Introduction

In aquatic ectotherms, environmental temperature fluctuations may have lethal effects at the extremes of environmental stress and minor temperature fluctuation may cause severe physiological disruption (see Wood & McDonald, 1997). It appears likely that such stress is particularly relevant for stenothermic ectotherms (Johnston & Bennett, 1996). The impact of acute thermal stress may also be compounded in the aquatic environment by the ability to physiologically respond to correlated changes in other water quality parameters, such as dissolved oxygen and unionized ammonia (Pennell & Barton, 1996). A number of different molecular chaperones appear to be involved in the thermal stress in fish (Iwama et al., 1998), suggesting a polygenic architecture for this trait. Conventional quantitative genetic theory has been used in salmonid fish (Osteichthyes: Salmonidae) to explore the heritability of the overall stress response (Pottinger & Pickering, 1997) and the response to specific stressors (Fevolden et al., 1999).

Backcrosses or F2 intercrosses of hybrids between lines with strong phenotypic divergence due to artificial selection have been used to detect QTL for complex physiological traits in agricultural (Knott et al., 1999) and laboratory (Cheverud et al., 1996) species. For example, backcrosses of phenotypically divergent rainbow trout (Oncorhynchus mykiss) lines have been used to detect QTL for upper thermal tolerance (UTT) (Jackson et al., 1998; Danzmann et al., 1999) and spawning time (Sakamoto et al., 1999). With the exception of cattle (see Stone et al., 1999), there has been little effort directed at QTL detection in outbred populations (see Poompuang & Hallerman, 1996). While theory predicts maximal power for QTL detection in intercrosses of selected lines (see Liu, 1998; Lynch & Walsh, 1998), the genomic constitution of such groups cannot be representative of wild populations, and domesticated salmonid populations derived from intensive single-trait selection are generally uncommon (Poompuang & Hallerman, 1996). The detection of QTL for certain traits in outbred populations might have greater relevance both to natural selection and artificial improvement, especially when taken in context of their effects against other sources of genetic variation (i.e. specific genetic background, parental effects, etc.). Suggestive evidence of QTL for growth (Gross & Nilsson, 1999) and disease resistance (Palti et al., 1999) exists in commercial salmonids but lacks documented pedigree information.

The objective of this study was to assess if previously reported associations between allelic variants at three SSR loci (Ssa20.19NUIG, Ssa14DU and Omy325UoG) (see refs in Jackson et al., 1998; Danzmann et al., 1999) and upper thermal tolerance in divergently selected lines for this trait were also detectable in unselected outbred rainbow trout. These SSR loci are linked to putative upper thermal tolerance QTL in backcrosses of the selected lines used by Jackson et al. (1998) and Danzmann et al. (1999). We verify the presence of a strong association between one of the putative QTL markers (Ssa20.19NUIG) and thermal tolerance in outbred fish. This marker was previously reported to have the strongest association with thermal tolerance among several markers tested. The use of a multigenerational grandsire QTL model incorporating phenotypic, allelic and pedigree information also supported this finding.

Materials and methods

Two commercial strains of rainbow trout, Spring Valley (SV) (Spring Valley Trout Farm Ltd, Petersburg ON) and Rainbow Springs (RS) (Rainbow Springs Trout Farm, Thamesford, ON) were used in this study. Two first-generation (G0) SV and RS ‘grandsires’ (G0SVM1, G0SVM2, G0RSM1 and GoRSM2) were mated within strain to 12 and 13 grandams, respectively, creating 35 pure-strain second-generation (G1) (‘parental’) families (Table 1). These fish were reared at the Ontario Ministry of Agriculture, Food and Rural Affairs (OMAFRA) Alma Aquacultural Research Station (AARS) (Alma, ON) until the 1996–97 breeding season when they were bred to form 11 2 × 2 diallel lots (44 inter- and intrastrain families). These were reared either at AARS or in the Axelrod Building Wet Lab (AWL), Department of Zoology, University of Guelph (Guelph, ON). Rearing was conducted in either 0.3 m tanks (AARS), washbasins (3.9 L/family) (AWL) or in subdivided sections (anterior to posterior from water inflow) of a 12.4-L raceway (3.1 L/family) (AWL). Raceway-reared lots were each housed in a single raceway, with each of the four families in one of the sequential units. Family order was random within each raceway.

Table 1 Two-generation pedigree (G0, G1) for the Spring Valley (SV) and Rainbow Springs (RS) rainbow trout experimental strains. Grandparental Spring Valley (‘G0SVM’) and Rainbow Springs (‘G0RSM’) grandsires are listed in each column within strain (i.e. G0SVM1). Spring Valley (‘G0SVF’) and Rainbow Springs (‘G0RSF’) granddams are listed at far left under each strain column (e.g. G0SVF1). G1 families (‘G1 fam.’, e.g. 93-2), individual and sex (‘G1 ID (Sex)’, e.g. 93-2-2 (F)) and heterozygous loci tested (‘H’) (in the order Omy325UoG/Ssa14DU/Ssa20.19NUIG, where ‘1’ indicates heterozygosity and ‘0’ homozygosity in that G1)

At 8–10 months postfertilization, 48–96 individuals per family (from each of the four families in each diallel lot) were acclimated for approximately two weeks at 10°C in the Hagen Aqualab, University of Guelph. The fish were then subjected by lot to a thermal challenge where water temperature was raised from 10°C to 25.7°C (critical thermal maximum (CTmax) for 10°C-acclimated rainbow trout) (Bidgood, 1980; Currie et al., 1998), over approximately one hour and maintained at that temperature thereafter. Temperature readings were taken throughout using a pair of inline sensors in the experimental tank. Oxygen saturation remained above 80% for the duration of the experiment. Effective time (ET) (Fry, 1971) of survival for individual fish was taken to be the time elapsed at CTmax when fish were unable to maintain equilibrium (see Jackson et al., 1998). Wet weight (FW, grams) and fork length (FL, length in millimetres from anterior tip of the snout to the fork in the caudal fin) were also recorded. Since variability in rate of temperature increase between experiments may have affected estimates of ET between lots, survival data was converted into cumulative exposure in degree · minutes (d · m) by summing all differences between the acclimation temperature (10°C) and the experimental temperature at each minute until the loss of equilibrium, considered here as upper thermal tolerance (UTT). ET was considered in comparison with previous work (Jackson et al., 1998; Danzmann et al., 1999).

DNA was extracted using a modification of Bardakci & Skibinski (1994) or the IsoQuick nuclear DNA extraction kit (ORCA Research). Variability at three SSR loci (Omy325UoG, Ssa14DU and Ssa20.19NUIG) was detected (see Jackson et al., 1998 for a general protocol) using the polymerase chain reaction (PCR) with radioactive end-labelled (γP33) primers. PCR products were separated in 6% denaturing polyacrylamide gels with fragment size determined using M13 sequencing ladder. Genotypes were manually scored using autoradiographs produced with Kodak BioMax film.

General linear modelling (PROC GLM; SAS Institute, 1996) was used to correct the effect of sequential ordering (by distance from inflow) of families in raceway G2 lots. Health and gross physiological state may be indicated by various growth traits (Pennell & Barton, 1996), such as wet weight (FW), fork length (FL) and condition factor (K) (the residual of log(FW)= log(a) + βlog(FL) (Pennell and Barton, 1996)). Since FL and FW were highly correlated (β=0.891–0.988) and collinear in each pair of half sib families (variance inflation factor (VIF) (SAS Institute, 1996) >10.0), only FL was used as a covariate in QTL analysis. This trait was strongly associated with UTT in previous work (Jackson et al., 1998). K was not collinear with either FL or FW in any of the half-sib families (correlation with FL: 0.220 − (−0.491), VIF < 10.0; correlation with FW: 0.601 − (−0.248), VIF < 10.0). FL and K were evaluated for inclusion as covariates in QTL modelling using sequential backward elimination (α for removal 0.05) on a fully saturated model within the half-sib progeny of each a common parent (SAS Institute, 1996).

Inheritance from each common G1 parent was tested separately at each locus (PROC GLM, SAS Institute, 1996) following the model,

where μ is the overall mean, yik is the UTT of G2 individual k from one of two half-sib families inheriting marker allele i at the locus of concern from the common G1 half-sib parent i, αi is the fixed effect of allelic inheritance at a given marker locus from the common G1 sire or dam, λj is the fixed effect of the non-common G1 parent i in the cross, αiλj is the interaction of alleles from the common G1 parent with the background of the non-common parents, βm is the regression coefficient for each of m covariates not rejected by stepwise backwards regression, Cmk is the value of covariate m for G2 half sib k and ɛijk is the random residual. Where residuals from QTL modelling were non-normally distributed (detected using the Shapiro–Wilk W-statistic (Shapiro & Wilk, 1965)) data were transformed using a likelihood-based macro (BoxGLM; M. Friendly, York University) to normalize residuals before reanalysis. Significance of locus effects were adjusted with simultaneous Bonferroni correction (Rice, 1989) by the total number of independent tests run, resulting in a significance threshold of P=0.05/82 independent G2 half-sib tests=0.0006, excluding cases where it was not possible to differentiate allele identity. Missing data and unassignable genotypes were ignored at all levels of analysis.

The existence of QTL for UTT in the G2 grandprogeny of each of the four G0 grandsires was tested using a modified grandsire model of the form

where yijgk is the UTT of G2 grandprogeny k, αi is the effect of allele i from the G0 grandsire at the SSR locus, αi(γj) is the effect of the G1 sire or dam of G2 individual k nested within the allele received from that grandsire, γj(φg) is the nested effect of the non-common G1 parent within the common G1 parent (i.e. full sib family-specific effects) in each pair of half sib families, βFLXFL and βKXK are the effects of covariates FL and K and ɛijgk is the random residual. BoxGLM was used to provide the optimal transformation of UTT data when non-normality in the model residuals was detected with the Kolmogorov–Smirnov D-statistic (Steel & Torrie, 1980). This test was used because of the large number of G2 grandprogeny per grandsire. Significance thresholds were corrected by the total number of tests, excluding the three loci that were homozygous in the four grandsires (P=0.05/9=0.00556). Missing data and cases where allelic inheritance from the grandsires could not be assigned with certainty in the G2 lots were excluded from analysis. Scheffe tests (in SAS Institute, 1996) were used to identify significant differences between UTT means for alleles from G0 sires in their G2 grandprogeny. Each possible form of inheritance at each locus from the G0 grandsires, including not inheriting either allele, were included as factor states in the allelic term.

Results

Family position within raceway had significant effects on UTT (F3,876=22.20, P=0.0001) and K (P=0.0005) but not on FL (P=0.2716) or FW (P=0.0639). Mean UTT of raceway-reared families increased within increasing distance downstream from the inflow (μnearest to inflow (1230.8 d · m)= μ2nd nearest (1221.7 d·m) < μ2nd furthest (1408.3 d · m) < μfurthest from inflow (1612.8 d · m)), while K of fish reared in the nearest (μ=0.993 log (g · mm–1)) and furthest (μ=0.989 log (g · mm–1)) cells were significantly lower than those in the middle two cells. UTT and K were adjusted for these effects. The correlation of FL and K with UTT depended strongly on half sib family pair. Departure of UTT data from normality, measured by non-normality in the model residuals (P < 0.05) (Shapiro & Wilk, 1965) varied with the G1 parent and the locus under consideration (see Table 2). BoxGLM was used to determine the best normalizing transformation (on the scale of UTT2 to UTT–2) where UTT was non-normally distributed in half sib families; only in one instance could an appropriate transformation not be found (see Table 2).

Table 2 Additive and interactive (locus-by-non-common parent) associations of SSR alleles (given by size in base pairs) from common G1 parents with upper thermal tolerance (UTT) in G2 half sib rainbow trout

Significant differences in UTT were detected between the G2 progeny of G1 sire 93-32-1 (a son of grandsire G0SVM2) inheriting different alleles at Ssa20.19NUIG from that sire (P=0.0001) (Table 2). Thermal tolerance of half-sibs inheriting different alleles at this locus was 2130.6 and 1836.5 d · m or 79.5 and 61.0 min of effective survival time (ET) at the CTmax, explaining approximately 7.5% of phenotypic variance in each of these traits. Although several tests were significant at a priori levels with inheritance at all three loci explaining a small to moderate proportion (1.52–7.50%) of phenotypic variance in UTT, no significant additive or interactive effects of the genomic regions marked by these SSR on UTT were detected for any other comparison after Bonferroni correction (Table 2). However, there was evidence of significant pairwise differences between particular combinations of alleles from common G1 parents and non-common parental backgrounds in some half sib groups, some of which were considerable (P=0.0001). Omy325UoG tended to be more frequently associated with UTT in G1 dams and Ssa20.19NUIG in G1 sires; the strength of marker–UTT associations appeared to be generally greater in male parents than in females overall, particularly at Ssa20.19NUIG (Table 2).

UTT was non-normally distributed in most G0 grandsire tests even after optimal transformation (P < 0.05) (although Box plots of transformed grandprogeny data appeared normal); the best possible transformation of UTT was used in grandsire QTL modelling. In the grandprogeny of G0SVM2, UTT was strongly associated with inheritance at Ssa20.19NUIG from him (Table 3; Fig. 1). Ssa20.19NUIG explained close to 1% of variation in the UTT of G0SVM2’s grandprogeny, which was considerably less than the nesting of non-common G1 parent within common G1 parent (i.e. full sib familial effect). FL also had a significant effect on UTT in the grandprogeny of G0SVM2 (Table 3). Scheffe tests for differences in UTT means indicated that grandprogeny inheriting allele 87 at Ssa20.19NUIG from G0SVM2 had a much higher thermal tolerance than those inheriting allele 83 or neither of the above alleles, supporting the findings from the half sib progeny of 93-32-1, which was a son of G0SVM2 (see also Fig. 1). A significant association between UTT and Omy325UoG nested by G1 parent was also found in the grandprogeny of G0SVM1 (F1,1274=2.93; P=0.0324), although no main effects of this locus were observed (P=0.2027). However, main effects of Ssa14DU on UTT were observed in the grandprogeny of G0SVM1 (F1,1264=3.15; P=0.0432). Neither of these general associations was significant after Bonferroni correction, however, and no other associations of the SSR loci with UTT were found (P > 0.05).

Table 3 Associations of grandparental alleles from G0 SV grandsire 2 (G0SVM2) at Ssa20.19NUIG with UTT in outbred G2 grandprogeny of rainbow trout
Fig. 1
figure 1

Mean Z-standardized upper thermal tolerance (UTT) of second generation (G2) grandprogeny of rainbow trout inheriting different alleles (83 or 87 bp; italics, joined by solid line) from Spring Valley grandsire G0SVM2. Alleles originating from G0 granddams are given (no italics, joined by dashed line) for comparison. Allelic identity in G1 dams 93-22-5, 93-25-4 and 93-30-5 and G1 sire 93-28-6 was not determinable; results from these individuals are not included.

Discussion

Our results indicate the linkage of an SSR locus (Ssa20.19NUIG) with a quantitative trait (upper thermal tolerance) in outbred rainbow trout. Previously, strong marker–trait associations (involving this marker) had only been reported in backcrosses of highly selected lines (Danzmann et al., 1999). Inheritance at this locus explained a considerable proportion (7.5%) of phenotypic variation in upper thermal tolerance in the progeny of an outbred G1 sire (93-32-1) approximating the definition of a ‘major’ QTL (Lynch & Walsh, 1998), although this experimental design was not able to separate the true estimate of effect for this QTL from marker–QTL recombination fraction. The association of Ssa20.19NUIG with UTT in this sire was supported by the use of a two-generational grandsire model in G0SVM2 including genotypic and phenotypic information in all of his G2 grandprogeny.

Differences in the recombination rates observed between male and female salmonids (Johnson et al., 1987; Sakamoto et al., 2000) may help to explain the differences in observed marker/trait associations by G1 sex and marker. Recombination rates in male salmonids appear to be higher towards telomeric regions than in centromeric regions of the chromosome, whereas recombination events appear uniformly distributed throughout the length of the female chromosome (Sakamoto et al., 2000). Block segregation of large chromosomal regions is thus common in males for most of the intertelomeric regions of a linkage group. This suggests both an increase in the probability of detecting QTL in centromeric regions and a limit on the usefulness of telomeric markers for QTL detection in sires. Omy325UoG (linkage group B) and Ssa20.19NUIG (linkage group S) appear to be telomeric while Ssa14DU (linkage group D) is centromeric (Danzmann et al. unpublished; Sakamoto et al., 2000). While the higher frequency of Omy325UoG/UTT associations in female G1 parents may suggest increased female-specific marker/QTL linkage, stronger associations of UTT with Ssa14DU would have been expected in male parents if pronounced QTL effects were present at this locus. Ssa20.19NUIG, while possibly telomeric, appeared to have strong evidence for QTL effects over multiple generations in male parents, which could indicate tight marker–QTL linkage in this region.

The amount of variation in thermal tolerance ascribed to Ssa20.19NUIG and overall tolerance of fish in the present study were both considerably less than in the backcross (BC) families studied previously (Jackson et al., 1998; Danzmann et al., 1999). The genetic background of the different groups might be responsible for a proportion of the difference between thermal tolerance in the BC and outbred families. Given the diverse sources of the BC families and the extremity of divergent selection on them (Ihssen, 1986), QTL segregants from the hybrid male of Jackson et al. (1998) and Danzmann et al. (1999) should have considerably greater mean effect than would be found in unselected populations. UTT variance in the BC families was roughly an order of magnitude greater than in the outbred groups (Perry et al. unpublished). Mean ET in the progeny of the low-line BC dam crossed to the hybrid male (Jackson et al., 1998; Danzmann et al., 1999) was roughly comparable to the outbred half-sibs in this work, suggesting little previous selection for increased thermal tolerance in the commercial strains. Sire and dam effects on UTT in the outbred half sibs also appeared to be considerably stronger than putative SSR-linked QTL, suggesting the greater role of polygenic background than these QTL in across-full sib family comparisons, although Ssa20 19NUIG seemed to be an exception to the above.

The fish stress response is common to a number of environmental stressors including various pathogens (Iwama et al., 1998), and the role of stress response candidates such as heat shock proteins (hsp) in transmembrane transport and protein conformation is well known (Morimoto et al., 1994). Improvement in upper thermal tolerance might result in correlated improvement in generalized stress tolerance and/or basal physiological efficiency for a number of traits including upper thermal tolerance. Moreover, the extensive use of sea-cages in salmonid aquaculture may result in the exposure of valuable production cohorts and/or broodstock to diurnal and seasonal environmental fluctuations, in which the direct and indirect consequences of acute thermal stress may be particularly serious due to the reduced oxygen carrying capacity of seawater (Pennell & Barton, 1996). Acute diurnal temperature fluctuation may also have diverse and serious impacts on the physiology of aquatic ectotherms, and may be a major driving force in their evolution (Johnston & Bennett, 1996; Wood & McDonald, 1997). Previous work has identified linkage between the heat shock protein (hsp) gene heat-shock cognate 71 (hsc71) and One14ASC (Sakamoto et al., 2000), a marker associated with UTT in the progeny of the low-line dam in the backcross families of Danzmann et al. (1999). A more complete comprehension of the genetic architecture of thermal tolerance (including those for correlated traits) might result in a clearer understanding of the evolutionary role and commercial significance of these genomic regions in thermal stress fitness and selection on this trait.