We conducted linkage analysis to follow up earlier work on microvascular complications of type 1 diabetes (T1D). We analyzed 415 families (2,008 individuals) previously genotyped for 402 SNP markers spanning chromosome 6. We did linkage analysis for the phenotypes of retinopathy and nephropathy. For retinopathy, two linkage peaks were mapped: one located at the HLA region and another novel locus telomeric to HLA. For nephropathy, a linkage peak centromeric to HLA was mapped, but the linkage peak telomeric to HLA seen in retinopathy was absent. Because of the strong association of T1D with DRB1*03:01 and DRB1*04:01, we stratified our analyses based on families whose probands were positive for DRB1*03:01 or DRB1*04:01. When analyzing the DRB1*03:01-positive retinopathy families, in addition to the novel telomeric locus, one centromeric to HLA was identified at the same location as the nephropathy peak. When we stratified on DRB1*04:01-positive families, the HLA telomeric peak strengthened but the centromeric peak disappeared. Our findings showed that HLA and non-HLA loci on chromosome 6 are involved in T1D complications’ expression. While the HLA region is a major contributor to the expression of T1D, our results suggest an interaction between specific HLA alleles and other loci that influence complications’ expression.
Retinopathy, nephropathy, and neuropathy are chronic microvascular complications responsible for much of the morbidity and mortality in type 1 diabetes (T1D). Evidence for familiarity in complications has been clearly demonstrated, suggesting a genetic contribution to these phenotypes [
Therefore, the aim of our study was to use the robust method of linkage analysis in a large well-characterized cohort of T1D families to identify gene-loci that predispose to type 1 diabetic complications. We focused our genome analysis on chromosome 6 to follow up our previous work showing the importance of loci on chromosome 6 to the genetic predisposition of T1D complications [
Families were ascertained through the presence of at least one family member with type 1 diabetes. A questionnaire was given to the proband or parents as well as to additional family members. The questionnaire included demographic, medical, genealogical, and familial information about T1D as well as complications. More details can be found in Lipner et al. [
Our dataset included 415 families (2,008 individuals) with T1D cases diagnosed before age 30 (Tables
Number of families with affected (T1D + complications)-unaffected (T1D only) members.
Affected-unaffected family members |
|
---|---|
1 affected-1 unaffected | 68 (16) |
2 affected-0 unaffected | 50 (12) |
0 affected-2 unaffected | 210 (51) |
Other | 87 (21) |
|
|
Total | 415 (100) |
Prevalence of clinical characteristics among 415 T1D families.
Clinical characteristic | Number (%) of individuals |
---|---|
Total | 2,008 (100.0) |
T1D + microvascular complications | 239 (11.9) |
T1D + retinopathy | 219 (91.6) |
T1D + nephropathy | 87 (36.4) |
T1D + neuropathy | 76 (31.8) |
T1D only | 629 (31.3) |
No T1D | 1140 (56.8) |
The accuracy of the self-reported information about complications was evaluated by the following. Including extra questions about complications-related conditions in the questionnaire. Reports of macular edema, vitrectomy, or complete or partial blindness were considered an indicator of retinopathy; reports of end-stage renal failure, kidney failure, or repeated high urinary albumin levels were considered an indicator of nephropathy. In cases of inconsistencies (e.g., report of macular edema but not retinopathy), further investigations were carried out through phone interviews. In order to avoid ambiguity, only the most obvious or severe cases of retinopathy or nephropathy were classified as “affected.” Data available from follow-up were used to confirm or update the presence/absence and progression of complications. 179 patients had medical records available allowing us to verify phenotype according to American Diabetes Association guidelines [ Information indicating
The accuracy of self-reported information was assessed in three ways. (1) Additional questions were included in questionnaires given to both patients and family members. (2) Follow-up telephone interviews were carried out by HBDI staff if the questionnaire was unclear. (3) Medical records were assessed on T1D patients that submitted medical records with the questionnaire (179 (2.3%)). (4) Follow-up questionnaires went to a subset of families for updated information about the development of complications, new cases of diabetes, and other related medical history. Twenty-three percent of the type 1 diabetics in the HBDI database responded with follow-up data and 10% of subjects included medical records with the questionnaire. On-going validation at HBDI has shown that questionnaire answers accurately reflect physician diagnosis in the medical records [
Since the majority of patients’ diagnoses are self-reported, T2D may have occasionally been misclassified as T1D. The presence of autoantibodies would confirm an autoimmune response. Autoantibody markers from a random sample of T1D study subjects (
Reliability of self-report questionnaires: self-reports of diabetes have demonstrated excellent agreement with the use of medical records [
Thus, if any T1D families were actually T2D, or if some patients with complications were misdiagnosed as complications-free, it is unlikely to have introduced bias into our results for two major reasons. (1) Misdiagnosing an affected person as “unaffected” decreases linkage evidence but does not lead to false linkage evidence [
The Center for Inherited Disease Research (CIDR) at the National Human Genome Research Institute did the genotyping. Average marker spacing was 0.58 cM. We restricted our analyses to the 402 marker SNPs on chromosome 6.
“Affected” phenotypes were (1) the presence of any microvascular complication, (2) the presence of retinopathy, and (3) the presence of nephropathy. Each phenotype was analyzed separately. The neuropathy phenotype yielded too little linkage information and no further analyses were done using that phenotype. T1D patients without complications were classified as “unaffected.” Individuals without T1D were excluded from these analyses (except parents). Families had at least one “affected” and one “unaffected” family member, or at least two affected members (e.g., at least two siblings with T1D, at least one of whom had complications).
Multipoint LOD (“logarithm of odds”) scores and heterogeneity LOD scores (HLOD scores) were calculated using the GeneHunter program [
We performed preliminary analyses on the phenotype “any complication” but our subsequent analysis classified only subjects with retinopathy (RET) as “affected” and, separately, only subjects with nephropathy (NEPH) as “affected.”
We previously showed [
With the affected phenotype defined as “presence of any complication,” a large linkage peak emerged centered in the HLA region (50–52 cM); the LOD and HLOD scores at 52 cM (HLA region location) were 4.0 and 5.3, respectively. Two separate, novel loci for complications were located
(a) Linkage analysis with “any complication” as the phenotype. (b) Linkage analysis with retinopathy as the phenotype. (c) Linkage analysis with nephropathy as the phenotype.
RET was the most common complication found in our dataset. We saw only minor differences between the ANY COMPLICATION and RET analyses. The maximum scores at the 42 cM peak for RET were LOD = 3.6 and HLOD = 5.0 (for ANY COMPLICATION, the scores were LOD = 2.6 and HLOD = 4.4 (compare Figures
The linkage results in the 45 NEPH families show 2 peaks: the first peak occurs over the HLA region at 52 cM (LOD = 1.3 and HLOD = 1.4 (Figure
These results reveal two novel loci that contribute to the expression of complications. These two loci appear to have differential influences on RET and NEPH: one influencing mostly NEPH (64 cM) and the other influencing only RET (42 cM). We then investigated possible interaction of these loci with HLA allele.
We previously showed [
(a) Linkage analysis with retinopathy as the phenotype, stratified by families whose probands are positive for DRB1*03:01/X = any allele. (b) Linkage analysis with retinopathy as the phenotype, stratified by families whose probands are positive for DRB1*03:01/X, where X≠DRB1*04:01. (c) Linkage analysis with retinopathy as the phenotype, stratified by families whose probands are positive for DRB1*04:01/X. (d) Linkage analysis with retinopathy as the phenotype, stratified by families whose probands are positive for DRB1*04:01/X = any allele, where X≠DRB1*03:01.
At the 42 cM locus, the unstratified RET analysis (above) had obtained a LOD = 3.6 and an HLOD = 5.0. The DRB1*03:01 stratification analysis still showed strong evidence for linkage, but the LOD and HLOD scores decreased (LOD = 3.0 and HLOD = 3.9), the decrease suggesting only that some families contributing to disease expression have been removed from the data. However, at the 64 cM locus, there was a significant
The above stratification analysis included DRB1*03:01 positive probands with the DRB1*03:01/04:01 genotype. Because our previous association analysis [
LOD score summary table.
LOD (HLOD) scores | |||
---|---|---|---|
Phenotype | 42 cM peak | 52 cM peak | 64 cM peak |
Any complication | 2.6 (4.4) | 4.0 (5.3) | −1 (2.6) |
Retinopathy | 3.6 (5.0) | 3.6 (5.0) | −1.5 (2.2) |
Nephropathy | −2.0 (0.0) | 1.3 (1.4) | 2.0 (2.2) |
Retinopathy + nephropathy analyzed together | 3.2 (4.8) | 4.0 (5.2) | −1.1 (2.3) |
|
|||
Stratification (retinopathy only) | |||
DRB1*03:01 | 3.0 (3.9) | 5.1 (5.3) | 3.1 (3.4) |
DRB1*04:01 | 4.1 (4.1) | 4.2 (4.2) | −0.5 (0.9) |
DRB1*03:01/X | 0.9 (1.6) | 2.0 (2.2) | 2.0 (2.2) |
(X≠DRB1*04:01) | |||
DRB1*04:01/X | 2.5 (2.5) | 1.4 (1.4) | −1.0 (0.0) |
(X≠DRB1*03:01) |
Numbers in parentheses are HLODs.
Numbers of families in the subgroups.
Complication | Stratification subgroup | Number of families included | Count of people included |
---|---|---|---|
Any complication | 159 families | 1015 people | |
|
|||
Retinopathy | All families | 144 families | 928 people |
DRB1*03:01 | 61 families | 409 people | |
DRB1*04:01 | 58 families | 336 people | |
Pure DRB1*03:01 | 37 families | 266 people | |
Pure DRB1*04:01 | 35 families | 199 people | |
|
|||
Nephropathy | 45 families | 325 people |
We analyzed the RET data using only families of probands carrying the DRB1*04:01 allele, including DRB1*03:03/DRB1*04:01 heterozygotes. The LOD and HLOD scores at the 42 cM locus in the DRB1*04:01-stratified analysis remained high (4.1 and 4.1, resp.). This suggests that heterogeneous loci contributing to RET were eliminated in the stratified sample (hence the increase in the LOD score) but some families contributing to the LOD score were also eliminated (thus the decrease in the HLOD). At the 64 cM locus, the HLOD decreased (HLOD = 0.9) compared to the unstratified analysis (HLOD = 2.2); the LOD scores remained negative (unstratified = −1.5 and stratified = −0.5). This is in sharp contrast to the DRB1*03:01 stratification findings, in which the evidence for linkage at the 64 cM locus was notably stronger than in the unstratified analysis. This suggests that the 64 cM locus interacts with the DRB1*03:01 allele to foster the expression of RET but that it does not interact with the DRB1*04:01 allele. This conclusion is strengthened when we look at the results of the “pure” DRB1*04:01 family analysis (see below).
When including only DRB1*04:01/X, (X≠03:01) families, the signal at the 42 cM location remains notable with no evidence of heterogeneity (LOD = 2.5 and HLOD = 2.5), despite the large drop in sample size (see Table
The stratification analysis results suggest that the 42 cM and the 64 cM loci interact epistatically and differentially with the DRB1*03:01 and *04:01 alleles, revealing evidence of locus heterogeneity for each phenotype. The unstratified analysis of the RET families at the 42 cM locus yields a LOD score of 3.6 and an HLOD of 5.0, indicating substantial locus heterogeneity in the data. Analysis of “pure” 03:01 families yields a LOD < 1 for RET at the 42 cM locus and an HLOD that still suggests heterogeneity (HLOD = 1.6). The “pure” 04:01 families yield a LOD = HLOD = 2.5 at the 42 cM locus. These results suggest that the 42 cM locus interacts positively with the 04:01 allele (to produce RET) and negatively with the 03:01 allele (to protect against RET). Furthermore, when only the 04:01 allele is present (and not 03:01), the 42 cM locus shows no evidence of heterogeneity.
The DRB1*03:01 stratification analyses suggest that the 42 cM locus influences RET less when *03:01 is present than when *04:01 is present. This is expected if *03:01 “protects” against RET. Like the 42 cM locus, the 64 cM locus shows evidence of interaction, but with the *03:01 allele. When stratifying on the *04:01 allele, there is almost no evidence for linkage at the 64 cM locus.
In this study, we used, for the first time, LOD score linkage analysis to identify loci that contribute to the expression of the microvascular complications of RET and NEPH. Linkage analysis has been shown to have the most power to detect loci important for disease expression and has the greatest ability to give us information about the genetic characteristics of the phenotype and the existence of heterogeneity [
The results of the “any complication” phenotype indicated the existence of three loci. The fact that the HLA locus appeared is unsurprising [
Statistically significant evidence for the existence of two loci for complications adds strong support for inherited influences on complications’ expression. The two loci appear to affect RET and NEPH differently. Both appeared to influence RET expression but the 64 cM locus appeared to influence only NEPH and the 42 cM locus had no influence on NEPH.
The significant change in the 64 cM locus’s LOD score among proband families with DRB1*03:01 strongly suggests that the presence of DRB1*03:01 increases the influence of the 64 cM locus on RET expression. The influence of the 64 cM locus virtually disappears when the proband has the 04:01 allele while the 42 cM peak is strengthened. When proband families are not selected for having a particular HLA allele, the observed evidence for heterogeneity is expected if the two HLA alleles contribute differentially to the phenotype. The positive differences in the LOD scores between the stratified and unstratified samples are strong indicators of interaction of the HLA alleles with the two loci [
The strong linkage evidence at the HLA region (52 cM) might indicate the influence of HLA on complications (known from association evidence) or merely cosegregation of HLA alleles with diabetes in general regardless of complications. Changes in the HLA linkage profile under stratification are not interpretable because we artificially altered the HLA allele structure by including or excluding specific alleles. However, finding that the DRB1*03:01 and *04:01 alleles interact with the novel loci confirms that HLA influences complications’ expression.
This study is not without ambiguities. The linkage region we have identified (30 cM–70 cM) is a relatively small one for most linkage analyses; yet we have observed three distinct loci with specific effects. One of those loci is the HLA region, which strongly affects T1D expression. Were we analyzing the T1D phenotype and not complications, the LOD score at HLA would be on the order of 40, thus swamping any other T1D-related signals. However, the narrowness of the region does not nullify the clear separateness of the linkage signals. No matter how the data are stratified and broken down, the consistency of the 42 cM and 64 cM peaks, even when the 42 cM peak disappears (as in the analysis of NEPH), argues strongly that these loci influence complications’ expression. The number of linkage analyses that we have performed may lead to the question as to whether the LOD scores for the two loci we identified should be subject to correction for genome-wide testing. All three peaks appear in the first analysis with notable (2.5–5) LOD scores and/or HLOD scores and the locations of these peaks were invariant. The information content of the genotypic data did not fall below 98% across the region. The usual criterion for significance of a LOD score (variously debated to be from 2.5 to 4.0) is for We used changes in the height of the peaks as indicators of the loci’s influence on complications’ expression. The question of the relative strengths of influence on gene expression as related to linkage peak height is not a well-studied area. Linkage will most likely only be observed with loci “necessary” for disease expression [ Previous work has demonstrated how the stratification technique we used can identify epistatically interacting loci [
The next step in this work is to analyze the whole genome, now that we know the importance of stratification loci in identifying interaction. While applying this study’s stratification approach may help us identify the specific genes in the linkage regions using association analysis of SNPs with retinopathy and/or nephropathy, the option also now exists to use next-generation sequencing to identify the disease-related variants. The difficulty, as with other common conditions, is identifying the disease-related variant if such variants do not occur in exons.
Replicating our work in other samples is highly desirable. However, since the wide adoption of GWAS as the genetic technique of choice and the accompanying decrease in the collection of family data, it is not clear how much family data exist for linkage of complications. Nevertheless, family studies are the best way to effectively use the newest genetic technologies [
The authors declare that there is no conflict of interests associated with this paper.
Ettie M. Lipner chose and implemented the statistical analyses and wrote the paper. Yaron Tomer contributed ideas and suggested analyses and reviewed the paper. Janelle A. Noble generated the HLA data and contributed valuable information to the paper. Maria C. Monti cleaned the data and did the initial familial analyses and reviewed the paper. John T. Lonsdale provided the data and supervised the updating of the clinical information. Barbara Corso did genetic data cleaning and database quality control. David A. Greenberg initiated the study, contributed analysis ideas, supervised the analyses, and contributed to the paper.
The authors thank Rebecca Yohannes (Columbia University), Junying Zhang (Columbia University), Ryan Subaran (Nationwide Children’s Hospital), and Abbie Neininger (Nationwide Children’s Hospital) for programming and analysis assistance. This project is funded, in part, under a grant with the Pennsylvania Department of Health. The Department specifically disclaims responsibility for any analyses, interpretations, or conclusions (John T. Lonsdale). This research was also supported in part by Grants 2T15 LM009451 (Ettie M. Lipner); NINDS NS27941, NIMH MH48858, NS61829, NS70323, Nationwide Children’s Hospital, and the New York State Psychiatric Institute (David A. Greenberg); DK61722 (Janelle A. Noble); DK61659, DK067555, and DK073681 (Yaron Tomer).