Identification of epistasis through a partial advanced intercross reveals three arthritis loci within the Cia5 QTL in mice


Identification of genes controlling complex diseases has proven to be difficult; however, animal models may pave the way to determine how low penetrant genes interact to promote disease development. We have dissected the Cia5/Eae3 susceptibility locus on mouse chromosome 3 previously identified to control disease in experimental models of multiple sclerosis and rheumatoid arthritis. Congenic strains showed significant but small effects on severity of both diseases. To improve the penetrance, we have now used a new strategy that defines the genetic interactions. The QTL interacted with another locus on chromosome 15 and a partial advanced intercross breeding of the two congenic strains for eight generations accumulated enough statistical power to identify interactions with several loci on chromosome 15. Thereby, three separate loci within the original QTL could be identified; Cia5 affected the onset of arthritis by an additive interaction with Cia31 on chromosome 15, whereas the Cia21 and Cia22 affected severity during the chronic phase of the disease through an epistatic interaction with Cia32 on chromosome 15. The definition of genetic interactions was a prerequisite to dissect the Cia5 QTL and we suggest the partial advanced intercross strategy to be helpful also for dissecting other QTL controlling complex phenotypes.


Essentially all human diseases are complex in the sense that incidence, severity and outcome are determined by interactions among multiple genes and environmental factors. Even for monogenic Mendelian traits, the genetic background can modify the expression of the phenotype.1 In autoimmune complex diseases like rheumatoid arthritis (RA) and multiple sclerosis (MS), characterized by chronic inflammation in the joints and the central nervous system, respectively, the genetic and environmental interactions are major obstacles to define the genetic control. Despite large efforts worldwide, the search for causative genes for complex phenotypes has not lived up to the expectations, in contrast to the identification of genes responsible for Mendelian traits, which has progressed rapidly. The explanation for this difference lies in genetic and phenotypic heterogeneity, polygenicity, strong environmental influences and epistasis. To increase the statistical power, animal models are used to limit environmental influences and, in addition, genetically defined inbred strains are available. In animals, it is also possible to test and study the biological role of candidate genes. Moreover, the genetic architecture of complex traits can be explored in order to identify general principles and pathways and to gain a broad understanding of the biology of these complex traits, as has been delineated by Wakeland and co-workers.2 When genes and pathways are identified, it will be possible to go back to the even more complex situation in humans.

Gene identification in animal models has also turned out to be more difficult than expected. One reason for this is that several of the identified disease loci have turned out to be a cluster of low penetrant genes.3, 4, 5 In order to dissect genes within a cluster, it is very important to find methods to improve the penetrance of the phenotype. This can be done by redefining the complex phenotype (eg susceptibility to disease) into subphenotypes (eg auto-antibody production)6, 7 and by utilizing information about genetic interactions in the experimental setup.8 It has been speculated that epistasis is ubiquitous in the genetic control and evolution of complex traits.9 An in-depth understanding of the epistatic interactions is essential in the genetic study of complex traits. In the current study, we use the statistical definition of the phenomenon where the epistatic (or interaction) deviation is the deviation of multilocus genotypic values from the additive combination of their single-locus components.10 The term interaction is used to describe when the phenotypic differences among individuals with various genotypes at one locus depend on their genotypes at other loci.

Owing to the nature of complex traits, it is usual for the minimal interval of a QTL—even in large family collections or experimental crosses—to be restricted to no less than 10 to 30 cM in primary genome screens for genetic linkage. This typically corresponds to 10 to 30 megabasepairs (Mb) of DNA, or 100 to 300 genes, which are far too many candidates to begin functional studies of each individual gene. Thus, fine mapping of the linked genetic region is a critical step for all gene-cloning projects. In this study, we use a partial advanced intercross (PAI) breeding strategy to fine map the Cia5/Eae3 locus on mouse chromosome 3, which is linked to collagen-induced arthritis (CIA) and experimental autoimmune encephalomyelitis (EAE), mouse models of RA and MS, respectively (reviewed in Holmdahl11 and Lindqvist et al12). We study arthritis in mice from a PAI between a Cia5/Eae3 congenic strain and a strain congenic for an interacting locus on chromosome 15, both on the B10.RIII background. Analyzing the genetic interactions in the PAI enabled us to define the genetic context where the phenotype has the highest penetrance, and thus, the genetic setting for the identification of new loci and the underlying genes.


Investigation of the original disease phenotypes controlled by Eae3/Cia5

To confirm previously found linkages to chromosome 3,13, 14 a congenic strain (denoted R3) was established by selectively breeding a fragment from the resistant RIIIS/J mouse strain containing the Cia5/Eae3 locus on to the susceptible B10.RIII background (Figure 1). The congenic strain was immunized for CIA and EAE (Figure 2a and b), and for both diseases, the disease-reducing effect of the locus could be confirmed, although with moderate effects. Thus, to be able to proceed with positional cloning, a phenotype with a stronger penetrance was needed. Two strategies were used to improve the gene finding possibilities: one aimed at improving the phenotype and the other at defining a more permissive genetic context.

Figure 1

Physical map of the Cia5/Eae3 region and the congenic strains. The markers are placed according to Mouse Ensemble built 33 ( The dark areas indicate the genetic regions included in the congenic strains. Cia5, Cia21 and Cia22 are placed according to results from the partial advanced intercross and experiments with the congenic strains.

Figure 2

Confirmation of the CIA and EAE phenotypes mapped in the original F2 intercrosses. (a) The mean arthritic score for female R3 (n=10) and B10.RIII (n=11) mice immunized with bCII in CFA. Day of onset in R3 was 44.0±3.6 and in B10.RIII 27.7±8.5, P=0.03. Three out of 10 R3 mice developed disease and six out of 11 B10.RIII mice. (b) Mean score for R3 (n=25) and B10.RIII (n=29) mice immunized with MBP 89–101 in CFA. In all, 16 of the R3 animals developed disease and 28 of the B10.RIII. The day of onset in B10.RIII was 29.6±17.1 and in R3 37.7±13.8, P=0.04. The mean maximal score in B10.RIII was 4.5±1.5 and in R3 3.0±1.9, P=0.006.

Increased penetrance by redefining the arthritis phenotype

Both EAE and CIA were induced using a protocol involving the use of mycobacterium tuberculosum cell wall as an adjuvant. To reduce the environmental complexity introduced by this protocol, we used incomplete Freund's adjuvant (IFA) without mycobacteria in the inducing emulsion. Immunization including IFA normally gives a reproducible but milder CIA.15 With this new protocol, termed CIA-IFA, there was a more pronounced and prolonged reduction of the arthritis severity in the congenic line, R3, as compared with the wild-type (wt) strain (Figure 3a). As no difference in the production of collagen reactive antibodies was found between the congenic and wt animals (data not shown), we concluded that the QTL exerted its main influence subsequent to the formation of arthritogenic antibodies. This led us to investigate the collagen antibody-induced arthritis (CAIA) model.16 Here, the effect was even more striking than with the CIA-IFA protocol; the R3 congenic mice where almost completely protected from arthritis as compared with the wt (Figure 3b).

Figure 3

Improved penetrance by redefining the arthritis phenotype. (a) CIA-IFA. The mean arthritic score for R3 (n=27) and B10.RIII (n=16) mice immunized with bCII in IFA. The incidence in the R3 strain was 30 vs 56% in B10.RIII. The mean maximal score in B10.RIII was 25.3±19.0 and 6.3±8.5 in R3, P=0.006. (b) CAIA. The mean arthritic score for R3 (n=15) and B10.RIII (n=10) mice injected with anti-CII antibodies. All of the B10.RIII mice developed arthritis and one of the R3 mice. Results are from two different experiments.

Reducing the genetic distance (through studies of arthritis models in subcongenic strains)

With these new, more penetrant phenotypes, we started the genetic fine mapping by inducing disease in the subcongenic strains R4 and R5 (Figure 1 and Table 1). When investigating the different subcongenic strains for CIA-IFA and CAIA, it became clear that more than one gene was involved in controlling arthritis in the original R3 congenic strain. The R4 strain showed a later onset of CAIA compared to the wt mice (Figure 4a); however, these mice showed no difference in onset or severity of CIA-IFA (Figure 4b). Lipopolysaccharide (LPS) injection 5 days after the antibody injection abolished the observed difference in CAIA development between the B10.RIII and R4 mice. Nevertheless, LPS did not affect the CAIA inhibition in the R3 mice (data not shown). The R5 mice, on the other hand, were not different from the wt when investigated for CAIA (Figure 4c) but developed CIA-IFA with a reduced severity (Figure 4d). For the R5 fragment, there was a minor influence by gender. In R5 males, there was a difference in the severity early in the disease (Figure 4e). Notably, the effect in males on CIA-IFA in the R5 strain was clearly recessive, since heterozygous mice developed the same disease as the B10.RIII mice. In females, which in general had a later onset of arthritis compared with males, an effect on maximal disease score rather than early severity was seen with the R5 fragment (Figure 4f). The female heterozygous mice, in contrast to the male heterozygous, were similar to the R5 congenic mice, in line with an RIIIS/J dominant inheritance pattern. Taken together, this indicates the involvement of more than one gene controlling arthritis in the R3 fragment: the upper R3 congenic region affects the early phase of arthritis (as illustrated by CIA-IFA in the R5 strain) and the lower region affects the effector phase of the disease (as illustrated by CAIA in the R4 strain). This gives important information on how the genes in the congenic fragment are operating, but once again, the phenotypes were too weak to proceed with a gene cloning project.

Table 1 Summary of arthritis development in the different chromosome 3 congenic strains compared with the parental strain, B10.RIII
Figure 4

Reduction of the genetic distance (through studies of arthritis models in subcongenic strains). (a) The mean arthritic score for R4 and B10.RIII mice injected with CII-specific monoclonal antibodies. The R4 strain had a later onset of CAIA (P=0.03) and six out of nine developed the disease compared to nine out of nine in the wt. (b) The mean arthritic score for R4 (n=30), R4 heterozygous (n=17) and B10.RIII (n=43) mice immunized with bCII in IFA. Eight, six and 11 mice developed disease of the R4, R4hz and B10.RIII, respectively. Results are from two different experiments. (c) The mean arthritic score for R5 (n=7) and B10.RIII (n=18) mice injected with CII-specific, monoclonal antibodies. Four R5 mice and 12 B10.RIII mice developed the disease. (d) The mean arthritic score for R5 (n=35), R5 heterozygous (n=18) and B10.RIII mice (n=100) immunized with bCII in IFA. The incidence was 33% in R5, 31% in heterozygous and 49% in B10.RIII mice. The R5 mice had a later day of disease onset (day 53.8±10.6 in R5 vs day 45.3±16.0 for B10.RIII, P=0.04) and a lower maximal disease score (7.7±13.3 in R5 vs 13.0±15.7 in B10.RIII, P=0.03). (e) The mean arthritic score for male R5 (n=15), R5 heterozygous (n=8) and B10.RIII (n=61) mice immunized with bCII in IFA. Five R5, six R5 heterozygous and 28 B10.RIII mice developed disease. The R5 mice had a lower disease severity than B10.RIII and the heterozygous mice during the early phase of the disease (AUC d18–50, P=0.03). The day of onset was 43.0±19.9 in B10.RIII vs 53.4±6.4 in R5, P=0.06. (f) The mean arthritic score for female R5 (n=20), R5 heterozygous (n=10), and B10.RIII (n=39) mice immunized with bCII in IFA. Six of the R5, none of the R5 heterozygous and 21 of the B10.RIII mice developed disease. The R5 mice had a lower maximal score than B10.RIII mice (5.6±9.7 vs 11.6±12.5, P=0.04).

Increased penetrance by defining the genetic background in a partial advanced intercross

To obtain a higher penetrance of the various subloci within the Cia5/Eae3 congenic fragment, we needed to better define the genetic context in which they operated. One way to do this is to determine which interactions could operate both within the fragment and through other loci in the genome. Data from the previous EAE experiment on an F2 intercross between the B10.RIII and RIIIS/J mouse strains indicated an interaction between Eae3 on chromosome 3 and the Eae2 locus on chromosome 15.13 To define the interactions and isolate the underlying genes into smaller fragments we created a PAI between Eae2 and Cia5/Eae3 congenic strains. This was done by selectively intercrossing mice with recombinations within Cia5/Eae3 and Eae2 for eight generations. All mice that were not used for further breeding were immunized for CIA-IFA, which we selected as our prototype phenotype. We then used the R/qtl software to investigate interactions in the PAI. Permutation tests (n=1000) were subsequently performed to establish the empirical significance thresholds for the interactions. Summarizing the data obtained with mice from the PAI, we could define three arthritis loci within the Cia5/Eae3 locus, denoted Cia5 (as the originally found susceptibility locus), Cia21 and Cia22. All three loci interacted with loci within the Eae2 region: Cia26, Cia30, Cia31 and Cia32 (Table 2).

Table 2 Summary of major interactions between the chromosome 3 and chromosome 15 loci for different arthritis phenotypes in mice from the partial advanced intercross

Cia5 controls onset and early severity

The Cia5 locus affected mainly the early phase of arthritis (days 18–50) by reducing the severity and by mediating a later day of onset in female mice. For both phenotypes, there was a strong interaction with Cia31 (Figure 5a and Table 2), where animals heterozygous for Cia31 showed the most significant differences depending on the genotype of Cia5 (Figure 6a). The effect could also be seen when mice were homozygous for B10.RIII at Cia31, but the difference was dramatically increased when the mice were heterozygous in this region. RIIIS/J alleles reduced the severity in an additive way. With the progression of disease into the late phase (days 53–92), the interaction with Cia31 was lost, but a significant effect of Cia5 was still maintained (Figure 5b). During the late phase, Cia5 interacted with Cia30, Cia32 and Cia26. This indicates that the Cia5 gene modulates arthritis severity during the entire disease course, whereas the chromosome 15 loci have effects on either the early or late phase of disease. Cia5 was linked to the markers in between and including D3Mit187 and D3Mit49, a distance of about 7 million base pairs.

Figure 5

Identification of three novel loci within the Cia5 locus through studies of interactions with loci on chromosome 15 in a partial advanced intercross. (a) Interaction graph for disease onset in female mice. The Cia5 locus interacted with Cia31. Significance levels according to 1000 permutation tests were P(0.99)=7.0, P(0.95)=6.0 and P(0.90)=5.4. (b) Interaction graph for the late phase of CIA (phenotype ‘AUC late’). The Cia5 locus interacted with Cia30, Cia32 and Cia26. Significance levels according to 1000 permutation tests were P(0.99)=7.5, P(0.95)=6.4 and P(0.90)=5.8. (c) Interaction graph for maximal arthritic score in female mice. The Cia21 and Cia22 interacted with Cia32. Significance levels according to 1000 permutation tests were P(0.99)=7.9, P(0.95)=6.2 and P(0.90)=5.7. The values written in the graphs are the LOD joint interaction values according to R/qtl. The distance between markers in cM are according to recombinations in the material that has been used and are not corresponding to the physical distance, which is written in brackets after each marker. For chromosome 15, the physical map is according to Celera ( (because of lack of information from Mouse Ensemble) and for chromosome 3 according to Mouse Ensemble built 33 (

Figure 6

Increased penetrance of Cia5, Cia21 and Cia22 through interactions with loci on chromosome 15. The mean arthritic score for mice from the PAI according to genotype on (a) Cia5 (D3Mit187), (b) Cia21 (D3Mit75) and (c) Cia22 (D3Mit215) when they were either B10.RIII/RIIIS/J heterozygous (left panel) or B10.RIII homozygous (right panel) on the interacting locus on chromosome 15 (Cia31=D15Mit21 and Cia32=D15Mit111).

Cia21 and Cia22 control severity during the chronic phase

The Cia21 locus had the strongest effect on arthritis severity in females during the late phase of disease, as illustrated by a lower area under the curve (AUC) and maximal disease score (Figure 6b). Again, RIIIS/J alleles reduced disease severity. There was a strong interaction with Cia32 (Figure 5b, c and Table 2). With heterozygosity at Cia32, RIIIS/J alleles at Cia21 were dominant protective, which will reduce the amount of breeding needed for positional cloning substantially. Cia21 was linked to the markers D3Mit75 and D3Mit284. This region is not covered in any of the subcongenic strains. However, since the R4, and not the R5 strain, had a later onset of CAIA but was not protected as the R3 strain, it is likely there is a second gene in this region controlling arthritis severity. Thus, Cia21 is most likely located between the markers D3Mit75 and D3Mit284, a distance of about 3 million base pairs.

The Cia22 locus was also linked to arthritis severity and the effect conferred by this locus was, as the Cia21, most pronounced in females during the late/chronic phase of the disease (Figure 5b and c). In contrast to Cia21, the effect was additive (Figure 6c). We believe that Cia22 is located between the markers A3R and D3Mit370. The Cia21 and Cia22 genes are phenotypically operating in the same direction. In addition, they are interacting with the same loci on chromosome 15 and are thus influencing each other in the PAI. The existence of a severity gene in this region is confirmed by that the R4 fragment, containing only the Cia22 locus, had an effect on the CAIA model.

Cia5 and Cia22 have additive effects on disease severity in females

The possibility of intrachromosomal interactions between the loci on chromosome 3 was also explored. The Cia22 locus interacted with Cia5 in the control of maximal disease score and AUC in female mice (LODjnt=8.7, where P(0.95) is 7.5). This is in line with the results in the subcongenic strains R4 and R5, which both had an incomplete penetrance compared with the congenic strain R3. The Cia5 locus was the strongest locus of the two (P=0.0001 vs P=0.02 for Cia22 under the additive model, q1+q2). However, there were more extensive interactions between the loci on chromosome 15, which are explained in detail in Karlsson et al.17

The nature of the interactions

Finally, we investigated the nature of the interactions between the different loci—whether they are additive or have ‘real’ epistatic components (statistically defined). We did so by comparing the interaction calculated using the complete model, which allows for both strict epistasis and additive interaction (q1*q2), with the calculations performed under the assumption of only additive interaction (q1+q2). The strongest true epistasis was found between Cia21 and Cia32, which interacted in an almost completely epistatic way (LODint=6.26, where P(0.95) is 5.4). The Cia5 interaction with Cia22 was more significant with the additive model than the complete model, and had an LODint value of 0.29 where P(0.95) is 6.1. The rest of the interactions had both epistatic and additive components (data not shown).

Nonobese diabetic (NOD) fragment of Cia5 does not influence the development of EAE and CIA

Since there are three diabetes loci within the original Cia5 region, previously identified in a cross between the NOD and C57Bl/6 mouse strains,8 we investigated whether an NOD fragment in this region would have the same effects on arthritis and EAE development as seen for the RIIIS/J fragment. In studies of an intercross between the CIA susceptible B10.Q and the resistant NOD.Q strains, we identified arthritis contributing loci, none of which were on chromosome 3.18 To confirm this finding, we made a B10.Q strain congenic for NOD.Q at Cia5. The congenic strain was tested for both CIA and EAE susceptibility, but showed no influence on the disease development of either CIA or EAE (Figure 7).

Figure 7

Cia5 from NOD does not influence the development of EAE and CIA. (a) The mean EAE score for B10.Q mice, congenic for Cia5 (n=8) and B10.Q wt mice (n=9) immunized with rMOG protein. All mice developed disease. (b) The mean arthritic score for B10.Q mice congenic for Cia5 (n=13) and B10.Q wt mice (n=10) immunized with CII in CFA. Six congenic mice and three wt mice developed disease.


Using a partial advanced intercross strategy, we have demonstrated a method to increase the penetrance of genes forming low penetrance QTL. We have found the original Cia5/Eae3 locus13, 14 to consist of three loci affecting separate phases of the progression of arthritis, which each interacted uniquely with loci on chromosome 15. Cia5 affects disease onset and early severity, whereas Cia21 and Cia22 affect the chronic/late phase of the disease.

There are numerous examples of chromosomal regions containing genes affecting complex phenotypes, but only a few where the actual gene has been identified.19 Therefore, strategies for improving the gene-finding possibilities are greatly needed. Since we know genes do not work as isolated units but rather in complex networks, one such strategy is to use interacting loci to define in what genetic context a particular susceptibility locus has the largest penetrance. We demonstrate that by altering the genotype of an interacting locus, the penetrance could be increased dramatically. Importantly, the penetrance of many Mendelian traits depends on the genetic background used and several so-called modifier loci have been identified.1

A PAI is an alternative method to the tedious work required with breeding and testing of subcongenic strains when the phenotype is not dominantly inherited. In order to fine map the locus, production of homozygous animals is needed, that is, both backcrossing and intercrossing for each fragment. Producing a large cross is a more efficient strategy to pinpoint the linked region/s and will, at the same time produce animals with valuable recombinations that can be used for further breeding. To find interacting loci, we are dependent on good F2 experiments where efforts of reliable phenotyping and genotyping have been made. In fact, by not studying interactions in an F2 cross there is a risk that susceptibility loci might be missed.20, 21

In addition, careful statistical analysis is important. We chose to use the R/qtl software22 to analyze genetic interactions. By using R/qtl on the original F2 intercross data,13 we were able to verify the finding that the main genetic interaction occurred between Eae3 on chromosome 3 and Eae2 on chromosome 15 (data not shown). Owing to the way the PAI was created, there was however a risk in utilizing a multipoint analysis. In the methods applied in R/qtl, genetic recombinations are assumed to be independent, which was not the case in the PAI since some recombination events happened in early generations and were selected for further breeding in all later generations. This is reflected in the genetic map created from the PAI (Figure 5) in which large deviations from the physical map can be seen. Therefore, we also analyzed the data with single-point analysis (ANOVA) (data not shown). No differences were detected between the two analysis strategies, showing that the regression and multiple imputation methods applied by R/qtl were robust enough against the non-Mendelian genotype distribution of the PAI.

We also show that the definition of the phenotype is crucial. By dividing the arthritis into well-defined subphenotypes, we observed that the early and late/chronic phases of the disease were differently regulated. It is evident that without the defined subphenotypes, we would not have detected all of the loci. This is also true concerning the disease induction where we excluded mycobacteria in the adjuvant to improve the disease induction protocol. The main reasons we needed to change the protocol were likely interactions between the Cia5 region and other genes and environment. In the congenic strain, the Cia5 locus was present in a different genetic context than in the original F2 intercross, namely all other chromosomes were of B10.RIII origin, which is why the effect of Cia5 was no longer seen.

The region on mouse chromosome 3 has earlier been found to interact with other loci. In a study by Mahler and Leiter,23 this region was associated with colitis, where interactions with multiple other loci were found. In addition, in the original CIA F2 intercross, the Cia5 locus was found to interact with Cia10 on chromosome 13.14 The region is also associated with autoimmune diabetes in the NOD strain. As for arthritis, there are three separate diabetes loci mapped in the region: Idd10, Idd17 and Idd18.4 Interestingly, they correspond physically to the Cia5, Cia21 and Cia22 loci and are also dependent on an interacting locus.24 It is tempting to speculate that the same genes control the diabetes effect. However, experimental data does not support this possi-bility. There was no CIA linkage to chromosome 3 in a cross between B10.Q and NOD.Q.18 Furthermore, we could not observe an effect on disease development for EAE and CIA when the NOD fragment of chromosome 3 was introduced on to the B10.Q background. This was also true when the Idd10/17/18 congenic strain was immunized for EAE; no differences in disease course between the congene and the wt were seen.25 The possibility that Idd10, Idd17 and Idd18 represents the same genes as Cia5, Cia21 and Cia22 still remains, but the effects are dependent on genes either in the MHC region or on chromosome 10, which are the only (known) regions that differs between the B10.RIII and B10.Q strains.26

Although mapping of susceptibility loci for complex diseases implies substantial work, we think it is important to identify naturally occurring allelic variants causing disease. They are likely to represent evolutionary selected alleles causing common diseases like RA in the ‘wrong’ genetic context and environment. The study of interactions between susceptibility loci is not only important during mapping but also once the genes are identified. While the use of knockout models and mutagenesis maybe faster and useful methods to identify members of a pathway, they do not allow the identification of naturally occurring alleles and studies of complex interactions. Additionally, the use of knockout models is a source of confusion because of 129 derived linked fragments, making it impossible to study only the effect of the knocked-out gene. Furthermore, methods for studying complex traits have improved markedly with the recent development of an extensive array of genome resources and technologies. This is reflected in the recent discoveries of genes affecting quantitative traits.27

In summary, we have demonstrated that by using a partial advanced intercross, we obtained enough statistical power to identify three loci linked to arthritis in the originally defined Cia5 region: Cia5, Cia21 and Cia22. Since loci containing clusters of interacting genes are likely to be the main obstacle in understanding the genetic influence on complex disease, we suggest that the direct experimental testing of advanced intercrossed congenic strains is a valuable tool for identifying the underlying genes.

Materials and methods

Production of B10.RIII chromosome 3 congenic strains of mice

Mice were bred and kept in the animal house of Medical Inflammation Research Unit, Lund University. The mice were kept in climate-controlled environment with 12-h light/dark cycles, fed with standard rodent chow and water ad libitum (as defined in Jan Klein, Tübingen, Germany, originally provided the B10.RIII mice and RIIIS/J mice were obtained from Jackson Laboratories (Bar Harbor, ME, USA). To be able to compare strains, all congenic strains and gene knockout mice produced at the Medical inflammation Research Unit are established on the B10 genetic background. To develop B10.RIII congenic mice that contain RIIIS/J genes on chromosome 3, mice from an (B10.RIII × RIIIS/J)F2 intercross heterozygous for the markers D3Mit72 and D3Mit101 were backcrossed to the susceptible strain, B10.RIII, for 10 generations. The mice were then typed with additional markers and mice heterozygous for all markers in between and including D3Mit187 and D3Mit370 were intercrossed to produce the congenic line R3 (Figure 1). The subcongenic lines R4 and R5 were derived from backcrossing and subsequent intercrossing of the R3 congenic mice (Figure 1).

Production of B10.Q chromosome 3 congenic strain

The B10.Q mice congenic for an NOD.Q fragment on chromosome 3 were made according to the speed congene protocol.28 Briefly, mice from an NOD.Q × B10.Q N2 backcross18 heterozygous for selected markers on chromosome 3 were backcrossed to B10.Q until other chromosomes were B10.Q (six generations). Mice heterozygous for markers in between and including D3Mit184 and D3Mit18 were intercrossed to produce a congenic line.

Production of Cia5 and Eae2—partial advanced intercross mice

The R3 congenic mice were crossed with Eae2 congenic mice.29 F2 animals were genotyped and selected mice were used for breeding. For each generation, this was repeated and mice not used for breeding were immunized with bovine type II collagen (CII) in IFA (see below). Data from 676 PAI mice (generation F2–F8), 315 single and subcongenic mice and 60 B10.RIII wt mice in eight CIA experiments have been collected. Single congenic and wt were used to control for disease variations in all eight experiments.

Induction and evaluation of CIA

Bovine type II collagen (CII) was prepared from calf nasal cartilage by pepsin digestion and subsequently purified as described previously.11 CIA was induced in mice 8–12 weeks old (experiments with single congenic mice) or 10–16 weeks old (PAI experiments) by intradermal (i.d.) immunization at the base of the tail with 100 μg bovine CII emulsified in CFA (complete Freund's adjuvant) or IFA (incomplete Freund's adjuvant) (Difco, Detroit, MI, USA) and subsequently boosted with 50 μg bovine CII emulsified in IFA 35 days later. The clinical arthritis severity was quantified as previously described using a 15 score system per limb.11 Briefly, each inflamed toe and knuckle was given a score of 1 (total 10 per limb), and an inflamed wrist or ankle was given a score of 5; combined score of 15 per limb and a maximum score of 60 per mouse. The area under the curve (AUC) values for the disease curves was calculated as the sum of scoring for each mouse. The AUC was also divided into separate phases of the disease, early (days 18–50) and late (days 53–92). The day of disease onset, maximal disease score and incidence were also evaluated.

Induction and evaluation of CAIA

CAIA was induced in 3- to 5-month-old male mice by an intravenous injection with 9 mg of a cocktail of arthritogenic, CII specific, monoclonal antibodies (M2139 and CIIC1) recognizing J1 and C1I epitopes of CII as described earlier.16 The development of arthritis was quantified daily according to the same scoring system as for CIA.

Induction and evaluation of EAE

The myelin basic protein (MBP) 89–101 peptide was synthesized on an Applied Biosystems peptide synthesizer model ABI430A using the FastMoc program. Mice were immunized i.d. at the base of the tail with 50 μg MBP 89–101 emulsified in IFA (Difco, Detroit, MI, USA) containing 100 μg Mycobacterium tuberculosis (H37Ra); 400 ng pertussis toxin (Sigma) was injected intraperitoneal (i.p.) immediately after immunization and 48 h later. Dr Robert Harris, Karolinska Hospital, generously provided the rat MOG recombinant protein, encompassing amino acids 1–120. Mice were immunized subcutaneously at the flank with 25 μg MOG emulsified in IFA containing 100 μg M. tuberculosis (H37Ra). Mice were examined for clinical signs of EAE based on an eight score scale as described previously.13


Genomic DNA was isolated from 1 mm tail or toe. The samples were dissolved in 500 μl 50 mM NaOH, in 95°C for 1–2 h, and then 100 μl 1 mM Tris buffer (pH 8) was added. The samples were centrifuged and 1 μl of the supernatant was used for PCR. Microsatellite variants were resolved on a MegaBACE DNA analysis system 1000 (Amersham Pharmacia Biotech, UK). SNP variants were sequenced with a PSQ 96 (Pyrosequencing, Sweden) according to the manufacturer's protocol.

Statistical analysis

Results are expressed as number (percentage) or mean±s.d. CIA and EAE scoring values were compared using Mann–Whitney U-test. Comparisons of group incidence were analyzed by Fischer's exact test. A P-value <0.05 was considered significant. The interaction data were calculated using R and the R/qtl software.22, 30 For the two-locus interaction analyzed in R/qtl, Haley–Knott regression, which utilizes regression of phenotypes on multipoint genotype probabilities and scanqtl (imputation model), was used. The calculated LODjoint score values compare a full model, if including covariates (y=μ+βq1+βq2+βq1 × q2+Aγ+Zδq1+Zδq2+Zδq1 × q2+ɛ) to a null model (y=μ+Aγ+ɛ). The epistasis, LODint score, compares the full model to an additive model (y=μ+βq1+βq2+Aγ+Zδq1+Zδq2 +ɛ). The q1 and q2 are unknown QTL genotypes at two different locations, A a matrix of covariates and Z a matrix of QTL interacting covariates. Permutation tests (n=1000) were carried out subsequently to establish empirical significance thresholds for the interactions. A threshold equal or above the 95th percentile was considered significant. Figures, illustrating the interactions, were created using the image function in scanqtl.


  1. 1

    Nadeau JH . Modifier genes in mice and humans. Nat Rev Genet 2001; 2: 165–174.

    CAS  Article  Google Scholar 

  2. 2

    Nguyen C, Limaye N, Wakeland EK . Susceptibility genes in the pathogenesis of murine lupus. Arthritis Res 2002; 4 (Suppl 3): S255–S263.

    Article  Google Scholar 

  3. 3

    Morel L, Blenman KR, Croker BP, Wakeland EK . The major murine systemic lupus erythematosus susceptibility locus, Sle1, is a cluster of functionally related genes. Proc Natl Acad Sci USA 2001; 98: 1787–1792.

    CAS  Article  Google Scholar 

  4. 4

    Podolin P . Localization of two insulin-dependent diabetes (Idd) genes to the Idd10 region on mouse chromosome 3. Mamm Genome 1998 1998; 9: 283–286.

    CAS  Article  Google Scholar 

  5. 5

    Serreze DV, Bridgett M, Chapman HD et al. Subcongenic analysis of the Idd13 locus in NOD/Lt mice: evidence for several susceptibility genes including a possible diabetogenic role for beta 2-microglobulin. J Immunol 1998; 160: 1472–1478.

    CAS  Google Scholar 

  6. 6

    Vingsbo-Lundberg C, Nordquist N, Olofsson P et al. Genetic control of arthritis onset, severity and chronicity in a model for rheumatoid arthritis in rats. Nat Genet 1998; 20: 401–404.

    CAS  Article  Google Scholar 

  7. 7

    Morel L, Croker BP, Blenman KR et al. Genetic reconstitution of systemic lupus erythematosus immunopathology with polycongenic murine strains. Proc Natl Acad Sci USA 2000; 97: 6670–6675.

    CAS  Article  Google Scholar 

  8. 8

    Lyons PA, Armitage N, Lord CJ et al. Mapping by genetic interaction: high-resolution congenic mapping of the type 1 diabetes loci Idd10 and Idd18 in the NOD mouse. Diabetes 2001; 50: 2633–2637.

    CAS  Article  Google Scholar 

  9. 9

    Cheverud JM, Routman EJ . Epistasis and its contribution to genetic variance components. Genetics 1995; 139: 1455–1461.

    CAS  PubMed  PubMed Central  Google Scholar 

  10. 10

    Falconer DS, Mackay TFC . Introduction to Quantitative Genetics 1996.

  11. 11

    Holmdahl R . Genetics of susceptibility to chronic experimental encephalomyelitis and arthritis. Curr Opin Immunol 1998; 10: 710–717.

    CAS  Article  Google Scholar 

  12. 12

    Lindqvist AK, Bockermann R, Johansson AC et al. Mouse models for rheumatoid arthritis. Trends Genet 2002; 18: S7–S13.

    CAS  Article  Google Scholar 

  13. 13

    Sundvall M, Jirholt J, Yang HT et al. Identification of murine loci associated with susceptibility to chronic experimental autoimmune encephalomyelitis. Nat Genet 1995; 10: 313–317.

    CAS  Article  Google Scholar 

  14. 14

    Jirholt J, Cook A, Emahazion T et al. Genetic linkage analysis of collagen-induced arthritis in the mouse. Eur J Immunol 1998; 28: 3321–3328.

    CAS  Article  Google Scholar 

  15. 15

    Svensson L, Nandakumar KS, Johansson A, Jansson L, Holmdahl R . IL-4-deficient mice develop less acute but more chronic relapsing collagen-induced arthritis. Eur J Immunol 2002; 32: 2944–2953.

    CAS  Article  Google Scholar 

  16. 16

    Nandakumar KS, Svensson L, Holmdahl R . Collagen type II-specific monoclonal antibody-induced arthritis in mice: description of the disease and the influence of age, sex, and genes. Am J Pathol 2003; 163: 1827–1837.

    CAS  Article  Google Scholar 

  17. 17

    Karlsson J, Johannesson M, Lindvall T et al. Genetic interactions in Eae2 control collagen induced arthritis and the CD4+/CD8+ T cell ratio. J Immunol 2005; 174: 533–541.

    CAS  Article  Google Scholar 

  18. 18

    Johansson AC, Sundler M, Kjellen P et al. Genetic control of collagen-induced arthritis in a cross with NOD and C57BL/10 mice is dependent on gene regions encoding complement factor 5 and FcgammaRIIb and is not associated with loci controlling diabetes. Eur J Immunol 2001; 31: 1847–1856.

    CAS  Article  Google Scholar 

  19. 19

    Glazier AM, Nadeau JH, Aitman TJ . Finding genes that underlie complex traits. Science 2002; 298: 2345–2349.

    CAS  Article  Google Scholar 

  20. 20

    Yi N, Xu S, Allison DB . Bayesian model choice and search strategies for mapping interacting quantitative trait loci. Genetics 2003; 165: 867–883.

    CAS  PubMed  PubMed Central  Google Scholar 

  21. 21

    Peripato AC, De Brito RA, Matioli SR et al. Epistasis affecting litter size in mice. J Evol Biol 2004; 17: 593–602.

    CAS  Article  Google Scholar 

  22. 22

    R development core team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing: Vienna, Austria, 2004.

  23. 23

    Mahler M, Leiter EH . Genetic and environmental context determines the course of colitis developing in IL-10-deficient mice. Inflamm Bowel Dis 2002; 8: 347–355.

    Article  Google Scholar 

  24. 24

    Podolin PL, Denny P, Lord CJ et al. Congenic mapping of the insulin-dependent diabetes (Idd) gene, Idd10, localizes two genes mediating the Idd10 effect and eliminates the candidate Fcgr1. J Immunol 1997; 159: 1835–1843.

    CAS  Google Scholar 

  25. 25

    Encinas JA, Wicker LS, Peterson LB et al. QTL influencing autoimmune diabetes and encephalomyelitis map to a 0.15-cM region containing Il2 [letter]. Nat Genet 1999; 21: 158–160.

    CAS  Article  Google Scholar 

  26. 26

    Dong P, Hood L, McIndoe RA . Detection of a large RIII-derived chromosomal segment on chromosome 10 in the H-2 congenic strain B10.RIII(71NS)/Sn. Genomics 1996; 31: 266–269.

    CAS  Article  Google Scholar 

  27. 27

    Korstanje R, Paigen B . From QTL to gene: the harvest begins. Nat Genet 2002; 31: 235–236.

    CAS  Article  Google Scholar 

  28. 28

    Markel P, Shu P, Ebeling C et al. Theoretical and empirical issues for marker-assisted breeding of congenic mouse strains. Nat Genet 1997; 17: 280–284.

    CAS  Article  Google Scholar 

  29. 29

    Jirholt J, Lindqvist AK, Karlsson J, Andersson A, Holmdahl R . Identification of susceptibility genes for experimental autoimmune encephalomyelitis that overcome the effect of protective alleles at the eae2 locus. Int Immunol 2002; 14: 79–85.

    CAS  Article  Google Scholar 

  30. 30

    Broman KW, Wu H, Sen S, Churchill GA . R/qtl: QTL mapping in experimental crosses. Bioinformatics 2003; 19: 889–890.

    CAS  Article  Google Scholar 

Download references


We thank Isabell Bohlin and Carlos Palestro for help with the animal care. This work was supported by grants from the Swedish Medical Research Council, the Swedish Foundation for Strategic Research, the Swedish Association against Rheumatism, the Crafoord, Lundberg, the Kock and Österlund Foundations and EU FP5 (QLG1-CT-2001-01407 ‘EUROME’). Dr Patrik Wernhoff was supported by the EU Grant HPMD-2000-00047.

Author information



Corresponding author

Correspondence to M Johannesson.

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Johannesson, M., Karlsson, J., Wernhoff, P. et al. Identification of epistasis through a partial advanced intercross reveals three arthritis loci within the Cia5 QTL in mice. Genes Immun 6, 175–185 (2005).

Download citation


  • partial advanced intercross
  • collagen-induced arthritis
  • quantitative trait locus

Further reading