-occurred with allele 1 (marker for the B*2705 haplotype) in 36% of B*2705-positive controls compared with 11% of patients with AS (P = 0.003, OR 0.21, 95% CI 0.08–0.57). MICA-A5 co-occurred with MICA-A4 in 21% of patients, versus 9% of B*2705-positive controls and 6% of random controls (P = 0.03). Interestingly, in the Sardinian population, this allele is also associated with HLA–B*2707 (data not shown), an allele predisposing to AS in some populations but not in others (15, 27). For the other markers (HCP5, MICB, NFKBIL, TNF, and DR), there was no significant difference between patients and B*2705-matched controls in the frequency of the alleles that could be assigned to the opposite chromosome (Table 4).


We describe herein a study performed in Sardinia in which we have defined markers spanning the HLA region in patients with AS and controls positive for either B*2705 or B*2709. The results show that B*2705- and B*2709-positive individuals share a central region downstream of the HLA–B gene toward the MICB region, whereas the other markers differ significantly. Another interesting observation is that these individuals also share a polymorphism in the HLA–E gene, which has been shown to influence the expression of the molecule on the cell surface (28). Allele A, encoding for R128, associates with both the B*2705 and B*2709 alleles. Genotype distribution, however, clearly shows a sharp increase in the frequency of allele G encoding for G128 in B*2705 and B*2709–positive controls (76%) compared with AS patients (47%) and randomly selected controls (46%), indicating that the presence of a G at position 128 is likely to protect against AS.

It has been shown that the HLA–E molecules bind self peptides mostly derived from the leader sequences of HLA class I molecules (29, 30). HLA–B27 molecules possess a leader sequence that is bound by the HLA–E molecules with low efficiency, especially by those with Arg-128. Therefore, patients with AS possessing the genotype AA are likely to express a lower amount of HLA–E molecules presenting the leader sequence of HLA–B27, compared with individuals expressing the AG genotype, as found in 76% of B27-positive controls. Moreover, an association with protection against the disease is also evident with marker D6S2674, which is located between the HLA–C and HLA–B genes. We did not find any significant difference in the frequency of HLA–B alleles coexpressed with HLA–B27. Conversely, in patients, there was increased homozygosity for HLA–Cw2, as well as a higher frequency of MICA-A5 in combination with MICA-A4, a marker of both the B*2705 and B*2709 haplotypes. Interestingly, in Sardinia, MICA-A5 is also associated with HLA–B*2707, an allele found in patients (15), although evidence that it is protective has been found in other populations (27). It is possible that this difference could be due to other genes mapping close to the B*2707 gene rather than to a different function of the HLA–B27 molecules per se (31).

Interestingly, allelic frequencies differentiating patients from B*2705-matched controls map in genes that are reported to be ligands for NK receptors, such as HLA–E, HLA–C, and MICA genes. There is a plethora of NK receptors that are positive or negative regulators of the NK activity expressed by T lymphocytes, as well. Therefore, it can be speculated that T effector cells restricted for HLA–B27 molecules might be modulated by the interaction of these receptors with nonclassic HLA molecules (32, 33).

A recent study by Lopez-Larrea et al has shown that the combination of a killer cell immunoglobulin-like receptor (KIR3DL1) that binds the HLA–B27 molecules and the presence in the co-occurring chromosome of the corresponding Bw4 epitope is protective against AS (34). We did not identify any significant variation in the HLA–A and B alleles expressed trans of the HLA–A2-B*2705 haplotype, or any strong difference in the expression of the Bw4 epitope, as derived from theoretical attribution based on HLA–B types. However, Lopez-Larrea and colleagues' data, as well as ours, indicate a balance of the inhibitory and activating NK receptors and their HLA counterparts as possible regulators of the disease.

HLA–E molecules have been shown to be relevant in recognition of viral peptides (35), in protection against viral infections (36, 37), and in presentation of Salmonella enterica peptides (38). Moreover, they have been reported to be associated with age at onset of type 1 diabetes mellitus (39). Therefore, a specific role of the HLA–E gene in AS would not be surprising, although it would need to be confirmed in additional studies.

The availability of a population such as that of Sardinia, in which most AS patients possess a highly conserved haplotype shared by B*2705-positive controls, allowed us to single out genes that might be responsible for protection rather than for disease susceptibility. It should be noted that the subjects used as healthy controls were blood donors and therefore showed no clinical signs of disease. However, it might be possible that some of them have subclinical disease at time of blood donation. Therefore, the data reported here, which indicate increased heterozygosity among HLA–B27–positive controls, might have more meaning if the controls could be evaluated longitudinally. In this respect, the findings in a cohort of B*2709-positive subjects, who can be reasonably considered as being healthy, lend support to the hypothesis that there are protective factors carried by the co-occurring chromosome, since they showed the highest frequency of Arg/Gly heterozygosity. Interestingly, we have recently identified a patient with AS who is B*2709 positive, and who is homozygous for allele 1 in D6S2674, a marker of the B*2705 haplotype, and has Arg/Arg at position 128 of HLA–E (40). While this is necessarily an anedoctical report, it might represent an interesting control for further studies.

It is also of note that HLA–E R128 is strongly associated with HLA–B27, which possess a leader sequence that is very weakly or not at all bound by the HLA–E R128 molecules. Besides the obvious explanation of a founder haplotype, the possibility should be taken into account, as recently demonstrated for the HLA–DR2 haplotype (41), that this configuration confers some selective advantage to this haplotype, which might have as a negative counterpart an association with increased susceptibility to seronegative rheumatic diseases, since HLA–B27 is associated not only with AS but also with other inflammatory rheumatic conditions (42). In this light, it is possible that the presence of HLA–E G128 might counteract this susceptibility.

In conclusion, we have demonstrated here the presence of markers that are significantly increased in B*2705-positive controls, suggesting a disease-protective role of genes likely to act as regulators of NK activity. Given the large spectrum of ligands as well as of receptors orchestrating NK functions, it would not be surprising that the same biologic effects could be obtained by a fine-tuning of different pairs of ligand–receptors. This could make it difficult to identify single genes associated with AS throughout all populations, because different common variants could be involved in different populations depending on their genetic stratification. For this reason, study of isolated populations might be more informative in suggesting biochemical pathways for further exploration. In this respect, our results direct attention to the NK function as a possible relevant cofactor in AS pathogenesis.


Dr. Sorrentino had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study design. Cascino, Mathieu, Sorrentino.

Acquisition of data. Paladini, Belfiore, Cauli.

Analysis and interpretation of data. Cascino, Fiorillo, Sorrentino.

Manuscript preparation. Cascino, Sorrentino.

Statistical analysis. Angelini.


The authors would like to thank the study patients for their cooperation, and Federica Lucantoni, and Eleuteria Lancia for technical assistance.