| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1From the Departments of Epidemiology and Biostatistics, and 3Ophthalmology, Case Western Reserve University, Cleveland, Ohio; and the 2Department of Ophthalmology and Visual Sciences, University of Wisconsin Medical School, Madison, Wisconsin.
 |
Abstract |
|---|
 Top Abstract MethodsResultsDiscussionReferences |
|---|
METHODS. To comprehend the role of pigmentary abnormalities (PA) and GA in AMD, the authors analyzed the data from a previous GWS on AMD (FARMS [Family Age-Related Maculopathy Study]sample of 34 extended families) looking only at PA. Presented are new GWS data from the full Beaver Dam Eye Study (BDES) family cohort, including longitudinal data at baseline and 5- and 10-year follow-up. A linkage analysis for PA/GA was performed on both samples for 338 markers covering all autosomes. Another linkage analysis using the rate of change along the PA/GA scale was performed with the BDES sample.
RESULTS. Analysis of the FARMS sample provided evidence for linkage with P < 0.01 in the 1q25, 5p13, 6q21-23, and 11q14 regions. The most significant peak was found on chromosome 1, near complement factor H (CFH), with P= 6.20 x 10–4. Analysis using the rate of change in BDES replicated the peaks in 5p13 and 6q21-23, suggesting that these loci may contribute to the rate of progression of PA/GA. Association analysis of CFH polymorphisms suggest that CFH may play a role in the development of pigmentary abnormalities and may modify the progression along the PA/GA scale.
CONCLUSIONS. These findings suggest a complex, heterogeneous model for PA/GA.
AMD prevalence increases with age to approximately 37% in persons 75 years of age and older in the Beaver Dam Eye Study (BDES) population-based sample.9 10 Although the pathogenesis of AMD remains unclear, age and family history are known to be important predictors of AMD risk. Environmental factors suggested to play a role in AMD development include smoking and, less consistently, exposure to light, blood pressure, and inflammation (reviewed by Schick et al.11 and Klein et al.1 ).
Several genetic epidemiology studies have established the importance of genetic factors in AMD.12 13 14 15 16 Several chromosomal regions were identified from genome scans for AMD.17 18 19 20 21 22 More recently, three genes have been identified as being strongly associated with AMD. Multiple studies have just recently shown that a polymorphism in the complement factor H gene on 1q is strongly associated with AMD.2 3 4 5 A variant in this region was reported earlier by Schultz et al.,6 who found a rare mutation in the hemicentin-1 (FIBL6) gene on chromosome 1 that segregated with AMD in a very large family. A variation in the toll-like receptor 4 (TLR4) gene on chromosome 9 was also found to be associated with AMD.7
Previous genome scans have found some additional regions in common for which the specific genes linked to AMD have yet to be found, but there has also been notable disagreement in the results. The wide variation in linkage peak locations and genes associated with AMD in different populations supports the hypothesis that AMD may be affected by several genetic and environmental variables that all influence the risk of progression to AMD, and no single factor alone explains most of the AMD cases. A genome-scan meta-analysis of these studies has shown that the 10q26 and, to a lesser extent, 1q, 2p, 3p, and 16, all met the criteria for genome-wide significance.23
Another explanation of the discrepancies among the different studies is that some of the regions of linkage may contribute to specific phenotypes and may be linked to only a single endstage form of the disease (e.g., GA or CNV), which has not been investigated extensively. The few studies looking at GA and CNV as specific outcomes have linked the 5p region as well as the 4q32 region to GA18 21 and the 9q and 22q regions to CNV.21 There have been no published reports, to our knowledge, suggesting any specific genes for development of retinal pigmentary abnormalities and GA.
Although antioxidant vitamin–zinc combinations have been shown to reduce the risk of progression to AMD by approximately 23%, a significant number of individuals still progress to severe visual loss.23 Because, in the evolution of AMD, different genes may play different roles in susceptibility at different stages, an understanding of the underlying pathophysiology of the development of pigmentary abnormalities (PAs) and GA, and of AMD in general, may aid in the design of more widely available or more effective treatments. In the current study, we investigated the underlying genetic factors in PA and GA, ignoring the CNV-related phenotypes, to determine whether specific loci may more strongly influence this AMD phenotype.
|
Methods |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
|
Phenotypic Evaluation
Each participant in both studies was interviewed and had stereoscopic 30° color fundus photographs of each eye taken. Two gradings were performed on photographs for each eye according to standardized protocols developed to classify and detect AMD.24 25 First, a preliminary masked grading was performed by one of two senior graders. A second independent, more detailed grading was then performed by one of three other experienced graders. These detailed gradings included subfield-by-subfield and lesion-by-lesion evaluation of each photograph set for each eye separately, according to the Wisconsin Age-Related Maculopathy Grading System.26 27 If clinically meaningful differences between the preliminary and detailed gradings remained after both gradings were completed, a third grader performed an edit.
In this study, we used a classification system with greater emphasis on the retinal PAs and GA. Based on the presence and extent of early and late AMD lesions, a PA/GA severity score from 1 to 6 was assigned to each eye, as shown in Table 2 . This severity scale is based on previous observations from the natural history of AMD showing that independent of drusen area, increased retinal pigment precedes and increases the risk of RPE depigmentation and that eyes with RPE depigmentation have a higher risk of GA than those with increased pigment only. This score reflects the amount of increased retinal pigment and RPE depigmentation and the presence of GA observed in the fundus photographs. A score of 6 was assigned when pure GA was present in the eye. Each step along the scale from 1 to 5 carries an increased risk of progression to step 6 (pure GA). Because CNV may itself contribute to pigmentary abnormalities that were not the focus of this study, individuals with documented CNV were excluded from analyses involving this scale. Scores for the left and right eyes for each individual were averaged to reflect a measure of the overall amount of PA present in an individual. In the cases in which the data for one eye were missing, the eye for which data were available was used in place of the average. For the initial PA/GA linkage analysis of the BDES, the PA/GA severity score for each period, baseline and 5- and 10-year follow-ups, was averaged over the available time points, to mimic what might be observed if evaluation had happened at a midpoint in the cohort. Analyses were performed with this score as a quantitative trait representing the severity of PA/GA, which is a more powerful alternative to using a dichotomous outcome indicating presence or absence of GA. The average PA/GA score for the FARMS as well as the average PA/GA scores for the BDES at baseline and 5- and 10-year follow-ups are shown in Table 1 . Approximately 80% of the individuals in the BDES sample have a PA/GA score of 1 in both eyes, bringing the sample-wide average score close to 1. Although some individuals were lost to follow-up in the course of the study, the percentage of individuals with higher PA/GA severity scores increased through time, as would be expected, thus raising the average PA/GA score.
|
Three SNPs in CFH and four SNPs in HMCN1 were genotyped in both samples (TaqMan; Applied Biosystems, Inc. [ABI], Foster City, CA). Two hundred ninety-seven individuals in the FARMS sample were genotyped, as were 2307 individuals in the BDES population. The details of the SNPs are given in Table 3 . The three SNPs in CFH cover the three linkage disequilibrium (LD) blocks in the CFH gene (see Fig. 2 ). The HMCN-1 gene is quite large. The four SNPs that were genotyped cover most of the gene, but there are regions of it that are not well covered by our SNPs (see Fig. 3 ). Genomic DNA was subjected to PCR amplification in a volume of 25 µL including 1x PCR master mix (TaqMan Universal Master Mix; ABI), 8% glycerol, and 1x PCR buffer (100 mM Tris [pH 8.0], 500 mM KCl; ABI) with 7.5 mM MgCl2, dNTPs (200 µM dATP, 200 µM dCTP, 200 µM dGTP, 400 µM dUTP), and 0.5 U of DNA polymerase; AmpliTaq Gold; ABI]. SNP genotyping assay mix (Assays-on-Demand; ABI) containing the two specific MGB probes (TaqMan probes; ABI) and forward and reverse primers, was also added. The 96- or 384-well plate containing the reaction mixture was then run on a sequence-detection system (model 7900; ABI). The overall genotyping error rate was estimated at <1% based on 609 replicates of all seven SNPs. No individual SNP showed error rates much higher than any other. Deviations from Hardy-Weinberg equilibrium were assessed by a 
2 goodness of fit test. All seven SNPs were found to be in Hardy-Weinberg proportion, with P > 0.05.
|
|
|
|
Before performing the linkage analysis on the BDES sample, any sib pairs found to be incorrectly reported were reclassified according to their likely true relationship as determined by RELTEST (SAGE 2005). Five pedigrees that were found to contain loops were cut so that they could be analyzed by methods that do not allow for pedigree with loops. MARKERINFO (SAGE 2005) was used to flag within-family Mendelian inconsistencies. All genotypes found to be inconsistent or unlikely were dropped and treated as missing. In total, 0.24% and 3.67% of the FARMS and BDES samples, respectively, were treated as missing.
Statistical Methods
Identification and Adjustment for Important Covariates.
Initially, a multiple linear regression on the PA/GA severity score was used to examine the environmental effects of gender, age, and smoking on PA/GA. In the population-based sample (BDES), age and age squared were the only covariates found to be significant predictors of PA/GA at the 95% confidence level. Therefore, the model best predicting the PA/GA severity score included age and age squared. The residuals from the linear regression were calculated for each individual. To remove the effect of age, the mean fitted value for age 80 years was calculated for each individual. This value represented the predicted score for that individual at age 80 years, which was then winsorized to make sure all values were positive for a Box-Cox power transformation.
To maximize power, a Box-Cox power transformation was performed on the data. We performed a segregation analysis, using the BDES to obtain a population-based transformation parameter with the program SEGREG (SAGE 2005). This analysis led to a transformation parameter 
1 = 0.119 under the two-mean-dominant model that was used to transform the BDES sample according to the Box-Cox formula.28 No transformation was performed for FARMS, as the results did not change with the transformation.
Linkage Analysis.
A model-free linkage analysis for PA/GA was performed for the FARMS sample and then for the BDES sample, using the Haseman-Elston regression as implemented in SIBPAL, with a revision to the method that transforms the sib pair trait values to a weighted combination of the squared-trait difference and squared-mean corrected-trait sum, adjusting for the nonindependence of both sib pairs and the squared-trait sums and differences (the w4 option in SIBPAL).29 Nominal probabilities are reported, but all critical results were verified by a permutation test.
This linkage analysis was then independently repeated for each of the 12 largest FARMS families to identify which of the families contributed to linkage at specific chromosomal locations, as previous in Iyengar et al.22 Any family with P < 0.1 in the region of interest was considered to contribute to the linkage signal in that region, as these families showed at least some evidence of linkage to that region. Once the contributing families were found, they were removed, and the Haseman-Elston regression was rerun, using the remaining sample to confirm that the linkage signal in that region was eliminated.
To investigate whether our results from the initial model-free linkage analysis were due to affecteds’ contributing to the linkage signal, we confirmed results using an affecteds-only approach. To accomplish this, we applied a covariate-based conditional logistic affected sib pair (ASP) LOD score method implemented in LODPAL (SAGE 2005) to the entire genome, using the BDES sample, for which several ASPs were available. Another advantage of implementing this method is that it has been extended to allow us to evaluate possible parent-of-origin effects, a hypothesis that has never been tested for AMD.
The ASP LOD scoring method requires all individuals to be defined as affected or unaffected. Because many younger individuals are likely to progress farther along the PA/GA severity scale, we used their mean fitted residual value at age 80 years as an estimate of their progression to pure GA, as described earlier, and then labeled the top 30% of the distribution of these values as affected and the bottom 30% of the distribution as unaffected. Although this eliminated 40% of the sample, it minimized misclassification for several reasons, including slower progression or measurement error.
Models with the addition of covariates can be compared with baseline models, and the results of this method will be reported as LOD scores, which are defined herein as the likelihood-ratio statistics divided by 2log10. Critical values can be computed to a 2 distribution with the degrees of freedom equal to the difference in the number of parameters for nested models. A statistically significant increase in LOD scores indicates that the covariate is significant in describing the difference between sib pairs. Because PA/GA severity scale values were previously adjusted for age and other covariates were found not to be statistically significant in describing PA/GA scores, the only covariate added to the baseline conditional logistic regression model was the parent-of-origin effects, modeled as a 0 for a mother’s allele and a 1 for a father’s allele.
Because longitudinal data were available for the BDES sample, we repeated the linkage analysis on the BDES sample, using the slope of the PA/GA scores as a surrogate for the progression to PA/GA. Data were collected at baseline and 5- and 10-year follow-ups. For those individuals for whom at least two scores were available (n= 1711), the slope was calculated by using the ordinary least squares (OLS) method. Those without scores for at least two time points were excluded from the analysis. A multiple regression was preformed on the OLS slope to account for age and age squared, and the residuals were mean fitted to age 80 years and transformed with the Box-Cox power transformation (here, 1 = 0.749), as in previous analyses.
Association Analysis.
To see whether linkage on chromosome 1 was due to variations in the CFH gene, we genotyped three SNPs in CFH (Table 3) . These SNPs were then compared to the average PA/GA scores for both BDES and FARMS and also to the OLS slope of the change in the scale for the BDES—the same outcomes used in the linkage analyses. Association analysis was performed via a linear regression, accounting for sibling and familial effects via the program ASSOC in the SAGE software package (SAGE 2005). For each regression, three different genotypic models (dominant, additive, and recessive) were tested. The best fitting model is reported in the results.
|
Results |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
|
Family-by-Family Linkage Analysis
In the FARMS sample, the four main linkage peaks were analyzed on a family-by-family basis to identify specific families contributing to the linkage signals at the four regions found with P < 0.01. The results of this analysis are shown in Figure 2 for chromosomes 5, 6, and 11.
Figure 2A shows the results for chromosome 5. Individual linkage results showed that families 440 and 461 were the only families to achieve a P < 0.1 in this region. Removing these two families only reduced the linkage signal to a P =0.434 at the D5S2500 locus.
Figure 2B shows a plot of the three families found to be contributing to the linkage signal on chromosome 6. Removing just these three families and running the Haseman-Elston regression on the remaining 31 families brought the remaining signal at the former peak at D6S474 to P = 0.568.
The results for chromosome 11 are shown in Figure 2C . Only two families were found to contribute to the linkage signal in this region, of which one was significant. Family 472 showed strong linkage to this region reaching a nominal P = 1.97 x 10–8 between markers D11S2006 and D11S2371. However, to reduce the remaining sample to a P > 0.1 in the region, family 461 was also removed, because it also contributed slightly to the signal.
The family-by-family analysis of chromosome 1 revealed a different story. It was not clear from the results of this analysis which families contributed to the signal. Six of the 12 families evaluated had a P in this region < 0.1. However, when any or all of these families were removed and the linkage analysis was repeated, the signal in this region remained strong. When the six largest families with P < 0.1 were removed, the nominal peak P only decreased in significance from 2.55 x 10–4 to 9.94 x 10–3(data not shown).
Affected Sib Pair LOD Score Method
The results of the conditional logistic ASP LOD score method for the BDES sample are shown in Table 5 . The most significant peak was found on chromosome 9 in the 9q33 region between markers D9S1825 and D9S2157 with an LOD score of 2.30. This peak is in the same region as a peak for AMD that was reported by Majewski et al.18
|
2 distribution with 1 df, corresponds to P =0.060. This does not quite meet the criteria for significance at the 95% confidence level, but it is close, suggesting that adding a parent-of-origin effect to the model may provide a better fit for the data at the locus on chromosome 9. Having a parent-of-origin effect in the model means that that not only is having a high PA/GA severity score dependent on a locus in this region, but it also may be differentially linked to PA/GA, depending on the parent from whom the risk allele was inherited. In particular, it appears that the paternal allele sharing is increased in this region in the BDES sample.
Longitudinal Analysis
The results of the longitudinal linkage analysis in BDES are shown in Figure 3 . The peak on chromosome 5 near marker D5S2500 found in the FARMS sample was also observed in this analysis, with a peak P = 3.23 x 10–3. A composite graph showing the results from the FARMS genome scan and BDES longitudinal genome scan for chromosome 5 together is shown is Figure 4A .
|
Association Analysis
The results of the association analysis of CFH with the PA/GA scale for BDES and FARMS as well as the OLS slope of the PA/GA scale for BDES show that CFH is statistically significantly associated with this outcome (Table 6) , particularly in the BDES sample, with a P much less than 1.0 x 10–7. The longitudinal association in BDES shows statistical significance with CFH as well.
|
|
Discussion |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
This study not only tests many genome scan markers and individual SNPs, it also includes a repetition of earlier studies of AMD, looking at only a subset of the phenotypes. This warrants the use of multiple testing. However, given the controversial nature and nonstandard methods for multiple testing, we have not adjusted, but have left it to the reader to interpret. No one locus achieved genome-wide significance as defined by Lander and Kruglyak31 in the genome scan. However, we believe that if we solely report findings using such stringent criteria, we will miss many potentially important associations that may have higher probability due to lower power, heterogeneity, or chance. As an example, in the original genome scan on the FARMS sample for AMD,22 the lowest probability found in the CFH region, which is now widely replicated as a disease susceptibility gene for AMD, is 5.2 x 10–3, similar to many of the probabilities that we think are significant enough to report in the FARMS sample in this scan.
There have been two other published genome scans for GA. Majewski et al.18 stratified their sample of AMD families into those with GA and CNV. In their subset of families with predominantly GA, they found a peak in the 4q32 region. This peak was not duplicated in this study, in which we examined the relationship along the PA/GA scale. The difference between their analyses and ours is also in the modeling of the trait—we used a PA/GA scale to give more information and increase power, whereas they used presence or absence of GA—which may explain the differences in results. Repeating the analyses using affected status, defined as clinically evident GA or not, showed results similar to those achieved with the scale variable, but we had few cases of end-stage GA in our sample, and so we do not think it appropriate to present the results. Abecasis et al.21 found peaks on chromosomes 5, 14, and 17 in families with GA in their two-point analysis, with only chromosome 5 showing evidence in their multipoint analysis. We also found a peak on chromosome 5, which was noted not only in our genome scan for PA/GA but also in the longitudinal analysis of progression along the PA/GA severity scale.
Another difference between our study and other studies is that we chose to use the average PA/GA severity score instead of the worse eye score as the outcome. Preliminary analysis of our data using the worse eye did not alter the results significantly.
Jun et al.32 presented the results of a genome scan for soft drusen in a subset of the BDES sample. Because persons with drusen are at a higher risk of development of PA/GA,32 one would expect that genes that predispose to PA/GA may also affect their drusen formation. Comparing the Haseman-Elston regression results from the drusen analysis and our PA/GA analyses using either FARMS or BDES, no overlapping linkage signals were found between the two analyses, with a P cutoff of 0.01, suggesting that separate genes may control the formation of soft drusen and PA/GA. Because, clinically, soft drusen formation usually precedes the development of pigmentary changes, it may be that the presence of the genes for PA is clinically irrelevant if the patient does not have the genes for soft drusen.
The peak on chromosome 1 has been duplicated frequently and is currently the center of much research. Both mutations in hemicentin-16 22 and the Y402H polymorphism in complement factor H,2 3 4 5 which are both in this region, have been implicated as causative mutations for AMD. Our results in the BDES sample show a strong association of CFH with PA/GA severity score. Despite the strong association, no peak was observed in the PA/GA genome scan in the BDES sample. Likewise, whereas no peak on chromosome 1 was observed in the longitudinal analysis, we saw associations with CFH with progression along the PA/GA scale during follow-up. This observation may be an illustration of the lack of power of a Haseman-Elston regression in this population-based sample, because of the low number of individuals at the high end of the scale.
Even though CFH was strongly associated with AMD in the BDES, it was of only borderline significance in the FARMS sample, in which the chromosome 1 peak was observed in the genome scan, which may be due to the lack of power for association in a sample of extended pedigrees because of the lack of independence of the individuals in the sample. In addition, a family-by-family analysis showed that when families with linkage at D1S1589, the locus closest to the CFH gene, were removed, the peak in the FARMS sample in this region remained strong, possibly because of several genes in the region affecting PA/GA. Although the previously reported mutation in hemicentin-1 is too rare to contribute to linkage in this sample, it may be that other, more frequent, variations in this gene contribute to a portion of the linkage in this region. In fact, previous association and haplotype analysis on the FARMS sample have linked variations in the hemicentin-1 gene to AMD.22 It is clear that the neither the hemicentin-1 nor the CFH gene alone contributes to a large portion of the signal in this region in the FARMS sample.
The peaks on chromosomes 5 and 6 were observed to be less significant in the analysis by Schick et al.19 of a subset of the BDES sample for AMD. This result suggests that this locus may contribute to overall AMD phenotypes, which include PA/GA. The reduced significance may be due to the reduced sample size used in their study, or may indicate that the loci contribute to PA/GA-specific phenotypes that were diluted when looking for AMD severity scale that included drusen type, area, and signs of CNV in addition to PA/GA. Very recently, data from the National Eye Institute (NEI) Age-Related Eye Disease Study (AREDS) has been released supporting a locus near the peak on chromosome 11 in this study.33 All peaks should be confirmed with independent data sets before concluding significance.
ASP LOD score analysis revealed a locus on 9q33 in the BDES sample as a region linked to PA/GA-affected status. The marker closest to our peak, D9S1825, is <10 Mb away from the toll-like receptor 4 (TLR4) gene that has recently been shown to be associated with AMD.7 Zareparsi et al.7 show that the D299G variant in the TLR4 gene is found more commonly in both the subgroups of patients with GA and of patients with CNV compared with the control subjects, but is slightly more common in the GA subgroup the CNV subgroup. Because pigmentary abnormalities are a hallmark of AMD, the association found in our analysis may be due to the TLR4 gene, but further work is necessary to make a final conclusion.
ASP LOD score analysis was performed on the FARMS sample but is not reported in this article because, despite the availability of extended pedigrees, making it ideal for linkage analysis, there were not enough sibling pairs for a reasonable ASP LOD score analysis. In contrast, the BDES sample had several sibling pairs but had only a small number of individuals with an average PA/GA severity score greater than 2. The fact that the BDES sample produced no significant peaks in the linkage analysis is probably due to the lower prevalence of severe PA/GA in the population and thus the reduced power to detect linkage in this sample.
In the family-by-family linkage analysis, we observed that only a few families contributed to the linkage at one of the four initially significant loci, and most of those families contributed to multiple locations. This suggests that these few families are driving the linkage signals and, in these families, the PA/GA score is oligogenic.
We did have a small number of individuals in each population with high PA/GA scores (Table 4) , which somewhat limited our power for linkage and association. However, we increased power by using the scale, which provided more information than a binary variable indicating a high amount of pigmentary abnormalities. In addition, having families in which siblings both have low scores (concordantly unaffected), provides information in the Haseman-Elston regression as does two siblings with high scores (concordantly affected) or two siblings with very different scores (discordant).
The results presented in this article provide further evidence of the genetic heterogeneity of AMD. Many genes may affect AMD and AMD-related phenotypes. In addition, variants that influence AMD in one population may be different from the variants found in another population. Often these genetic variations attributed to AMD risk in a sample have a very low population frequency, such as the hemicentin-1 mutation.6 In addition, it is possible that multiple variations contribute to the high prevalence of AMD on a population level. Future work is needed to confirm linkage of all proposed regions to specific variations in genes in the FARMS and BDES samples as well as in independent populations.
|
Footnotes |
|---|
Submitted for publication November 3, 2006; revised January 29, 2007; accepted April 20, 2007.
Disclosure: C.L. Thompson, None; G. Jun, None; B.E.K. Klein, None; R. Klein, None; J. Capriotti, None; K.E. Lee, None; S.K. Iyengar, None
The publication costs of this article were defrayed in part by page charge payment. This article must therefore be marked "advertisement" in accordance with 18 U.S.C. 
1734 solely to indicate this fact.
Corresponding author: Sudha K. Iyengar, Department of Epidemiology and Biostatistics, Case Western Reserve University, Wolstein Research Building 1315, 10900 Euclid Avenue, Cleveland, OH 44106-7281; ski{at}case.edu.
|
References |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  A. O. Edwards, D. Chen, B. L. Fridley, K. M. James, Y. Wu, G. Abecasis, A. Swaroop, M. Othman, K. Branham, S. K. Iyengar, et al. Toll-like Receptor Polymorphisms and Age-Related Macular Degeneration Invest. Ophthalmol. Vis. Sci., April 1, 2008; 49(4): 1652 - 1659. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. L. Thompson, B. E.K. Klein, R. Klein, Z. Xu, J. Capriotti, T. Joshi, D. Leontiev, K. E. Lee, R. C. Elston, and S. K. Iyengar Complement factor H and hemicentin-1 in age-related macular degeneration and renal phenotypes Hum. Mol. Genet., September 1, 2007; 16(17): 2135 - 2148. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
