Quantitative trait loci influencing several phenotypes were assessed using a genetically heterogeneous mouse population. The 145 individuals were produced by a cross between (BALB/cJ × C57BL/6J)F1 females and (C3H/HeJ × DBA/2J)F1 males. The population is genetically equivalent to full siblings derived from heterozygous parents, with known linkage phase. Each individual in the population represents a unique combination of alleles from the inbred grandparents. Quantitative phenotypes for eight T cell measures were obtained at 8 and 18 mo of age. Single-marker locus, repeated measures analysis of variance identified nine marker-phenotype associations with an experimentwise significance level of P < 0.05. Six of the eight quantitative phenotypes could be associated with at least one locus having experiment-wide significance. Composite interval, repeated measures analysis of variance identified 13 chromosomal regions with comparisonwise (nominal) significance associations of P < 0.001. The heterozygous-parent cross provides a reproducible, general method for identification of loci associated with quantitative trait phenotypes or repeated phenotypic measures.

ANALYSIS of the genetic regulation of quantitative phenotypes under complex multilocus control has typically employed a system in which progenitor animals, often from inbred stocks that differ in the trait in question, are used to produce a segregating F2 or backcross generation. Associations between marker loci and the levels of the trait are then used to count and to map the effector loci. This strategy, though productive in some situations, will fail to identify loci of potential interest that are not polymorphic between the two chosen progenitors or that require a specific context of interacting genes (Lander and Schork 1994; Frankel 1995).

The genetic analysis of mammalian aging poses particular challenges. The traits of interest, such as the incidence of late life diseases, the pace of age-related changes in a variety of organ systems, and life span, cannot be measured inexpensively even in short-lived species (such as mice), because the tests must be preceded by months or years of costly husbandry. In addition, differences among the inbred strains in mean life span or in age-sensitive traits may well represent differential susceptibility to specific early-life diseases, rather than more general alterations in the aging process. Although many single gene mutations are known to shorten mean life span in mice or in humans, each of these typically operates by interference with one or a small number of physiological systems and may thus provide little insight into mechanisms that might coordinately regulate age-related changes in many cell and tissue types. Mutant-hunting approaches, which can lead to the discovery of new loci with developmental or morphological effects, are generally not feasible for analysis of phenotypes that require years to develop.

We have attempted to circumvent some of these difficulties by a strategy in which each mouse in an aging population is tested for several age-sensitive traits. To increase the level of potential allelic variability, we have generated our test population by a heterozygous-parent breeding method, so that each animal can be considered a full sibling derived from parents with known linkage phase. This approach provides a reproducible heterogeneous population in which a range of late-life illnesses develop and influence longevity (Milleret al. 1994; Chrispet al. 1996). The breeding system is amenable to rapid, automated, DNA marker-based genetic analysis using simple sequence length polymorphism (SSLP) markers already known to discriminate among the grandparental strains. Within the population, we have tested each mouse at two ages for the proportions of eight T cell measures, including five that have been shown to vary with age in previous cross-sectional and longitudinal studies (Miller 1996, 1997). The analytical approach, based on repeated measures analysis of variance procedures and composite interval mapping (Zeng 1994; Xu 1996), can identify loci that influence the age-independent levels of T cell subsets as well as loci whose influence on the trait varies with animal age. Although the heterozygous-parent cross method does not create a formal homozygous class of the sort obtained in F2 crosses and backcrosses, interactions between maternal and paternal alleles can, in practice, reveal the effects of recessive alleles. The candidate quantitative trait loci (QTL) should help to guide further confirmatory analysis using a larger population of mice and a higher-density genetic map and provide points of departure for genetic analysis of aging, age-associated phenotypic changes, and T cell subset regulation.


Mouse handling: (BALB/cJ × C57BL/6J)F1 females and (C3H/HeJ × DBA/2J)F1 males were purchased from the Jackson Laboratories and mated to produce the study population. The animals for this study were derived from 12 birth cages, each containing a single male and two females. Offspring were weaned at 4 wk and housed segregated by sex in standard mouse cages, each initially containing four or five mice. In most cases, all members of a litter entered the population at weaning. Throughout the study, all mice were housed in a single suite of specific-pathogen-free (SPF) rooms under identical environmental conditions (12:12 hr light:dark cycle, 23°) and given ad libitum access to water and laboratory mouse chow; quarterly tests of sentinel mice showed that the facility remained SPF throughout the period of the study.

Immunophenotyping assays have been described in detail previously (Miller et al. 1994, 1996).

Genotyping assays: Polymorphic marker loci were selected for PCR-based genotyping using data provided by the Mouse SSLP Database, Whitehead/MIT Center for Genome Research (Cambridge, MA; http://www.genome.wi.mit.edu/cgi-bin/mouse/) or the Mouse Genome Database 3.1, Mouse Genome Informatics, the Jackson Laboratory (Bar Harbor, ME; http://www.informatics.jax.org/). Primer pairs were either purchased from Research Genetics or synthesized locally (University of Michigan Molecular Biology Core Facility). Genomic DNA was isolated from each adult animal by removal of ≈1 cm of tail and preparation using a standard phenol-extraction method (Sambrooket al. 1989). Final DNA preparations were tested for concentration, ability to sustain PCR amplification under standard conditions, and electrophoretic size distribution.

Amplification of genomic DNA was performed using 20 ng mouse DNA in 10 mm Tris-HCl pH 8.3, 50 mm KCl, 1.5 mm MgCl2, 200 μm each dNTP, 100 nm each primer, and 0.2 units Taq DNA polymerase. After a 3-min denaturation at 94°, 35 cycles of denaturation at 94° (30 sec), annealing at 55° (30 sec), and extension at 72° (30 sec) were performed, followed by a final extension at 72° (5 min). Reactions were held in 96-well, thin-wall polycarbonate plates (Costar, Cambridge, MA), and temperature controlled using a 96-well UNO thermocycler (Biometra, Tampa, FL). Amplification products were mixed with formamide to 50%, heat denatured, and electrophoresed on 0.5-mm-thick, 7 m urea, 6% polyacrylamide gels. After electrophoresis, bands were visualized by silver staining (Silver Sequence; Promega, Madison, WI). Briefly, gels were fixed in 10% acetic acid, rinsed in water, and stained in 5.9 mm AgNO3, 0.056% formaldehyde. After a water rinse, gels were developed at 4° in 0.28 m Na2CO3, 0.056% formaldehyde. Development was terminated with 10% acetic acid, and gels were rinsed in water and dried onto cellophane. Gels were scored independently by two people, each of whom entered scores by hand into an Excel spreadsheet (Microsoft). Scoring conflicts were automatically flagged for rescoring. Rescoring was performed independently by both original scorers. Verification of independent segregation was performed on the final genotype scores using χ2-analysis of expected and observed genotypes. Chromosome lengths were defined as the distance from the most centromeric to the most telomeric marker locus in the Whitehead/MIT database. Chromosomal localization and order of markers were calculated using both an Excel macro and the MapMaker program package (Whitehead Institute, MIT).

An ALFexpress automated sequence analyzer (Pharmacia, Piscataway, NJ) was used for genotyping several markers near the conclusion of the study. Amplification reactions were performed as above, except that one primer was fluorescently tagged with the dye Cy-5. Amplification reactions were mixed with formamide (50% final concentration) and internal DNA size standards, heat denatured, and loaded onto 0.3-mm-thick, 6 m urea, 6% polyacrylamide gels (LongRanger; FMC, Rockland, ME). Migration information was captured using the ALFexpress AM v3.02 software. Alleles were scored independently by two people using Fragment Manager software (v1.20; Pharmacia) and the resulting data were exported to Excel spreadsheets for record keeping and analysis.

Statistical methods for marker-phenotype analysis: For the initial analysis, a single-marker model was used, with no inclusion of information from flanking markers or markers in other intervals. This is a simple linear regression method for associating markers with QTL (Solleret al. 1976; Haley and Knott 1992; Knott and Haley 1992). Because repeated measurements were taken on each animal and are correlated, a mixed linear model was used to identify associations between the single markers and longitudinal traits. The Mixed procedure in SAS (version 6.11; SAS Institute) was used to perform mixed-model analyses. This parametric procedure uses a ridge-stabilized Newton-Raphson algorithm to optimize a maximum-likelihood function. To account for the correlated repeated time data an unstructured residual covariance matrix was specified in addition to the fixed effects. The linear model for single timepoint analysis (8 mo or 18 mo) specified yk=μ+C+S+αim+αjp+δij+εkfori,j=1,2, where yk is the phenotype for the kth individual, μ is the overall mean, C is the birth cage effect, S is the effect of gender and is used only for traits influenced by gender, αim and αjp are the additive effects of the maternal alleles Mim and the paternal alleles Mip, respectively, of marker M, δij is the maternal-by-paternal allelic interaction (dominance effect), and ϵk is the error term with N(0, σε2). C is a random effect; all other specified effects are fixed. For loci that are not fully informative, the second additive effect and the dominance effect are deleted from the model. Dummy variables were created for the maternal and paternal indicators.

For the repeated measures analysis model (combined 8- and 18-mo values), additional terms for timepoint (T) and, where a sex effect is observed, gender-by-timepoint (ST) interactions are included: yk=μ+C+S+T+ST+αim+αjp+δij+εkfori,j=1,2.

Critical values for experimentwise significance were assessed using permutation testing (Churchill and Doerge 1994). For permutation testing, 1000 shuffled data analyses were performed using the same regression models and experimental data. The program recorded the highest test statistic obtained in each analysis, and these were ordered at the end of the procedure. The 90, 95, and 99th percentile values were reported as the experimentwise significance level at α = 0.1, 0.05, and 0.01, respectively.

Composite interval analysis was performed using Proc Mixed adapted to the heterozygous parent cross-breeding structure. The model was derived from the “four-way cross” QTL analysis method developed by Xu (1996) and is an extension of the established F2 composite interval mapping method of Zeng (1994). The linear model and conditional probability frequencies are described in detail in Xu (1996, Table 1 and Equation 7) and have been implemented with no significant modifications when both flanking markers were fully informative. To account for the correlated repeated time data an unstructured covariance matrix was specified in addition to the fixed effects. Again, the birth cage of each individual was included as a random effect. When at least one flanking marker was missing maternal (or paternal) information, the corresponding paternal (or maternal) and dominance probabilities were not calculated, and these terms were deleted from the model.

Post hoc comparison of means was performed using the least significant difference or planned comparisons option in Statistica (release 5.0b; StatSoft, Tulsa, OK). In all analyses, with the exception of a logarithmic transformation of CD3 cell levels, primary trait data were not transformed.


A defined heterogeneous mouse population: The genetically heterogeneous mouse population, designated UM-HET3, was derived from the progenitor inbred strains BALB/cJ (C), C57BL/6J (B6), C3H/HeJ (C3), and DBA/2J (D2). Female CB6F1 mice were bred with C3D2F1 males to generate a population of animals. For an autosomal locus with detectable polymorphisms among the progenitor genomes, four possible genotypes are obtained: C/C3, C/D2, B6/C3, and B6/D2. Alleles from the four progenitor chromosomes are expected to be represented equally, and this was confirmed by visual observation of the phenotypes associated with coat color loci brown and agouti (not shown).

The starting population for this analysis contained 174 animals (88 male, 86 female) born between 9/14/93 and 2/16/94, and housed in same-sex cages. Six cages containing male mice (22 animals) were eliminated from the study because some of the animals had sustained injuries from fighting. Four males and two females were excluded from analysis because they died before the first phenotyping analysis (<8 mo), and one male was killed in error during the study. The final analyzed population consisted of 145 animals (84 females and 61 males) that were kept in a single room under controlled SPF conditions. All animals entered into the UM-HET3 population were genotyped and were assayed for T cell subset phenotypes at 8 and 18 mo of age. The animals were maintained for their natural life span, or killed when judged by an experienced technician to be moribund. At necropsy, the animals exhibited a range of pathologies and presumptive causes of death, as expected for a genetically heterogeneous population (Chrispet al. 1996).

View this table:

T cell subset phenotypes

Longitudinal phenotype analysis: Animals were tested for eight T cell phenotypes (Table 1) at 8 and 18 mo of age, using flow cytometry on a sample of tail venous blood (Milleret al. 1996). Eighteen male and 8 female animals entered into the study at 8 mo did not survive to the 18-mo timepoint and consequently provide no longitudinal information. Overall, a small number (<5%) of the immunotyping assays were deemed unreliable on technical grounds; these were not included in the analysis. The CD4V T cell subset assays at the 8-mo timepoint were successful on less than half of the animals and were not included in any analysis.

A previous study of T cell subsets in UM-HET3 mice (Miller 1997) has shown that these mice, like most inbred and F1 hybrid mice, exhibit an increase with age in proportions of CD8M, CD4M, CD8P, and CD4P cells, and a decline in CD4V cells. CD4 cells also decrease with age in the peripheral blood of UM-HET3 mice, as well as in some, but not all, studies of inbred and F1 hybrid mice. Many of these T cell subsets exhibit correlated variation. Individual UM-HET3 mice with relatively high levels of CD4M cells, for example, also tend to have high levels of CD8M cells and low levels of CD4V cells (Miller 1997). It is not known whether these correlations have any genetic basis. Gender was found to have a significant effect on the CD3, CD4, CD4P, and CD8P phenotypes.

Genotype analysis: Dietrich et al. (1994) have examined the four inbred progenitor strains at ∼6000 SSLP loci, and maintain the associated allele size information in a common database (Whitehead/MIT Center for Genome Research). In this study, 83 SSLP marker loci were used for genome-wide genotyping of the full UM-HET3 population. Sixty-eight of the marker loci are informative for all four progenitor alleles, 13 are informative for maternal or paternal segregating alleles only, and 2 are X-linked (Table 2). At the time of statistical analysis, unambiguous genotype data were lacking for ∼6% of the animal-locus data points.

Marker map order within the UM-HET3 population was in agreement with the Mouse Genome Database (The Jackson Laboratory), with two exceptions. First, the marker D8Mit112 did not map to chromosome 8. On typing with the Jackson Laboratory Interspecific Backcross panel (Roweet al. 1994), it was placed on chromosome 9 at ≈17 cM. Second, the markers D12Mit63 and D12Mit34 show linkage to each other but not to the other markers on chromosome 12. Interspecific backcross mapping using the Jackson Laboratory backcross panel placed D12Mit34 on chromosome 12 at ≈19 cM and D12Mit105 at ≈7 cM. As a consequence, the UM-HET3 map order inconsistencies on chromosome 12 remain unresolved. Segregation analysis of the UM-HET3 population indicated a single significant (P < 0.01) departure from expected random segregation. The locus D11Mit2 is favored for the C3 paternal allele by ∼2:1 in male UM-HET3 animals only.

At the current genotyping density, ∼95% of the autosomal genome lies within 30 cM of a marker. Six autosomes have only two four-way informative loci; also, the X chromosome is significantly underrepresented relative to its genetic length.

Associations between marker loci and traits: The goal of the genetic analysis was to identify QTL associated with either the level of a particular trait, or with the degree that the trait changed over time. The phenotype and genotype data were characterized by two statistical methods.

The initial analysis sought associations between alleles at individual marker loci and each specific T cell subset using a general, parametric statistical model (Proc Mixed; SAS version 6.11) to perform repeated measures analysis of variance. The Proc Mixed analytical model examined each of the eight phenotypes as a dependent variable, and set four independent variables: maternal-derived genotype, paternal-derived genotype, age at phenotype measurement, and, for traits CD3, CD8, CD4P, and CD8P, sex of the individual. This approach can reveal the dependence of the T cell subset on interactions among the independent variables, such as an age-dependent effect of a locus on phenotype, or an interaction between maternal and paternal alleles. The Proc Mixed analysis generates a locus-specific F-statistic for each of four hypotheses: (1) overall segregation of a QTL using all genotypes, (2) segregation of a QTL using maternal-derived genotypes, (3) segregation of a QTL using paternal-derived genotypes, and (4) segregation of a QTL using maternal-by-paternal allelic interaction. In each hypothesis, the model included adjustment for birth cage effects and, where applicable, gender. A permutation test was performed on 1000-repetition shuffling of the phenotype data to obtain experimentwise empirical threshold significance values, as suggested by Churchill and Doerge (1994). The associated P values are given in Table 3. Single-locus Proc Mixed analysis identified nine loci associated with T cell subsets with experimentwise P < 0.05.

View this table:

SSLP genetic marker loci

The second analysis also used a repeated measures analysis of variance method, but included the influence of flanking markers and QTL effects in other chromosomal intervals (i.e., composite interval mapping). The “four-way cross” parameters defined by Xu (1996) were incorporated into the statistical model. In the four-way cross design, the overall phenotype distribution is a mixture of four component distributions. Each component distribution is defined by one of the four possible genotypic pairs at a given QTL. The regression model of Xu (1996) approximates the mixture of these distributions, where the portion of the variance not explained by marker uncertainty is combined into the residual error. The hypotheses tested under this repeated measures model were the same as for the single locus method: (1) overall segregation of a QTL using all genotypes, (2) segregation of a QTL using maternal-derived genotypes, (3) segregation of a QTL using paternal-derived genotypes, and (4) segregation of a QTL using maternal-by-paternal allelic interaction. Again, the mixed model can reveal the dependence of the T cell subset on interactions among the independent variables, such as an age-dependent effect of a chromosomal interval on phenotype or an interaction between maternal and paternal chromosomal intervals. The Proc Mixed F-statistic for each hypothesis was generated across each of 64 chromosomal intervals. Intervals were calculated at a density reflecting the distance between flanking markers as follows: markers <5 cM apart, 1-cM intervals; 5 to 20 cM, 2-cM intervals; 21 to 30 cM, 3-cM intervals; 31 to 40 cM, 4-cM intervals; and ≥40 cM, 5-cM intervals. Results of the composite interval analysis are reported in Table 4. The composite interval mapping method–performed within a repeated measures, mixed model framework using eight phenotypes assessed separately–is computationally demanding (estimated at >400 hr for permutation analysis). Consequently, a 1000-repetition permutation test to determine experimentwise critical values was not performed; rather, all comparisonwise P values <0.001 are reported.

The composite interval method confirmed three candidate QTL associations observed in the single-marker analysis: CD4 with chromosomes 9 and 15, and CD8P with chromosome 9. In addition, new associations were identified that indicate candidate loci within intervals. For example, the candidate QTL for CD8M levels on chromosome 11 is detected 4 cM from D11Mit2 and 5 cM from D11Mit83.

Post hoc analysis of candidate marker-phenotype associations: Post hoc tests were used to gain a more detailed understanding of the QTL associated with the phenotypes. As an example, single-marker statistical analysis identified an association between locus D12Mit105 and CD8M levels (P = 0.01; Table 3). While the analysis identified an effect of the maternal-derived allele, it does not specify which maternal allele is associated with high levels of CD8M cells. Figure 1 shows the UM-HET3 population distribution for CD8M segregated by maternal-derived allele (B6 vs. C). At 8 mo of age those mice with the B6 allele at D12Mit105 have higher mean values of CD8M than those with the C allele (post hoc t-test, P = 0.00005). The B6 allele is also associated with higher levels of CD8M cells at 18 mo (post hoc t-test, P = 0.0031). It is thus likely that D12Mit105 is linked to a gene (or genes) that varies between the B6 and C progenitors and affects CD8M levels in an age-independent fashion in a population of genetically heterogeneous siblings.

The repeated measures analysis is able to identify loci where the effect of the allele on a trait varies with age. The association between D12Mit105 and CD8 levels identified in the composite interval analysis presents a clear example. Figure 2 illustrates the post hoc analysis of the 8- and 18-mo phenotype distributions, segregated by maternal-derived allele (B6 vs. C). The B6 allele is associated with higher levels of CD8 T cells, but only later in life (post hoc t-test, P = 0.00029). The QTL linked to D12Mit105 is a novel locus that interacts in a life-course-specific manner with the CD8 T cell levels.

Finally, some interactions between maternal and paternal alleles may be the result of “recessive” effects, where only one of the four possible allelic pairs is associated with a phenotype. The locus D17Mit221 provides an example. Table 3 shows an association between D17Mit221 and CD4M levels with an experimentwise P = 0.02, but only with maternal-by-paternal allelic interaction. Post hoc analysis indicated that only one allelic pairing of the maternal- and paternal-derived genotypes has a phenotypic effect (Figure 3), with C/C3 heterozygotes having a consistently higher mean value for CD4M levels, with post hoc P-values between 0.007 and 0.00003.


The intent of this project has been (1) to develop a generalized system for experimental QTL analysis in heterozygous-parent cross mice, (2) to show how the system can be used to examine several traits simultaneously within a single population of animals, and (3) to identify novel loci that affect T lymphocyte subsets in an age-associated manner. Unlike conventional F2 or backcross analysis systems, the heterozygous-parent method uses inbred progenitor strains that are not chosen for variation of a specific phenotype. Rather, the strains were selected, from among the well-characterized inbred lines, to provide significant genetic heterogeneity. The progenitor strains have been examined for genetic relationships by historical analysis and protein polymorphism, and are known to exhibit substantial interstrain variation at standard SSLP loci (Atchley and Fitch 1991; Dietrichet al. 1992).

View this table:

Single-marker analysis: associations with experimentwise significance of P < 0.05

View this table:

Composite interval mapping: associations with comparisonwise significance P < 0.001

Figure 1.

–Association between locus D12Mit-105 and CD8M levels. The effect of the maternal-derived allele (B6 vs. C) is shown for the 8-mo and 18-mo timepoints. The histogram gives the distribution of animals in each phenotype range. In each panel, the P value is a post hoc comparison of the two genotypic groups. At both assay time-points a significant difference is seen in the group means (B6 allele > C allele).

T cell subset populations provide an attractive complex system for analysis, since some subset populations undergo reproducible change across the life span (Miller 1996). The levels of multiple subtypes can be assayed from a single, small blood sample with little impact on the animal. Identification of novel genetic loci affecting T cell subsets–and their regulation over the life course–may be useful for understanding the components of age-dependent change in immune function. Studies of age-associated changes in T cell subsets in both humans and mice strongly suggest that aging leads to an increase in the proportion of memory T cells and a reciprocal decrease in the proportion of virgin T cells; the proportion of cells expressing P-glycoprotein also increases in both the CD4 and CD8 populations (Milleret al. 1994). The QTL identified in this study represent candidates for factors influencing CD8M cells (CH1, CH5, CH11, CH12, CH17), CD4M cells (CH11, CH12, CH17), and CD4V cells (CH4), as well as CD4P (CH4, CH9) and CD8P cells (CH9, CH12).

The initial survey did not identify loci that were associated with levels, or rates of change, of groups of age-sensitive T cell subsets. Thus even though high CD4M, high CD8M, and low CD4V levels tend to be associated with one another in UM-HET3 mice (Miller 1997) and in a different stock of heterozygous-parent-derived mice (Milleret al. 1994), we did not note any loci that had strong effects on two or more of these three subsets. Composite interval mapping did identify regions of chromosomes 11 and 17 that may affect CD4M and CD8M; however, the allelic influence was not consistent across the phenotypes. A larger population of HET animals should provide increased statistical power to detect QTL that may coordinately influence groups of traits. The data do suggest that a locus close to D12Mit34 may influence CD4M levels in an age-dependent manner. High levels of CD4M cells are associated with relatively short life span in UM-HET3 mice (Milleret al. 1997), and it will be of interest to see if loci that influence CD4M levels also affect life span and disease risk in these mice.

Figure 2.

–Association between locus D12Mit-105 and CD8 T cell levels changes over time. See legend to Figure 1 for an explanation of the panels. The post hoc comparison of the two genotypic groups shows a significant difference in means only at 18 mo (B6 allele > C allele).

Figure 3.

–Association between locus D17Mit221 and CD4M levels. Memory CD4 T cell level phenotype distributions are given as “box-and-whisker” plots for each of the four genotypic groups in the population (box, standard error of mean; whisker, 1 SD). Post hoc comparison of C/C3 heterozygotes with the other three genotype groups shows a significant difference in mean values, in each case.

Even in a relatively modest population of 145 UM-HET3 animals, candidate QTL of biological interest have been identified. In the analysis using single locus, repeated measures analysis of variance, six of the eight immune cell phenotypes are associated with at least one marker locus at experimentwise significance P < 0.05. Caution is needed in the interpretation of these QTL, however. First, the associations may not be independent, because genetic and nongenetic factors could lead to correlations among the traits. For example, a QTL associated with CD8M maps to mouse chromosome 17 in the region of H-2, the major histocompatibility region (≈20 cM). Next, the large number of marker-phenotype comparisons presents a challenge for assessing statistical significance. This has been addressed using a permutation test for the single locus repeated measures model (Table 3). However, the repeated measures composite interval mapping model (Table 4) generated only nominal significance levels without adjustment for multiple comparisons. The determination of experimentwise significance levels by permutation analysis is computationally impractical at this time. The simplest treatment would reduce the incidence of false-positive associations by setting an experimentwise α= 0.05 using the Bonferroni correction for multiple comparisons. In the composite interval model, eight traits, three time-dependent values, and 64 interval locations were examined, giving 0.05/(8 × 64 × 3) = 3.3 × 10–5 = 0.000033. However, because the assumption of independence of measures does not hold either for the markers or phenotypes, this must be taken as an extremely conservative value. At the initial, exploratory stage of analysis we have used an arbitrary significance level of P < 0.001 to identify candidate regions for future analysis.

Additional phenotypes may be obtained in longitudinal heterozygous-parent sibships, including measures of serum components, morphology, behavior, or response to chemical treatments. Several issues relate to using the system, however. First, the progenitor strains must harbor genetic variants at relevant loci. Preliminary phenotypic analysis of inbred or F1 animals may not be useful in this regard, because the sibship progeny will not reestablish either the F1 heterozygote or inbred homozygote allelic relationships at any locus. Second, because each animal has a unique combination of the progenitor alleles across its genome, the candidate loci must have significant phenotypic effects even within a genetically heterogeneous population. Third, because no formal homozygote class exists in the progeny, only “dominant” effects may be observed unless two progenitor strains contribute similar recessive alleles, presumably reflecting shared ancestry (see Figure 3 for an example). Last, in the current CB6F1 × C3D2F1 breeding scheme, only four of the six possible allelic pairings are obtained. An orthogonal cross such as CD2F1 × B6C3F1 might be used to ensure complete sampling of the progenitor variation.

In this initial study, two methods of statistical analysis were used to identify chromosomal regions with an effect on quantitative values for immune cell populations. Neither analysis is ideal, however. The single-marker analysis using Proc Mixed is straightforward and takes advantage of permutation tests to assess experimentwide significance. The method is a general parametric model and can accept data that exhibit unequal size classes and unequal variance among classes. The method also does not eliminate individuals with incomplete data. This is particularly important in longitudinal analysis, where some individuals have phenotype information at only one of the two assay timepoints. The method does not take advantage of flanking marker information and does not account for possible multiple-linked QTL. Composite interval mapping methods (Zeng 1994), on the other hand, combine interval mapping (Lander and Botstein 1989) with multiple regression analysis to use all the available mapping data and obtain a more precise localization of the QTL. The composite interval method provides a test statistic on a marker interval that is designed to be unaffected by QTL located outside the defined interval. However, because the composite interval method uses more genotypic data it can result in lower power to detect QTL, especially in studies with small sample size (Zeng 1994).

The strengths and weaknesses of the two analysis models can provide an explanation for the differences seen between Tables 3 and 4. For example, composite interval mapping detected a candidate QTL associated with CD8M levels on chromosome 5, at ∼41 cM. No chromosome 5 QTL was observed in single-point analysis. The QTL location is roughly midway across the interval D5Mit205-D5Mit25, 8 cM from D5Mit205 and 5 cM from D5Mit25. Locations at a distance from typed markers are weakly characterized by single-point analysis, as expected (Lander and Botstein 1989). Conversely, some candidate QTL observed in single-point analysis do not reach comparable significance levels using composite interval mapping methods. The CD4M single-point analysis candidate QTL on chromosomes 12 and 17 (Table 3) were each observed during composite interval analysis, but at much lower significance (not shown). In these cases, incorporation of the flanking data information increased the quantity of data used in the test and, presumably, decreased the overall confidence.

Ultimately, the statistical ambiguities associated with any initial QTL result require experimental confirmation. Because the UM-HET3 population is generated from commercially available, highly fecund F1 stocks, producing additional full-sibling mice is simple and inexpensive. If phenotypic distributions are replicated in additional populations, the data may be merged. The availability of a high-density mouse SSLP marker map–with known polymorphisms among the four progenitor strains–should allow rapid genetic refinement of candidate QTL. Together, the increased power should provide a rational process for improving QTL significance and accurate genetic localization. Phenotype information from a second replicate CB6F1 × C3D2F1 population (128 animals) shows T cell population distributions consistent with the UM-HET3 population and preliminary confirmation of D12Mit105 association with CD8M cell levels (not shown).

The heterozygous-parent cross system is particularly suitable for analysis of the genetic basis of aging and age-associated traits. It is not yet clear to what extent the various aspects of the aged phenotype, including changes in many physiological systems and biochemical variables, are coordinately regulated. It is probable that many of the individual traits that distinguish young from aged individuals may be under separate genetic control, but it also seems possible that some loci may influence the extent or rate of change in several age-sensitive systems. Multiple phenotypes have been used to examine common influencing QTL in standard F2 or backcross strategies (Jiang and Zeng 1995; Korolet al. 1995). Selective breeding studies in Drosophila (Rose 1991) and mice (Covelliet al. 1989), and analyses of recombinant inbred nematodes (Johnson 1987) and mice (Gelmanet al. 1988) have all shown that life span can be regulated by genetic variants within these species. The ways in which the timing of late-life events might be regulated synchronously or asynchronously by genetic factors deserve greater consideration. Although genome-wide scans and assessment of late-life phenotypes are inherently costly for large populations, the marginal cost of measuring additional traits can be quite low. The power of the heterozygous-parent cross approach for discovering new genetic influences thus can be increased dramatically by parallel assessment of multiple traits of interest.


The authors acknowledge the professional staff of the University of Michigan Unit for Laboratory Animal Medicine; technical assistance from Erin Belloli, Julie Kim, Luann Linsalata, Jennifer McGrath, and Lisa Mullins; and advice from numerous National Institute on Aging, Longevity Assurance Gene Consortium members and Dr. S. Xu at the University of California, Riverside. The research was performed with funding from the National Institutes of Health (NIH), National Institute on Aging, grants R01-AG11687 and R01-AG08808. A.U.J. is supported by an NIH predoctoral training grant in genomic sciences T32-HG00040; D.T.B. was funded in part by a Searle Scholar Fellowship from the Chicago Community Trust.


  • Communicating editor: T. F. C. Mackay

  • Received July 28, 1997.
  • Accepted October 27, 1998.


View Abstract