Remarkable heterogeneity revealed in the genetic architecture of resistance to a key bacterial pathogen in two commercial rainbow trout populations

Abstract

Background

Bacterial cold-water disease (BCWD), caused by Flavobacterium psychrophilum, remains a major challenge for rainbow trout (Oncorhynchus mykiss) aquaculture, due to the absence of effective vaccines and increasing concerns over antibiotic use. Genetic selection for disease resistance offers a sustainable alternative. In this study, we investigated the genetic architecture of BCWD resistance in two French commercial rainbow trout populations using a standardized waterborne infection model and high-density SNP genotyping.

Results

Survival following experimental infection varied significantly between populations, with population B showing higher resistance (71.3% vs. 50.7% of survival at 29 days post-infection). Genome-wide association studies (GWAS) were performed using a Bayesian sparse linear mixed model (BSLMM), separately in each population and in a combined dataset. Eleven quantitative trait loci (QTLs) were identified across the analyses, with limited overlap between populations, highlighting the complexity and partial divergence of resistance architectures. Several candidate genes located within QTL regions were involved in immune signalling, inflammation, macrophages/neutrophils biology, and soluble factors important for antibacterial defences. Notably, two QTLs contained genes from the complement system (e.g., C3, Cfb), highlighting their central role in resistance to F. psychrophilum.

Conclusions

Our findings underscore the polygenic nature of BCWD resistance, the influence of host genetic background, and provide valuable targets for selection for BCWD resistance in rainbow trout breeding programs.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12864-025-12231-3.

Background

Salmonid aquaculture has experienced remarkable growth over the past decades [1]; however, its expansion remains significantly hindered by infectious diseases. Although vaccination against several major bacterial diseases has been highly successful, resulting in a substantial reduction in antibiotic use [2], many infectious diseases still lack effective commercial vaccines and continue to pose serious threats to fish health. As an alternative to vaccine development, genetic selection for disease-resistant, robust, or tolerant fish has gained increasing interest. This approach has catalyzed substantial efforts to understand the genetic architecture of resistance to various fish pathogens. To date, numerous studies across a wide range of fish species have explored the genetic basis of disease resistance using genome-wide association studies (GWAS) [3].

In fish, resistance to infectious diseases generally appears to be polygenic, involving multiple genomic regions. In rare cases, pathogen resistance can be explained by few major quantitative trait loci (QTLs) - a notable example include Atlantic salmon resistance to IPNV [4–7]. However, studies rarely compare the genetic architecture of resistance across different populations. It is often assumed that the genetic architecture underlying resistance to a specific pathogen is largely conserved across populations within a species. While this may be true at the level of biological pathways or gene functions, resistance to a specific pathogen is more likely to involve a shared set of genes, however, allele frequencies at different loci are expected to diverge across populations. As a result, the genetic targets available for selection may vary between populations, even if the underlying mechanisms are conserved.

Bacterial cold-water disease (BCWD), also known as rainbow trout fry syndrome, is a devastating disease in farmed salmonids in freshwater, caused by the Gram-negative bacterium Flavobacterium psychrophilum [8–10]. F. psychrophilum outbreaks affect fish health and welfare, cause rearing issues and lead to significant economic losses worldwide. Rainbow trout is a particularly susceptible host species [11, 12]. Diseased fish show signs of tissue erosion, often involving the caudal peduncle or caudal fin, but also necrotic lesions, exophthalmia, anemia, lethargy or abnormal swimming. The bacterium primarily resides in skin lesions, surrounding muscle tissues, and is strongly associated with phagocytes in lymphoid organs, particularly the spleen, which often exhibits hemorrhages, enlargement, and high bacterial loads [12]. In rainbow trout fry, the infection results in septicaemia leading to extremely rapid high mortality [13–15]. Although recent developments are ongoing, there is still no efficacious commercial vaccine against F. psychrophilum [2, 16]. The main approach to infection control relies on improved rearing practices [17, 18] and antibiotic treatment; however, the latter raises concerns regarding the development of antibiotic resistance [9, 19, 20]. Additionally, the dissemination of the causative agent through global fish and egg trade is suspected to further exacerbate the issue [21, 22].

Genetic resistance to F. psychrophilum infections has been established with moderate heritability estimated in rainbow trout breeding populations exposed to the pathogen by injection [23–26], by immersion of hydrogen peroxide-treated fish [27] or during natural outbreaks [28]. Selective breeding has been shown to increase rainbow trout resistance to BCWD [24, 29, 30], making it an attractive strategy to reduce the impact of outbreaks. Nevertheless, in most studies, resistance was evaluated using experimental infection models that bypass the external barriers (e.g. skin, mucus) [31, 32]. These methods reflect only partially the natural physiopathology. The influence of the infection route was substantial in a previous study [33] where we compared genetic traits associated with BCWD resistance following experimental infections via injection or immersion in experimental rainbow trout lines. The detection of QTL aims to the identification of genetic markers that are associated with BCWD resistance, which can then be used to improve the accuracy of either pedigree selection [34] or genomic selection [35] or to provide genetically improved eggs and fry to multipliers and growers through marker-assisted selection [36]. Moreover, identifying candidate genes within QTL regions helps to better understand the mechanisms involved in BCWD resistance. Previous GWAS for BCWD resistance in various rainbow trout populations have identified QTLs that are associated with survival to F. psychrophilum infection. However, confidence intervals for these QTL were broad, due to the low density of genetic markers used (e.g [24–28, 33, 37, 38]). With the recent development of a new 665 K SNP genotyping array for rainbow trout [39], the detection and mapping of QTLs can be now much precise.

In this context, this study aimed to characterize and compare the genetic architecture of host response to F. psychrophilum infection in two French commercial populations of rainbow trout using a waterborne experimental infection model by taking advantage of recent advances in genomic tools available for this species.

Methods

Bacterial strain and culture conditions

F. psychrophilum strain FRGDSA 1882/11 was grown aerobically at 18 °C in tryptone yeast extract salts (TYES) broth [0.4% (w/v) tryptone, 0.04% yeast extract, 0.05% (w/v) MgSO₄ 7H₂O, 0.02% (w/v) CaCl₂ 2H₂O, 0.05% (w/v) D-glucose, pH 7.2] or on TYES agar supplemented with 5% fetal calf serum (FCS). The strain was streaked on TYES FCS agar from − 80 °C freezer stock, and plates were incubated 4 days at 18 °C. The pre-cultures were prepared by inoculation of one colony into 5 ml of TYES broth and grown during 36 h. Cultures were performed by dilution of the pre-culture in Erlenmeyer flasks in TYES broth at an initial OD at 600 nm of 0.02 measured using a Biochrom WPA CO8000 spectrophotometer. Broth cultures were incubated at 18 °C with shaking at 200 rpm. Late-exponential phase cultures (OD of 1.0 ± 0.1 equivalent to 10⁹ CFU mL⁻¹) were used for infection challenges. Bacterial concentration was determined by colony forming unit (CFU) counting after plating 10-fold serial dilutions of the culture on TYES FCS agar and incubation 4 days at 18 °C.

Production of the fish

Fish were derived from two French commercial populations from two breeding companies, namely Les Aquaculteurs Bretons (Pop. A) and Viviers de Sarrance (Pop. B), performing multi-traits selection combining mass selection of type PROSPER [40] for growth and improvement in gutted carcass yield measured by ultrasound tomography [41]. Additionally, reproductive traits were also selected for population B [42].

In November 2021, two-year-old females (n = 119 and n = 98 for populations A and B, respectively) were mated with neomales (sex-reversed XX females used as sires; n = 100 and n = 92 for populations A and B, respectively) in independent full-factorial blocks of ~ 10 sires mated with ~ 10 dams. Fin samples were collected from all parents and stored in 99% ethanol for later genotyping. The fertilized eggs were incubated in routine conditions in cylindrical-conical incubators at 9.5 °C. At 29 days post-fertilisation (dpf), eyed-stage eggs were transported into the INRAE-IERP experimental unit (Jouy-en-Josas, France) and disinfected (Romeiode^®). The absence of F. psychrophilum was checked both before and after disinfection, as well as after hatching, as follows: Egg lysates and whole-body homogenates were prepared using a FastPrep apparatus (MP Biomedicals), streaked on TYES FCS agar, then the plates were examined for bacterial colonies after incubation at 18 °C for 4 days. A PCR-based identification of F. psychrophilum was conducted for each yellow colonies as described by Rochat et al. [43] to confirm the absence of the pathogen, and overall health status of the fish. Fish were reared in the BSL2 zone at 10 °C in 300-L tanks with UV-treated and dechlorinated tap water (1 renewal per hour). Commercial fish feed (@Le Gouessant) was delivered ad libitum, 4 to 6 times per day.

Rainbow trout experimental infections

Three experimental infections were performed for each commercial population (Table 1). Trials I and II were conducted on a small subset of fish to evaluate the BCWD resistance of each population after infection by bath (Trial I) or injection (Trial II). Trial III was conducted by bath on the large cohort (1200 fish per population) to produce data for GWAS analyses. Each trial was done concomitantly on both populations using the same bacterial culture, but fish were maintained in separated tanks. For each trial, uninfected fish from the same batch were maintained in the facility under the same environmental conditions to serve as controls, and no mortality or clinical signs of disease were observed in these groups.

Table 1.

BCWD experimental infections parameters

Trials	Infection route	Population	Bacterial dilutiona	Infectious dosesb				Number of fishc
				T0		T24
I	Bath	A	1/200	8.5 × 106	7.7 × 106	1.0 × 107	9.4 × 106	60
			1/1000	1.8 × 106	2.5 × 106	1.1 × 106	1.1 × 106	60
		B	1/200	8.6 × 106	7.7 × 106	1.2 × 107	9.0 × 106	60
			1/1000	1.5 × 106	1.9 × 106	5.5 × 105	3.0 × 105	60
II	Injection	A and B	-	5 doses from 1.0 × 102 to 1.0 × 106				20
III	Bath	A	1/1000	1.3 × 106		8.0 × 106		1200
		B	1/1000	1.0 × 106		1.0 × 107		1200

^a For bath challenges, fold dilution of the bacterial culture in aquarium water

^b For Trials I and III, bacterial concentration is indicated as CFU/mL of water for each tank at the beginning (T0) and the end of immersion (T24); for Trial II, 10-fold serial dilutions of culture used to infect 5 groups of fish by intramuscular injection and the bacterial doses are indicated as CFU/fish

^c Total number of fish for each dose-group

For Trial I, 150 fish (average body weight of 0.8 g) of each population were distributed into 5 aquariums of 15-L (30 fish/tank), supplied with flow-through water at 10 °C at 1 renewal per hour. Fish were fasted for 48 h prior infection, then were exposed to bacteria by immersion. Two infectious doses were tested using 2 duplicated aquaria/dose; sterile TYES was used for the mock-infected control group. Biological independent bacterial cultures were prepared and diluted 100- or 1000-fold into 10 L of aquarium water. Bacteria were maintained in contact with fish for 24 h, with the water flow interrupted during the exposure period. Throughout the experiments, water was maintained at 10 °C under continuous aeration (O₂ > 8 mg L⁻¹), and physico-chemical parameters (temperature, NH₄⁺, pH) and bacterial concentrations were monitored immediately after the beginning and at the end of bacterial exposure (see Supplementary file 1, Table S1). After complete water flush, fish were maintained in flow-through water at 10 °C (1 renewal per hour), observations of endpoints (scoring grid) and mortality was recorded twice a day for 29 days. Dead fish from each group were examined for the presence of F. psychrophilum in the spleen by plating tissue homogenates on TYES FCS agar and visually inspecting the appearance of bacterial yellow colonies. The Kaplan-Meier method was used to draw survival curves for each group of fish using combined data from the 2 replicates.

For Trial II, 5 groups of 20 fish of each population (average body weight of 5.3 g) were challenged at different doses, as follows: 5 suspensions were prepared by 10-fold serial dilutions (10⁻² to 10⁻⁶) of a bacterial culture in TYES, fish were anesthetized in 100 mg/l tricaine (MS222, Sigma), then 50 µl of the suspension was injected in the dorsal muscle. The median lethal dose (LD50) of strain FRGDSA 1882/11 was estimated for each rainbow trout population using the Reed and Muench method with cumulative mortality rates at 14 days post-infection (dpi). The mean death time was calculated at each dose by summing the times of death for all fish and dividing by the total number of dead fish. Curves were drawn by plotting mean death time as function of log dose (CFU/fish).

Trial III was conducted after 4 months of rearing in flow water at 10 °C, at an average weight of 3.9 g for population A and 3.3 g for population B. After 48 h of acclimation (in the same aquaculture system) in two 300-L tanks (one tank per population), 1200 fish were infected by immersion as described for Trial I. Briefly, 300 mL of bacterial culture was 1000-fold diluted in the water of each tank, and fish were maintained in contact with bacteria during 24 h. Bacterial concentrations (Table 1) and physico-chemical parameters (see Supplementary file 1, Table S1) were monitored. After water flush, fish were kept at 10 °C with water flow (1 renewal per hour), observations of endpoints (scoring grid) and mortality was monitored twice a day during 29 days. The Kaplan-Meier method was used to estimate the probability of death. Dead fish from each tank were examined for the presence of F. psychrophilum in spleen. At 29 dpi, surviving fish were euthanized by anaesthetic overdose (300 mg/L of tricaine). A piece of caudal fin was clipped from all individuals at the time of death/euthanasia, identified by tube barcoding and stored in 100% ethanol for DNA extraction.

Genotyping

The collected fin samples from the 100 sires, the 119 dams, the 1148 offspring of population A, and the 92 sires, the 98 dams, the 1038 offspring of population B were genotyped on the INRAE genotyping platform Gentyane (Clermont-Ferrand, France). All the offspring of population A and population B, and some parents (population A: 30 dams and 94 sires, population B: 4 dams and 92 sires) were genotyped for 57,501 SNPs (medium-density (MD) genotypes) with the 57 K SNP AxiomTM Trout Genotyping array from Thermo Fisher [44]. Most of the dams from population B (n = 94) and 89 dams and 6 sires from population A were genotyped for 664,531 SNPs (high-density (HD) genotypes) with the 665 K SNP AxiomTM Trout Genotyping array from Thermo Fisher [39]. Preliminary quality controls were applied to select SNPs that were identified as polymorphic in each population by the Axiom Analysis Suite software with cut-off values of 95% for SNP calling rate and 90% for sample calling rate. SNPs with probe polymorphisms and multiple locations on the Arlee strain rainbow trout genome assembly (accession number: GCA_013265735.3 USDA_OmykA_1.1; [45]) were also discarded, as in Bernard et al. [39]. Two other filtering steps were applied to the genotypes of challenged individuals from populations A and B with the PLINK software (v.1.9) [46, 47], to remove (1) SNPs with a MAF less than 5% and (2) SNPs with a p-value for the Hardy–Weinberg test less than the 10⁻⁶ threshold. Parentage assignment was done using the R package APIS [48], using 1000 SNPs from the MD chip with the highest MAF with the positive assignment error rate set to 5%. Parentage assignment reached 99% for population A and 91% for population B. In population A, the mean number of phenotyped and genotyped progenies per sire, per dam, and per full-sib family were respectively 11.4 ± 5.7 [min-max: 1–26], 9.6 ± 4.6 [min-max: 1–23], and 1.6 ± 0.8 [min-max: 1–5] (unassigned individuals excluded). In population B, the mean number of phenotyped and genotyped progeny per sire, per dam, and per full-sib family were respectively 10.3 ± 3.4 [min-max: 3–19], 9.7 ± 4.2 [min-max: 1–24], and 1.7 ± 0.9 [min-max: 1–5] (unassigned individuals excluded).

Based on the quality-filtered genotypes and pedigree information, the FImpute3 software [49] was used to impute missing genotypes for both the offspring and parents, and to impute the offspring’s MD genotypes to HD based on the parent reference HD genotypes.

Ultimately, following the quality control procedures and imputation, 1148 offspring with genotypes for 404,429 SNPs were retained for population A. Similarly, 1038 offspring with genotypes for 402,713 SNPs were retained for population B for further analysis.

Bayesian sparse linear mixed model for genome-wide association studies

The GWAS were conducted using a Bayesian sparse linear mixed model (BSLMM) that assumes that each of the SNPs has at least a relatively small effect but some sparse SNPs may have an additional large effect [50]. Therefore, BSLMM is capable of adapting to different genetic architectures of the studied trait, from the infinitesimal polygenic model to a model that assumes that only a very small proportion of all variants affect the phenotype. The BSLMM was fitted separately for each population (hereinafter referred to as GWAS ‘Pop. A’ or GWAS ‘Pop. B’) considering the survival status at the end of the challenge (29 dpi), and also at 10 dpi for population A (hereinafter referred to as GWAS ‘Pop. A 10 dpi’). We performed this analysis exclusively for population A because it is the only population in which we observed a continuous decrease in survival after 10 dpi. In addition, to capture shared QTLs across the two populations, a combined analysis of the two populations (hereinafter referred to as GWAS Comb. ‘A + B’). To account for systematic differences between populations, phenotypes were pre-corrected for the population mean effect estimated from a linear mixed model, which is mathematically equivalent to including population as a fixed effect in the GWAS model. This approach allowed us to test for loci influencing survival across both populations under a unified framework. In each case, a threshold model was fitted to describe the survival status (dead or alive) of individuals. The observed status can be interpreted as an expression of a non-observable, gaussian phenotype (called liability) below or above a threshold that is derived according to the probability of observing the status dead versus alive in the real world [51]. The fitted model under the underlying liability scale was:

where is the n-vector of liability phenotypes, is an n-vector of 1 s, µ is a scalar representing the phenotype mean, X is an n x p matrix of genotypes measured on n individuals at p SNPs, β is the corresponding p-vector of the SNP effects; is a vector of random additive genetic effects that are assumed distributed according to ), where the additive genetic variance and is the genomic relationship matrix; and is a n-vector of residuals that are assumed distributed as ), where is the variance of the residual errors. Assuming /p, the SNP effects can be decomposed into two parts: a p-vector α, which captures the small effects that all SNPs have, and the p-vector β, which captures the additional effects of some large-effect SNPs. In the above model, can be viewed as the combined effect of all small-effect SNPs, and the total effect size for a SNP is . The individual SNP effects were sampled from a mixture of two normal distributions, where is the variance of the small SNP effects, is the variance associated with large SNP effects, and is the proportion of sparse SNPs with large effects.

There are two important parameters that are derived from BSLMM. The first one, PVE, corresponds to the genomic heritability of the liability and is defined as the proportion of the phenotypic variance explained by the genetic variance (i.e.

The second interesting parameter, PGE, reflects how well one could predict phenotype using only the sparse large effects, and is derived as the proportion of the genetic variance explained by the variance of the sparse large effects terms:

The BSLMM was implemented using the genome-wide efficient mixed model association (GEMMA) software [52]. In GEMMA, we applied a threshold BSLMM model using a Markov chain Monte Carlo (MCMC) method. The MCMC algorithm was run with 10 million cycles and a burn-in period of 500,000 cycles. Results were saved every 100 cycles for further analysis. To ensure convergence of the distribution of the hyper-parameter , the minimum and maximum numbers of SNPs that were sampled to be included in the model with large effects were set to 1 and 50, respectively. These values were based on an initial run with default values that indicated that the median number of SNPs that had a large effect was ranged between 10 and 20.

QTL identification and annotation

The BSLMM uses a MCMC algorithm for sampling from the posterior distribution to obtain all parameters values, including SNP effect estimates , the hyper-parameter , and a posterior inclusion probability (PIP) for each SNP that indicates the proportion of samples in which that SNP was classified as having a large effect. The Bayes factor (BF) was calculated to quantify the degree of association between a SNP and the survival phenotype [53]:

As proposed by Kass and Raftery [54], the transformation of the BF (hereafter named logBF) was computed as twice the natural logarithm of the BF to produce values within the same range as the usual likelihood ratio test values, thus facilitating the determination of thresholds to declare QTL. Strong evidence for a QTL was provided by a value logBF ≥ 15 at a SNP position. A credibility interval for the QTL was computed around the peak SNP including all SNPs for which logBF ≥ 6 in a sliding window of 250 kb.

Each GWAS was repeated 10 times using different random seeds for each dataset, and the resulting lists of significant SNPs were compared (see Supplementary file 2, Tables S2 and S3). We considered only the QTLs that were detected with strong evidence in at least 6 out of 10 seeds for a given dataset. To define the list of associated genes, we determined the positions of the selected QTLs by taking the minimum and maximum positions of the credibility intervals, considering all seeds. Genes within QTL credibility intervals (see Supplementary file 3, Table S4) were annotated with the NCBI O. mykiss Arlee genome assembly (GCA_013265735.3 USDA_OmykA_1.1; [45]).

Results

BCWD resistance phenotypes

The BCWD resistance of the two populations was assessed using experimental infections on a small subset of fish using F. psychrophilum FRGDSA 1882/11, a virulent strain isolated in France from diseased rainbow trout and belonging to the clonal complex CC-ST90 (ST108) [22, 43]. The two populations varied in their responses to bath infection. For population A, survival rates differed depending on the infectious dose (p = 0.0027), while no dose effect was observed for population B (p >0.05, Fig. 1A). Fish exposure to an initial bacterial concentration of ~ 2 × 10⁶ CFU/mL resulted in final survival rates of 53.3% and 61.7% for population A and population B, respectively, with mortality starting at 8 days post-infection.

Fig. 1.

BCWD resistance phenotypes of the two rainbow trout populations following infection by F. psychrophilum FRGDSA 1882/11. A Kaplan-Meier survival curves of rainbow trout fry (average body weight of 0.8 g) infected via immersion at two bacterial concentrations (Trial I); experimental infectious doses are indicated in Table 1; colored areas correspond to 95% confidence intervals. B Comparison of death time values following intramuscular injection of various infectious doses in the two populations: y axis, mean death time (days); x axis, log dose indicated as CFU/fish (Trial II). Dashed lines fitted mean death time values using semilog nonlinear regression function of the GraphPad Prism 8 software

In contrast to the bath infection trial, groups infected by intramuscular injection experienced high mortality (100% after injection with 10⁶ CFU; Supplementary file 1, Fig. S1), highlighting the critical role of external defence barriers and bacterial entry in BCWD resistance. No difference was observed in the median lethal dose (LD50) which was estimated to be 5.5 × 10² and 4.0 × 10² CFU for population A and B, respectively. The kinetic of response to infection was slightly faster in population A, with lower time to death values for all dose groups tested (Fig. 1B), indicating that bacteria might be better controlled in population B.

The experimental infection conducted for QTL discovery (Trial III) used the bath infection model with 1200 offspring for each population (Table 1; see Supplementary file 1, Table S1). The presence of F. psychrophilum was confirmed on dead fish during the infection, validating BCWD-related deaths. The peak of mortality was observed at 9 dpi for the two populations, and survival rates at 29 dpi were 50.7% for population A and 71.3% for population B (Fig. 2). Interestingly, the survival dynamics following the mortality peak varied between the two populations. In population A, survival gradually declined from 77.9% at 10 dpi to 50.7% at 29 dpi. Conversely, low mortality was observed after day 9 in population B, with survival rates of 77.6% and 71.3% at 10 and 29 dpi, respectively. This observation is also in favour of a lower susceptibility of population B.

Fig. 2.

BCWD challenge conducted for QTL detection in the two French rainbow trout populations. Kaplan-Meier survival curves of fish infected by immersion (Trial III). Bacterial doses and water quality parameters are indicated in Table 1 and Supplementary file 1, Table S1, respectively. Average body weight was 3.9 g for population A and 3.3 g for population B. Mortality and morbidity events were recorded over 29 days following bacterial exposure. Colored areas correspond to 95% confidence intervals

Genetic architecture of host response to F. psychrophilum infection

The estimates of the genomic heritability of survival (i.e. status at 29 dpi) on the underlying liability scale (PVE) were 0.52 (CI: 0.42–0.61) in population A and 0.65 (CI: 0.53–0.76) in population B. The estimated proportion of genetic variance explained by the large-effect SNPs (PGE) was ~ 58% for population A and ~ 74% for population B. The BSLMM estimated an average of 17 (CI: 4–39) and 12 (CI: 8–19) SNPs with large effects (n.γ) for populations A and B, respectively, suggesting that only a few loci have substantial contributions to the genetic variance (Table 2).

Table 2.

Average estimates of the number of SNPs with large effects (n.γ), the proportion of genetic variance explained by the large-effect SNPs (PGE) and the genomic heritability on the underlying liability scale (PVE) from 10 GWAS using probit BSLMM runs with 10 million cycles considering the survival status at 29 dpi (95% confidence intervals in brackets)

Population	n.γ	PGE	PVE
A	17 (4–39)	0.58 (0.36–0.85)	0.52 (0.42–0.61)
B	12 (8–19)	0.74 (0.57–0.89)	0.65 (0.53–0.76)

Four GWAS were performed to detect QTLs for BCWD resistance using various phenotypic datasets, as follows: the survival status at the end of the challenge (29 dpi) for each population A and B (GWAS ‘Pop. A’ and ‘Pop. B’), or the survival status of the two of them combined to capture shared QTL and refine the QTL regions (GWAS ‘Comb. A + B’); finally, the survival status at 10 dpi for population A (GWAS ‘Pop. A 10 dpi’) to investigate QTL associated with the mortality peak only.

A total of 37 regions were detected at least once (see Supplementary file 2, Tables S2 and S3), of which 30 were specific of a single GWAS (Supplementary file 1, Fig. S2). Of the detected QTLs, only 11 showed high occurrence (i.e. detected in at least 6 out of 10 seeds for a given GWAS) and were retained for downstream analysis (Table 3). The GWAS ‘Pop. A’ identified 3 QTL regions in Omy3, Omy17 and Omy29 with a large effect on BCWD resistance (Fig. 3A, Table 3). Another QTL located in Omy3 was identified in this population using the survival status at 10 dpi (GWAS ‘Pop. A 10 dpi’; Fig. 3B). This QTL was located in close proximity to the one identified on the same chromosome for survival at 29 dpi. A total of 6 QTLs were identified for population B in Omy1, Omy11, Omy12 and Omy24 (GWAS ‘Pop. B’; Fig. 3C). No overlapping QTLs were found among these 3 GWAS. By combining the two datasets in GWAS ‘Comb. A + B’, 4 QTLs were detected, 3 of which also identified in GWAS ‘Pop. A’ (Table 3). The additional QTL was located in Omy31 (corresponding to Omy25 in the Swanson rainbow trout genome assembly; Fig. 3D).

Table 3.

Location of the QTLs detected for BCWD resistance assessed as the survival status of the fish in the two French rainbow trout populations in the 4 GWAS

GWAS	QTL_IDa	Occur. (%)b	Chr.	QTL start (Mb)c	QTL end (Mb)c	logBFd	Number of genes
Pop. A	Omy3_54.387	100	3	54.270712	54.862004	18.68–19.62	9
	Omy17_32.578	100	17	32.575001	32.626696	15.16–16.35	2
	Omy29_33.441	90	29	32.952883	34.405456	15.05–15.57	40
Pop. A 10 dpi	Omy3_53.505	100	3	53.465989	53.98674	17.26–18.50	13
Pop. B	Omy1_24.646	100	1	24.572341	25.3825	≥ 22.07	38
	Omy1_58.069	70	1	57.293434	58.069571	15.71–20.46	5
	Omy1_59.302	80	1	58.80904	59.411582	15.32–29.06	13
	Omy11_43.809	60	11	43.804699	44.070523	19.22–19.88	1
	Omy12_31.030	60	12	30.649112	31.030327	17.87–24.52	12
	Omy24_22.618	60	24	22.267241	22.627181	16.95–20.36	3
Comb. A + B	Omy3_54.387	100	3	54.053907	54.685772	18.25–20.11	13
	Omy17_32.578	100	17	32.123483	32.626696	16.10–18.08	25
	Omy29_33.441	100	29	33.434218	34.453517	17.91–20.14	29
	Omy31_25.233	100	31	24.852072	25.233141	16.28–19.42	11

^a names indicate the genomic position of the more frequent SNP pic (Mb) among the 10 runs of each GWAS using different random seeds. For QTLs identified in both ‘Pop. A’ and ‘Comb. A + B’ analyses with different SNP pics, the name was set according to GWAS ‘Pop. A’ results. Omy31 corresponds to Omy25 in the Swanson genome assembly

^b QTL detection occurrence in percent for the 10 runs of each GWAS using different random seeds

^c start and end positions in the Arlee line genome assembly GCA_013265735.3 are min and max values resulting from the 10 runs of each GWAS using different random seeds

^d logBF intervals are determined by the min and max values resulting from the 10 runs of each GWAS using different random seeds

Fig. 3.

Representative Manhattan plots of QTLs detected for BCWD resistance trait in the two rainbow trout populations from the 4 GWAS. In (A) ‘Pop. A’, survival status at 29 dpi for population A; and (B) ‘Pop. A 10 dpi’, survival status at 10 dpi for population A; (C) ‘Pop. B’, survival status at 29 dpi for population B; (D) ‘Comb. A + B’, combined survival data at 29 dpi. The red horizontal line corresponds to the QTL evidence threshold (logBF ≥ 15)

Effects of QTL on BCWD resistance and comparison between the two populations

In population A, the overall survival post-challenge was 50.7%, while individuals with the most advantageous homozygous genotypes at peak SNPs showed higher survival rates, ranging from 66.1% to 77.3%. In contrast, individuals carrying homozygous genotypes deemed unfavourable at these SNPs exhibited markedly lower survival rates, specifically between 37.4% and 37.9% (Fig. 4). The peak SNPs were associated with relatively high minor allele frequencies (MAF) between 0.35 and 0.45, indicating that beneficial alleles were relatively common within the population (Table 4). Survival rates for heterozygous individuals at these SNPs ranged from 46.4% to 57.9%.

Fig. 4.

Phenotypic effets of QTLs at peak SNPs in the two populations. Barplots present the percentage of survival at one SNP peak for 3 sub-groups of genotypes (homozygotes “A/A” and “B/B” and heterozygotes “A/B”). The width of the bars is proportional to the number of individuals for a given genotype. Values for all peak SNPs are given in Table 4. The x-axis is set at the average survival value of the overall population. The phenotypes for the effect are represented for the genotype at the following SNP peaks: GWAS Pop. A, Omy3_54.387 (Affx-1237719653), Omy17_32.578 (Affx-1237476495), Omy29_33.441 (Affx-88914235); GWAS Pop. A 10 dpi, Omy3_53.505 (Affx-1237709824); GWAS Pop. B, Omy1_24.646 (Affx-88955346), Omy1_58.069 (Affx-1237304154), Omy1_59.302 (Affx-88929515), Omy11_43.809 (Affx-1237377317), Omy12_31.030 (Affx-88937839), Omy24_22.618 (Affx-1248569909); GWAS Comb. A + B, Omy3_54.387 (Affx-1237719653), Omy17_32.578 (Affx-1237476544), Omy29_33.441 (Affx-1237691146), Omy31_25.233 (Affx-1237631285)

Table 4.

Effects of the QTLs detected for BCWD resistance in the two French rainbow trout populations

GWAS	QTL_ID	peak SNP_ID	Allele		Position (Mb)	Population	MAF (%)	Number of fish			Survival (%)
			A	B				A/A	A/B	B/B	A/A	A/B	B/B
Pop. A	Omy3_54.387	Affx-1237707186	A	C	54.387425	A	0.45	344	584	220	37.79	48.29	77.27
		Affx-1237719653	C	T	54.440401	A	0.45	220	584	344	77.27	48.29	37.79
	Omy17_32.578	Affx-1237476495	C	T	32.578302	A	0.35	118	568	462	66.10	57.92	37.88
	Omy29_33.441	Affx-88914235	A	G	33.434218	A	0.45	348	571	229	66.38	46.41	37.55
		Affx-1237700882	A	C	33.441494	A	0.45	346	572	230	66.47	46.50	37.39
Pop. A 10 dpi	Omy3_53.505	Affx-1237709824	C	A	53.505016	A	0.45	203	613	332	88.18	79.93	67.77
Pop. B	Omy1_24.646	Affx-88955346	A	C	24.64612	B	0.11	827	194	17	68.32	81.44	100.00
		Affx-1237315044	A	G	24.6648	B	0.15	760	243	35	66.97	81.48	94.29
		Affx-1237318664	T	C	24.821422	B	0.14	764	252	22	72.25	67.86	77.27
	Omy1_58.069	Affx-1237303793	A	G	57.358354	B	0.46	300	526	212	69.33	72.24	71.70
		Affx-1237303799	G	A	57.384799	B	0.46	212	526	300	71.70	72.24	69.33
		Affx-1237304154	A	G	57.523824	B	0.49	239	532	267	68.62	71.62	73.03
		Affx-1237304167	G	A	58.069571	B	0.49	267	532	239	73.03	71.62	68.62
	Omy1_59.302	Affx-1237313868	A	G	58.991253	B	0.49	240	531	267	68.75	71.56	73.03
		Affx-1237313875	A	C	58.997236	B	0.49	240	531	267	68.75	71.56	73.03
		Affx-1237317094	A	G	59.302668	B	0.44	308	543	187	67.53	70.17	80.75
		Affx-88929515	T	C	59.311866	B	0.5	242	558	238	67.77	68.46	81.51
	Omy11_43.809	Affx-1237377317	T	C	43.809155	B	0.42	333	540	165	69.97	65.56	92.73
		Affx-88956542	A	G	43.818821	B	0.42	333	540	165	69.97	65.56	92.73
		Affx-1237377318	C	A	43.821986	B	0.42	165	540	333	92.73	65.56	69.97
	Omy12_31.030	Affx-88937839	G	T	30.649112	B	0.44	315	533	190	67.94	69.79	81.05
		Affx-88910175	T	G	31.030327	B	0.47	210	566	262	73.81	66.78	79.01
	Omy24_22.618	Affx-1248569909	A	G	22.566388	B	0.07	901	133	4	71.48	70.68	50.00
		Affx-1248569917	G	A	22.59737	B	0.07	4	133	901	50.00	70.68	71.48
		Affx-1248569920	C	A	22.618322	B	0.07	4	135	899	50.00	71.11	71.41
		Affx-1237626334	T	C	22.627181	B	0.07	901	133	4	71.48	70.68	50.00
Comb. A+B	Omy3_54.387	Affx-1237712934	G	T	54.270712	A	0.38	171	528	449	81.87	48.67	41.20
						B	0.22	632	348	58	72.47	70.40	63.79
		Affx-1237719653	C	T	54.440401	A	0.45	220	584	344	77.27	48.29	37.79
						B	0.36	428	478	132	71.73	73.01	63.64
		Affx-1237722563	G	T	54.591781	A	0.45	218	589	341	77.06	48.39	37.83
						B	0.44	336	495	207	72.32	73.94	63.29
	Omy17_32.578	Affx-1237476538	A	C	32.617741	A	0.38	423	572	153	36.64	57.69	63.40
						B	0.47	295	505	238	62.71	73.47	77.31
		Affx-1237476542	G	A	32.625675	A	0.38	153	572	423	63.40	57.69	36.64
						B	0.47	238	505	295	77.31	73.47	62.71
		Affx-1237476543	G	T	32.625952	A	0.38	153	572	423	63.40	57.69	36.64
						B	0.47	238	505	295	77.31	73.47	62.71
		Affx-1237476544	A	G	32.626696	A	0.38	423	572	153	36.64	57.69	63.40
						B	0.47	295	505	238	62.71	73.47	77.31
	Omy29_33.441	Affx-1237691146	T	G	34.136594	A	0.45	343	578	227	65.89	46.54	38.33
						B	0.46	222	511	305	83.78	71.82	61.31
		Affx-1237691151	G	A	34.15054	A	0.45	227	578	343	38.33	46.54	65.89
						B	0.46	305	511	222	61.31	71.82	83.78
	Omy31_25.233	Affx-1237631285	G	A	25.233141	A	0.46	251	551	346	60.16	53.54	39.31
						B	0.46	312	505	221	77.56	70.10	65.16

This pattern was also evident when examining early survival rates at 10 dpi. Heterozygous individuals at the peak SNP of the QTL detected on Omy3 had a survival rate of 79.9%, closely mirroring the population average of 77.6%. In contrast, those with the favourable homozygous genotype at this SNP exhibited a higher survival rate of 88.2%.

In population B, the MAF for peak SNPs at QTLs identified in GWAS ‘Pop. B’ were generally between 0.42 and 0.49, except for the QTLs Omy1_24.646 and Omy24_22.618, where MAFs ranged from 0.07 to 0.15. Except for Omy1_24.646, survival rates for heterozygotes at most peak SNPs were generally near the population mean (71.3%), with values between 65.6% and 72.2%, while favourable homozygous genotypes led to the highest observed survival rates, reaching up to 100% (Fig. 4; Table 4).

Survival rates of each genotype at peak SNPs in the QTLs identified in the combined analysis (GWAS ‘Comb. A + B’) were analysed separately in each population (Fig. 4). The peak SNPs associated to QTL detected in both the combined ‘Comb. A + B’ and ‘Pop. A’ GWAS showed consistent trends on survival rates in population A. In population B, at these peak SNPs, individuals carrying unfavorable homozygous genotypes exhibited markedly lower survival rates (ranging from 61.3% to 63.6%), while the overall survival of the population was 71.3%. For the QTL in Omy31, individuals with the unfavourable homozygous genotype exhibited a lower survival rate (39.3% and 65.2% for population A and B, respectively) compared to the mean survival rates of populations A and B (50.7 and 71.3%, respectively). These results confirm that the QTLs identified in the combined analysis contribute to BCWD resistance in both populations, except Omy3_54.387 that is mostly specific to population A.

Figure 5 provides a comparative analysis of the logBF values obtained for each SNP within the primary QTL regions identified independently for each population (GWAS ‘Pop. A’ and ‘Pop. B’) and in the combined analysis (GWAS ‘Comb. A + B’). This detailed analysis can be found for the other QTL regions in Supplementary file 1, Figs. S3–S5. Notably, for two QTLs identified in population A (Omy17_32.578 and Omy29_33.441), the maximum logBF values observed in the corresponding regions of population B were 8.7 and 10.2, respectively, suggesting that these genomic regions also contribute to BCWD resistance in population B, albeit to a lesser extent (Figs. 5A and Fig. S3). In contrast, a different pattern was observed for the QTL detected on Omy3, associated with early survival phenotypes at 10 dpi, where the logBF values in population B did not exceed 1.9 (Fig. 5B). A similar pattern was observed for all QTLs detected in population B (Fig. S3), such as the QTL Omy1_24.646 where the logBF values in population A remained ≤ 2.3 (Fig. 5C). For the QTL Omy31_25.233 identified by combining the two populations (GWAS ‘Comb. A + B’), the logBF values were notably high in both populations, peaking at 14.8 and 11.2 in GWAS ‘Pop. A’ and ‘Pop. B’, respectively (Fig. 5D). Although this QTL did not reach genome-wide significance in either population when analysed independently, it showed consistent suggestive evidence, particularly in Pop. A. The increased detection power from combining the datasets enabled this signal to pass the genome-wide threshold, highlighting the utility of the combined analysis for consolidating borderline QTL across populations. Overall, this comparison highlights that certain QTL regions seem to contribute to BCWD resistance across both populations, although their relative impact and significance vary.

Fig. 5.

Comparative analysis of logBF values in QTL regions across two rainbow trout populations. Panel (A) shows QTL detected in Pop. A for survival status at 29 dpi, and panel (B) for survival status at 10 dpi. Panel (C) presents QTL detected in Pop. B at 29 dpi, while panel (D) displays results from a GWAS combining both populations (Comb. A + B) for survival status at 29 dpi. A similar analysis is provided for all QTLs in Supplementary file 1, Figs. S2-S4. The positions of key candidate genes (see Table 5) are marked on the right side of each graph

Putative functional implications of gene sets located in the QTL regions

To explore underlying mechanisms of resistance to BCWD, we selected genes located in each QTL, i.e. around the peak SNP including all SNPs for which logBF ≥ 6 in a sliding window of 250 kb. A total of 176 genes were identified across the 11 main QTLs (Table 3 and Supplementary file 3, Table S4), with each QTL region containing between 1 and 40 genes. Based on the molecular functions suggested by their protein domain composition, and on functional knowledge available for their mammalian orthologs, we looked for relevant connection of genes located in QTL with known resistance or pathogenesis mechanisms. A selection of 36 genes potentially involved in the antibacterial response are listed in Table 5 with a short description of the mechanisms of their human or mouse ortholog, and the GWAS analysis which led to the identification of the corresponding QTL. These genes could be classified into four main functional classes: (1) genes involved in immune signalling pathways, (2) genes mediated inflammation and/or acute phase response, (3) genes associated to macrophage biology and activation, and (4) soluble factors potentially important for antibacterial defences (Fig. 6).

Table 5.

Selected immune genes in the QTL regions. Gene positions are provided in Supplementary file 3, Table S4

Chr	QTL_ID	GWAS				HGNC	Descriptiona	References
		Pop. A	Pop. B	Comb. A + B	Pop. A 10 dpi
1	Omy1_24.646		•			PRMT7	protein arginine methyltransferase 7 [induces extravasation of monocytes]	[55]
			•			CIAPIN1	cytokine induced apoptosis inhibitor 1 [cytokine-induced inhibitor of apoptosis]	[56, 57]
			•			SPATA2L	spermatogenesis associated 2 like [negative regulator of TLR4/TRIF-dependent signal transduction]	[58]
			•			PLCG2	phospholipase C gamma 2 [inflammation signalling]
1	Omy1_58.069		•			PLPP4	phospholipid phosphatase 4 [inflammation signalling]
1	Omy1_59.302		•			SRF	serum response factor [neutrophil migration; myeloid development]	[59]
			•			KLC1	kinesin light chain 1 [inflammation regulation (transepithelial migration)]	[60]
			•			ECHS1	enoyl-CoA hydratase, short chain 1 [positive regulator of sphingosine and prostaglandin pathways]	[61]
3	Omy3_54.387	•		•		PTPN4	protein tyrosine phosphatase non-receptor type 4 [negative regulator of TLR4/TRIF-dependent signal transduction]	[62]
		•		•		RAMP1	receptor activity modifying protein 1 [down regulates inflammation and reduces antibacterial response via IL10 induction]	[63, 64]
		•		•		SLC6A6	solute carrier family 6 member 6 [SlC6A6 induction controls M1 polarization]	[65]
		•		•		CD22	CD22 molecule [down regulates signalling via mIgM; controls microglial phagocytosis]	[66, 67]
		•				KLF7	Kruppel like factor 7 [increases expression of proinflammatory cytokines via NF-kB in rheumatoid arthritis]	[68]
3	Omy3_53.505				•	RND3	Rho family GTPase 3 [induces proinflammatory cytokines; promotes macrophage activation]	[69]
					•	SIGLEC14	sialic acid binding Ig like lectin 14 [positive regulator of NLRP3 inflammasome activation]	[70]
					•	PARP14	poly(ADP-ribose) polymerase family member 14 [complex role in IFN response]	[71]
					•	FASTKD3	FAST kinase domains 3 [RNA-binding protein regulating mitochondrial RNA metabolism and PI3K activation]	[72]
					•	TMEM163	transmembrane protein 163 [critical modulator of P2XRs and ATP signalling]	[73]
11	Omy11_43.809		•			COX7C	cytochrome c oxidase subunit 7 C [oxidative phosphorylation]
12	Omy12_31.030		•			DSCAM	DS cell adhesion molecule [promotes macrophage phagocytosis of bacteria in fly]	[74]
17	Omy17_32.578	•				CYP2J2	cytochrome P450 family 2 subfamily J member 2 [oxidative phosphorylation]
		•		•		C3	complement C3 [central component of the complement cascade]
24	Omy24_22.618		•			BARX2	BARX homeobox 2 [SRF binding protein]	[75]
29	Omy29_33.441	•				BRD8	bromodomain containing 8 [controls the secretion of beta-defensin 1 and multiple chemokines]	[76]
		•				PPP3CA	protein phosphatase 3 catalytic subunit alpha [negatively regulates TLR-mediated activation pathways in macrophages]	[77]
		•				UBE2D2	ubiquitin conjugating enzyme E2 D2 [likely contributes to antibacterial autophagy]	[78]
		•		•		COX8A	cytochrome c oxidase subunit 8 A [oxidative phosphorylation]
		•		•		RASGRP2	RAS guanyl releasing protein 2 [controls neutrophil chemotaxis and regulates platelet and T-cell adhesion]	[79]
		•		•		CFB	complement factor B [activated cleaved CFB associates with C3b to form the alternative pathway C3 convertase]
		•		•		CAPN1	calpain 1 [promotes inflammation]	[80]
		•		•		PACS1	phosphofurin acidic cluster sorting protein 1 [regulates ER and oxidative stress]	[81]
		•		•		DPP3	dipeptidyl peptidase 3 [negatively controls the production of inflammatory cytokines]	[82]
		•		•		IL2RG	interleukin 2 receptor subunit gamma [common gamma chain (γC), a key signalling component for IL2, IL4, IL7, IL9, IL15, and IL21]	[83]
		•		•		ZIC3	Zic family member 3 [transcription factor of cathelicidin expression in chicken]	[84]
31	Omy31_25.233			•		CD74	CD74 molecule [invariant polypeptide of major histocompatibility complex, class ii antigen-associated, but also MIF (Macrophage migration inhibitory factor) receptor]	[85, 86]
				•		CSF1R	colony stimulating factor 1 receptor [receptor for CSF1, a macrophage- and monocyte-specific growth factor]	[87]

^a The functional annotation is mainly based on data obtained from mouse and human genes designated by the HGNC in the table. The rainbow trout genes located in the QTL may have different functions. Additionally, these human and rainbow trout genes do not correspond to 1 to 1 orthologs

Fig. 6.

Selected immune genes located in QTLs conferring BCWD resistance in two French rainbow trout populations

Discussion

Methodological considerations and QTL detection power

Large numbers of densely genotyped individuals allow for fine mapping of QTLs thanks to the high SNP density along the genome, which builds strong linkage disequilibrium (LD) between SNPs and causative mutations [88]. Imputation has therefore become a standard practice to increase genome coverage and improve GWAS resolution, as large cohorts can be genotyped at lower density (and cost) and imputed up to denser marker panels [89]. This approach has already proven powerful for QTL detection in binary traits in fish, such as spontaneous masculinisation in rainbow trout [90] or VNN resistance in European seabass [91]. At the intermarker distance of the rainbow trout HD chip - approximately one SNP every 4 kb on the Arlee reference genome [39] - linkage disequilibrium (LD) was estimated to be relatively high, with average r² values of 0.42 and 0.39 in populations A and B, respectively (formerly LB and LC in Bernard et al. [39]). However, r² values vary substantially both between chromosomes and within different regions of the same chromosome, as well as across populations, raising two key challenges.

First, such LD heterogeneity generates strong uncertainty about the location of the causative variant, often resulting in the definition of large QTL regions. This is particularly illustrated by the QTLs detected on Omy3 in population A, where two regions associated with survival at different time points (10 dpi vs. 29 dpi) were located in close proximity (within a few Mb), raising the question of whether they correspond to a single QTL with a broad signal or to distinct linked resistance loci.

The LD variability across the genome also affects the estimation of SNP effects. The estimated effect of a SNP is, in expectation, equal to the true effect of the underlying causative variant multiplied by r (the square root of the r² between the marker and the unknown causative variant). Since this LD is unknown and varies across populations for the same two positions, the estimated SNP effects may differ markedly between populations. Furthermore, because QTL detection power is influenced by the minor allele frequency (MAF) of the marker, some QTLs may be population-specific, which is consistent with the limited number of shared QTL regions identified between populations A and B in our study.

Combining the two populations was not intended to replace the within-population GWAS, but rather to strengthen the robustness of QTL signals across different genetic backgrounds and to confirm the suggestive signals detected within each population through a multi-population analysis.The rationale for this analysis was based on simulation results in cattle breeds [92] as well as findings from real datasets in human genetics [93]. Considering GWAS across human populations for 31 physiological and health traits, Miao et al. [92] demonstrated that local genetic correlations are highly heterogeneous along the genome. Notably, among the traits they analyzed, basophil count exhibited the lowest cross-population genetic correlation (rg = 0.23), yet substantial regional correlations (rg = 0.83) were identified in some genomic regions. These high local genetic correlations helped pinpoint portable genetic effects between populations despite the low global genetic correlation estimated from genome-wide data. In a multi-breed cattle context, van den Berg et al. [93] confirmed through simulations based on real datasets that multi-breed GWAS generally detect more QTL with improved accuracy than within-breed GWAS, particularly when QTL have similar effects and allele frequencies across breeds. When allele frequencies differ or variants are breed-specific, multi-breed GWAS perform at least as well or often better than within-breed analyses, demonstrating the advantages of multi-breed GWAS under the assumption of rg = 1. Conversely, when the causal variant segregates only in one breed, multi-breed GWAS do not lead to an increase in power for that QTL, explaining the limited advantage observed in breed-specific scenarios.

From an applied breeding perspective, loci that consistently contribute to resistance across multiple populations are of particular interest, as they are more likely to be relevant for broader selection programs. In this context, assuming rg = 1 between populations was a pragmatic choice in our study to increase detection power for shared signals, while acknowledging that population-specific effects also exist. The combined analysis therefore provided complementary evidence to the independent population analyses.

While differences in LD structure partly explain the inconsistency of QTL signals across analyses, epistatic interactions offer another important layer of complexity. In particular, Fraslin et al. [33], using the F. psychrophilum waterborne experimental infection model in rainbow trout isogenic lines, through RAD‑sequencing, identified QTLs whose individual effects on resistance were conditional on the genotypes at interacting loci. In such cases, the contribution of each locus to the trait is not constant, but depends on the genetic configuration at other loci. As a result, the significance and apparent effect of a given QTL may vary across analyses, especially when these interactions are not explicitly modelled. This may explain why certain QTL regions appear unstable across BSLMM runs, with only 11 QTL reached a high occurrence threshold across BSLMM runs among the 37 QTL detected, as Bayesian models may alternate between alternative but equivalent explanatory loci, depending on how the underlying epistatic signal is captured in each MCMC chain [50]. Taken together, these observations highlight that QTL detection instability is not solely due to LD variation or sampling noise, but can also result from context-dependent genetic architecture, such as epistasis. These methodological considerations are essential for interpreting the patterns of QTL detection observed in our study. They provide a framework for understanding both the shared and population-specific QTLs identified, as well as the variability in their apparent effects across analyses.

Genetic architecture of BCWD resistance

In a previous study, we identified QTLs for BCWD resistance using families derived from crossing two gynogenetic doubled haploid lines [94] with highly contrasted resistance phenotypes, using RAD sequencing genotyping technology. The genetic basis of resistance was found to be polygenic and strongly influenced by the infection route. A total of 15 QTLs were identified, with only 3 shared between cohorts challenged via injection or immersion [33].

In the present study, we performed GWAS to investigate and compare the genetic architecture of host resistance to F. psychrophilum in two French commercial rainbow trout populations using the waterborne experimental infection model [33]. The results confirmed the polygenic nature of resistance, characterized by the involvement of several main QTLs with limited overlapping loci between the two populations. The estimated proportion of variance explained by additive genetic effects (PVE) was 0.52 in population A and 0.65 (Table 2) in population B, consistent with previously reported moderate to high heritability estimates for F. psychrophilum resistance, which generally range from 0.3 to 0.8 under liability scale (or transformed from the observed scale according to Dempster and Lerner [95]) depending on the strain, and infection protocol used (e.g [23, 25, 28, 96]). The proportion of genetic variance explained by SNPs with large effects (PGE) also differed between populations A and B (0.58 vs. 0.74), highlighting differences in the underlying genetic architecture.

An update of the integrated list of QTLs for BCWD resistance from Vallejo et al. [25] encompassing those identified here is available in Supplementary file 4, Table S5; it comprises regions located in 28 out of the 29 chromosomes of the Swanson genome assembly [25–28, 33, 37, 38, 97–99]. The most frequently reported chromosome was Omy25 (whose p arm corresponds to Omy31 on the Arlee line rainbow trout genome assembly), Omy3, Omy8, Omy17, Omy7, then Omy19. Recent advances in genotyping methods showed that at least 12 out of 28 chromosomes carry multiple QTLs, thus increasing the number of regions of interest. For instance, QTLs involved in BCWD resistance on Omy3 [28, 33, 38, 98] have been recently located around 18 Mbp [26], 57 Mbp, 61 Mbp and 77 Mbp [25]. Two QTLs were identified within the 57 Mbp region of chromosome Omy3 in population A when considering resistance phenotypes at 10- and 29-days post-infection (dpi). These QTLs were detected at SNPs Omy3_53.505 and Omy3_54.387, located at 56.68 Mbp and 57.63 Mbp, respectively, on the Swanson reference genome. Linkage disequilibrium between SNP peaks was high, with an r² value exceeding 0.90, suggesting that they may belong to the same haplotype block, and thus to a same QTL. The QTL Omy31_25.233 (Omy25 in Swanson genome assembly) from this study coincides with genomic regions recently reported for other rainbow trout populations [25, 27]. This QTL was detected only when data from the two populations were combined, as this increased the power of QTL detection, and it had been previously identified by Fraslin et al. [33] when the effect of a main QTL on Omy3 was first adjusted. The QTLs Omy24_22.618 in population B and Omy29_33.441 in population A were also located within regions reported in previous studies [33, 38]. The QTL Omy17_32.578 likely corresponds to the one identified in INRA doubled haploid lines by Fraslin et al. [33], but it is located far away from those reported on Omy17 in other studies [28, 99, 100]. Finally, while we found 4 QTLs on Omy1 in population B (Omy1_24.646, Omy1_58.069, Omy1_59.302, Omy11_43.809), they were distinct from the regions previously reported in Omy1 in the population NCCCWA [37, 38], and likely represent new QTLs.

The number of QTL regions and the striking difference between genetic structures of the resistance in populations A and B illustrate the complexity and diversity of the mechanisms underlying this trait. A similar trend was also observed by Vallejo et al. [26] comparing two rainbow trout breeding populations (NCCCWA and Troutlodge). Fourteen QTLs for BCWD resistance were detected, with only 4 shared by the two populations. In the same vein, the QTLs were partly different between two closely-related breeding populations derived from the same stock [25]. The limited overlap between QTLs detected in populations A and B, as well as in other more distant rainbow trout populations, may reflect original differences in their population genetic diversity [101] but also their distinct recent selective pressures. For instance, QTLs on Omy8 have been reported only in North American populations (NCCCWA, Troutlodge Inc., Clear Springs Foods Inc [26, 38, 99]), whereas QTLs on Omy21 and Omy29 were detected only in French populations (INRAE crossed line [33], Pop. A and B; in this study). Variations in the BCWD pressures they experience could shape the selective forces acting on their genomes, whether or not resistance to F. psychrophilum is incorporated into selection programs. Indeed, F. psychrophilum isolates causing severe outbreaks in rainbow trout farms worldwide belong in large majority to the clonal complex CC-ST10, in which some sequence types (ST) are more prevalent in certain geographic area (e.g. ST10 and ST78 in North America, ST2 in Europe). Other genotypes have also been reported in diseased rainbow trout, and strains belonging to CC-ST90 have been frequently reported in France and in other countries [22, 102–105].

In addition, the results of GWAS could also be affected by the challenge model. Notably, the infection models used in the different studies differs by the bacterial genotype (i.e. strain CSF259-93 belonging to CC-ST10 in North America studies versus strain FRGDSA 1882/11 belonging to CC-ST90 in Fraslin et al. [33] and this study) and by the infection route. Experimental infections provide a controlled setting that helps establish link between QTL and a specific pathogen, while natural outbreaks introduce confounding factors such as environmental variability and putative multiple pathogen exposures. This may partly explain why for population A, QTLs identified here were different from those identified during a natural disease outbreak associated with F. psychrophilum [28] in a previous cohort of the same population. Future studies investigating the effects of F. psychrophilum strain on the genetic architecture of BCWD resistance will enhance our understanding of molecular mechanisms underlying host-pathogen interactions.

QTL-associated genes and putative resistance mechanisms

Resistance to F. psychrophilum infection likely involves immune pathways relying on components encoded in different genomic regions. For some of them, multiple paralogous genes sharing a same function but different genomic locations, may exist. This likely contributes to the complexity of the genetic architecture of this trait. Improved precision in QTLs mapping will help to identify the genes - or gene combinations - involved in the resistance/susceptibility to this disease. In this study, we have identified a number of candidate genes located within QTL regions, with functions associated to immune signalling, inflammation/acute phase factors, activation of macrophages/neutrophils and to the complement cascade and other soluble factors (Supplementary file 4, Table S5).

Immune signalling

Three negative regulators of TLR signalling were present (Spata2l in Omy1_24.646; Ptpn4 in Omy3_54.387 and Ppp3ca in Omy29_33.441). Fastkd3, a gene encoding a RNA-binding protein that regulates mitochondrial RNA metabolism and PI3K activation, was in Omy3_53.505. Many PARP proteins play a role in type I IFN response, but human/mouse PARP14 catalyses mono-ADP-ribosylation of STAT1, thus prevents its phosphorylation, inhibiting the JAK/STAT pathway. Whether trout parp14 (in Omy3_53.505) has the same function remains to be demonstrated.

Inflammation and acute phase

QTL regions contain several genes of the oxidative burst (Cox7c: Omy11_43.809, Cox8a: Omy29_33.441, Cyp2j2: Omy17_32.578), as well as regulators (Ciapin1: Omy1_24.646; Pacs1: Omy29_33.441). Other genes regulate the expression of proinflammatory cytokines like Dpp3 (Omy29_33.441) and Capn1 (Omy29_33.441); or the production of sphingosines/eicosanoids or DAG (Echs1: Omy1_59.302; Plpp4: Omy1_58.069). In human, SIGLEC14 is a positive regulator of NLRP3 inflammasome activation, but the trout Siglec14 gene (Omy3_53.505) may have a different function linked to antibacterial defences, since SIGLECs are found primarily on the surface of immune cells and play multiple roles in immunity. TMEM163 (Omy3_53.505) regulates purinergic receptors and Ca²⁺ role in cell activation and inflammation.

Macrophage/neutrophil biology and activation

Macrophages and neutrophils are key cellular effectors of the antibacterial response, and genes involved in their activation and biology appear as natural candidates for resistance/susceptibility factors. QTL regions contain genes involved in macrophage activation (Rnd3: Omy3_53.505; Csf1r: Omy31_25.233) or polarisation (Slc6a6: Omy3_54.387), genes regulating the production of key cytokines by macrophages such as Ramp1 (Omy3_54.387) which promotes IL10 production and reduces antibacterial response, and genes controlling migration and phagocytosis like Srf (Omy1_59.302), Barx2 (Omy24_22.618), Prmt7 (Omy1_24.646), Cd22 (Omy3_54.387), Cd74/mifR/li (Omy31_25.233) and Rasgrp2 (Omy29_33.441). Cd74 may also be important through its key function in Ag presentation by MHC class II molecules. UBE2D2 (Omy29_33.441) controls antibacterial autophagy.

Soluble factors potentially important for antibacterial defences

The most important soluble factors identified in QTL regions are probably the complement genes Cfb (Omy29_33.441) and C3 (Omy17_32.578), as the complement cascade plays essential roles in antibacterial responses. Besides, Zic3 and Brd8 (both in Omy29_33.441) control the production of cathelicidins.

The considerable divergence between the BCWD resistance QTL array detected between two commercial populations in our study parallels to the high heterogeneity of regions identified in previous studies using multiple approaches; the imprecision of older methods makes it even more difficult to discuss the gene content and associated functions of all these regions. However, common QTLs between this study and the most recent ones [25–27] point to potentially important factors for the immune response against F. psychrophilum, whose functions have been discussed above: Omy3_53.505 and Omy3_54.387 in population A encompassing fastkd3, tmem163, ptpn4, ramp1 and cd22; and Omy31_25.233 in populations A and B encompassing csf1r and Cd74/mifR/li. Besides, different genes from the complement cascade have been repeatedly found within QTL regions from multiple studies: the c3.2 gene (LOC100136952) is the only gene in QTL Omy17_32.578 contributing to survival in populations A and B, while its paralog c3.1 (LOC 110489027) on Omy2 was in a QTL reported by Fraslin et al. [33]. C3 complement encoding genes were also upregulated in multiple trout genetic backgrounds during F. psychrophilum infection [15, 106, 107]. Cfb is located in QTL Omy29_33.441, which contributes to survival in populations A and B. C1ql2 and c1ql3 were associated to QTL of resistance to BCWD in American populations, respectively on Omy8 (Troutlodge Inc.; [25]) and Omy13 (NCCCWA; [38]). Finally, a homolog of complement C8 gamma was found in another QTL of NCCCWA trout on Omy11 [37]. However, this gene may not be directly involved in the bacterial lysis by the complement: indeed, human C8G is not required for the bactericidal activity of the complement membrane attack complex, but it plays a regulatory role in (neuro)inflammation [108]. Taken together, these findings support the involvement of the complement pathway in resistance to BCWD, and the direct interactions between its protein components offer a mechanistic basis to understand epistatic effects.

Conclusions

This study highlights the complex and partly population-specific genetic architecture underlying resistance to F. psychrophilum in rainbow trout. Using a standardized infection model and high-density SNP genotyping, we identified several QTLs with strong effects on survival in both commercial populations, including one shared region detected in the combined analysis. Although the overlap between the QTL detected in the two populations is limited, certain regions likely contributed to BCWD resistance in both, though with different strength. These results confirm that resistance to BCWD is a polygenic trait shaped by both major and minor effect loci, whose contributions may vary across populations. Although the QTLs differ, genes and functional pathways associated with QTLs can converge in different populations.

While these QTLs represent promising tools for improving disease resistance through genetic selection, it is important to first consider the potential influence of pathogen genetic variability on host resistance. Therefore, to effectively enhance the resistance of rainbow trout populations by integrating these QTLs into routine breeding programs, further research should investigate the extent to which host-pathogen genetic interactions affect QTL relevance and stability across environments and epidemiological contexts. Such knowledge is essential to ensure that resistance-associated markers are effectively and reliably applied, either to enhance the precision of pedigree- or genomic-based selection strategies, or to guide the dissemination of improved genetic material via marker-assisted selection.

Authors’ contributions

SP curated the data, performed formal analyses and visualizations, and wrote the original draft of the manuscript. DR and BL contributed to the methodology and conducted the investigations. ES and AD provided resources. JA curated the data, developed software, and provided resources. YF was responsible for project administration. PB contributed to the conceptualization, performed formal analyses, and contributed to writing the original draft and reviewing the manuscript. ED contributed to the conceptualization and methodology, acquired funding, and administered the project. FP contributed to the conceptualization and methodology, developed software, supervised the project, and contributed to manuscript reviewing. TR contributed to the conceptualization, methodology, investigation, formal analyses, and visualizations, administered the project, and wrote the original draft of the manuscript. All authors read and approved the final manuscript.

Funding

This study was supported by the European Maritime and Fisheries Fund and France AgriMer (Flavocontrol project, n° PFEA470020FA1000004). PB was supported by ANR (ANR-21-CE35-0019, Lipofishvac) and by the European Union’s Horizon 2020 research and innovation program under grant agreement No 817923 (AQUAFAANG).

Data availability

The authors declare that all data necessary to support the conclusions of this research article are included within the article or are available upon reasonable request to the corresponding author (simon.pouil@inrae.fr). The genotype datasets contain proprietary commercial information and will be provided for research purposes only and under a Material Transfer Agreement (MTA). In line with Springer Nature’s policy on proprietary and third-party data, metadata records are publicly available here: (10.57745/WV8XJY).

Declarations

Ethics approval and consent to participate

Animal experiments and sampling were performed at the INRAE-IERP fish facilities (doi: 10.15454/1.5572427140471238E12, Jouy-en-Josas, France) authorised for animal experimentation under the French regulation C78-720. The experiment was carried out according to the European guidelines (Directive 2010–63-EU); the protocols were evaluated and approved by the ethical committee COMETHEA n°45 and authorized by the French Ministry of Higher Education and Research (APAFIS # 36460-2021112509411919 v3).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.