SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

Objective

To further analyze the association of HLA–DRB1 alleles with disease susceptibility in recent-onset rheumatoid arthritis (RA).

Methods

One hundred sixty-seven Caucasian RA patients and 166 healthy controls were typed for HLA–DRB1.

Results

The association of susceptibility to RA with the group of alleles encoding the shared epitope susceptibility sequences (SESSs) was confirmed in recent-onset RA. Among non-SESS alleles, DRB1*07, *1201, *1301, and *1501 showed significant protective effects. Even after correction for the influence of SESS alleles, significant independent protective effects of DRB1 alleles were observed. Protective alleles shared a third hypervariable region motif. Independent homozygosity effects were observed both for susceptibility and for protective alleles.

Conclusion

Nonsusceptibility alleles differ significantly with regard to RA risk. Protective alleles show clear homology at positions 67–74, often encoding isoleucine at position 67 or aspartic acid at position 70. Susceptibility and protective alleles both show homozygosity effects. Based on these results and on data reported in the literature, in order to incorporate the finding of differential risks among nonsusceptibility alleles, we propose to reshape the shared epitope hypothesis as follows: HLA-associated risk for RA is encoded by amino acid substitutions at positions 67–74 of the HLA–DRB1 molecule.

While original studies described an association of HLA–DR4 with rheumatoid arthritis (RA), later studies showed that HLA–DR1 was also associated with RA (for review, see ref. 1). When information on the amino acid sequences became available, Gregersen et al (2) observed that the disease-associated HLA–DRB1 alleles showed a striking sequence homology within the third hypervariable region (amino acids 67–74) and stated that “Susceptibility to rheumatoid arthritis may be due to a group of related epitopes found in common among non-Dw10 subtypes of DR4 as well as in some DR1 alleles.” Those investigators also made it clear that this should be regarded as “a conceptual framework,” where “still more knowledge is required to understand the complex patterns of Ia gene associations.” Later, a more strict interpretation of this hypothesis led to the concept that susceptibility to RA is encoded by the amino acid sequences QKRAA and QRRAA, as well as by the sequence RRRAA (encoded by DR10), in the third hypervariable region of the HLA–DRB1 molecule (1). Weyand et al (3) reported that homozygosity for these shared epitope susceptibility sequences (SESSs) increased the risk for RA even further.

However, several investigators have reported results that cannot be fully explained by this strict, simple, and straightforward interpretation. First, HLA associations seem to differ depending on the stage of the disease. For instance, while Thomson et al (4) reported that the DRB1*0404 allele conferred the highest risk in a population-based study of early RA, Weyand et al (3) showed that DRB1*0401 conferred a higher risk among patients with longstanding RA accompanied by extraarticular features. This indicates that analyses of the effects of HLA on susceptibility might be confounded by HLA associations with severity, specific complications, or treatment effects. Therefore, studies on HLA-encoded susceptibility should preferably be conducted on patients with recent-onset RA. Second, risk for the disease is not influenced by the number of SESSs (0, 1, or 2) alone. Among SESS homozygotes, the compound heterozygote DRB1*0401/0404 appears to confer a very high risk for RA (5), even in recent-onset RA (6). Third, the strength of association between HLA–DR1 and RA appears to differ from one population to another (6–8), although in this case, confounding by disease duration cannot be excluded.

Apart from these complexities, the shared epitope hypothesis in its strict form also fails to explain the effects of nonsusceptibility HLA–DRB1 alleles. Investigators in a study in Chile (9) and in a large study in Japan (10) have reported an association of HLA–DR9 with RA. In Kuwait (11), an association of HLA–DR3 with RA was observed. Investigators in other studies report protective effects of DRB1*0402 (12) or other HLA–DRB1 alleles (13). Some investigators have interpreted these findings as evidence that a gene in linkage disequilibrium with HLA–DRB1 may influence RA risk (14), although these hypotheses have been called into question (15, 16).

In this study of a Caucasian population, we show that HLA–DRB1 alleles that do not encode the SESSs differ significantly with regard to the risk they confer for RA. Furthermore, our study indicates that protective alleles show structural homologies as well as a homozygosity effect. Based on the discrepancies in the literature and on our results, we propose that the strict version of the shared epitope hypothesis is incomplete. In our view, the shared epitope hypothesis should be generalized, as follows: HLA-associated RA risk is encoded by amino acid substitutions at positions 67–74 of the HLA–DRB1 molecule. Here, data reported in the literature seem to indicate that RA risk and progression are not encoded by the same amino acid sequences. Exact definition of the effects of individual sequences and of homozygosity on susceptibility, severity, and treatment effects will have to await larger (pooled) studies, including large studies of patients with recent-onset RA.

PATIENTS AND METHODS

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

A cohort of 167 unrelated Dutch Caucasian patients with recent-onset (<1 year) RA according to the 1987 revised criteria of the American College of Rheumatology (ACR; formerly, the American Rheumatism Association) (17) and 166 unrelated healthy Caucasian controls were typed for the HLA–DRB1 alleles described in the 1991 HLA Workshop report (18). End-labeled sequence-specific oligonucleotide probes were used, as described previously (15). Sixty-five percent of the patients were women, with a median age of 55 years (range 17–86 years), and 78% were positive for IgM rheumatoid factor (>10 IU/ml). Clinical examination exactly 1 year after study entry showed that 139 of the 167 patients (83%) fulfilled the ACR criteria at that time, despite treatment; 17 of the other 28 patients fulfilled the ACR criteria at several subsequent examinations at a later stage. Overall, only 11 of the 167 patients (7%) did not consistently fulfill the ACR criteria after the first year.

Forty percent of the controls were women. Their median age was 37 years.

The allele frequencies in controls were checked against the frequencies obtained in the German population described in the HLA Workshop (19). Only the frequency of DRB1*0803 was lower in our controls (0 of 332 alleles) than in the German controls (3 of 180 alleles) (P = 0.04, by Fisher's exact test); however, after correcting for the number of alleles tested (34), this difference was not significant. The absence of DRB1*0402 in our control population was in line with the low frequency (0.6%) of this allele in the German population.

Homozygosity for a given allele was assumed when no other HLA–DRB1 allele could be detected. Weyand et al (3) chose to verify homozygosity by sequence analysis and found only 1 assignment error in 13 patients. In this study, homozygosity was verified by typing these patients for HLA–DRB3, DRB4, DRB5, DQA1, and DQB1 alleles. For DQA1 and DQB1 typing, the typing system described by Khalil et al (20) was used, with minor modifications and additions, while the typing procedure was identical to that described above for HLA–DRB1. All typing results in HLA–DRB1–homozygous individuals were compatible with the known DRB1,3,4,5;DQA1;DQB1 haplotypes in Caucasians. Furthermore, in healthy controls, 2 individuals were homozygous for DRB1*0401, *07, and *1501, while 1 individual was homozygous for DRB1*0101, *0301, *1101, *1302, and *1401. Given the allele frequencies (see Table 1), these numbers are not excessive and are well within the expectations under Hardy-Weinberg equilibrium.

Table 1. Frequencies of HLA–DRB1 alleles in 166 controls and 167 RA patients*
DRB1Controls (n = 332 alleles)Patients (n = 334 alleles)OR95% CIP
n%n%2-tailed1-tailed
  • *

    RA = rheumatoid arthritis; OR = odds ratio; 95% CI = 95% confidence interval.

  • If only the first two digits are given, data are pooled for all subtypes.

  • By Fisher's exact test. Only P values that were significant are shown.

010129946141.671.02–2.730.0490.026
01020031
010341100.250.03–2.21
0301411243131.050.66–1.66
0401401274222.081.37–3.160.0010.000
040310000.00
04041032062.050.95–4.450.0910.047
040510100.990.06–15.96
040731100.330.03–3.18
0408311033.381.01–12.410.0890.046
0738111960.470.26–0.830.0080.006
080162620.990.32–3.11
08030010
080421100.500.04–5.49
090152931.810.60–5.46
100131511.670.40–7.03
11011751440.810.39–1.67
11020010
110310211.990.18–22.10
110462210.330.07–1.63
120193210.220.05–0.990.0370.031
1301237820.330.15–0.750.0060.004
13021961130.560.26–1.20
130352210.390.08–2.05
14011131341.180.52–2.68
150143132780.590.36–0.980.0440.027
150241310.740.17–3.35
160172931.290.47–3.49
160210000.00

In the text and tables, amino acid sequences at HLA–DRB1 positions 67–74 are represented by the standard single-letter amino acid code, showing only the polymorphic residues at positions 67, 70, 71, 73, and 74. Relative risk (RR) was calculated as the odds ratio (OR) in 2 × 2 tables, and 95% confidence intervals (95% CIs) are given. For 2 × 2 tables, a Fisher's 2-tailed exact test was used, unless indicated otherwise. A chi-square test was used for larger tables. Bonferroni correction for multiple testing was considered, but was not applied if associations have been described before. Investigators in earlier studies reported susceptibility effects of DRB1*0101, DRB1*0401, DRB1*0404, DRB1*0408, the shared epitope, and homozygosity (for review, see ref. 1), as well as HLA–DR3 (where *0301 is the major subtype associated) in Kuwait (11, 21). Furthermore, protective effects were reported for DRB1*07 (3, 13, 22–25), *1201 (26), *1301 (24, 26–28), and *1501 (29, 30). Multivariate logistic regression analysis was performed using the SPSSX package (SPSS, Chicago, IL).

RESULTS

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

Susceptibility alleles and RA. The association of RA with DRB1*0101, *0401, *0404, and *0408 was confirmed (see Table 1). A significantly lower frequency in RA patients compared with controls was confirmed for the alleles DRB1*07 (*0701 and/or *0702), *1201, *1301, and *1501 (See Table 1 and Discussion). The SESSs were considered to be present in HLA–DRB1 alleles encoding the amino acid sequence LQKAA (*0401 or *0409), LQRAA (*0101, *0102, *0404, *0405, or *0408), or LRRAA (*1001) at amino acid positions 67, 70, 71, 73, and 74. The allele frequency of SESS was 86 of 332 in controls (26%) and 159 of 334 in patients (48%) (OR 2.60, 95% CI 1.88–3.60; P = 6 × 10−9). Among individuals positive for SESSs, 13 of 73 controls (18%) and 39 of 120 patients (33%) were homozygous for SESSs, showing that homozygosity for SESSs increases RA risk significantly (OR 2.22, 95% CI 1.09–4.52; P = 0.03). The RR for SESS-homozygous individuals compared with SESS-negative individuals was 6.17 (P = 4 × 10−7). The compound heterozygote 0401/(0404 or 0408) was observed in 2 of 166 controls and in 6 of 167 patients (OR 3.06, 95% CI 0.61–15.37; P = 0.28).

Different RA risks encoded by nonsusceptibility alleles. To analyze whether individual nonsusceptibility alleles encode different RA risks, we performed a multivariate logistic regression analysis using the alleles as predictors. In this analysis, the SESS alleles were pooled. A significant protective effect of DRB1*1301 was observed (OR 0.36, P = 0.02). This effect was not influenced by the inclusion or exclusion of age, sex, or homozygosity effects of SESS alleles. Thus, independent of the influence of the susceptibility alleles, DRB1*1301 has a significant protective effect regarding RA compared with other nonsusceptibility alleles.

In another approach, all individuals in whom either or both HLA–DRB1 alleles encoded SESSs were excluded. Subsequently, an allelic frequency analysis in patients and controls was performed. In the resulting 47 RA patients, the allele frequency of DRB1*0301 was significantly higher (27 of 94) compared with controls (28 of 186) (OR 2.27, 95% CI 1.25–4.15; P = 0.007). This result indicates that DRB1*0301 encodes a higher RA risk than do other nonsusceptibility alleles.

Structural homologies among protective alleles. The alleles DRB1*07, *1201, *1301, and *1501 have a protective effect regarding RA risk (see Table 1). Like the susceptibility alleles, these protective alleles showed homology in the third hypervariable region (amino acids 67–74) (see Table 2). The 3 alleles with the lowest ORs (DRB1*07, *1201, and *1301) all encode isoleucine (I) at position 67 and aspartic acid (D) at position 70. Another allele present at reasonable frequency and sharing these characteristics is DRB1*1302; this allele also tends to have a protective effect (OR 0.56, 95% CI 0.26–1.20; P = 0.13). DRB1*1501, the fourth allele with a significant protective effect, shares the isoleucine (I) at position 67, but not the aspartic acid (D) at position 70. Actually, among our controls, the alleles that encode isoleucine at position 67 all showed a protective effect (OR <1.0) (see Table 1).

Table 2. Amino acid sequences of susceptibility, neutral, and protective HLA–DRB1 alleles in the third hypervariable region
  • *

    The third hypervariable region (HVR3) corresponds to the amino acids at positions 67, 70, 71, 73, and 74 of the HLA–DRB1 molecule.

Allele risk group
HVR3 amino acid  sequence*HLA–DRB1 alleles encoding this HVR3 sequence
Susceptibility alleles
LQRAA0101 0102 0404 0405 0408 0410 1402 1406 1409
LQKAA0401 0409
LRRAA1001
Neutral alleles
LQKGR0301 0302 0303
FRRAE0901
FDRAA0805 1101 1104 1105 1202 1305 1601
LQRAE0403 0406 0407 0411
LRRAE1401 1404 1405 1407 1408 1410
Other1602 1103 1403 0801 0802 0804
Protective alleles
IDEAA0103 0402 1102 1301 1302 1304
IQAAA1501 1502 1503
IDRAL0412 0803
IDRAA1201 1306
IDKAA1303
IDRGQ0701 0702

For a further descriptive analysis, all alleles encoding isoleucine at position 67 were grouped together as the protective alleles (P), while the alleles encoding the SESSs were grouped together as the susceptibility alleles (S), and all other HLA–DRB1 alleles were grouped together as the neutral alleles (N). Table 3 shows the frequencies of the different genotypes in patients and controls. As expected, the P alleles showed a highly significant protective effect regarding RA risk (OR 0.37, 95% CI 0.26–0.52, P < 0.001).

Table 3. Risk group genotypes in 166 controls and 167 RA patients*
GenotypeControlsPatientsOR95% CIP
n%n%
  • *

    The HLA–DRB1 alleles encoding the shared epitope susceptibility sequences at amino acid positions 67–74 (i.e., DRB1*0101, *0102, *0401, *0404, *0405, *0408, *1001) were grouped together as the susceptibility alleles (S), while all alleles encoding isoleucine at position 67 were grouped together as the protective alleles (P); the other alleles were grouped together as the neutral alleles (N). Using these allele groups, the corresponding genotypes were determined for each control and patient, and the total frequency of each genotype was determined. See Table 1 for other definitions.

  • By Fisher's 2-tailed exact test. Only P values that were significant are shown.

S/S13839233.591.83–7.010.000
N/S311943261.510.90–2.55
P/S291738231.390.81–2.39
N/N161017101.060.52–2.18
P/N382323140.540.30–0.950.034
P/P3923740.140.06–0.330.000

Individuals homozygous for the protective alleles (P/P) showed an even lower RA risk than heterozygous individuals (P/N and P/S) (OR 0.20, 95% CI 0.08–0.47; P = 2 × 10−4). One might argue that the higher RA risk in heterozygous individuals than in homozygous P/P individuals is the result of the S allele (and its association with RA) in the P/S heterozygote. However, even after excluding all individuals carrying an S allele, the homozygosity effect was still significant for P/P versus P/N individuals (OR 0.30, 95% CI 0.11–0.77; P = 0.02). This analysis therefore suggests that an independent homozygosity effect is also present for the protective alleles. As expected, differential risk was maximal between the P/P genotype and the S/S genotype (RR 16.7, 95% CI 6.0–46.4; P = 2 × 10−9).

In the above discussion, protective alleles were defined by the presence of isoleucine at position 67. This definition best fits our data. However, the substitution leucine-67-isoleucine is a conservative one, and it may be argued that steric hindrance alone is not likely to explain the resulting difference in RA risk. Alternatively, one could argue that the presence of aspartic acid (D) at position 70 is the major determinant of protectivity. This involves a nonconservative substitution of polar (but neutral) glutamine (Q) to negatively charged aspartic acid (D) at position 70. We analyzed the effects of these alleles, grouping alleles that encode aspartic acid at position 70 as the protective alleles (D), the alleles encoding the SESSs as the susceptibility alleles (S), and all other HLA–DRB1 alleles as the neutral alleles (M). D alleles showed a highly significant protective effect regarding RA risk (OR 0.44, 95% CI 0.31–0.61; P < 0.001). Excluding all individuals carrying the S allele showed that the D alleles had an independent homozygosity effect (for D/D versus D/M individuals OR 0.38, 95% CI 0.15–0.94; P = 0.04). Again, differential risk was maximal between the D/D genotype and the S/S genotype (RR 12.4, 95% CI 4.6–33.5; P = 1 × 10−9).

DISCUSSION

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

This study in patients with recent-onset RA shows that, as in longstanding, erosive RA, the HLA–DRB1 alleles DRB1*0101/0102, *0401, and *0404 are associated with disease (8, 19, 31). Furthermore, a clear and significant homozygosity effect was observed. These data confirm the reported associations of early RA with DRB1*0404 (4, 6) and DRB1*0401 (6, 32, 33).

Patients with recent-onset RA show a relatively low frequency of the susceptibility alleles. This, and the size of our study group, made it possible to look more closely at the nonsusceptibility alleles. Several of these alleles were present in significantly decreased frequencies among patients compared with controls (e.g., DRB1*07, *1201, *1301, and *1501). Earlier studies have shown low risks associated with HLA–DR7 (3, 13, 22–25), DRB1*1201 (26), DRB1*1301 (24, 26–28), and DRB1*1501, the last being the major molecular subtype of HLA–DR2 in Caucasians (29, 30).

One may argue that these low allele frequencies among the patients reflect the absence of susceptibility. According to this view, all nonsusceptibility alleles will be present in reduced frequencies among patients compared with controls, with the more frequent nonsusceptibility alleles showing significant “protective effects” in a straightforward analysis using Fisher's exact test. However, another explanation for these reduced frequencies is the existence of independent protective effects of individual nonsusceptibility alleles. To eliminate the influence of SESS alleles, we used logistic regression analysis as our first approach, and as our second approach, we excluded individuals encoding SESS alleles. In both approaches, (groups of) HLA–DRB1 non-SESS alleles showed independent, significant effects on susceptibility to RA. These observations are supported by the results of Wakitani et al (13). In 852 patients with longstanding RA, those investigators showed that after exclusion of DRB1*0405 and *0101, a significant negative association was still observed for the alleles *0701, *1302, *0802, and *1405. However, not all susceptibility alleles were excluded from the analysis, and since not all genotype frequencies of the HLA–DRB1 alleles were available, that report does not allow us to rank DRB1 alleles for their effect on RA risk.

We subsequently studied the structure of the third hypervariable region (amino acids 67–74) to analyze which characteristics were associated with low and neutral risk, respectively. Like susceptibility alleles, low-risk alleles showed strong homology in the third hypervariable region. At amino acid positions 67–74, DRB1*1201 encodes IDRAA, *1301 encodes IDEAA, DRB1*07 encodes IDRGQ, and DRB1*1501 encodes IQAAA. Thus, all of these alleles encode isoleucine (I) at position 67, while the 3 alleles with the lowest ORs for RA also encoded aspartic acid (D) at position 70. In fact, all alleles encoding isoleucine at position 67 showed a protective effect (OR <1). Thus, our study results suggest that we can differentiate susceptibility alleles encoding the SESSs, protective alleles encoding isoleucine at position 67 and/or aspartic acid at position 70, and the other neutral alleles. Such a protective effect of aspartic acid at position 70 was recently proposed by Del Rincón and Escalante (34), Mattey et al (35), and Reviron et al (36).

Our study indicates that the protective alleles, defined in this way, also display homozygosity effects, decreasing RA risk even further. This is not explained by the absence of the SESSs among individuals homozygous for protective alleles, but instead, represents an independent protective effect. Consequently, the homozygotes for susceptibility and protective alleles show large differences in relative risk for RA. This underscores the importance of the HLA class II region in determining disease susceptibility. The results presented here indicate an etiologic fraction of ∼80% in recent-onset RA (i.e., 80% of cases would not have existed if everyone were homozygous for the protective alleles). However, the conclusions from our data regarding the protective homologies should be approached with caution, since these alleles were grouped a posteriori. Further studies in patients with recent-onset RA will be needed to confirm this classification of alleles.

Excluding the SESS alleles from our analysis, we observed a significant association of RA with DRB1*0301 (OR 2.27, 95% CI 1.25–4.15; P = 0.007), while DRB1*0901 showed the highest RA risk (OR 2.71, 95% CI 0.59–12.37; P not significant). This shows that DRB1*0301, and possibly *0901, encode a higher RA risk than other nonsusceptibility alleles. This observation explains the reported associations of DR3 and DRB1*0901 alleles with longstanding RA in populations where SESS alleles are present at low frequencies. In Chile, where HLA–DR4–encoding DRB1 alleles (predominantly *0403 and *0406) do not encode susceptibility to RA (37), HLA–DR9 was associated with RA (9). In Japan, DRB1*0901 was associated with RA after exclusion of the major SESS alleles *0101 and *0405 (10). In Kuwait, the association of HLA–DR3 with RA (11) is explained by a high frequency of HLA–DR3 together with a relatively low frequency of HLA–DR4 alleles (21). Furthermore, subtyping of HLA–DR4, performed in Israeli Arabs, shows that 80% of the HLA–DR4 alleles are encoded by non-SESS alleles (38). Although the observations discussed may be perceived to be in conflict with the strict interpretation of the shared epitope hypothesis, we think that they are compatible with, and well explained by, the version proposed here.

What does this tell us about the disease mechanism? Assuming differential risks for nonsusceptibility alleles, this hypothesis is able to explain data in the existing literature by referring to mutations in amino acid sequences at positions 67–74 in the HLA–DRB1 molecule. Thus, this version supports the notion that this sequence is central in conferring susceptibility to RA and does not indicate a role for another gene within the HLA complex. The fact that local amino acid substitutions and homozygosity induce differential risks suggests that the interaction of this region with presented peptides, superantigens, or (as has been recently suggested) invariant chain (35) may be important in the pathogenesis of this disease.

Finally, we would like to stress that HLA associations with RA differ depending on the stage of the disease. For example, Salmon et al (39) showed an association of DRB1*0401 (Dw4) with progression from early symmetric polyarthritis to RA. Weyand et al (3) showed an increased frequency of DRB1*0401 in patients with longstanding RA, especially in the presence of extraarticular features. However, Thomson et al (4) showed an association of DRB1*0404 with very early RA. This suggests that while DRB1*0404 and probably HLA–DR1 may be associated with early RA, it is DRB1*0401 that may be associated with progression of disease to severe erosive RA.

In conclusion, if the shared epitope hypothesis is strictly defined by the well-known amino acid susceptibility sequences, this hypothesis cannot explain the following: 1) differences in strength of association with specific susceptibility alleles and genotypes (DRB1*0101 versus *0401 versus *0404) depending on stage of disease, 2) associations of nonsusceptibility DRB1 alleles with RA as identified in large studies, and 3) differential risk observed among non-SESS alleles. Although others have taken this as an argument for a role of other genes within the HLA complex, we prefer the interpretation that the shared epitope hypothesis in its strict definition is incomplete. To facilitate further research, the following redefinition is indicated: HLA-associated RA risk is encoded by amino acid substitutions at positions 67–74 of the HLA–DRB1 molecule. Here, data reported in the literature seem to indicate that RA risk and progression are not encoded by the same amino acid sequences. Clearly, collaborative international studies are needed to rank the susceptibility effects of individual HLA–DRB1 alleles.

Acknowledgements

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

We gratefully acknowledge the help of Dr. W. Allebes in obtaining samples from the controls, of Mrs. J. Schipper in the typing procedure, and of Dr. M. A. van 't Hof for statistical advice.

REFERENCES

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES