Abstract
The major locus for multiple sclerosis (MS) susceptibility is located within the class II region of the Major Histocompatibility Complex (MHC). HLA-DRB1 alleles, constituting the strongest MS susceptibility factors, have been widely exploited in research including construction of transgenic animal models of MS. Many studies have concluded that HLA-DRB1*15 allele itself determines MS-associated susceptibility. If this were true, haplotypes bearing this allele would confer equal risk. If HLA-DRB1*15 bearing haplotypes differed for risk, roles for other loci in this region would be implied and further study of the fine structure of this locus would be compelling. We have tested the hypothesis comparing haplotypes stratified by HLA class I tagging. We show here that HLA-DRB1*15-bearing-haplotypes in 1970 individuals from 494 MS families are indeed heterogeneous. Some HLA-DRB1*15 haplotypes determine susceptibility while others do not. Three groups of class I tagged HLA-DRB1*15 haplotypes were not over-transmitted: (i) HLA-DRB1*15-HLA-B*08 (TR = 25, NT = 23, Odds Ratio = 1.09), (ii) -HLA-B*27 (TR = 18, NT = 17, Odds Ratio = 1.06), and (iii) rare HLA-DRB1*15 haplotypes (frequency <0.02). Rare haplotypes were significantly different from common haplotypes, and transmissions were remarkably similar to those for class-I-matched non-HLA-DRB1*15 haplotypes. These results unambiguously indicate that HLA-DRB1*15 is part of a susceptibility haplotype but cannot be the susceptibility allele itself, requiring either epistatic interactions, epigenetic modifications on some haplotypes, or nearby structural variation. These findings strongly imply that differences among HLA-DRB1*15 haplotypes will furnish the basis for MHC-associated susceptibility in MS and raise the possibility that the MHC haplotype is the fundamental unit of genetic control of immune response.
Keywords: MHC, class II, transmission
The origins of multiple sclerosis (MS) susceptibility, as with many complex genetic diseases, remain uncertain. It is widely believed that MS is a CD4+ Th1-mediated autoimmune disease (1, 2). This is supported by many studies based on experimental autoimmune encephalomyelitis (EAE) models (3, 4), and by genetic studies showing HLA class II genes represent the strongest risk for MS (5–9). It is often assumed that the MS DR-association is secondary to a role in antigen-presentation to CD4+ T cells responding to a myelin protein (2).
The evidence that familial risk in MS is determined by genes is overwhelming, although broadly acting and as yet unidentified environmental factors are undeniable in accounting for the disease geography (10). Associations between MS and specific HLA alleles are clear, but MS does not display the classical Mendelian inheritance patterns attributable to a single locus (5, 11). Genome scans have shown few consensus linkages (5, 12–15) or associations (16) additional to HLA genes. The highest non-Major Histocompatibility Complex (MHC) lod score previously reported was for markers in chromosome 5p in the initial Canadian cohort (5), which was also positive in a Scandinavian study (17). Zhang and colleagues (2005) found association with the 5p candidate interleukin 7 (IL-7) receptor (15), subsequently confirmed with supporting functional data (18, 19). A whole genome association study has shown few significant effects outside what had been previously highlighted by linkage, and the effect sizes are very small at a population level (16), but larger effects may pertain to individual families. Methods used to estimate the contribution from the MHC may be inappropriate since the assumptions made are not sustainable in MS (20). It has been suggested that some 50% of susceptibility remains to be found (21).
These efforts highlight the MHC as the main susceptibility locus for MS. HLA-DRB1 allelic subtypes have been widely explored, including the construction of transgenic animal models of MS transgenic for HLA-DRB1*15. Complexity in this region clearly indicates an epistatic hierarchy of resistance and susceptibility alleles at HLA-DRB1, which interact in cis and trans to influence overall risk (8). Dense single nucleotide polymorphism (SNP) typing localized association to the immediate region of HLA class II (9). Reports from several case-control studies suggesting an independent association with the HLA class I region (22–24) appear to be secondary to linkage disequilibrium (LD) in family-based material (25). The problems of correcting for LD and for concomitant epistatic interactions, which may be haplotype-specific, are formidable for case-control studies.
Complex HLA-DRB1 interactions have been demonstrated, and specific susceptibility and resistance alleles and interactions among them in trans have been identified (8). The roles of these interactions have been extended in a recent study by evaluating HLA-DRB1, HLA-A, and HLA-B haplotypes requiring the construction of 2-locus haplotypes (25). HLA-DRB1*15 haplotypes were equally over-transmitted regardless of what common allele was present at HLA-A and -B in cis or in trans, illustrating no detectable independent effect of class I on susceptibility. These findings did not rule out a class II-dependent functional interaction for class I indeed the long-range LD surrounding HLA class I and II seen in this study as has characterized many other systems has remained unexplained (25).
Analogies with murine immunogenetic studies imply that HLA-DRB1 loci themselves are involved directly in autoimmune susceptibility but the data remain circumstantial in the case of MS. Importantly, the probability of haplotype sharing of non-HLA-DRB1*15 alleles between affected sibling pairs unexpectedly does not differ (26). Based on findings from pooled low frequency haplotypes (25), we hypothesized that constructing HLA class I and II haplotypes might differentiate among HLA-DRB1*15 alleles, imply MS susceptibility is not solely HLA-DRB1*15 based and provide a tool for identifying the distinguishing features. Here we show that HLA class I alleles present in cis allow construction of haplotypes, which demonstrate marked risk heterogeneity, imply the HLA-DRB1 alleles do not mediate risk themselves and highlight the potential for regulatory or epigenetic effects in this region.
Results
Two-Locus Haplotype Transmissions of HLA-DRB1*15-HLA-A and HLA-DRB1*X-HLA-A.
Overall, HLA-DRB1*15-HLA-A haplotypes were 529 times transmitted and 260 times not transmitted (P = 1.00 × 10−21; Table 1). Most HLA-DRB1*15-HLA-A haplotypes, but not all, appeared to be over-transmitted from parents to affected offspring. Haplotypes that were significantly over-transmitted included HLA-DRB1*15-HLA-A*01 (P = 0.0040), -A*02 (P = 3.64 × 10−09), -A*03 (P = 5.67 × 10−09), -A*11 (P = 0.034), -A*24 (P = 1.07 × 10−11), and -A*25 (P = 0.0019) haplotypes (Table 1). Some haplotypes were especially over-transmitted (for example: HLA-DRB1*15-HLA-A*02, -A*03, -A*24), but even the large size of this sample does not allow for definitive comparison among over-transmitted haplotypes given the number of haplotypes and potential comparisons.
Table 1.
Two-locus haplotype transmissions of HLA-DRB1*15−HLA-A and HLA-DRB1*X−HLA-A from HLA-DRB1*15 positive (n = 536) and negative parents (n = 452)
|
HLA-DRB1*15 Positive Parents (n = 536) |
HLA-DRB1*15 Negative Parents (n = 452) |
Comparison |
|||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Haplotypes | TR | NT | OR | χ2 | P | Haplotypes | TR | NT | OR | χ2 | P | χ2 | P |
| 15-1 | 51 | 26 | 1.96 | 8.27 | 0.0040 | X-1 | 119 | 113 | 1.05 | 0.16 | 0.69 | 5.21 | 0.022 |
| 15-2 | 136 | 55 | 2.47 | 34.81 | 3.64 × 10−9 | X-2 | 145 | 157 | 0.92 | 0.48 | 0.49 | 25.67 | 1.051 × 10−7 |
| 15-3 | 140 | 59 | 2.37 | 33.95 | 5.67 × 10−9 | X-3 | 93 | 73 | 1.27 | 2.41 | 0.12 | 8.048 | 0.0046 |
| 15-11 | 41 | 24 | 1.71 | 4.50 | 0.034 | X-11 | 50 | 42 | 1.19 | 0.70 | 0.40 | 1.19 | 0.28 |
| 15-23 | 4 | 1 | 4.00 | 1.93 | 0.17 | X-23 | 13 | 12 | 1.08 | 0.04 | 0.84 | —† | 0.22 |
| 15-24 | 92 | 22 | 4.18 | 46.20 | 1.07 × 10−11 | X-24 | 51 | 63 | 0.81 | 1.26 | 0.26 | 31.53 | 1.96 × 10−8 |
| 15-25 | 20 | 5 | 4.00 | 9.64 | 0.0019 | X-25 | 13 | 7 | 1.86 | 1.80 | 0.18 | —† | 0.14 |
| 15-26 | 11 | 9 | 1.22 | 0.20 | 0.65 | X-26 | 12 | 27 | 0.44 | 5.77 | 0.016 | —† | 0.046 |
| 15-29 | 4 | 16 | 0.25 | 7.71 | 0.0055 | X-29 | 14 | 12 | 1.17 | 0.15 | 0.69 | 5.44 | 0.020 |
| 15-30 | 6 | 20 | 0.30 | 7.95 | 0.0048 | X-30 | 9 | 15 | 0.60 | 1.50 | 0.22 | 1.24 | 0.27 |
| 15-31 | 2 | 4 | 0.50 | 0.68 | 0.41 | X-31 | 17 | 12 | 1.42 | 0.86 | 0.35 | —† | 0.19 |
| 15-32 | 10 | 4 | 2.50 | 2.66 | 0.10 | X-32 | 12 | 15 | 0.80 | 0.33 | 0.56 | —† | 0.071 |
| 15-33 | 2 | 10 | 0.20 | 5.82 | 0.016 | X-33 | 5 | 11 | 0.45 | 2.25 | 0.13 | 0.78 | 0.38 |
| 15-68 | 10 | 5 | 2.00 | 1.70 | 0.19 | X-68 | 24 | 37 | 0.65 | 2.77 | 0.10 | —† | 0.039 |
| TOTAL | 529 | 260 | 2.035 | 91.71 | 1.00 × 10−21 | TOTAL | 577 | 596 | 0.97 | 0.31 | 0.58 | 61.16 | 5.26 × 10−15 |
HLA-DRB1*X refers to any allele other than HLA-DRB1*15.
†Fisher's exact test was used when the expected transmissions in any of the cells of the table were below 10. Uninformative MS families were excluded in this analysis, which may contain parents homozygous for both class I/II, and/or HLA identical parents for both class I/II, and/or with missing class I/II typing.
Unexpectedly, some haplotypes such as HLA-DRB1*15-HLA-A*26 (P = 0.65) were neutral-transmitted, whereas other haplotypes were significantly under-transmitted including HLA-DRB1*15-HLA-A*29 (P = 0.0055), -A*30 (P = 0.0048), and -A*33 (P = 0.016) haplotypes (Table 1). Increased transmission of HLA-DRB1*15-associated HLA-A alleles was not present when parents lacked HLA-DRB1*15, and no significant transmission distortions were observed for the totalled HLA-DRB1*X-HLA-A haplotypes (where X is any HLA-DRB1 allele other than HLA-DRB1*15) (P = 0.58) except for HLA-DRB1*X-HLA-A*26 (P = 0.016), which was significantly under-transmitted (Table 1).
Significant differences were found between most of the commonly over-transmitted HLA-DRB1*15-HLA-A haplotypes and their paired corresponding HLA-DRB1*X-HLA-A haplotypes. Conversely, neutral-transmitted and under-transmitted HLA-DRB1*15-HLA-A haplotypes had similar transmission patterns as compared with their paired HLA-DRB1*X-HLA-A haplotypes (Table 1).
Two-Locus Haplotype Transmissions of HLA-DRB1*15-HLA-B and HLA-DRB1*X-HLA-B.
Mirroring the transmissions of HLA-DRB1*15-HLA-A haplotypes, most HLA-DRB1*15-HLA-B haplotypes also appeared to be over-transmitted including HLA-DRB1*15-HLA-B*07 (P = 3.19 × 10−17), -B*18 (P = 5.20 × 10−05), -B*37 (P = 0.041), -B*40 (P = 0.0030), -B*44 (P = 0.0059), -B*49 (P = 0.039), -B*51 (P = 0.00021), and -B*57 (P = 0.00018) haplotypes (Table 2).
Table 2.
Two-locus haplotype transmissions of HLA-DRB1*15−HLA-B and HLA-DRB1*X−HLA-B from HLA-DRB1*15 positive (n = 536) and negative parents (n = 452)
|
HLA-DRB1*15 Positive Parents (n = 536) |
HLA-DRB1*15 Negative Parents (n = 452) |
Comparison |
|||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Haplotypes | TR | NT | OR | χ2 | P | Haplotypes | TR | NT | OR | χ2 | P | χ2 | P |
| 15-7 | 263 | 104 | 2.53 | 71.22 | 3.19 × 10−17 | X-7 | 60 | 58 | 1.03 | 0.03 | 0.85 | 17.39 | 3.044 × 10−5 |
| 15-8 | 25 | 23 | 1.09 | 0.08 | 0.77 | X-8 | 68 | 61 | 1.11 | 0.38 | 0.54 | 0.0056 | 0.94 |
| 15-13 | 4 | 10 | 0.40 | 2.66 | 0.10 | X-13 | 11 | 16 | 0.69 | 0.93 | 0.34 | 0.59 | 0.44 |
| 15-14 | 6 | 4 | 1.50 | 0.40 | 0.53 | X-14 | 14 | 12 | 1.17 | 0.15 | 0.69 | —† | 0.28 |
| 15-15 | 6 | 2 | 3.00 | 2.09 | 0.15 | X-15 | 29 | 26 | 1.12 | 0.16 | 0.69 | —† | 0.16 |
| 15-18 | 31 | 7 | 4.43 | 16.37 | 5.20 × 10−5 | X-18 | 31 | 21 | 1.48 | 1.92 | 0.17 | —† | 0.016 |
| 15-27 | 18 | 17 | 1.06 | 0.03 | 0.87 | X-27 | 33 | 23 | 1.43 | 1.79 | 0.18 | 0.49 | 0.48 |
| 15-35 | 19 | 11 | 1.73 | 2.16 | 0.14 | X-35 | 67 | 82 | 0.82 | 1.51 | 0.22 | 3.37 | 0.066 |
| 15-37 | 12 | 4 | 3.00 | 4.19 | 0.041 | X-37 | 11 | 15 | 0.73 | 0.62 | 0.43 | —† | 0.031 |
| 15-38 | 2 | 6 | 0.33 | 2.09 | 0.15 | X-38 | 3 | 10 | 0.30 | 3.77 | 0.052 | —† | 0.39 |
| 15-39 | 3 | 0 | 0.00 | 4.16 | 0.041 | X-39 | 9 | 24 | 0.38 | 6.82 | 0.0090 | —† | 0.031 |
| 15-40 | 21 | 6 | 3.50 | 8.83 | 0.0030 | X-40 | 14 | 17 | 0.82 | 0.29 | 0.59 | —† | 0.0090 |
| 15-41 | 2 | 4 | 0.50 | 0.68 | 0.41 | X-41 | 5 | 4 | 1.25 | 0.11 | 0.74 | —† | 0.29 |
| 15-44 | 47 | 24 | 1.96 | 7.59 | 0.0059 | X-44 | 71 | 66 | 1.08 | 0.18 | 0.67 | 3.94 | 0.047 |
| 15-49 | 20 | 9 | 2.22 | 4.28 | 0.039 | X-49 | 6 | 17 | 0.35 | 5.26 | 0.022 | —† | 0.0020 |
| 15-51 | 22 | 4 | 5.50 | 13.72 | 0.00021 | X-51 | 26 | 26 | 1.00 | 0.00 | 1.00 | —† | 0.0022 |
| 15-52 | 1 | 6 | 0.17 | 3.96 | 0.047 | X-52 | 1 | 1 | 1.00 | 0.00 | 1.00 | —† | 0.39 |
| 15-55 | 3 | 0 | 0.00 | 4.16 | 0.041 | X-55 | 2 | 8 | 0.25 | 3.60 | 0.058 | —† | 0.035 |
| 15-56 | 1 | 4 | 0.25 | 1.93 | 0.17 | X-56 | 2 | 0 | 0.00 | 2.00 | 0.16 | —† | 0.14 |
| 15-57 | 20 | 3 | 6.67 | 14.07 | 0.00018 | X-57 | 19 | 15 | 1.27 | 0.47 | 0.49 | —† | 0.011 |
| 15-58 | 6 | 4 | 1.50 | 0.40 | 0.53 | X-58 | 3 | 7 | 0.43 | 1.60 | 0.21 | —† | 0.15 |
| 15-60 | 1 | 4 | 0.25 | 1.93 | 0.17 | X-60 | 16 | 18 | 0.89 | 0.12 | 0.73 | —† | 0.22 |
| 15-62 | 3 | 0 | 0.00 | 4.16 | 0.041 | X-62 | 27 | 24 | 1.13 | 0.18 | 0.67 | —† | 0.16 |
| 15-64 | 2 | 4 | 0.50 | 0.68 | 0.41 | X-64 | 14 | 22 | 0.64 | 1.78 | 0.18 | —† | 0.34 |
| TOTAL | 538 | 260 | 2.069 | 96.85 | 7.49 × 10−23 | TOTAL | 542 | 573 | 0.95 | 0.86 | 0.35 | 55.46 | 9.54 × 10−14 |
HLA-DRB1*X refers to any allele other than HLA-DRB1*15.
†Fisher's exact test was used when the expected transmissions in any of the cells of the table were below 10. Uninformative MS families were excluded in this analysis, which may contain parents homozygous for both class I/II, and/or HLA identical parents for both class I/II, and/or with missing class I/II typing.
Haplotypes such as HLA-DRB1*15-HLA-B*08 (P = 0.77) and -B*27 (P = 0.87) were neutral-transmitted, whereas other haplotypes including HLA-DRB1*15-HLA-B*13 (P = 0.10), -B*38 (P = 0.15), and -B*52 (P = 0.047) seemed to be under-transmitted (Table 2). The totalled transmission of HLA-DRB1*15-HLA-B haplotypes was 538 times transmitted and 260 times not transmitted (P = 7.49 × 10−23; Table 2).
HLA-DRB1*X-HLA-B haplotypes were not over-transmitted from HLA-DRB1*15 negative parents, and there were no significant transmission distortions observed for the totalled HLA-DRB1*X-HLA-B haplotypes (P = 0.35) except for HLA-DRB1*X-HLA-B*39 (P = 0.0090) and -B*49 (P = 0.022) haplotypes, which were significantly under-transmitted (Table 2).
Significant differences were found between the common HLA-DRB1*15-HLA-B haplotypes and their paired HLA-DRB1*X-HLA-B haplotypes. On the other hand, as with HLA-A haplotypes, the rare HLA-DRB1*15 bearing haplotypes had similar transmission patterns as compared with their corresponding HLA-DRB1*X-HLA-B haplotypes (Table 2).
Two-Locus Haplotype Transmissions of HLA-DRB1*15-HLA-A/B and HLA-DRB1*X-HLA-A/B: Categorized by Their HLA Class I Broad Serological Equivalents.
In the HLA-A broad serological equivalent analyses, HLA-A9 (with closely related HLA-A*23, and -A*24 allele groups; P = 2.20 × 10−11), and -A10 (with closely related HLA-A*25, and -A*26 allele groups; P = 0.011) carrying HLA-DRB1*15 were significantly over-transmitted, whereas HLA-A19 haplotypes carrying HLA-DRB1*15 (with closely related HLA-A*29, -A*30, -A*31, -A*32, and -A*33 allele groups; P = 0.00068) were significantly under-transmitted. No significant transmission distortions of non-HLA-DRB1*15 haplotypes were observed (Table 3).
Table 3.
Two-locus haplotype transmissions of HLA-DRB1*15/X−HLA-A and HLA-DRB1*15/X−HLA-B from HLA-DRB1*15 positive (n = 536) and negative parents (n = 452) pooled by their HLA class I broad serological equivalents
| Broad Serological Equivalents |
HLA-DRB1*15 Positive Parents (n = 536) |
HLA-DRB1*15 Negative Parents (n = 452) |
Comparison |
|||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Haplotypes | TR | NT | OR | χ2 | P | Haplotypes | TR | NT | OR | χ2 | P | χ2 | P | |
| HLA-A9 | 15-23 | 4 | 1 | 4.00 | 1.93 | 0.17 | X-23 | 13 | 12 | 1.08 | 0.04 | 0.84 | —† | 0.22 |
| 15-24 | 92 | 22 | 4.18 | 46.20 | 1.07 × 10−11 | X-24 | 51 | 63 | 0.81 | 1.26 | 0.26 | 31.53 | 1.96 × 10−8 | |
| Total | 96 | 23 | 4.17 | 44.78 | 2.20 × 10−11 | Total | 64 | 75 | 0.85 | 0.87 | 0.35 | 32.64 | 1.11 × 10−8 | |
| HLA-A10 | 15-25 | 20 | 5 | 4.00 | 9.64 | 0.0019 | X-25 | 13 | 7 | 1.86 | 1.80 | 0.18 | —† | 0.14 |
| 15-26 | 11 | 9 | 1.22 | 0.20 | 0.65 | X-26 | 12 | 27 | 0.44 | 5.77 | 0.016 | —† | 0.046 | |
| Total | 31 | 14 | 2.21 | 6.42 | 0.011 | Total | 25 | 34 | 0.74 | 1.37 | 0.24 | 7.22 | 0.0072 | |
| HLA-A19 | 15-29 | 4 | 16 | 0.25 | 7.71 | 0.0055 | X-29 | 14 | 12 | 1.17 | 0.15 | 0.69 | 5.44 | 0.020 |
| 15-30 | 6 | 20 | 0.30 | 7.95 | 0.0048 | X-30 | 9 | 15 | 0.60 | 1.50 | 0.22 | 1.24 | 0.27 | |
| 15-31 | 2 | 4 | 0.50 | 0.68 | 0.41 | X-31 | 17 | 12 | 1.42 | 0.86 | 0.35 | —† | 0.19 | |
| 15-32 | 10 | 4 | 2.50 | 2.66 | 0.10 | X-32 | 12 | 15 | 0.80 | 0.33 | 0.56 | —† | 0.071 | |
| 15-33 | 2 | 10 | 0.20 | 5.82 | 0.016 | X-33 | 5 | 11 | 0.45 | 2.25 | 0.13 | 0.78 | 0.38 | |
| Total | 24 | 54 | 0.44 | 11.54 | 0.00068 | Total | 57 | 65 | 0.88 | 0.53 | 0.47 | 5.024 | 0.025 | |
| HLA-B15 | 15-51 | 22 | 4 | 5.50 | 13.72 | 0.00021 | X-51 | 26 | 26 | 1.00 | 0.00 | 1.00 | —† | 0.0022 |
| 15-52 | 1 | 6 | 0.17 | 3.96 | 0.047 | X-52 | 1 | 1 | 1.00 | 0.00 | 1.00 | —† | 0.39 | |
| Total | 23 | 10 | 2.30 | 5.12 | 0.024 | Total | 27 | 27 | 1.00 | 0.00 | 1.00 | 3.25 | 0.071 | |
| HLA-B16 | 15-38 | 2 | 6 | 0.33 | 2.09 | 0.15 | X-38 | 3 | 10 | 0.30 | 3.77 | 0.052 | —† | 0.39 |
| 15-39 | 3 | 0 | 0.00 | 4.16 | 0.041 | X-39 | 9 | 24 | 0.38 | 6.82 | 0.0090 | —† | 0.031 | |
| Total | 5 | 6 | 0.83 | 0.091 | 0.76 | Total | 12 | 34 | 0.35 | 10.52 | 0.0012 | —† | 0.13 | |
| HLA-B17 | 15-57 | 20 | 3 | 6.67 | 14.07 | 0.00018 | X-57 | 19 | 15 | 1.27 | 0.47 | 0.49 | —† | 0.011 |
| 15-58 | 6 | 4 | 1.50 | 0.40 | 0.53 | X-58 | 3 | 7 | 0.43 | 1.60 | 0.21 | —† | 0.15 | |
| Total | 26 | 7 | 3.71 | 10.94 | 0.00094 | Total | 22 | 22 | 1.00 | 0.00 | 1.00 | —† | 0.0068 | |
| HLA-B22 | 15-55 | 3 | 0 | 0.00 | 4.16 | 0.041 | X-55 | 2 | 8 | 0.25 | 3.60 | 0.058 | —† | 0.035 |
| 15-56 | 1 | 4 | 0.25 | 1.93 | 0.17 | X-56 | 2 | 0 | 0.00 | 2.00 | 0.16 | —† | 0.14 | |
| Total | 4 | 4 | 1.00 | 0.00 | 1.00 | Total | 4 | 8 | 0.5 | 1.33 | 0.25 | —† | 0.28 | |
HLA-DRB1*X refers to any allele other than HLA-DRB1*15.
†Fisher's exact test was used when the expected transmissions in any of the cells of the table were below 10. HLA-A*23 and -A*24 are serological equivalents of HLA-A9. HLA-A*25 and -A*26 are serological equivalents of HLA-A10. HLA-A*29, -A*30, -A*31, -A*32, and -A*33 are serological equivalents of HLA-A19. HLA-B*51 and -B*52 are serological equivalents of HLA-B15. HLA-B*38 and -B*39 are serological equivalents of HLA-B16. HLA-B*57 and -B*58 are serological equivalents of HLA-B17. HLA-B*55 and -B*56 are serological equivalents of HLA-B22.
In the HLA-B broad serological equivalent analyses, HLA-B15 (with closely related HLA-B*51, and -B*52 allele groups; P = 0.024) and -B17 (with closely related HLA-B*57, and -B*58 allele groups; P = 0.00094) were significantly over-transmitted in the presence of HLA-DRB1*15, whereas HLA-B16 (with closely related HLA-B*38, and -B*39 allele groups; P = 0.76) and -B22 (with closely related HLA-B*55, and -B*56 allele groups; P = 1.00) were neutral-transmitted with HLA-DRB1*15. No significant transmission distortions of non-HLA-DRB1*15 haplotypes were observed, except for HLA-DRB1*X-HLA-B16 (with closely related HLA-B*38, and -B*39 allele groups; P = 0.0012), which was significantly under-transmitted (Table 3).
Transmission of Common and Rare HLA-DRB1*15 Haplotypes.
Two-locus (HLA-DRB1*15-HLA-A and HLA-DRB1*15-HLA-B) haplotypes were separated into 2 groups: (i) common, and (ii) rare, based on their haplotype frequency (greater or less than 0.02). The common HLA-DRB1*15-HLA-A and HLA-DRB1*15-HLA-B haplotypes were significantly different as compared with their corresponding rare HLA-DRB1*15 bearing haplotypes (P = 1.60 × 10−10, P = 1.12 × 10−08, respectively). The common HLA-DRB1*15 positive haplotypes were found to be significantly over-transmitted, whereas the rare HLA-DRB1*15 positive haplotypes were mostly neutral-transmitted or under-transmitted (Table 4).
Table 4.
A comparison of common and rare HLA-DRB1*15−HLA-A and HLA-DRB1*15−HLA-B haplotypes transmission
| Haplotypes | Common Haplotypes |
Rare Haplotypes |
Comparison |
|||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| TR | NT | OR | χ2 | P | TR | NT | OR | χ2 | P | χ2 | P | |
| HLA-DRB1*15−HLA-A | 480 | 191 | 2.51 | 124.47 | 6.64 × 10−29 | 49 | 69 | 0.71 | 3.39 | 0.066 | 40.90 | 1.60 × 10−10 |
| HLA-DRB1*15−HLA-B | 470 | 184 | 2.55 | 125.07 | 4.91 × 10−29 | 68 | 76 | 0.89 | 0.44 | 0.50 | 32.63 | 1.12 × 10−8 |
Two-locus (HLA-DRB1*15−HLA-A, and HLA-DRB1*15−HLA-B) haplotypes were separated into two groups: 1) Common haplotypes with haplotype frequency greater than 0.02, and 2) Rare haplotypes with haplotype frequency less than 0.02.
Discussion
A strong association between the HLA-DRB1*15 allele and MS has been shown in studies of northern Europeans and their descendants. MS susceptibility has been variably attributed to alleles at the HLA-DR (27, 28) and HLA-DQ loci (29), but epistasis (gene-gene interactions) in this region is demonstrably strong (8, 29). Unexpectedly, sharing of haplotypes between affected sibling pairs from non-HLA-DRB1*15 bearing families is not less than in HLA-DRB1*15 bearing families (26). This finding suggested heterogeneity among HLA-DRB1*15 haplotypes. The concept that risk is dependent on factors peculiar to some but not all haplotypes thereby directs attention away from the structural elements of HLA-DRB1 gene itself.
This study has incorporated family data with haplotype transmission disequilibrium test (TDT), which simultaneously assesses linkage and association. Haplotypes produce more definitive transmissions than do the alleles encompassing them, and this tends to increase power. However, the larger number of haplotypes relative to alleles at individual loci tends to decrease power due to the additional degrees of freedom required for the analysis (30) and may account for why the findings reported here have not previously been detected.
To overcome these limitations, the current study included an exceptionally large sample of MS families, focusing analysis on particular haplotypes and groups of haplotypes. HLA-DRB1*15 haplotypes have been assembled into their HLA-A or HLA-B broad serological equivalents (see Table 3), and have also been assigned as either common or rare based on their transmission frequencies (see Table 4).
In the analyses of HLA-A broad serological equivalents, we found that the closely related HLA-A*23 and -A*24 allele groups (HLA-A9), and HLA-A*25 and -A*26 allele groups (HLA-A10) carrying HLA-DRB1*15 were significantly over-transmitted, whereas HLA-A19 haplotypes (with closely related HLA-A*29, -A*30, -A*31, -A*32, and -A*33 allele groups) carrying HLA-DRB1*15 was significantly under-transmitted. For the HLA-B broad serological equivalent analyses, HLA-B15 (with closely related HLA-B*51, and -B*52 allele groups) and -B17 (with closely related HLA-B*57, and -B*58 allele groups) were significantly over-transmitted in the presence of HLA-DRB1*15, whereas HLA-DRB1*15 bearing HLA-B16 and -B22 haplotypes were neutral-transmitted (including closely related HLA-B*38, -B*39 allele groups, and HLA-B*55, -B*56 allele groups respectively; see Table 3).
We have shown in MS that alleles at HLA-DRB1, HLA-A and HLA-B are not randomly transmitted, instead alleles tend to be associated with other alleles in a set or cassette of common haplotypes. This long range and discontinuous LD between HLA class I and II has already been described previously (25). Since transmissions to affected offspring among parental HLA-DRB1*15 haplotypes are significantly different, it is plausible that HLA-DRB1*15 bearing HLA-A9 (HLA-A23, and -A*24), -A10 (HLA-A*25, and -A*26), -B15 (HLA-B*51, and -B*52) and -B17 (HLA-B*57, and -B*58) haplotypes had different functional properties under selection as compared with HLA-DRB1*15 bearing HLA-A19 (HLA-A*29, -A*30, -A*31, -A*32, and -A*33), -B16 (HLA-B*38, and -B*39) and -B22 (HLA-B*55, and -B*56) haplotypes. These observations may well reflect more general characteristics of this region, which may be present in many other autoimmune related diseases.
General over-transmission of HLA-DRB1*15 positive 2-locus haplotypes have been observed for most of the common haplotypes (see Table 4). However there were notable exceptions, with HLA-DRB1*15-HLA-B*08 and HLA-DRB1*15-HLA-B*27, 2 of the common haplotypes, along with pooled rare haplotypes taken together. Not only were the transmissions of rare haplotypes significantly different from the common haplotypes, but their transmissions were also remarkably similar as compared with non-HLA-DRB1*15 haplotypes matched at HLA class I. For example, when parents are positive for HLA-DRB1*15, the transmission of HLA-DRB1*15 haplotypes carrying HLA-B*07 was significantly increased, but the transmission of HLA-DRB1*15 haplotypes bearing HLA-B*08 was neutral. Furthermore, the transmission of HLA-DRB1*15-HLA-B*08 (Odds Ratio [OR] = 1.09) was no different from its paired HLA-DRB1*X-HLA-B*08 haplotype (OR = 1.11). This differential transmission was also evident among other 2-locus HLA-DRB1*15/X haplotypes.
These observations indicate unequivocally that HLA-DRB1*15 haplotypes are heterogeneous and not all HLA-DRB1*15 haplotypes are associated with MS susceptibility. This contradicts expectations of disease-related antigen-specific HLA-DRB1 allele restriction as the basis for MHC association. TDT analysis of HLA class I and class II haplotypes reveals at least 3 populations of HLA-DRB1*15 haplotypes, including those that are over-transmitted (susceptible), neutral-transmitted/under-transmitted (non-susceptible). The differential transmissions of these susceptible and non-susceptible HLA-DRB1*15 haplotypes were confirmed by the contingency table analysis for common and rare haplotypes (see Table 4). This is not accounted for by differences in the alleles themselves and places focus on adjacent variation in regulatory regions and on the importance of extended haplotypes containing alleles in LD.
In this study we have demonstrated that HLA-DRB1*15 is part of a susceptibility haplotype, but cannot be viewed as a MS susceptibility allele itself. The alleles or haplotypes can be considered as markers for differential susceptibility carried by these haplotypes, which would generate a ready approach to identify haplotype-specific variation responsible for disease risk in a region of the genome characterized by extraordinary polymorphism. Furthermore, increased haplotype sharing accompanied by the absence of distorted haplotype transmission in the HLA-DRB1*15 negative families highlights the possibility of epigenetic modification of HLA-DRB1 haplotypes.
These novel findings strongly imply that differences between susceptible (over-transmitted) and non-susceptible (neutral-transmitted/under-transmitted) HLA-DRB1*15 haplotypes in HLA-DRB1*15 positive families, complemented by increased haplotype sharing in the absence of increased haplotype transmission in HLA-DRB1*15 negative families will contain the basis for MHC-associated risk in MS and may shed light on the nature of HLA class I−II LD. We also believe that these findings will be a general phenomenon, with wider relevance for other MHC-associated complex disease phenotypes, and animal model construction.
The difference between common and rare haplotypes is intriguing and suggests the differences among them in terms of disease risk may relate to evolutionary age and the timing of divergence of the rarer haplotypes from common ancestral trees. Additionally, contributions from selection might reasonably be expected. The findings in this study suggest that the fundamental unit of immunogenetics may be a haplotype-based functional cassette rather than a single histocompatibility allele and its ability to bind specific peptides as has been widely believed. Although the findings are derived from a single autoimmune disease they further raise the possibility that epistatic interactions of HLA class II alleles with as yet unidentified linked variation (such as seen in the HLA class I loci) determine immune responses more generally.
Methods
Subjects.
We selected 1970 individuals from 494 MS families as part of an ongoing Canadian Collaborative Project on the Genetic Susceptibility to MS (CCPGSMS), for which the methodology has been described previously (31). Informed consent was obtained from all subjects and the experiments performed for this investigation comply with current guidelines and ethics. All families were Canadian and of European descent. The family types included are type 1 [families with both parents positive for HLA-DRB1*15 (either heterozygous or homozygous for HLA-DRB1*15)], type 2 [families with one parent positive for HLA-DRB1*15 (either heterozygous or homozygous for HLA-DRB1*15), and one parent negative for HLA-DRB1*15], and type 3 [families with both parents negative for HLA-DRB1*15].
This study used an expanded cohort of families from a previous study (25). In that previous study, a total of 1258 individuals from 294 MS families were included. In this current study, an additional independent cohort consisting 712 individuals from 200 MS families were included, which confirms the findings from the previous study (25). When comparing the 2 cohorts, we found no differences for the HLA class I/II haplotype transmissions, and the haplotype TDT patterns were very similar between the 2 datasets (data not shown). Therefore, we have decided to pool the 2 cohorts together for this study. The initial dataset (25) generated the hypotheses tested in this study, which asks if there are any HLA-DRB1*1501 haplotypes tagged by HLA class I that are different for risk. By using this enlarged cohort of 1970 individuals from 494 MS families, we were able to adequately address this question.
HLA Typing.
The genotyping for the HLA-DRB1 was performed using either low- or high-resolution allele-specific PCR amplification method (8). Low-resolution HLA-DRB1 genotypes were obtained by a combination of 24 PCR reactions, and high-resolution HLA-DRB1 genotypes were obtained with an additional 48 PCR reactions. In each individual reaction, positive control primers were designed to amplify a second non-polymorphic genomic control segment. Amplified products were separated by electrophoresis in 2% agarose gels containing ethidium bromide after the addition of loading buffer, and visualized them using UV illumination.
The genotyping for the HLA-A and HLA-B were performed using a low-resolution allele-specific PCR amplification method (32). Low-resolution HLA-A and HLA-B genotypes were obtained by a combination of 96 PCR reactions. PCR products were electrophoresed in 1% agarose gels containing ethidium bromide after the addition of loading buffer and visualized using UV illumination.
Statistical Methods.
The family pedigree files were first tested using the PEDCHECK program (33) for the presence of errors in Mendelian transmission.
Two-locus (HLA-DRB1-HLA-A, and HLA-DRB1-HLA-B) haplotypes were constructed. Transmissions of HLA-DRB1*15 haplotypes from HLA-DRB1*15 positive parents, and transmissions of HLA-DRB1*X haplotypes (where X refers to any allele other than HLA-DRB1*15) from HLA-DRB1*15 negative parents were analyzed. To eliminate the main affect of HLA-DRB1*15, transmissions of HLA-DRB1*X haplotypes from HLA-DRB1*15 heterozygous parents (for example: 15/X, where X refers to any allele other than HLA-DRB1*15) were not counted. Transmissions of HLA-DRB1*X haplotypes were only counted from HLA-DRB1*15 negative parents (for example: X/X, where X refers to any allele other than HLA-DRB1*15).
Parents positive for HLA-DRB1*15 can either be heterozygous or homozygous for HLA-DRB1*15, whereas HLA-DRB1*15 negative parents can bear any HLA-DRB1 allele except HLA-DRB1*15. The “Total HLA-DRB1*15 Positive Parents” includes all parents from Type 1 families plus the HLA-DRB1*15 positive parents from Type 2 families, whereas the “Total HLA-DRB1*15 Negative Parents” includes all parents from Type 3 families plus the HLA-DRB1*15 negative parents from Type 2 families (see Subjects section for details).
TDT was performed for each locus individually and also to multilocus haplotypes using the UNPHASED program (ref. 34 and www.hgmp.mrc.ac.uk). Since this study reports no independent associations of HLA class I alleles, correction for multiple testing was not applied, and all P-values in tables were presented as uncorrected P-values (Puncorrected).
Two-locus HLA-DRB1*15 bearing haplotypes (HLA-DRB1*15-HLA-A and HLA-DRB1*15-HLA-B) were categorized by their HLA-A and HLA-B broad serological equivalents, and the totals of the allele/haplotype groups were calculated. Furthermore, 2-locus HLA-DRB1*15 bearing haplotypes (HLA-DRB1*15-HLA-A and HLA-DRB1*15-HLA-B) were then separated into 2 groups: (i) common haplotypes with frequency >0.02, and 2) rare haplotypes with frequency <0.02. This threshold of 0.02 used to assign common or rare haplotypes is only arbitrary and was considered based on the sample size and the haplotypes transmission probabilities.
Acknowledgments.
We thank our colleagues at the Wellcome Trust Centre for Human Genetics and the Oxford Transplant Centre for their help and support. We would also like to thank all patients who generously participated in this study, and physicians participating in the CCPGSMS including: J. J-F. Oger, D. W. Paty, S. A. Hashimoto, V. Devonshire, J. Hooge, J. P. Smythe, and T. Traboulsee (Vancouver); L. Metz (Edmonton); S. Warren (Calgary); W. Hader (Saskatoon); R. Nelson and M. Freedman (Ottawa); D. Brunet (Kingston); J. Paulseth (Hamilton); G. Rice and M. Kremenchutzky (London); P. O'Connor, T. Gray, and M. Hohol (Toronto); P. Duquette and Y. Lapierre (Montreal); J-P. Bouchard (Quebec City); T. J. Murray, V. Bhan, and C. Maxner (Halifax); W. Pryse-Phillips and M. Stefanelli (St. Johns). This work was funded by the Multiple Sclerosis Societies of Canada and Great Britain and Northern Ireland. This study was made possible by the Canadian Collaborative Project on the Genetic Susceptibility to MS (CCPGSMS).
Footnotes
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
References
- 1.Martin R, McFarland HF, McFarlin DE. Immunological aspects of demyelinating diseases. Annu Rev Immunol. 1992;10:153–187. doi: 10.1146/annurev.iy.10.040192.001101. [DOI] [PubMed] [Google Scholar]
- 2.Hafler DA. Multiple sclerosis. J Clin Invest. 2004;113:788–794. doi: 10.1172/JCI21357. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Zamvil SS, Steinman L. The T lymphocyte in experimental allergic encephalomyelitis. Annu Rev Immunol. 1990;8:579–621. doi: 10.1146/annurev.iy.08.040190.003051. [DOI] [PubMed] [Google Scholar]
- 4.Gregersen JW, et al. Functional epistasis on a common MHC haplotype associated with multiple sclerosis. Nature. 2006;443:574–577. doi: 10.1038/nature05133. [DOI] [PubMed] [Google Scholar]
- 5.Ebers GC, et al. A full genome search in multiple sclerosis. Nat Genet. 1996;13:472–476. doi: 10.1038/ng0896-472. [DOI] [PubMed] [Google Scholar]
- 6.Haines JL, et al. Linkage of the MHC to familial multiple sclerosis suggests genetic heterogeneity. Hum Mol Genet. 1998;7:1229–1234. doi: 10.1093/hmg/7.8.1229. [DOI] [PubMed] [Google Scholar]
- 7.Dyment DA, Ebers GC, Sadovnick AD. The genetics of MS. Lancet Neurol. 2004;3:104–110. doi: 10.1016/s1474-4422(03)00663-x. [DOI] [PubMed] [Google Scholar]
- 8.Dyment DA, et al. Complex interactions among MHC haplotypes in multiple sclerosis: Susceptibility and resistance. Hum Mol Genet. 2005;14:2019–2026. doi: 10.1093/hmg/ddi206. [DOI] [PubMed] [Google Scholar]
- 9.Lincoln MR, et al. A predominant role for the HLA class II region in the association of the MHC region with multiple sclerosis. Nat Genet. 2005;37:1108–1112. doi: 10.1038/ng1647. [DOI] [PubMed] [Google Scholar]
- 10.Ebers GC. Environmental factors and multiple sclerosis. Lancet Neurol. 2008;7:268–277. doi: 10.1016/S1474-4422(08)70042-5. [DOI] [PubMed] [Google Scholar]
- 11.Ebers GC, Sadovnick AD. The role of genetic factors in multiple sclerosis susceptibility. J Neuroimmunol. 1994;54:1–17. doi: 10.1016/0165-5728(94)90225-9. [DOI] [PubMed] [Google Scholar]
- 12.Hains JL, et al. A complete genomic screen for multiple sclerosis underscores a role for the major histocompatability complex. Nat Genet. 1996;13:469–471. doi: 10.1038/ng0896-469. [DOI] [PubMed] [Google Scholar]
- 13.Sawcer S, et al. A genome screen in multiple sclerosis reveals susceptibility loci on chromosome 6p21 and 17q22. Nat Genet. 1996;13:464–468. doi: 10.1038/ng0896-464. [DOI] [PubMed] [Google Scholar]
- 14.Kuokkanen S, et al. Genomewide scan of multiple sclerosis in Finnish multiplex families. Am J Hum Genet. 1997;61:1379–1387. doi: 10.1086/301637. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Zhang Z, et al. Two genes encoding immune-regulatory molecules (LAG3 and IL7R) confer susceptibility to multiple sclerosis. Genes Immun. 2005;6:145–152. doi: 10.1038/sj.gene.6364171. [DOI] [PubMed] [Google Scholar]
- 16.Sawcer S, et al. A whole genome screen for linkage disequilibrium in multiple sclerosis confirms disease associations with regions previously linked to susceptibility. Brain. 2002;125:1337–1347. doi: 10.1093/brain/awf143. [DOI] [PubMed] [Google Scholar]
- 17.Oturai A, et al. Linkage and association analysis of susceptibility regions on chromosome 5 and 6 in 106 Scandinavian sibling pair families with multiple sclerosis. Ann Neruol. 1999;46:612–616. [PubMed] [Google Scholar]
- 18.Gregory SG, et al. Interleukin 7 receptor α chain (IL7R) shows allelic and functional association with multiple sclerosis. Nat Genet. 2007;39:1083–1091. doi: 10.1038/ng2103. [DOI] [PubMed] [Google Scholar]
- 19.Lundmark F, et al. Variation in interleukin 7 receptor α chain (IL7R) influences risk of multiple sclerosis. Nat Genet. 2007;39:1108–1113. doi: 10.1038/ng2106. [DOI] [PubMed] [Google Scholar]
- 20.Risch N, Merikangas K. The future of genetic studies of complex human disease. Science. 1996;273:1516–1517. doi: 10.1126/science.273.5281.1516. [DOI] [PubMed] [Google Scholar]
- 21.Peltonen L. Old suspects found guilty—the first genome profile of multiple sclerosis. N Engl J Med. 2007;357:927–929. doi: 10.1056/NEJMe078147. [DOI] [PubMed] [Google Scholar]
- 22.Fogdell-Hahn A, Ligers A, Gronning M, Hillert J, Olerup O. Multiple sclerosis: A modifying influence of HLA class I genes in an HLA class II associated autoimmune disease. Tissue Antigens. 2000;55:140–148. doi: 10.1034/j.1399-0039.2000.550205.x. [DOI] [PubMed] [Google Scholar]
- 23.Harbo HF, et al. Genes in the HLA class I region may contribute to the HLA class II-associated genetic susceptibility to multiple sclerosis. Tissue Antigens. 2004;63:237–247. doi: 10.1111/j.0001-2815.2004.00173.x. [DOI] [PubMed] [Google Scholar]
- 24.Yeo TW, et al. A second major histocompatibility complex susceptibility locus for multiple sclerosis. Ann Neurol. 2007;61:228–236. doi: 10.1002/ana.21063. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Chao MJ, et al. Transmission of class I/II multi-locus MHC haplotypes and multiple sclerosis susceptibility: Accounting for linkage disequilibrium. Hum Mol Genet. 2007;16:1951–1958. doi: 10.1093/hmg/ddm142. [DOI] [PubMed] [Google Scholar]
- 26.Ligers A, et al. Evidence of linkage with HLA-DR in DRB1*15-negative families with multiple sclerosis. Am J Hum Genet. 2001;69:900–903. doi: 10.1086/323480. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Ghabanbasani MZ, et al. Importance of HLA-DRB1 and DQA1 genes and of the amino acid polymorphisms in the functional domain of DR beta 1 chain in multiple sclerosis. J Neuroimmunol. 1995;59:77–82. doi: 10.1016/0165-5728(95)00027-y. [DOI] [PubMed] [Google Scholar]
- 28.Haegert DG, Swift FV, Benedikz J. Evidence for a complex role of HLA class II genotypes in susceptibility to multiple sclerosis in Iceland. Neurology. 1996;46:1107–1111. doi: 10.1212/wnl.46.4.1107. [DOI] [PubMed] [Google Scholar]
- 29.Spurkland A, et al. The HLA-DQ (alpha 1*0102, beta 1*0602) heterodimer may confer susceptibility to multiple sclerosis in the absence of HLA-DR (alpha 1*01, beta 1*1501) heterodimer. Tissue Antigens. 1997;50:15–22. doi: 10.1111/j.1399-0039.1997.tb02828.x. [DOI] [PubMed] [Google Scholar]
- 30.Seltman H, Roeder K, Devlin B. Transmission/disequilibrium test meets measure haplotype analysis: Family-based association analysis guided by evolution of haplotypes. Am J Hum Genet. 2001;68:1250–1263. doi: 10.1086/320110. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.The Canadian Collaborative Study Group. Sadovnick AD, Risch NJ, Ebers GC. Canadian collaborative project on genetic susceptibility to MS, phase 2: Rationale and method. Can J Neurol Sci. 1998;25:216–221. doi: 10.1017/s0317167100034041. [DOI] [PubMed] [Google Scholar]
- 32.Bunce M, et al. Phototyping: Comprehensive DNA typing for HLA-A, B, C, DRB1, DRB3, DRB4, DRB5 and DQB1 by PCR with 144 primer mixes utilizing sequence-specific primers (PCR-SSP) Tissue Antigens. 1995;46:355–367. doi: 10.1111/j.1399-0039.1995.tb03127.x. [DOI] [PubMed] [Google Scholar]
- 33.O'Connel JR, Weeks DE. PedCheck: A program for identifying genotype incompatibilities in linkage analysis. Am J Hum Genet. 1998;63:259–266. doi: 10.1086/301904. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Dudbridge F. Pedigree disequilibrium tests for multilocus haplotypes. Genet Epidemiol. 2003;25:115–221. doi: 10.1002/gepi.10252. [DOI] [PubMed] [Google Scholar]