Background: The current state of affairs

Hatchery fish have lower fitness than wild fish:

 Genetic change via  captive breeding has been documented in a wide variety of taxa, and it poses a serious problem for ex-situ conservation (captive breeding and release) programs (Frankham, 2008; Montgomery et al., 2010). This phenomenon is particularly well documented in salmonid fish.  Traditional hatchery stocks (those propagated for many generations using returning hatchery fish as parents) are genetically different from wild populations for a wide variety of adult and juvenile traits (reviewed in Fleming and Petersson, 2001, and in Bilby et al., 2005).  Fish from traditional hatcheries usually have much lower fitness than wild fish when they breed in the wild (often < 10% that of wild fish: reviewed in Berejikian and Ford, 2004, Araki et al., 2008, and Recovery Implementation Science Team, 2009). 

 What had not been appreciated until recently is how rapidly this fitness decline occurs.  For example, hatchery-produced Chinook salmon juveniles were less aggressive and more vulnerable to predators than wild juveniles after only a single generation in a state-of-the-art hatchery(Fritts et al., 2007; Pearsons et al., 2007).  Here the juveniles were raised in identical environments - they differed only in parentage.  Reisenbichler and McIntyre (1977) made all possible crosses among wild and hatchery steelhead trout (HxH, HxW, WxH and WxW), where the hatchery fish had been derived from the same wild population only two generations earlier.  Offspring were released into fenced off streams as eyed-embryos or fry, and recaptured 12 months later (parentage assigned using allozymes).  Juveniles produced from hatchery parents showed approximately 85-90% the survival of juveniles produced using wild parents, while hybrids were intermediate (Reisenbichler and McIntyre, 1977; Reisenbichler and Rubin, 1999).  When the same families were reared in a hatchery pond, the HxH offspring survived and grew best.

Consistent with the above results, three studies have now shown first-generation hatchery fish to have significantly lower lifetime reproductive success than wild fish (from two populations of steelhead and a population of coho).  E. Berntson and colleagues found the relative reproductive success (RRS) of first-generation steelhead in Little Sheep Creek, Oregon, to be consistently between 0.3 and 0.6 that of wild fish in six run years (E. Berntson, in preparation).  In a similar study, V. Theriault and colleagues studied three run years of coho in Calapooya creek, a tributary of the Umpqua River.  The RRS for first-generation hatchery fish averaged 0.56 to 0.77 that of wild fish (Theriault et al., unpub. data).  We found similar results for winter-run steelhead in the Hood River, as detailed below.

 We now have a three-generation pedigree of all the anadromous, winter-run steelhead (Oncorhynchus mykiss) that have spawned in the Hood River since 1991 (see study design and further details under “Major Accomplishments”). The older hatchery stocks of steelhead in the Hood River had 10-30% the fitness of wild fish when both spawn in the wild (Araki et al., 2007a).  First generation hatchery fish (i.e. fish created using wild parents as broodstock) have about 15% lower fitness than wild fish (Araki et al., 2007b).  A comparison between first and second-generation fish raised side-by-side in the same hatchery showed that each generation spent in the hatchery substantially reduced the performance in the wild of the hatchery fish (Araki et al., 2007b).  Because both types of fish experienced identical environments, and because the reciprocal crosses showed no evidence of maternal effects, the difference is almost certainly genetic.  We estimated an initial average decline in fitness of 37.5% per generation of hatchery rearing. 

 Data such as these are disturbing because hatchery and wild fish are allowed to mix freely in many rivers.  Furthermore, many hatcheries have adopted the mission of supplementation. The assumption had been that using local, wild-born fish as parents every year would circumvent the domestication effects that are so apparent in multi-generation stocks.  However, if that assumption is not true, then supplementation efforts may actually drag down the fitness of the wild stocks they are meant to support (Ford, 2002; Goodman, 2005; Lynch and O’Hely 2001).  Last year we analyzed the fitness of wild-born fish that differed in parentage, and found that wild-born offspring of naturally occurring HxH, WxH and WxW crosses also differed substantially in fitness (Araki et al., 2009).  Because the fish of different parentages presumably experienced identical environments (this time in the wild), the difference between them is probably genetically based.  Thus, these first-generation fish probably are dragging down the fitness of the wild population, although the extent of the effect still needs to be estimated accurately.

Objective I – finish and fully analyze the pedigrees

What still needs to be done?

In addition to improving our point estimates of RRS by adding more run years for the winter stock analyses, we can now test whether summer run H_new differ from wild summers. This will be an independent test of the hypothesis that first-generation fish differ from wild.  However, one interesting difference between the summer-run and winter-run data sets is that the winter population had very little influence from any old hatchery stocks over the last two decades.  The summer run fish, on the other hand, had a  large influence from the highly domesticated Skamania stock, which returned to the Hood River by the thousands until the mid-1990’s. Therefore, the wild summer-run fish probably started with a big genetic handicap (i.e., high levels of hatchery introgression), relative to the wild winters.  So one interesting prediction is that there will be less of a fitness difference between wild and H_new fish in the summer run than there was in the winter run. 

 In 1997 the last Skamania fish were passed over the dam and an H_new stock of summer run fish was started.  In 1998-2000 only wild fish were deliberately let over the dam, and in 2001 the first of the H_new summer stock began returning.  When we finish the 2009 run year of wild fish we will be able to estimate the fitnesses of each type of fish in run years 2001 to 2004 (>98% of offspring returned for RYs 01-03, and ~70% should have returned for RY 04).  One particularly interesting analysis we can do with the summer run is to estimate RRS of wild fish that differ in parentage, as we did for winter-run fish of wild and H_new ancestry (Araki et al.2009).  However, here the hatchery influence is from the old Skamania stock, not an H_new stock.  We estimate that in RY 2001 approximately 82% of the gene pool in the wild fish originated from Skamania fish.  That percentage drops to 57% in RY 2002, 21% in RY 2003 and 0% in RY 2004.  So it will be interesting to test whether the fitness of wild fish increases relative to that of the H_new stock over the 2001-2004 period because the wild population will become “purer” (i.e. a smaller fraction of wild-born individuals descended from Skamania each year).  This analysis should give us an idea of how quickly a population rebounds once the influence of an old, out-of-basin stock is removed.   

Another question  we can now address  in both summer and winter programs is the effect of having individuals of mixed ancestry in the broodstock.  In other words, although a program may use only unclipped (wild-born) fish as broodstock, the managers cannot know the proportion of hatchery ancestry in those fish without an extensive pedigree.  One prediction is that wild-born broodstock with high hatchery ancestry will perform best in the hatchery (i.e. produce more surviving hatchery offspring), but that their offspring will perform worse than other hatchery fish in the wild.  If so, this effect would exacerbate the variance in performance among families both in the wild and in the hatchery.  In other words, broodstock fish that produce the most surviving offspring also produce the worst-performing offspring in the wild – a positive feedback that would enhance the negative effect of the hatchery on the wild population.  We have two pieces of data in support of this hypothesis.  First, the ratio of variance in family size (V_k) to mean family size (k) among wild fish in the Hood River was higher than typical values reported for other salmonid populations (Araki et al., 2007c).  We attribute this high variance to the presence of fish of varying hatchery background.  Second, we recently performed  the same analysis for broodstock used in the hatchery, and found a similarly large index of variability measured at their returning adult offspring.  In fact, the ratio of effective number of breeders (N_b) to actual number of breeders (N_c, for “census”) among hatchery broodstock was ~  0.50 across each of several brood years – not much higher than what we observe in wild fish.  Furthermore, egg take per female is not correlated with the number of offspring that return.  Figures 1 and 2 illustrate the high variability among broodstock pairs in their production of returning offspring.

Figure 1.  Distribution of number of returning offspring per hatchery broodstock pair (winter-run H_new program) over nine brood years.  The distribution is as skewed as one observes in wild populations, with most pairs producing fewer than 5 returning offspring, and a few pairs producing dozens.  We suspect that selection is involved and will test whether the background of the broodstock parents explains some of this variance.

Figure 2.  Principal coordinates analysis illustrating drift in the winter-run H_new hatchery fish each year that results from the skewed (non-random) success among broodstock pairs in the hatchery.  Each point represents the allele frequencies in either the wild population (dark triangles) or in the returning hatchery fish (open triangles; several hundred fish each year) in each of 11 run years. The wild population has very constant allele frequencies from year to year, while there is a large amount of drift in the hatchery offspring each year.  Note that allele frequencies in the broodstock parents of those hatchery fish are indistinguishable from those in the wild (data not shown), so the drift results from the reproductive skew illustrated in figure 1 above.

This was a well-run, conservation hatchery program, so the variance in family sizes shouldn’t be so large.  We therefore suspect that some sort of selection is involved.  As part of our continued analysis of the pedigree, we will test whether the ancestry of the broodstock explains some of that variance, and will also test for a negative correlation between performance in the hatchery (production of returning offspring) and performance in the wild (RRS of those returning offspring).  We will also explore phenotypic correlates of those two types of performance, including body size, run timing, cross date and MHC genotype.  An added twist to the analysis is that some fish will have resident trout for parents, and so we will test whether a resident parent is essentially equivilant to a wild anadromous parent, or has other effects.

  In summary, our first objective will be to complete the 19 year pedigrees for both winter  summer run fish.  We will analyze both pedigrees to tease apart the effects of hatchery ancestry on performance in the wild and in the hatchery. At the end of the next grant cycle we will have points estimates of RRS for H_new vs. wild for multiple run years in each run, multiple estimates of the fitness of wild-born fish as a function of their ancestry (wild, Hnew, Skamania, and resident parentage), and an analysis of whether ancestry is responsible for the large skew in performance of broodstock.  At that point we will have completed the study that relied on complete samples of returning offspring each year.

 Because the Powerdale dam has been removed, all future pedigree work on the Hood River will have to use samples collected at weirs in particular tributaries in the Hood River (the ODFW will begin sampling weirs this year; R. French, pers. comm.).  Possible questions we might address with these samples include the effects of increased numbers of hatchery fish that will now be able to travel upstream (previously controlled at the dam),  including hatchery strays from other rivers (which could be substantial; Keefer et al., 2008).  We may also examine whether there is a genetic component to where fish go to spawn, and perform more in-depth analyses of interactions between anadromous and resident fish in particular tributaries.  For now, we will archive the samples until we have enough run years in hand to ask some interesting questions.

Objective II: Mechanisms – the next big question

 It is now clear that something about hatcheries causes a rapid and profound genetic change in steelhead, and probably in salmonids in general.  Unfortunately, we do not yet understand the mechanisms by which captive rearing causes such precipitous and genetically-based declines in fitness. 

 Why is it important to figure out the mechanism causing low fitness of hatchery fish? Given what we now know about the fitness of even first-generation hatchery fish, managers are now in a quandary given the competing goals of protecting native populations and insuring sufficient production for recreational and commercial fisheries.  One solution would be to physically segregate wild and hatchery fish.  This would be difficult and might require shutting down some hatcheries.  Alternately, one could move towards integrated broodstock programs and let the hatchery fish continue to mix with wild fish. The problem here is that even though first-generation fish (H_new) are much more fit than old hatchery stocks, there is much more gene flow between H_new fish and wild fish than between old stocks and wild fish.  So given they are still less fit than wild fish, the overall impact of H_new fish on wild populations could still be important (as we showed in the Hood River).  Alternately, one could deliberately select for highly domesticated hatchery stocks (for example, stocks with very early run times) on the assumption that these changes would result in minimal gene flow into the wild populations with which they mix.  Difficulties associated with this strategy include unintended selection on correlated traits and some continued gene flow into wild populations, and fish that may be of less economic and societal value.  Each of these solutions obviously has its problems. 

 One way out of this situation would be to figure out how to make hatchery fish that don’t impose a genetic load on wild populations.  Thus, identifying what traits are under selection is a critical area of research.  Hypotheses for why hatchery fish are less fit than wild fish in nature include the following:

 (1) Relaxed natural selection:  Survival from egg to smolt is usually 85-95% in hatcheries versus 1-5% in the wild (Reisenbichler et al., 2004).  Thus, relaxed purifying selection during the egg-to-smolt stage could result in the accumulation of mutations that are neutral in the hatchery but harmful in the wild (Lynch and O’Hely, 2001).  Similarly, because adults are spawned artificially in the hatchery, there can be relaxed natural selection on adult reproductive traits.  For example, hatchery breeders do not choose or compete for mates, find suitable spawning sites, defend nests, and so on.  Nevertheless, relaxed natural selection seems an unlikely explanation for such dramatic declines after just a single generation of hatchery culture unless salmon carry a very high standing genetic load at many traits (e.g. as appears to be the case in oysters; Launey and Hedgecock, 2001).  Because typical rates of mutation to deleterious mutations are around one mutation per genome per generation, and the average effect of such a mutation in the heterozygous state is around 2% (Lynch et al., 1999), it should take at least a few generations for mutational effects to show up.  There is no evidence that salmon have unusually high mutation rates (e.g. rates at neutral microsatellite loci appear to be typical; Steinberg et al., 2002).  So unless the hatchery environment somehow induces a huge increase in the deleterious mutation rate, relaxed natural selection seems an unlikely explanation for such rapid declines in the first generation of hatchery culture, at least for typical traits under stabilizing selection.  On the other hand, for unusual traits that have a high standing genetic load, a single generation of relaxed selection might have an effect.  Such traits include those under cyclical selection, such as immune relevant genes (Hedrick, 2002; Tennessen and Blouin, 2008).  Under this scenario, the population can never optimally adapt to an environment that is a constantly moving target – in this case, pathogens.  Thus, part of what we propose below is to ask whether relaxed selection on mate choice causes hatchery fish to have unfavorable genotypes at the major histocompatibility loci (MHC), which are known to be under sexual selection and to influence disease resistance in salmon (more below).  We propose estimating the fitness costs of having suboptimal MHC genotypes, and estimating the gain in hatchery fish performance that could be achieved by choosing broodstock based on MHC genotype.

(2) Strong domestication selection: This is the traditional explanation for hatchery fitness declines.  Here positive selection for adaptation to the hatchery environment comes at the expense of adaptation to the wild environment (Ford, 2002).  A theoretical analysis by Araki et al. (2008) showed that strong selection on phenotypic variation generated among fish by the hatchery is indeed sufficient to get fitness declines of the magnitude observed in the Hood River.  The fact that after just one or two generations hatchery fish perform better than wild fish in hatchery environments (Reisenbichler and Rubin, 1999) also points to selection, rather than to some generalized genomic deterioration as the source of genetic change in hatchery fish.  If we find a negative correlation between performance in the hatchery and performance in the wild (as outlined above), then that would be further support for the hypothesis of domestication selection.  What traits might be under strong directional selection remains a mystery, but a plausible case has been made for growth rate (Reisenbichler et al., 2004).  We are currently testing that hypothesis by comparing the growth of fish of hatchery and wild backgrounds raised in a common garden experiment.  Below we also propose a novel approach for identifying traits that were under selection in the hatchery: searching for genes that are differentially expressed between fish of hatchery and wild background.

 (3) More exotic mechanisms: Two intriguing possibilities are: (1) the hatchery induces a high mutation rate in the offspring (which then makes the mutation accumulation hypothesis more plausible), and (2) heritable epigenetic changes that are induced by the hatchery environment.  Epigenetic changes are mitotically or meiotically heritable changes in gene expression that result from modifications to the DNA without change in the DNA sequence (Richards, 2006; Bird, 2007; Jirtle and Skinner, 2007).  Most importantly, epigenetic changes can be transmitted to the next generation (Reik 2007; Richards, 2006; Jirtle and Skinner 2007).  It is therefore conceivable that rearing in a hatchery environment during the early part of the salmon lifecycle could cause epigenetic changes that eventually affect the fitness of those individuals and their offspring.  We recently showed that hatchery and wild adult steelhead from the Hood River do not differ in overall levels of genomic methylation (Blouin et al., 2010).   However, altered methylation of of a few DNA sites or other epigenetic processes (e.g. histone modifications) could still be important, so the epigenetics hypothesis deserves further study.  Rather than propose further studies at this time on genome-wide epigenetic differences between types of fish, we will be guided by  the results of the gene expression studies proposed below.  Epigenetic modifications exist to regulate gene expression, so it is conceivable that differential expression we observe for some genes might actually result from environmentally induced epigenetic effects rather than traditional selection.  We may return to this hypothesis in the future, depending on the outcome of the gene expression studies (see below).

 One final comment to put the work on mechanisms into context:  Whatever is going on in hatcheries almost certainly involves multiple traits, with some under directional selection and some under relaxed selection (e.g. Araki et al., 2008).  Thus, a multipronged approach to modifying hatchery practice will be necessary.  Here we are asking (1) can we identify particular traits that were the targets of directional selection, and (2) could reversing the relaxed selection on immunity genes contribute significantly as part of that multipronged approach.

Objective II, Deliverable 1. Gene expression study

Why look at gene expression?

There are two approaches to figuring out which traits are under selection.  First, one can make educated guesses and then test those candidate traits (the “top down” approach).  For example, there is a plausible argument that hatcheries select for excessively fast growth rate, and we are currently testing that hypothesis.  However, many other traits could be involved, such as adult breeding behaviors, sperm competitive abilities, adaptation to water temperatures or crowding levels that differ from those in the wild, and so on.  There may be traits we haven’t even considered.

 The second, “bottom-up”, approach to figuring out which traits are under selection is to figure out what changed in the genome of hatchery fish, and to then ask what traits those genes control.  In the first attempt of this kind, Roberge et al. (2006) used microarrays to perform gene expression profiling in two stocks of farmed Atlantic salmon (5-7 generations in captivity) and in their wild progenitor populations.  They found many differentially-expressed genes, including seven that had changed in parallel in the two farmed stocks.  Several of these genes were in growth-regulating pathways, but several were novel and may point to unknown traits under selection.  Indeed, gene expression studies frequently reveal new and unexpected genetic pathways involved in phenotypic variation (Whitehead and Crawford, 2006). 

 Finding the genes that responded to the hatchery environment  should then point us to potential traits under selection.  That knowledge will help us decide how to modify hatchery practices to reduce those selection pressures (it is also possible that certain genes are differentially expressed, not because of traditional directional selection, but because of heritable, environmentally-induced epigenetic effects.  We may return to this hypothesis in the future).

 Here we will use the gene expression profiling approach to identify potential traits that differ genetically between hatchery and wild fish.  However, we will use a new technology for gene expression analysis that is an improvement over the traditional microarray approaches.  We will use the Illumina next generation sequencing technology available at OSU (see below) to quantify expression in the entire transcriptomes of individual fry that differ only in parentage (first-generation hatchery parents versus wild parents).  We will compare fish reared in the same environment, but whose parents were either wild or first-generation hatchery fish, and do this for stocks from two independent river systems.  Response by the same genes in two independent hatchery stocks would be strong evidence for parallel evolution under similar selective pressures (Derome et al., 2006; Roberge et al., 2006). 

Gene expression analysis involves comparing the extent to which individual genes are “turned on” by counting the number of times we observe messenger RNAs (mRNA) transcribed from each gene.  Messenger RNA transcription level (“expression”) is highly correlated with protein production.  Thus, changes in transcription rate at individual genes are a major way in which differences in phenotype (traits) evolve among individuals.  Thus, it is most likely that the genetic change in hatchery fish involves increased or decreased transcription at some key genes.  So this is a reasonable  first place to look. 

 Expression profiling involves counting the number of times we see mRNA from each gene in a tissue sample.  Here we will use a new method for counting these mRNAs, which involves sequencing the transcriptome (i.e. all the mRNA in the sample) of each fish (the mRNA-seq procedure; Mortazavi et al., 2008) (http://en.wikipedia.org/wiki/RNA-Seq).  We extract total mRNA from a fish, convert it into cDNA, and then sequence that cDNA library using the new, high-throughput sequencing technologies that are now available.  The result is a digital count of the number of times each gene was represented in the complete mRNA sample.  The identity of each read (what gene it came from) is deduced by comparing your reads to a reference genome for the species or, for species with out a complete genome,to a reference transcriptome (all the known coding regions). These counts are analyzed by ANOVA with corrections for false discovery rate to decide if your types of fish (e.g. H vs. W) differ in expression for a particular locus.  OSU’s Center for Genome Research and Biocomputing has an Illumina 1G paired-end sequencing system that we use for this procedure (http://www.illumina.com/technology/mrna_seq.ilmn).  The sample preparations and bioinformatics analyses will be conducted in collaboration with my colleague at OSU, Dr. Todd Mockler, who specializes in Illumina-based expression profiling and downstream gene ontology analysis (e.g.  Filichkin et al., 2009; Fox et al., 2009) (see letter of support by Mockler in "Project Documents"). 

 The samples to use are already in hand:

We froze in liquid nitrogen the offspring from 15 HxH and 17 WxW fish families created from Hood River fish in 2009 (~12 offspring per family frozen just after yolk absorption). In 2010 we created and froze offspring from seventeen 2x2 matrices, each created by crossing an H and W male with an H and W female (siblings of fish used in the growth rate experiments described under “Accomplishments”).  The reciprocal “hybrids” in this design will allow us to separate additive genetic from non-additive and maternal effects.  These samples are all stored at -80 Celsius and await processing.  We also have similar samples and sample sizes created using wild and first-generation hatchery steelhead from the Siletz River, which we will use to see if the same selection pressures operated in independent hatcheries.

 Why look at gene expression in fry, as opposed to other life stages?

We froze the fry just after yolk absorption, when they would begin feeding.  This is a major transition period in development and a time of high mortality in the wild. Also, evidence points to differential survival in the wild among juveniles as the key life stage in which genetic differences result in fitness differences.  For example, studies have shown reduced survival of hatchery fish during the egg to parr or egg to 1-year old stages (Reisenbichler and McIntyre, 1977; Reisenbichler and Rubin, 1999).  Several studies have demonstrated behavioral differences between juveniles of hatchery and wild origin (e.g. aggression, vulnerability to predators; Fritts et al., 2007; Pearsons et al., 2007).  Therefore, we hypothesize that gene expression differences are most likely to be expressed in the early juvenile stages.  This is also the stage analyzed by Roberge et al. (2006) who found consistent gene expression differences in two stocks of farmed salmon and their wild progenitors.  In the future we may follow this study by looking at expresson differences in other life stages.

 Preliminary data and progress so far:

The genome of O. mykiss has not yet been sequenced.  Therefore, the first step in this project was to sequence Hood River fish transcriptomes very deeply and with extra-long reads, and then align those reads with the genomes of the five fish species for which we do have completed sequences.  This allows us to build a reference transcriptome for O. mykiss that we can use for the actual expression analysis of individual fish (each fish sequenced much less deeply than the ones used to build the reference).  We successfully sequenced four Hood River fish with 80 bp paired-end runs, each to a depth that would average 10X coverage of all genes assuming a ~30,000 gene genome and typical distribution expression differences among genes. We are now assembling the reference transcriptome.  However, we have recently learned that a complete genome for O. mykiss is in progress  and may be available in around a year’s time (M. Miller, Univ. Oregon, pers. comm.).  When the complete genome  is available, we can re-analyze the data we are now generating to see if we can identify more of our reads (a certain percentage are always binned as ‘unknown’) and identify more of the rarer transcripts.

 We currently have a small grant ($30,000) from the Oregon Department of Fisheries and Wildlife for supplies to complete preliminary work on this question. That money allows us to build the reference transcriptome and should provide for running one or two fish from each of 30 families (15 HxH and 15 WxW).  We just finished preparing our first 8 expression samples and will be doing preliminary runs this month to optimize the number of samples to multiplex per lane.

 Here we are requesting funding from BPA to perform this project in more depth, including both years of crosses and, eventually, more fish per family.  We also would like to test whether loci identified as differentially expressed in the Hood River also differ between fish of H and W background from the Siletz River (for this latter step we would use quantitative PCR to test the most interesting individual genes).  We expect this work to take about 2 years, and propose picking it up again in the first year of this project.  By that time we’ll have finished and analyzed the initial samples funded by ODFW, and we hope, have an even better reference transcriptome in the form of a completed O. mykiss genome.  Also, the next generation sequencing technology is advancing so rapidly that in two years we’ll probably be able to examine more individuals at lower cost.

 Current estimates of cost:

Based on our preliminary runs, we calculate the costs of reagents for sample preparation and the fees for using the Illumina to currently be approximately $500 per fish.  This includes: Illumina mRNA-seq kit (RS-100-0801) = 8 mRNA-Seq samples for $1,940.00

= $242.5 per fish for the sample prep. Paired-end 40 cycle sequencing = $1,044 per lane (includes sequencing reagents, backup tapes, central services lab fees) assuming we 4-plex in each lane = $261/fish for the sequencing.  If we do one fish per family from each of the seventeen 2x2 matrices we created in 2010, that would be 68 fish for $34,000 at today’s prices (which we, again, expect to come down substantially).  With the ODFW-funded samples from 2009, this would give us unrelated fry from a minimum of 32 HxH families and 32 from WxW families, and 17 each from HxW and WxH families.  Any strongly up or down regulated genes between fish types should be apparent with those sample sizes.

Objective II, Deliverable 2. Traits involved in relaxed natural selection – MHC?

As mentioned above, the effect of relaxed selection should not be evident after just one or two generations of hatchery culture unless the there is a high standing genetic load (many suboptimal alleles segregating) for the traits involved. Such a high genetic load seems implausible for most traits (which should be under stabilizing selection) but could be true for loci under cyclical or fluctuating selection (Hedrick, 2002; Tennessen and Blouin, 2008).  In the case of immunity genes, temporally fluctuating selection can also act in concert with traditional frequency-dependent selection, in which rarity of alleles per se is favored (Hedrick, 2002).  Now consider the choice of a small number of wild broodstock for a conservation hatchery.  If choice is random, then first, one is likely to get the most common alleles (often the least fit) in the broodstock, and second, there is no opportunity for sexual selection to sort among those alleles. All individuals get to breed, no matter their allelic status. Also, the lack of mate choice means there is no opportunity for creating the most favorable combinations of alleles in heterozygous offspring.

 Loci of the major histocompatibility complex (MHC) are an essential part of vertebrate immune systems.  They are among the most polymorphic coding genes known, and have been shown to be under strong selection in many species, including salmon and trout (reviewed in Bernatchez and Landry, 2003). It is well documented that females of many species prefer males with different MHC genotypes (e.g. Penn, 2002; Bernatchez and Landry, 2003; Mays and Hill, 2004; Ziegler et al., 2005), as heterozygous offspring tend to be fitter than homozygotes, and heterozygotes for the most dissimilar genes have the greatest fitness.  For example, non-random mating with regard to MHC genotypes has been demonstrated in Chinook salmon (Neff et al., 2008), brown trout (Forsberg et al., 2007) and Atlantic salmon (Landry et al., 2001).  It is also well established in many species that MHC genotype influences disease resistance, with many examples found in salmonid fish (reviews in Bernatchez and Landry, 2003, and Spurgin and Richardson, 2010).  For example, MHC genotype controls resistance to infectious hematopoietic necrosis virus (IHNV) in chinook salmon (Arkush et al., 2002), to furunculosis (Aeromonas salmonicida) and infectious salmon anemia virus (ISAV) in Atlantic salmon (Lohm et al., 2002; Grimholtz et al., 2003), to Pseudomonas fluorescens in whitefish (Wedekind et al., 2004), and to bacterial coldwater disease (Flavobacterium psychrophilum) in O. mykiss (Johnson et al., 2008), which is our study species.  In O. mykiss it was also observed the IHNV upregulates expression of certain MHC loci (Hansen and Patra, 2002).  Note that IHNV is a pathogen of concern in the Hood River, and one for which all broodstock are tested (C. Banner, ODFW). Lastly, MHC loci in salmon and steelhead show patterns of allele frequency divergence among populations that differ from those at neutral loci (e.g. Evans et al., 2010; Aguilar and Garza, 2006), which further suggests that these loci are under selection.

 Relaxed selection by hatcheries on which fish get to reproduce could result in major fitness effects at MHC loci.  In one compelling example from Atlantic salmon  (Consuegra and Garcia de Leaniz, 2008), MHC alleles appeared to be paired randomly in local-origin, first-generation hatchery fish, but alleles in heterozygotes were more dissimilar than expected by chance in their wild cousins (which implies disassortative mating or non-random survival with respect to MHC).  Furthermore, MHC allele dissimilarity was negatively correlated with parasite levels within each type of fish, and hatchery fish were more heavily parasitized than wild.  In another example, which used a lab-reared factorial cross among 11 male and 11 female Chinook salmon of known MHC genotype, Pitcher and Neff (2007) observed substantial non-additive effects on offspring survival and growth to 80 days post fertilization.  They estimated that pairing adults assortatively by MHC would have increased offspring performance by 6% - and this was just to the fry stage.  The payoff over the entire life cycle could be much larger.

 Thus, a great deal of evidence suggests that either letting broodstock choose their own mates (not very practical), or having managers choose broodstock based on MHC as a proxy might give an immediate payoff by creating fitter first-generation hatchery fish.  Whether that benefit would extend into the second, natural born generation depends on whether the allelic effects are mainly additive (some alleles are better than others, on average) or non-additive (it is the particular pair of alleles in a heterozygote that matters) (“good genes “ vs. “compatible genes” in the terminology of Neff and Pitcher, 2005).  Evidence for each type of effect runs about 50:50 in the literature.  For example, only additive effects of particular MHC alleles are observed in the Atlantic-salmon-furunculosis system (Lohm et al. 2002), while Consuegra and Garcia de Leaniz (2008) observed only non-additive effects.  If the effects are mainly additive, then they would indeed carry over to the next generation.  If effects are mostly non-additive, then pairing parents disassortatively by MHC would get you a boost in fitness of the first-generation hatchery fish, but any effect would vanish in the next generation as those allelic combinations are broken up.

 We propose testing whether MHC influences the fitness of hatchery and wild fish in the Hood River.  Here we want to know if there is any evidence to support the idea that choosing broodstock based on MHC and/or deliberately mating MHC-dissimilar males and females would (1) improve hatchery production (i.e. produce hatchery fish that grow better and are more likely to return as adults), and (2) produce hatchery fish that are more like wild fish (i.e. have better reproductive success in the wild, given they survive to return as adults).  A direct, experimental test of these hypotheses would take over a decade to complete. However, we have already estimated the lifetime fitness of thousands of fish of varying hatchery and wild background, including the performance of hundreds of fish as broodstock in the hatchery.  Therefore, we simply have to sequence MHC loci in a selected set of those fish to answer the following questions. 

 For fish arriving at the dam:

1) Is the RRS of steelhead in the Hood River correlated with their MHC genotypes, and if so, do the effects appear to be additive or non-additive?

 2)  Do hatchery fish wind up, on average, with less favorable MHC genotypes than wild fish?

 For matings conducted in the hatchery:

1) Did broodstock pairs having either better individual alleles (additive effect), or better matched alleles (non-additive effect) produce more returning hatchery offspring (here “better” is as inferred from 1 & 2 above)?

 2) Within full-sib families, were the survivors a non-random subset of the possible sibling genotypes?

 Direct tests of disease resistance:

We also propose doing two direct tests of the effects of MHC on disease resistance in Hood River fish.  These include:

 1) We will also sequence MHC in the fish used for the disease challenge experiments described in the “Accomplishments” section.  Here we will test if MHC influences survival differences among sibships, or among siblings of different genotype within sibships.

 2) We propose harvesting 200 returning winter-run hatchery fish and quantifying their parasite loads (we can’t use wild fish because the population is federally protected).  Then we will ask if MHC genotypes correlate with their naturally-acquired parasite load (as in Consuegra and Garcia de Leaniz, 2008).  These fish will be collected by ODFW staff at a weir that they will run at the base of the East fork of the Hood River for the purpose of keeping hatchery fish out of that fork.  The parasitology work will be done by Kym Jacobson’s lab (NOAA Fisheries, Hatfield Marine lab; e.g. Baldwin et al., 2008; Jacobson et al., 2008)(see letter of support in "Project Documents"). 

 Methods

Classical MHC loci are involved in distinguishing foreign from ‘self’ molecules, and in identifying pathogens for destruction.  Class I MHC loci recognize intracellular pathogens, while class II MHC loci recognize extracelluar pathogens.  Given that both class I and class II loci have been demonstrated to be under fluctuating and directional selection, yet target different types of pathogens, we will sequence individuals at both types of loci.  Furthermore, because there are two exons (α and β) associated with class II MHC loci that exhibit different levels of polymorphism, we will sequence individuals at both of these loci.  There are multiple loci associated with class I MHC in O. mykiss. We will focus on MHC Iα, which is most commonly associated with disease resistance (Dijkstra et al. 2007).  Importantly, the class II MHC loci have no paralogues (homologous gene copies to which the primers could bind) (Gomez et al. 2010), which means that we can reliably amplify our region of interest. 

 We propose sequencing 450 fish from the disease-challenge trails, the 200 returning adult fish collected for parasite surveys, and 2000 fish from the main pedigree (including broodstock) to answer the questions posed above.  Each will be sequenced at the MHCIα, MHCIIα and MHCIIβ loci.

Preliminary data

To validate our proposed methodology, we sequenced 30 winter-run, Hood River steelhead at the MHCIα, MHCIIα and MHCIIβ loci.  Primers for these three loci were identified from previous studies (Miller et al.1997, Gomez et al. 2010) and verified by matching primer sequences to the curated collection of O. mykiss MHC sequences found at the immuno-polymorphism database (http://www.ebi.ac.uk/ipd/mhc/fish/index.html).  All loci give clear, consistently readable electropherograms (e.g. Fig. 3), demonstrating that these loci and primer pairs will be suitable for the proposed study.  Note also that the PCR works equally well on DNA from fin snips and on our oldest scales from the early 1990’s.  Therefore we will be able to use the entire pedigree for this study.  We identified 19 polymorphic sites with 10 nonsynonymous substitutions at the MHCIIβ locus.  Interestingly, at the MHCIα locus, each sequenced individual had at least one unique allele, which is not surprising given the high polymorphism observed at this locus (Dijkstra et al. 2007).  MHCIIα was the least polymorphic locus, with 8 polymorphic sites comprising 3 unique alleles.  The methods work well in our lab and we already have  the necessary samples in hand, which makes us well prepared to complete this project.  Furthermore, as noted above, our extensive pedigree data set uniquely positions us to address these important questions. 

 Figure 3. Electropherograms of MHCIIβ sequence data from five Hood River steelhead. 

Timeline

Here we request four years of funding.  By that time we should have answered all the key questions we originally set out to answer in terms of RRS of summer and winter fish in the Hood River. For the future I think it is most interesting to focus on the mechanisms of fitness decline.  For now we have four main datasets to complete over four years: finish and analyze the summer-run fish, finish and analyze the winter-run fish, do the gene expression study, and do the MHC study.  We could start with any of them in 2012, but in order to keep the yearly budgets at about the same amount as in previous years, I propose spreading the four projects out as in the figure below.  In the “Deliverables and budget” section I assigned about a quarter of the total four-year budget to each of these four main datasets, which is probably pretty close to what each will cost.  In FY 2011, which is the last year of our current funding, we will focus on the summer-runs and the growth rate and disease challenge experiments (see “Accomplishments” for more info on those later two in progress). We will continue in the new funding cycle with the goal of completely finishing the summer run dataset in fy 2013.  Then, we will return to the winter run and finish up that analysis by 2015 (we have already published extensively on the winter run, and so want to ask the same questions on the summers before returning to the winter run).  We’ll  perform the gene expression study in 2012-2014, and the MHC study in 2014-2015.  Doing the MHC later makes sense because by then we’ll have estimated lifetime fitness for more of the summer fish. 

We have not requested a fifth year of funding at this point because it is hard to predict what the key questions will be that far out.  It largely depends on what we find out in the next few years.  For example, we hope to identify some key genes that differ in expression between fish of H and W background.  If so, we would then want to design tests of hypotheses about the traits those genes control.  We may also return to the epigenetics hypothesis.  Finally, by then we will have accumulated five years of samples from the weirs that will be operated by ODFW in the Hood River, and could ask several interesting questions using those samples (e.g. there may be a large influx of strays now that the dam is gone; Keefer et al., 2008.  What effect might they have?).