A genome screen of systemic lupus erythematosus using affected-relative-pair linkage analysis with covariates demonstrates genetic heterogeneity


Systemic lupus erythematosus (SLE) appears to be the consequence of complex genetics and of only partly understood environmental contributions. Previous work by ourselves and by others has established genetic effects on 1q, 2q, 4p, 6p, and 16p using SLE as the phenotype. However, individual SLE affecteds are extraordinarily different from one another by clinical and laboratory measures. This variation may have a genetic basis; if so, it is advantageous to incorporate measures of between-family clinical variability as covariates in a genetic linkage analysis of affected relative pairs (ARPs) to allow for locus heterogeneity. This approach was applied to genome scan marker data from 160 pedigrees multiplex for SLE and containing 202 ARPs. Because the number of potential covariates was large, we used both ad hoc methods and formal principal components analysis to construct four composite covariates using the SLE classification criteria plus age of onset, ethnicity, and sex. Linkage analysis without covariates has detected evidence for linkage at 1q22–24, 2q37, 4p16, 12p12–11, and 17p13. Linkage analysis with these covariates uncovered linkage at 13p11, 17q11–25, and 20q12 and greatly improved evidence for linkage at 1q22–24, 2q37, 12p12–11, and 17p13. Follow-up analysis identified the original variables contributing to locus heterogeneity in each of these locations. In conclusion, allowing for locus heterogeneity through the incorporation of covariates in linkage analysis is a useful way to dissect the genetic contributions to SLE and uncover new genetic effects.


Systemic lupus erythematosus (SLE) is a chronic, heterogeneous autoimmune disorder characterized by the production of self-reactive antibodies. SLE is believed to be multifactorial and most likely involves complex interactions between genetic, environmental, and hormonal factors. Multiple autoantibody specificities that contribute to disease either by directly binding to self-antigens or by induction of inflammation following tissue deposition of antibody-antigen immune complexes may be present.1,2 Clinical manifestations of SLE patients may be constitutional or result from inflammation in various organ systems, including skin and mucous membrances, joints, kidney, brain, serous membranes, lung, heart, and occasionally gastrointestinal tract.3 Organ systems may be involved singly or in any combination. Hallmark features of SLE are the production of autoantibodies against nuclear components and the classic indurated and erythematosus malar rash.3 Classification of SLE requires that a patient meet any four of 11 criteria established by the American College of Rheumatology (ACR).4,5 These 11 criteria include 20 clinical manifestations identified by clinical examinations, laboratory tests, and patient self-report.

SLE is at least nine times more prevalent in women but the underlying cause of the gender effect is not clearly established. A higher prevalence of renal dysfunction in affected males has been reported in numerous studies.6,7,8,9,10 Recently, Jedrey et al,11 using the multicase families analyzed for the present paper, demonstrated that SLE females with affected male relatives had a higher prevalence of renal disease than SLE females with no affected male relative, suggesting that SLE families that include a male affected member may be a distinct subpopulation.

In the present paper, we apply an affected-relative-pair linkage method that allows covariates to be included.12 This method is designed to allow for locus heterogeneity measureable by additional phenotypic information obtained from affected individuals. The primary purpose of the analysis is therefore to gain power in linkage analysis by allowing for covariate-related locus heterogeneity and, in so doing, begin to identify subsets of families defined by clinical characteristics and other phenotypic characteristics as being linked to particular genetic locations. Because the number of potential covariates is large, we first combine ACR criteria according to organ system and then apply principal component (PC) analysis to obtain a small set of variables that explain much of the between-family variation. After scanning the genome using these variables, we then reanalyze significant regions using the original variables. By so doing, we obtain greater evidence of linkage to some previously reported locations, detect linkage in some previously undetected locations, and describe characteristics of the ‘linked’ subset of families for all these locations.


The eigenvalues and proportions of explained variance for the PCs are given in Table 1. PCs 1, 2, 4, and 6 explain 82.6% of the between-family variability and were thus chosen as covariates for the genome screen. We considered including PC 10 in the set to be scanned because it contributes a proportion of between-family variability not much different from that of PC 6. However, because the eigenvalue associated with PC 10 was one of the smallest and because we were concerned about excess Type I error, we decided to exclude PC 10 from further consideration. The coefficients for these PCs are shown in Table 2.

Table1 Eigenvalues, proportion of explained variance, and proportion of between-family variance explained by principal components
Table2 Coefficients of principal components 1, 2, 4, and 6.

Lod scores for the baseline model (ie, without covariates) and models with each of the four PCs as covariates are shown in Figure 1. In Table 3, we give summary information at 13 locations for which either the overall lod score or the covariate effect alone was significant at the 0.001 level. For the eight signals for which the total lod score exceeded 2.79 (α = 0.001), five occurred in locations for which the baseline model showed substantial evidence for linkage. (For the signal on chromosome 2, the baseline lod score exceeded 1.0 just 10 cM from the location of the maximum with PC 4.)

Figure 1

Genome scan results for baseline model and four models with a principal component as a covariate. The lower dashed line indicates the α = 0.001 significance level for the baseline model; the upper dashed line indicates the α = 0.001 significance level for a model with a single covariate.

Table3 Significant baseline and principal component (PC) linkage signals

For these eight regions, we obtained the best, most-parsimonious models using the original variables. In Table 4, we illustrate part of this process for the signal in the FcγR region on chromosome 1q. First, the lod-score contributions of the four variables with the highest absolute loadings on PC 2 were assessed; the indicator denoting the presence in the family of a male affected (MF) gave the highest lod-score contribution and was the only variable significant at the α = 0.05 level of significance. The incremental effects of the remaining covariates to the model with MF were then assessed; race alone was significant (P = 0.0406) and no further improvement was obtained. The interaction between MF and race was not significant and thus the best, most-parsimonious model was taken to be one that included the main effects of those two covariates. The values of the parameter estimates (Table 5) indicate that affected relative pairs (ARPs) most likely linked are non-European-American (non-EA) families without a male affected member.

Table4 Some multiple regression analyses of the 1q23 (FcγR) region
Table5 Final models for eight regions

In Table 5, the final models for the eight regions are summarized. For the region on chromosome 13, no significant model was found using the original variables. For the region on chromosome 4p, no covariate significantly improved the baseline lod score. For three of the remaining regions, race was included in the final model. At 17q, the EA group showed stronger evidence for linkage; at 1q and 2q, the non-EA group showed stronger evidence for linkage. We divided the non-EA group into African-Americans (AA) and other ethnicities; no additional effects were found. The signals on chromosomes 1q and 20 had significant gender effects; in both cases, families with male affecteds were less likely to be linked.

We performed additional analyses on selected regions. For the second dermatologic criterion (DM2) signal on 12p, we fit separate models for its components oral ulcers and photosensitivity. Because hematologic criterion (HM) also had a substantial individual effect in this region (lod score = 2.35), we also fit separate models for its components hemolytic anemia, leukopenia, lymphopenia, and thrombocytopenia. The results of some of these analyses are in Table 6. For the four HM components, the covariate parameters are negative, indicating that ARPs without hematological manifestations are more likely linked. Thrombocytopenia has the largest effect of any single component and has a larger effect than HM itself; addition of other HM components does not improve the thrombocytopenia model. Of all the one-covariate signals, the signal using the combined variable DM2 is largest. The best, most parsimonious model includes both DM2 and thrombocytopenia; ARPs with oral ulcers and/or photosensitivity but lack thrombocytopenia are the most likely to be linked.

Table6 Follow-up models for 12p12–11

Because the signal on 4p16 previously showed epistasis with 5p15 in EAs,13 we fit various covariate models to both regions, the largest of which are reported in Table 7. Only age-of-onset significantly improved the lod score at 5p15; no significant covariates were found at 4p16. In both locations, however, neurological criterion (NR) had one of the largest (though nonsignificant) contributions; ARPs with neurological and/or psychiatric manifestations were the most likely to be linked to these locations. Epistasis implies that the same subset of families is segregating for two or more loci; it may be that this subset is characterized in part by the presence of NR manifestations.

Table7 Regression models for 2q37, 4p16, and 5p15

It is interesting to note that the region we detected on 2q (Table 5; reproduced in Table 7) also shows ARPs with NR manifestations are more likely to be linked. One can therefore hypothesize that there may be epistasis among these three regions; the small subset of ARPs with NR manifestations is more strongly linked to all three locations. However, the role of ethnicity appears to differ among the three locations. As a result, we computed race-specific sibling recurrence risk ratios (λs) for two different values of the NR covariate. For both ethnic groups, estimates of λs for ASPs with an NR score equal 4 (corresponding to individual values of (2,2) or (1,3) for the ASP; no pairs had higher scores) were substantially larger than 1 in all three regions. EAs have estimates of λs equal to about 3.7 for all three regions, suggesting roughly equal contributions from the three regions. For non-EAs, estimates of λs are much more variable, suggesting unequal contributions from the three regions, with the largest from the locus on 2q. For ARPs with no indication of NR, evidence for linkage was considerably weaker in both ethnic groups.


We used a linkage method that includes covariates to allow for locus heterogeneity in a collection of SLE families. We were thus able to improve evidence for linkage in several genomic regions and determine the characteristics of families most likely linked to each region. Thirteen regions showed significant (at the α = 0.001 level) overall lod scores and/or covariate effects. Three additional regions (chromosomes 6, 16, and 18) showed signals that just missed those cutoffs. No evidence for linkage was found using additional covariate models to the HLA region and to regions on chromosomes 10, 14, and 16 that have been implicated by independent studies. One region (chromosome 13) gave a significant lod score with a PC but not with any of the original variables; the reason for this result is unknown, although Type I error cannot be ruled out.

Many of our most significant signals were in regions where the baseline model (ie, the model without covariates) showed substantial evidence for linkage, suggesting that the baseline model has not been ineffective in detecting linkage in these data. In several locations, evidence for linkage was greatly improved, however, with the addition of covariates. For example, the lod score at 2q increased from about 1.0 to over 4.0 with the addition of clinical and demographic covariates. This region of 2q has been also been identified in independent studies.14,15

We obtained a large new linkage signal with a peak at about 20p11.2, with substantial linkage extending throughout chromosome 20. The 20p12 region has been identified in independent studies as being linked to SLE14,16 and to psoriasis.17,18 The 20q12 region has been identified as being linked to SLE14,16 and to Grave’s disease.19 However, because of the imprecise nature of location estimates in linkage studies, it is not at present possible to determine if, in fact, two separate susceptibility regions exist and, if so, to which region our families map. An earlier report using a subset of our families gives the strongest evidence in 20q12.20I n any event, it is notable that autoimmune diseases often cluster in families;21 these results suggest that susceptibility genes for autoimmunity in general may be present on chromosome 20.

At 17p13, we observed a baseline lod score of 2.2 and a lod score of 4.4 with age-at-onset as a covariate. In a recent report of results obtained using 16 of these pedigrees that were of EA ancestry and had one or more members with vitiligo, Nath et al22 reported a maximum lod score of 4.0 at D17S974-D17S1303, about 12 cM from our peak at D17S1298. As vitiligo was not among the criteria we analyzed, we fit a model using as a covariate an indicator variable denoting the presence or absence of vitiligo in any family member and observed an effect of vitiligo (lod score = 1.53, P = 0.008) at D17S974-D17S1303 but no improvement at D17S1298. When vitiligo is added to the model containing age-of-onset, it improves the lod score at D17S974-D17S1303 from 0.6 to 2.4 but did not improve the lod score of 4.4 at D17S1298. We observed no effect of race at either location, either alone or in conjunction with the other covariates. As location estimates obtained from linkage analysis are not precise, it is unclear if these findings represent two separate peaks or only one and, if the latter, which model and which location estimate more accurately describes the underlying genetic model.

In our study, three regions (2q, 4p, and 5p) show some evidence that families with NR manifestations are more likely to be linked. Because of the strength of linkage in ARPs with NR manifestations, and because NR manifestations are uncommon, we might find that the same subgroup of families is linked to all three regions. Whether this suspicion would survive formal multilocus modeling has yet to be determined.

We examined all models for influential observations. For the age-at-onset signal on 17p, we noted that one ARP with very late ages-at-onset contributed 0.90 to the total lod score. We fit the model excluding this pair; the lod score was now 3.50, but the parameter estimates were virtually identical to those obtained when that pair was included. We concluded that although this pair contributed substantially to the lod score, it did not influence the fit of the model.

In any study in which several covariates are analyzed, inflation of Type I error is potential problem. Appropriate corrections for multiple testing are unavailable in this context, but would likely require prohibitively small significance levels because the multiple testing adjustment required in the context of a genome scan of a even single variable gives significance levels on the order of 10−5. In this paper, we used PC analysis solely to reduce the number of covariates to be subjected to testing in order to help control Type I error. Overall, these results should be considered exploratory and will require independent confirmation. In general, we favor careful selection of covariates for analysis and, perhaps more important, complete reporting of all covariates and genomic locations analyzed to allow the reader to place the highlighted results in context. Nonetheless, we have shown that considerably more information can be obtained from a linkage analysis if additional phenotypic information is included to allow for locus heterogeneity.

A more long-term goal of these analyses, in addition to improving evidence for linkage, is to identify subgroups of families or ARPs that segregate for different sets of loci. For example, a significant MF covariate indicates that ARPs with a male affected and ARPs without a male affected differ significantly in the degree to which they are linked. As a result, follow-up studies such as fine mapping and candidate gene analysis might be restricted to ARPs in the more highly linked group. Alternatively and more generally, the fitted model might be used to calculate, for each ARP, the probability that the pair is linked, conditional on the pair’s allele sharing and covariate values. These probabilities might then be used as weights in subsequent linkage analyses (to detect additional loci) and association analyses (to evaluate candidate genes) to effectively isolate a subgroup of families and, hopefully, increase power. These and other methodological improvements (such as multilocus models with covariates) currently under development are possible next steps in the genetic dissection of SLE.

Materials and methods


Procedures for recruitment of families and genotyping of 324 microsatellite markers have been described previously.20 In addition to the data from 126 families used by Rao et al,23 data from 34 additional SLE multiplex families were included in this analysis. Individuals who met the ACR criteria for a diagnosis of SLE were considered to be affected with SLE. This data set included 117 affected sibling pairs (ASPs) and 85 affected half-sibling, avuncular, and cousin pairs, giving a total of 202 affected relative pairs (ARPs). Relationship testing24 was used to confirm the genetic relationships within nuclear families. Our sample of 372 affecteds was predominantly female (345:27) and had a mean age-at-onset of 33.2 years, consistent with previous observations in sporadic cases.25 Of the 160 families, 92 were European-American, 56 were African-American, two were Asian-American, two were Native American, seven were Hispanic-American, and one was Middle-Eastern-American. Because African-Americans have a higher prevalence of both SLE and SLE-related renal disease, we separated race into two classifications: European-American (EA) and non-EA. The numbers of Hispanic, Asian, and Native American families were too small to warrant separate ethnic categories. When the non-EA group was observed to be the more strongly linked, we further separate the non-EA group into subcategories for further analysis.

Both medical records and patient (or patient surrogate) interviews were used to score SLE patients for the presence or absence of each ACR manifestation. All evaluations were performed by a nurse with review by a physician with specialized training and extensive experience in rheumatic disease. Medical records and patient awareness may not definitively report manifestations possibly consistent with SLE in a way that fits the format and definition given in the ACR Criteria for Classification.4,5 To cope with this problem, we used a scoring system to assess the evidence supporting a given ACR criterion for SLE. A score of 3 was given if the medical record or patient self-report (or surrogate) convincingly supported the criterion. A score of 2 was given if the medical record or patient self-report strongly suggested that the criterion was present, but there were reservations that prevented or precluded the evaluator from being completely convinced. A score of 1 indicated that the evaluator doubted that the criterion was present, but some aspect of the available evidence prevented a confident conclusion that the criterion had been absent. A score of 0 meant that there was no evidence supporting the criterion or that the criterion was convincingly negative. Only rarely did a patient self-report or surrogate provide convincing evidence of the presence of a characteristic in the absence of medical record evidence.

Two dermatological scores were obtained by taking the highest value (0, 1, 2, or 3) from malar rash and discoid rash (DM1) and from photosensitivity and oral ulcers (DM2); renal score (RN) by taking the highest value from proteinuria and cellular casts; immunological (IM) score by taking the highest value from LE cell, anti-DNA, and anti-Sm; hematologic score (HM) by taking the highest value from hemolytic anemia, leukopenia, lymphopenia, and thrombocytopenia; neurological score (NR) was obtained by taking the highest value from seizures and psychosis; cardiopulmonary score (CP) by taking the highest value from pericarditis and pleuritis; and arthritic score (AR) by taking the single measure of arthritis.

Analysis strategy

Because the number of potential covariates is large, we were concerned that analysis using the original variables would generate large numbers of Type I errors. To reduce the impact of this problem, we first used PC analysis to obtain a smaller set of linear combinations of the original variables that accounted for at least 80% of the between-family variability. We then performed a genome screen using each of these PCs in turn. When a large signal was identified, we analyzed linkage in that genomic region using the original variables as covariates, focusing on those covariates with large contributions (in absolute value) to the significant PC. Using stepwise multiple regression, we chose a best-fitting most parsimonious final model.

Principal component scores

PC scores26 for inclusion as covariates in ARP linkage analysis were obtained from the correlation matrix of the eight SLE-related phenotypes (using the ordinal 0, 1, 2, 3 scale), plus ethnicity, age-of-onset, gender, and an indicator variable (MF) coded as 1 if the SLE patient was a member of a family that included a male affected member and 0 otherwise. We initially obtained a set of PC scores without including MF and a set without including gender. The analysis that included both gender-related variables generated PCs more meaningfully identified contrasts between subgroups of families, particularly with respect to the gender-related variables. Consequently, this is the analysis reported herein.

Linkage analysis

As our linkage analysis model, we used the one-parameter conditional logistic ARP model described in Goddard et al.12 The model is in essence a regression model that includes all types of ARPs. The mode of inheritance parameter is constrained to an optimal value roughly halfway between a dominant and a recessive model27 and thus the model requires only one additional parameter for each covariate. Let λ12) be the recurrence risk ratio for a pair of relatives that share exactly 1 (2) alleles identical-by-descent (IBD) and put β1 = ln(λ1) (β2 = ln(λ2)). The Whittemore and Tu27minmax constraint is implemented by putting λ2 = 3.634λ1 − 2.634 and covariates are added to the model by putting

where the xi, i = 1, 2, . . ., K, are the covariates to be included in the model, and the γi are the corresponding parameters. This model is currently implemented in the SAGE28 program LODPAL and is currently available as a beta version from http://darwin.cwru.edu. In this analysis, multipoint IBD sharing estimates for autosomal loci were obtained using the GENIBD program, also in SAGE.

In this analysis, inclusion of a covariate allows for linkage heterogeneity due to the covariate. For example, a binary covariate indicating population membership allows for population heterogeneity in linkage to a particular genomic location. Under genetic constraints, including such a covariate is equivalent to analyzing each subpopulation separately and adding the lod scores. Continuous covariates have a similar interpretation in that they allow for heterogeneity in linkage due to the covariate. It should be noted that these analyses are performed using the entire set of ARPs and is an alternative to the more common but less efficient and less powerful practice of analyzing subsets of pairs. Using the parameter estimates, one can then calculate sibling recurrence risk ratios at particular values of the covariate:

subject to the minmax constraint described above. In our analysis, we analyzed each PC separately and included as a covariate the sum of the PC values for the two members of the ARP. We assumed genetic constraints29 to hold at the sample mean covariate value (which is zero for a PC), but not necessarily at other covariate values.30 With the exception of race and the presence of a male affected, which were included as family-specific binary covariates, the original variables were also included as the sum of values for the two members of the ARP. Additional details on this method of analysis, including the asymptotic distributions of likelihood ratio statistics (here referred to as ‘lod scores’), and an application to a prostate cancer genome screen are given in Goddard et al.12


  1. 1

    Maddison PJ, Reichlin M . Deposition of antibodies to a soluble cytoplasmic antigen in the kidneys of patients with systemic lupus erythematosus Arthritis Rheum 1979 22: 858–863

  2. 2

    Skinner RP, Maddison PJ . Analysis of polyethylene glycol precipitates from SLE sera: antibody enrichment in association with disease activity Clin Exp Rheumatol 1990 8: 553–560

  3. 3

    Gladmann DD, Urowitz MD . Systemic lupus erythematosus: clinical features. In: Klippel JH, Dieppe PA (eds), Rheumatology Time Mirror International Publishers Ltd: London 1997

  4. 4

    Tan EM, Cohen AS, Fries JF et al. The 1982 revised criteria for the classification of system lupus erythematosus Arthritis Rheum 1982 25: 1271–1277

  5. 5

    Hochberg MC . Updating the American College of Rheumatology criteria for systemic lupus erythematosus Arthritis Rheum 1997 40: 1725

  6. 6

    Blum A, Rubinow A, Galun E . Predominance of renal involvement in male patients with systemic lupus erythematosus Clin Exp Rheumatol 1991 9: 206–207

  7. 7

    Ward MM, Studenski S . Systemic lupus erythematosus in men: a multivariate analysis of gender differences in clinical manifestations J Rheumatol 1990 17: 220–224

  8. 8

    Molina JF, Drenkard C, Molina J et al. Systemic lupus erythematosus in males: a study of 107 Latin American patients Medicine 1996 25: 124–130

  9. 9

    Jacobsen S, Petersen J, Ullman S et al. A multicentre study of 513 Danish patients with systemic lupus erythematosus. I. Disease manifestations and analyses of clinical subsets Clin Rheumatol 1998 17: 468–477

  10. 10

    Mok CC, Lau CS, Chan TM, Wong RWS . Clinical characteristics and outcome of southern Chinese males with systemic lupus erythematosus Lupus 1999 8: 188–196

  11. 11

    Jedrey CM, Olson JM, Gray-McGuire C, Bruner GR, Harley JB, Moser KL . Increased prevalence of renal disease in SLE families with affected males Arthritis Rheum 2002 46: 428–435

  12. 12

    Goddard KAB, Witte JS, Suarez BK, Catalona WJ, Olson JM . Model-free linkage analysis with covariates confirms linkage of prostate cancer to chromosomes 1 and 4 Am J Hum Genet 2001 68: 1197–1206

  13. 13

    Gray-McGuire C, Moser KL, Gaffney PM et al. Genome scan of human systemic lupus erythematosus by regression modeling: evidence of linkage and epistasis at 4p16–15.2 Am J Hum Genet 2000 67: 1460–1469

  14. 14

    Gaffney PM, Ortmann WA, Selby SA et al. Genome screening in human systemic lupus erythematosus: results from a second Minnesota cohort and combined analyses of 187 sib-pair families Am J Hum Genet 2000 66: 547–556

  15. 15

    Lindqvist AKB, Steinsson K, Johanneson B et al. A susceptibility locus for human systemic lupus erythematosus (hSLE1) on chromosome 2q J Autoimmun 2000 14: 169–178

  16. 16

    Shai R, Quixmorio FP Jr, Li L et al. Genome-wide screen for systemic lupus erythematosus susceptibility genes in multiplex families Hum Mol Genet 1999 8: 639–644

  17. 17

    Nair RP, Henseler T, Jenish S et al. Evidence for two psoriasis susceptibility loci (HLA and 17q) and two novel candidate regions (16q and 20p) by genome-wide scan Hum Mol Genet 1997 6: 1349–1356

  18. 18

    Trembath RC, Clough RL, Rosbotham JL et al. Identification of a major susceptibility locus on chromosome 6p and evidence for further disease loci revealed by a two stage genome-wide search in psoriasis Hum Mol Genet 1997 6: 813–820

  19. 19

    Tomer Y, Barbesino G, Greenberg DA, Concepcion E, Davies TF . A new Graves disease-susceptibility locus maps to chromosome 20q11.2. International Consortium for the Genetics of Autoimmune Thyroid Disease Am J Hum Genet 1998 63: 1749–1756

  20. 20

    Moser KL, Neas BR, Salmon JE et al. Genome scan of human systemic lupus erythematosus: evidence for linkage on chromosome 1q in African-American pedigrees Proc Natl Acad Sci 1998 95: 14869–14874

  21. 21

    Bias WB, Reveille JD, Beaty TH, Meyer DA, Arnett FC . Evidence that autoimmunity in man is a mendelian dominant trait Am J Hum Genet 1986 39: 584–602

  22. 22

    Nath SK, Kelly JA, Namjou B et al. Evidence for a susceptibility gene, SLEV1, on chromosome 17p13 in families with vitiligo-related systemic lupus erythematosus Am J Hum Genet 2001 69: 1401–1406

  23. 23

    Rao S, Olson JM, Moser KL et al. Linkage analysis of human systemic lupus erythematosus-related traits: a principal component approach Arthritis Rheum 2001 12: 2807–2818

  24. 24

    Olson JM . Relationship estimation by Markov-process models in a sib-pair linkage study Am J Hum Genet 1999 64: 1464–1472

  25. 25

    Klippel JH . Systemic lupus erythematosus: demographics, prognosis, and outcome J Rheumatol 1997 24): (Suppl 48): 67–71

  26. 26

    Johnson RA, Wichern DW . Applied Multivariate Statistical Analysis Prentice Hall: Englewood Cliffs, New Jersey 1988

  27. 27

    Whittemore AS, Tu I-P . Simple, robust linkage tests for affected sibs Am J Hum Genet 1998 62: 1228–1242

  28. 28

    S.A.G.E. Statistical Analysis for Genetic Epidemiology, Release 4.0 A computer program package available from the Department of Epidemiology and Biostatistics, Case Western Reserve University, Cleveland, Ohio 2001

  29. 29

    Holmans P . Asymptotic properties of affected-sib-pair linkage analysis Am J Hum Genet 1995 52: 362–374

  30. 30

    Greenwood CMT, Bull SB . Analysis of affected sib pairs, with covariates—with and without constraints Am J Hum Genet 1999 64: 871–885

Download references

Author information

Correspondence to J M Olson.

Additional information

This work was supported in part by US Public Health Service grants HG01577 from the National Center for Human Genome Research, RR03655 and RR15577 from the National Center for Research Resources, AR42460 and AR45231 from the National Institute of Arthritis and Musculoskeletal and Skin Diseases, and AI24717 and AI31584 from the National Institute of Allergy and Infectious Diseases and the US Department of Veterans Affairs. The 101 pedigrees (Cohorts A, B, and C) were obtained from the Lupus Multiplex Registry and Repository (AR52221) (see http://mrf.ouhsc.edu/lupus). Some of the results in this paper were obtained using the program package SAGE, supported by US Public Health Service Resource Grant RR03655 from the National Center for Research Resources.

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Olson, J., Song, Y., Dudek, D. et al. A genome screen of systemic lupus erythematosus using affected-relative-pair linkage analysis with covariates demonstrates genetic heterogeneity. Genes Immun 3, S5–S12 (2002). https://doi.org/10.1038/sj.gene.6363860

Download citation


  • systemic lupus erythematosus
  • genetic linkage analysis
  • affected relative pairs

Further reading