Methodology for genetic evaluation of disease resistance in aquaculture species: challenges and future prospects


Correspondence: J Ødegård, Nofima Marin, PO Box 5010, NO-1432 Ås, Norway. E-mail:


Resistance against specific diseases affecting aquaculture species often show moderate to high heritabilities, and there is thus a large potential for genetic improvement. However, genetic evaluation of disease resistance based on survival under challenge testing still face some challenges that may complicate both recording and statistical analysis of disease resistance, e.g., susceptibility and time until death may be different aspects of resistance, or survival may be an inadequate measure of resistance. Hence, for some diseases, more advanced statistical modelling and/or supplementing indicators besides survival would be an advantage. Furthermore, tested individuals are usually excluded as selection candidates as they might be potential disease carriers, and thus pose a health risk. Under classical selection, this restriction reduces both accuracy and intensity of selection, as little or nothing is known about how sib selection candidates deviate from the family means. Thus, selection methods allowing within-family selection are of particular importance in selection for improved disease resistance based on disease challenge testing. Examples of such methods are indirect selection on correlated traits measurable on selection candidates, selection on identified quantitative trait loci and genomic selection. In the future, genomic information has the potential to substantially improve selective breeding for disease resistance traits, given that this information can be acquired on a massive scale and at an affordable cost.


In recent years, selective breeding for disease resistance has received increasing importance in aquaculture breeding programmes worldwide, including species such as Atlantic salmon (e.g., Thodesen & Gjedrem 2006; Houston, Haley, Hamilton, Guy, Tinch, Taggart, McAndrew & Bishop 2008), rainbow trout (Dorson, Quillet, Hollebecq, Torhy & Chevassus 1995), Atlantic cod (Ødegård, Kettunen Præbel & Sommer 2010), rohu carp (Nguyen & Ponzoni 2006) and Pacific white shrimp (Argue, Arce, Lotz & Moss 2002; Cock, Gitterle, Salazar & Rye 2009). In aquaculture, as well as in animal production in general, maintaining a high survival rate is crucial to economy, animal welfare and sustainability of the industry. Selective breeding is one out of several preventive measures that can be taken to improve survival. However, early studies on survival at different life stages in Atlantic salmon and rainbow trout showed low heritabilities, and genetic correlation over longer time spans were low and non-significant (Kanis, Refstie & Gjedrem 1976; Rye, Lillevik & Gjerde 1990). However, Standal and Gjerde (1987) reported an underlying heritability of 0.17 for field survival over three year classes for Atlantic salmon, where the main cause of mortality was Vibrio salmonicida. In a recent study on rainbow trout, Vehviläinen, Kause, Quinton, Koskinen and Paananen (2008) estimated highly variable heritabilities of survival for different generations and station-specific cohorts, with genetic correlations between them ranging from significantly negative to significantly positive values. Hence, based on these results, crude survival not taking the underlying causes of death into account is of limited value in genetic selection. Because of the early discouraging results on field survival, the focus was shifted towards resistance against specific diseases. The latter is commonly measured as survival under challenge testing with the pathogen. In some of the first studies, high heritabilities were estimated for several diseases, such as furunculosis (Aeromonas salmonicida) and infectious salmon anaemia (ISA) in Atlantic salmon (Gjedrem, Salte & Gjøen 1991; Gjedrem & Gjøen 1995; Gjøen, Refstie, Ulla & Gjerde 1997). Several later studies on a wide range of aquaculture species have shown that many diseases have moderate to high heritabilities under challenge test conditions (Tables 1–3).

Table 1.   Recent heritability estimates of resistance to bacterial diseases in aquaculture species
  • *

    Survival, liability scale.

  • †Survival time.

  • ERM, enteric redmouth disease; RTFS, rainbow trout fry syndrome.

A. salmonicida (furunculosis)Atlantic salmon (Salmo salar)0.46±0.13–0.59±0.14Ødegård et al. (2007a)*
Atlantic salmon (Salmo salar)0.43±0.02Ødegård et al. (2007b)*
Atlantic salmon (Salmo salar)0.59±0.06Olesen, Hung and Ødegård (2007)*
Atlantic salmon (Salmo salar)0.62Kjøglum et al. (2008)*
Atlantic salmon (Salmo salar)0.47±0.05Gjerde, Evensen, Bentsen and Storset (2009)*
Brook charr (Salvelinus fontinalis)0.51±0.03Perry, Tarte, Croisetière, Belhumeur and Bernatchez (2004)
Y. ruckeri (ERM)Rainbow trout (Onchorhynchus mykiss)0.42Henryon et al. (2005)*
F. psychrophilum (RTFS)Rainbow trout (Onchorhynchus mykiss)0.43Henryon et al. (2005)*
A. hydrophilaRohu carp (Labeo rohita)0.03±0.06–0.39±0.12Das Mahapatra, Gjerde, Sahoo, Saha, Barat, Sahoo, Mohanty, Ødegård, Rye and Salte (2008)*
Common carp (Cyprinus carpio)0.04±0.03Ødegård, Olesen, Dixon, Jeney, Nielsen, Way, Joiner, Jeney, Ardó, Rónyai. & Gjerde (2010)*
V. anguillarum (vibriosis)Atlantic cod (Gadus morhua)0.08–0.17Kettunen, Serenius and Fjalestad (2007)
Atlantic cod (Gadus morhua)0.17±0.12Gjerde and colleagues (in manuscript)*
Table 2.   Recent heritability estimates of resistance to viral diseases in aquaculture species
  • *

    Survival, liability scale.

  • †Transformed to liability scale.

  • ISAV, infectious salmon anaemia virus; IPNV, infectious pancreatic necrosis virus; VHSV, viral haemorrhagic septicaemia virus; KHV, koi herpes virus; WSSV, white spot syndrome virus; TSV, taura syndrome virus.

ISAVAtlantic salmon (Salmo salar)0.32±0.02Ødegård et al. (2007b)*
Atlantic salmon (Salmo salar)0.24±0.03Olesen et al. (2007)*
Atlantic salmon (Salmo salar)0.37Kjøglum et al. (2008)*
Atlantic salmon (Salmo salar)0.40±0.04Gjerde et al. (2009)*
IPNVAtlantic salmon (Salmo salar)0.43

Guy, Bishop, Brotherstone, Hamilton, Roberts, McAndrew and Woolliams (2006)
Guy, Bishop, Woolliams and Brotherstone (2009)*
Atlantic salmon (Salmo salar)0.50Wetten, Aasmundstad, Kjøglum and Storset (2007)
Atlantic salmon (Salmo salar)0.55Kjøglum et al. (2008)*
VHSVRainbow trout (Oncorhynchus mykiss)0.57Henryon et al. (2005)*
KHVCommon carp (Cyprinus carpio)0.79±0.15Ødegård et al. (2009)*
NodavirusAtlantic cod (Gadus morhua)0.75±0.11Ødegård et al. (2010)*
WSSVShrimp (Penaeus vannamei)0.04±0.02Gitterle et al. (2006)*
TSVShrimp (Penaeus vannamei)0.30±0.13Argue et al. (2002)*
SPDVAtlantic salmon (Salmo salar) 0.21±0.005Norris et al. (2008)*
Table 3.   Recent heritability estimates of resistance to parasites in aquaculture species
  • *

    Number of parasites.

  • †Survival, liability scale.

  • ‡Linear days to death, observed scale (partly censored trait).

  • §

    §Binary median test-survival, observed scale.

  • ¶Survival category grouped by days to death and gill score at trial termination, observed scale.

Sea louse (Caligus elongates)Atlantic salmon (Salmo salar)0.22Mustafa and MacKinnon (1999)*
Sea louse (Lepeophtheirus salmonis)Atlantic salmon (Salmo salar)0.14±0.02Kolstad, Heuch, Gjerde, Gjedrem and Salte (2005)*
Atlantic salmon (Salmo salar)0.07±0.02Glover, Aasmundstad, Nilsen, Storset and Skaala (2005)*
Gyrodactylus salarisAtlantic salmon (Salmo salar)0.32±0.10Salte et al. (2010)
Neoparamoeba spp. (amoebic gill disease)Atlantic salmon (Salmo salar)0.49±0.09Taylor, Kube, Muller and Elliott (2009)

The current study intends to review the different methods of genetic evaluations for disease resistance in aquaculture species. Many of the referenced studies will be on salmonids (Atlantic salmon and rainbow trout) as these are among the most widely studied aquaculture species. However, the methods and problems described and discussed are, to a large extent, equally relevant across species.

Genetic evaluation of disease resistance

Challenge testing

Selective breeding for improved disease resistance in farmed fish is commonly based on challenge testing for specific pathogens (one pathogen at a time, using different subsamples of all breeding nucleus families). In such tests, fish are exposed to the pathogen in controlled environments, often through cohabitant testing or injection of the pathogen. Mortalities are recorded daily, or in some cases more frequently, and the test is typically continued until at least 50% of the individuals have died (Fjalestad, Gjedrem & Gjerde 1993) or until mortality naturally levels off. Hence, for such tests, both survival at the end of test and time until death are usually recorded.

Statistical analysis of disease-related traits

Cross-sectional models

The simplest statistical models used for analysing challenge test data are the cross-sectional models, which can be defined as models analysing disease resistance measured as a single record (e.g., alive/dead) at a fixed point in time. Hence, time of death is not taken into account. The cut-off point may be chosen at a cumulative mortality of approximately 50% (at which the phenotypic variance of a binary trait is maximized) or when mortality naturally levels off.

Although simple, cross-sectional models have often been shown to give accurate predictions of family breeding values (Gjøen et al. 1997; Ødegård, Olesen, Gjerde & Klemetsdal 2006; Ødegård, Olesen, Gjerde & Klemetsdal 2007a), especially for balanced data with large families kept in common test environments. Often, linear models are used in analysis of such data (treating observations as normally distributed), even though observations are binary. To account for the binary nature of data, generalized linear mixed models can be used; e.g., the widely used threshold (probit link function) model (Gianola & Foulley 1983). With respect to accuracy of predicted breeding values, limited differences have usually been found between linear models and the more advanced threshold models (Gitterle, Ødegård, Gjerde, Rye & Salte 2006; Ødegård et al. 2006, 2007a). Further, the relative difference is expected to decrease with increasing heritability and family sizes and when mortality approaches 50%.

Longitudinal models

By ranking families based on whether or not the fish were alive at a fixed point in time, the time perspective is ignored; i.e., at which time the fish dies. If time of death to some extent reflects resistance to disease (e.g., the first fish to die are the least resistant ones), time until death may be a more appropriate trait definition than crude survival/death at a fixed point in time. However, time until death is only known for non-survivors, while only the minimum survival time is known for the survivors (right-censored data). Further, survival times are usually non-normally distributed, the effect of environmental factors may not be constant over time (time-dependent effects) and (given the genetic variation in survival) survivors at the end of test are not a random sample of the population. Proportional hazards frailty models have been suggested for analysis of lifetime data (Ducrocq & Casella 1996), modelling risk of mortality (hazard) as a function of time, as well as a function of genetic factors. Furthermore, the method also accounts for censoring (some fish still alive at the end of test). A simpler method that approximates proportional hazards models is a survival score model (Veerkamp, Brotherstone, Engel & Meuwissen 2001). Here, lifespan of the individual (within the test period) is split into several sub-periods (e.g., days), and the individual is scored as ‘alive’ or dead for each period. Number of records will then reflect time of death, while individuals surviving the entire test will have all survival scores set to ‘alive’. The binary survival scores can be analysed using threshold models or even with linear models (treating the binary survival scores as normally distributed observations). These models have substantial similarities with the proportional hazards models, and both types of models may take time-dependent effects and censoring into account. For analysis of challenge test data, these models have, in some cases, been shown to yield a somewhat better precision of estimated breeding values, compared with simpler cross-sectional models (Gitterle et al. 2006; Ødegård et al. 2006, 2007a).

Analysis of survival data using longitudinal models (proportional hazards or survival score) relies on an assumption that resistance is associated with survival time and that censoring (survival at end of test) itself does not contain any information besides that the animal has survived up to a given time. In real data, these assumptions may not hold. The analyses may be biased if some individuals are not at risk (non-susceptible), which may be the case if the bacteria, virus or parasites involved appear to be non-pathogenic or at least non-lethal to parts of the population (i.e., resistant or tolerant fish). Furthermore, if survival is recorded at a fixed point in time, rather than when mortality naturally levels off, the survivors will be a rather unpredictable mixture of non-susceptible fish and susceptible fish with censored survival times, which may yield bias in both simple cross-sectional models and more advanced longitudinal models. As an example, Salte, Bentsen, Moen, Tripathy, Bakke, Ødegård, Omholt & Hansen (2010) estimated a genetic correlation of only 0.32±0.10 between survival at the end of test (after mortality had levelled off) and time until death of Atlantic salmon challenge tested with the ectoparasite Gyrodactylus salaris. Hence, an earlier forced termination of the testing period (e.g., at 50% mortality) would have caused severe re-ranking of families. In medical research, longitudinal survival models (called cure models) have been developed to account for the existence of non-susceptible individuals (Farewell 1982; Kuk & Chen 1992). New methods of accounting for such factors in challenge test data on aquaculture species are currently under development (Norwegian research council project no. 192331).

Indirect recording of disease resistance

Some diseases may not necessarily be terminal, but can still cause substantial losses through reduced growth, impaired product quality or losses due to secondary infections. One example is pancreas disease in Atlantic salmon, where 10–15% of the survivors fail to grow and substantial mortality may be observed several months after the acute phase of the disease (McLoughlin & Doherty 1998). For such diseases, selective breeding based on survival during challenge testing may not be appropriate as survivors may still be affected by the disease, and other types of data besides survival should thus be utilized. One possibility is to use indicator traits such as immune parameters, although such indicators have shown variable results (Lund, Gjedrem, Bentsen, Eide, Larsen & Røed 1995; Sahoo, Mahapatra, Saha, Barat, Sahoo, Mohanty, Gjerde, Ødegård, Rye & Salte 2008) or growth records, if affected individuals fail to grow or exhibit reduced growth rate. Growth of survivors may be observed during challenge tests or in commercial test environments with generally high incidence rates of the disease. Hence, disease resistance may be inferred, not only by survival but also by the surviving individuals' ability to maintain growth after exposure to the pathogen. Indirect recording of health traits through growth rate is, however, hampered by the fact that there usually exists substantial variation in growth potential among healthy as well as affected individuals. The challenge is thus to distinguish ‘natural’ variation in growth rate (in healthy individuals) from variation in growth due to infections or other environmental stressors. Furthermore, differences in growth rate between healthy and diseased individuals may be more evident over time, and a long recording period might thus be necessary. In dairy cattle, mixture models have previously been developed with the aim of distinguishing somatic cell count in healthy and mastitic cows (Ødegård, Madsen, Gianola, Klemetsdal, Jensen, Heringstad & Korsgaard 2005), and this methodology may also be used for identification of diseased fish based on recorded growth or other relevant indicator traits. Generally, favourable genetic relationships have been estimated between growth in the field and survival in both field and challenge tests (Gjedrem & Olesen 2005; Nielsen, Ødegård, Olesen, Gjerde, Ardo, Jeney & Jeney 2010). Hence, mixture models may even be used for predicting breeding values for health based on field growth data.

Genetic parameters of disease resistance

The high heritabilities for disease traits often obtained in challenge tested fish populations (Tables 1–3) stand in stark contrast to many of the diseases currently included in selective breeding programmes in livestock. One example is clinical mastitis in dairy cattle, which has an estimated underlying heritability of only 0.07 (Heringstad, Rekaya, Gianola, Klemetsdal & Weigel 2003). Further, calculation of breeding values for such traits is also hampered by the binary nature of data (healthy/diseased). Despite this, clinical mastitis has been successfully implemented in breeding programmes in the Nordic countries (Heringstad et al. 2003). Hence, for highly heritable disease traits (which is typical for aquaculture species), selection is expected to be effective, especially in combination with the high fecundity of both males and females.

Given that genetic correlations exist between resistance against different diseases, genetic improvement with respect to one disease will also indirectly affect resistance against other diseases. Negative genetic correlations indicate that genetic improvement of one disease would have an adverse effect on other diseases (depending on the size of the correlation), while positive genetic correlations would have the opposite effect. In a study on Atlantic salmon by Gjøen et al. (1997), positive genetic correlations (0.10–0.91) were estimated between resistance against the bacteria A. salmonicida, V. salmonicida and Vibrio anguillarum, while weakly negative genetic correlations were estimated between resistance against these pathogens and ISA (−0.24 to −0.05). However, the latter of these findings are not supported by more recent and extensive studies. Ødegård, Olesen, Gjerde and Klemetsdal (2007b) estimated a small, but significant favourable genetic correlation between resistance to furunculosis and ISA, while Kjøglum, Henryon, Aasmundstad and Korsgaard (2008) estimated genetic correlations around zero (−0.05 to 0.07) between resistance against different diseases in Atlantic salmon [furunculosis, ISA and infectious pancreas necrosis (IPN)]. Furthermore, in a study on rainbow trout, weak genetic correlations were found between resistance against the bacteria Yersinia ruckeri and Flavobacterium psychrophilum, and the VHS virus (Henryon, Berg, Olesen, Kjaer, Slierendrecht, Jokumsen & Lund 2005). Hence, there are few indications for adverse genetic correlations between resistance for different diseases; therefore, simultaneous improvement of resistance against various diseases in selection programmes appears realistic.

Genetic improvement of disease resistance and vaccination programmes

Vaccination programmes have been imperative in developing efficient prophylactic programmes for a wide range of diseases in aquaculture species, and is currently considered the single most important factor for disease control in Atlantic salmon (Håstein, Gudding & Evensen 2005). For diseases such as furunculosis (A. salmonicida), farmed Atlantic salmon are routinely vaccinated. However, selective breeding for improved furunculosis resistance has so far been based exclusively on testing of unvaccinated fish. The question then remains whether selection for improved resistance in unvaccinated fish would yield a similar genetic response in vaccinated fish. Unfortunately, a genetic analysis of disease resistance of vaccinated and unvaccinated fish from the same families indicated a low genetic correlation (0.32±0.13) between resistance in the two groups (Drangsholt, Gjerde, Bentsen & Ødegård 2009). Hence, selection for improved resistance in unvaccinated fish would have limited effect on vaccinated fish. However, selection on unvaccinated fish would still be effective if the long-term goal is to produce a fish that survive the infection without vaccination. If vaccination against furunculosis is considered inevitable also in the long run, challenge testing of vaccinated fish would be much more efficient. Although vaccination programmes may be important also in the future, the vaccines in use may change considerably over time. Hence, selective breeding adapted to a specific type of vaccine might be a rather risky approach.

Classical selection for improved disease resistance

Selective breeding for disease resistance using challenge testing often faces some major restricting factors. Besides testing capacity, testing costs and other practical limitations, one important factor is the fact that tested individuals are commonly excluded as selection candidates because they might be potential disease carriers (e.g., Vike, Nylund & Nylund 2009). Classical selective breeding based on phenotypic and pedigree information is thus restricted to selection among untested individuals based on the performance of tested relatives, mainly full- and half-sibs (sib selection). In such cases, classical selection is unable to distinguish between untested individuals within the same full-sib family, and is therefore reduced to between-family selection. This has two major effects: (1) reduced intensity of selection and (2) restricted accuracy of selection. Selection intensity is reduced because the lack of individual information forces the breeder to choose random individuals from high-ranking families, rather than the highest ranking individuals across families, while accuracy is reduced because no information is acquired about within-family variation (half of the total genetic variation). Hence, both accuracy and selection intensity are likely to increase if survivors from disease challenge tests could be considered as selection candidates. However, this is not commonly allowed, in spite of the fact that some pathogens are generally and widely present in many farming environments (e.g., IPNV, ISAV and A. salmonicida in Atlantic salmon, and Aeromonas hydrophila in carps). For some of the diseases, the survivors from challenge testing may not pose a significant threat with respect to disseminating the disease, in particular if measures are taken to reduce the risk (e.g., through medical treatment of survivors from challenge testing with bacteria or a period of quarantine until screening for virus) and if tested selection candidates are used only within already heavily infected areas. Preferably, the survivors should be reared together with the non-challenged candidates so that contemporary and thus comparable trait records for the other important traits can be recorded on the challenged candidates as well. Without this, the benefit of using the survivors as breeders will be lower due to reduced accuracy and intensity of selection for the non-disease traits. Another drawback is that individuals are usually tested for a single disease only, which reduces the advantage of individual selection for multiple diseases. Further research should be undertaken to investigate the potential benefits and risks of using survivors from challenge tests.

Genomic information in selection for improved disease resistance

Genomic information offers new possibilities for selective breeding for disease resistance. Information on quantitative trait loci (QTL) may be implemented in breeding programmes either through gene- or marker-assisted selection (MAS), or dense genome-wide marker resources may be used in genomic selection. Using these methods, candidates may be selected based on individual markers or actual causative genes (QTL), rather than phenotypes of sibs only. Furthermore, gene expression profiles may also be used as an indirect measure of disease resistance.

Genomic selection

Genomic selection is based on utilizing data on individual phenotypes linked with genomic data from genome-wide dense marker maps (Meuwissen, Hayes & Goddard 2001), i.e., pedigree relationships may be replaced by genomic similarities estimated through information on genetic markers. Genomic selection does not require any prior knowledge about the effect and location of specific QTL. Accurate individual breeding values may be calculated for all genotyped fish, even though phenotypic information only exist on a subsample of the population (particularly relevant for challenge test data). Individual breeding values can be calculated as soon as the fish are genotyped, implying a potential for shortening of generation intervals, depending on the reproductive characteristics of the species. Further, using genomic selection, data on full-sibs of the selection candidates may, to a large extent, completely replace individual data (given that the association between marker alleles and phenotypes are re-estimated regularly), which is never the case in classical selection (Ødegård, Yazdi, Sonesson & Meuwissen 2009). Simulations by Sonesson and Meuwissen (2009) confirmed that genomic selection yields high genetic gain, accuracy of selection and lower rates of inbreeding. However, high cost of genotyping may limit practical application in the near future. Indeed, the SNP arrays required for genomic selection are presently unavailable for the vast majority of aquaculture species (although such a chip is now available for Atlantic salmon).

QTL for disease resistance

The mapping of QTL for disease resistance traits has been the focus of a number of studies in aquaculture species, given the potential for additional genetic improvement through within-family selection for resistance on live breeding candidates using makers linked with QTL. Characterization of the number and strength of QTL affecting resistance can also reveal some of the underlying genetic architecture behind resistance; e.g., whether it is controlled by one or few loci with large effect, or many loci with small effect. Typically, genome scans use the binary trait dead or alive or days to death with data from challenge tests (e.g., Moen, Sonesson, Hayes, Lien, Munck & Meuwissen 2007) or natural disease outbreaks (Houston et al. 2008). In some cases, it has been demonstrated that QTL mapping methods for quantitative traits can be applied to binary survival data under the assumption that the underlying resistance is a continuous variable (e.g., Visscher, Haley & Knott 1996). Moen et al. (2007) compared the suitability of survival models for analysing challenge test data to binary models, and somewhat surprisingly found that the best-performing analysis methods were those based on the binary data. However, as alluded to earlier, overall survival and time until death may actually represent different aspects of disease resistance, and it may be possible to detect different QTL for these traits if experimental designs are powerful enough and appropriate models are used. The use of selective genotyping of a subset of fish (a common approach to reduce costs) with challenge test data where the test has been terminated before complete mortality is compromised to a certain degree by the fact that extremes within the surviving fraction cannot be identified (as compared with the first animals to die), and therefore, a random sample of survivors is generally taken.

Because of the relatively limited molecular marker resources available in aquaculture species, published studies mapping QTL for disease resistance have generally used within-family linkage analysis to identify relatively broad chromosomal regions harbouring QTL with moderate to large effects (Table 4). Some caution should be exercised when comparing results of the various studies, as a number of different QTL analysis methods have been used, including different methods for the calculation of significance thresholds (that define a significant QTL) and amount of phenotypic variance explained by the QTL. Additionally, QTL effects may be overestimated in some cases (Darvasi & Soller 1992; Xu 2003). Nevertheless, these studies have provided some interesting insights into the underlying genetic architecture of disease resistance traits, with implications for selective breeding.

Table 4.   QTL mapped for disease resistance traits in aquaculture species
SpeciesPathogen/parasiteQTLs found (PVE)*References
  • *

    Number of significant QTL (percentage of phenotypic variance explained, PVE).

  • Genome-wide significant QTL, one additional suggestive QTL found with 8.9% PVE.

  • Genome-wide significant QTL, 13 additional suggestive QTL found, each with <1% PVE.

  • §

    §10 individual marker trait linkages for parasite counts at different days post infection. Largest PVE is given (marker-QTL linkages on day 30).

  • ¶Causing the disease MSX.

  • ∥Causing the disease Dermo.

  • **

    ** No estimate of the amount of phenotypic variance explained by the QTL provided (NP).

  • ††

    ††Chromosome-wide significant QTL based on regression interval mapping.

  • ISAV, infectious salmon anaemia virus; IPNV, infectious pancreatic necrosis virus; QTL, quantitative trait loci.

Atlantic salmon (Salmo salar)IPNV2 (24.6%, 18.2%)Houston et al. (2008)
 1 (50.9%)Houston et al. (2009)
Atlantic salmon (Salmo salar)IPNV2 (23.4%, 0.9%)Moen et al. (2009)
Atlantic salmon (Salmo salar)ISAV1 (6%)Moen et al. (2007)
Atlantic salmon (Salmo salar)Gyrodactylus salaris parasite10 (27.3% total)§Gilbey, Verspoor, Mo, Sterud, Olstad, Hytterod, Jones and Noble (2006)
Eastern oyster (Crassostrea virginica)Perkinsus marinus parasiteand Haplosporidium nelsoni parasite12 (NP)**Yu and Guo (2006)
European flat oyster (Ostrea edulis)Bonamia ostreae parasite4 (NP)††Lallias, Gomez-Raya, Haley, Arzul, Heurtebise, Beaumont, Boudry and Lapègue (2009)
Japanese flounder (Paralichthys olivaceus)Lymphocystis disease virus1 (50%)Fuji et al. (2006)

In particular, the widely accepted assumption that disease resistance is a complex, polygenic trait has been challenged by results for IPN in Atlantic salmon. A major QTL that explains the majority of the genetic variance for resistance to IPN has been identified by two independent studies in Scottish (Houston et al. 2008) and Norwegian (Moen, Baranski, Sonesson & Kjøglum 2009) populations. In both populations, this QTL has been narrowed down to a relatively small chromosomal region and is estimated to explain 80–98% of the genetic variance for resistance, with a similar effect at both fry and post-smolt stages. Interestingly, the QTL was found to be extremely heterozygous in the Norwegian population, suggesting little directional selection at this QTL in the natural environment of their wild ancestors. This QTL is now being implemented in within-family selection in both Scotland (Houston, Haley, Hamilton, Guy, Mota-Velasco, Gheyas, Tinch, Taggart, Bron, Starkey, McAndrew, Verner-Jeffreys, Paley, Rimmer, Tew & Bishop 2009) and Norway (Moen et al. 2009). Although resistance to other diseases appears to be more polygenic, with multiple QTL detected (Table 4), evidence for a single locus explaining the majority of genetic variance has also been found in Japanese flounder for lymphocystis disease virus (Fuji, Kobayashi, Hasegawa, Coimbra, Sakamoto & Okamoto 2006). In these cases, where a single QTL explains such a large proportion of the genetic variance for resistance, it is reasonable to assume that the high-resistance allele will move towards fixation using conventional selection (Sehested & Mao 1992). The application of MAS would, however, accelerate this process substantially and allow selection on other traits without compromising genetic gain for resistance. It may be desirable in some cases to deliberately retain the low-resistance allele at low frequency in the population as a ‘safety measure’ in the case of unknown advantageous effects of this allele for other traits.

Candidate gene studies

In addition to the mapping of unknown QTL via genome scans, a number of studies have been carried out to investigate associations of polymorphisms in candidate genes for disease resistance, and the majority of these have focused on the major histocompatibility complex (MHC) gene family in salmonids. The MHC genes are known to be crucial elements of adaptive immunity and have been linked to numerous diseases in other species. However, studies on a number of pathogens in salmonids have revealed somewhat conflicting evidence for the role of MHC in resistance, with strong associations found for ISA and furunculosis (Grimholt, Larsen, Nordmo, Midtlyng, Kjoeglum, Storset, Saebø & Stet 2003; Kjøglum, Larsen, Bakke & Grimholt 2006) and IHNV (Miller, Winton, Schulze, Purcell & Ming 2004), but weaker/suggestive associations for amoebic gill disease (Wynne, Cook, Nowak & Elliott 2007), sea louse (Glover, Grimholt, Bakke, Nilsen, Storset & Skaala 2007; Gharbi, Glover, Stone, MacDonald, Matthews, Grimholt & Stear 2009) and bacterial cold-water disease (Johnson, Vallejo, Silverstein, Welch, Wiens, Hallerman & Palti 2008). It certainly seems apparent that MHC polymorphisms do not explain the majority of genetic variance for resistance for some diseases, and that other genes play a more important role, such as the case in IPN where the locus explaining nearly all the genetic variation is not one of the MHC loci. Genome-wide expression profiling using microarrays has also been used to identify specific genes involved in disease infection and resistance, such as whirling disease in rainbow trout (Baerwald, Welsh, Hedrick & May 2008) and ISAV in Atlantic salmon (Jørgensen, Afanasyev & Krasnov 2008). These studies provide greater insight into the molecular determinants, responses, pathways and interactions related to infection and resistance. While not providing direct markers that can be used in MAS, this information can nevertheless help identify key genes and potentially highlight responses that may be of important value as ‘phenotypes’ for different aspects of disease infection and resistance.

Gene expression profiling as a selection tool

Apart from the use of gene expression profiles to better understand the function and regulation of genes, Robinson, Goddard and Hayes (2008) have proposed the use of gene expression profiles in indirect selection for disease resistance. The concept is that certain cells (e.g., macrophages or leucocytes) will respond to the disease agent, and the resulting cascades of gene expression changes in these cells could potentially differ between animals that are able to resist versus those that are more susceptible to the disease. As tissue for gene expression profiling can potentially be collected from live candidates, the profiles could then be used as a selection tool. The use as a selection tool assumes that a method of eliciting a gene expression response from breeding candidates is available, e.g., by challenging cells derived from these individuals to disease. Robinson et al. (2008) evaluated a method for formulating prediction equations using random regression with cross-validation on a set of gene expression data from breast cancer patients, and found a moderate correlation between the predicted and the actual phenotype (0.32±0.06). Based on simulations using gene expression data in a selective breeding programme for Atlantic salmon, Robinson and Hayes (2008) found that use of such data in combination with sib challenge testing would improve both the accuracy and intensity of selection and would thus speed up genetic gain for disease resistance.

Currents application of MAS and future prospects

Apart from the case of IPN in Scotland (Houston et al. 2009) and Norway (Moen et al. 2009), and marker-assisted selection for a lymphocystis disease-resistant Japanese flounder in Japan (Fuji, Hasegawa, Honda, Kumasaka, Sakamoto & Okamoto 2007), there are no documented examples of MAS for disease resistance traits for aquaculture species. This is likely due to a number of factors. Firstly, the limited marker resources have typically resulted in the identification of relatively broad chromosomal regions, requiring the linkage phase of the QTL and marker alleles to be established in every family before MAS can be applied. Although this is considered to be the least efficient form of MAS (Dekkers 2004), Sonesson (2007) showed that the relative ease of producing large full-sib families in aquaculture species means that it can be a useful strategy. Nevertheless, there are considerable economic and practical downstream advantages for MAS if QTL are fine mapped to identify markers (or marker haplotypes) in population-wide association with the QTL. Such an association has been demonstrated for the previously discussed IPN QTL in Atlantic salmon. Indeed, the application of genomic selection relies on such an association. As third-generation sequencing technology approaches, promising dramatically faster and higher throughput genotyping than current-generation instrumentation, the use of genomic information for selective breeding for improved disease resistance represents a big step forward, given that this type of information can be acquired on a massive scale at an affordable cost.


Resistance against specific diseases often show moderate to high heritabilities in aquaculture species, and there is therefore a huge potential for improvement of disease resistance through genetic selection. However, genetic evaluation of disease resistance based on survival under challenge testing still faces challenges that may complicate both recording and statistical analysis of resistance, e.g., susceptibility and time until death may actually represent different aspects of disease resistance, mortality may be an inadequate measure of resistance (e.g., if problems associated with the disease is observed among the survivors) and the exclusion of surviving challenged individuals as selection candidates slows down the rate of genetic change. Genomic information can be used to overcome the latter specific problem and to improve genetic evaluations in general. Given that high-throughput genotyping will be available in the future at an affordable cost, genomic information is likely to substantially increase the efficiency of selective breeding for improved disease resistance in aquaculture species.


Jørgen Ødegård acknowledges The Scottish Aquaculture Research Forum (SARF) for the invitation as speaker to the international symposium ‘Scottish Aquaculture: A Sustainable Future’, for covering travelling costs and for the invitation to write this paper. The study was partly funded by the Norwegian Research Council project no. 192331.