Our goal was to study general resilience of layer chickens, which is expected to be related to the capacity of an animal to cope with different types of challenges: disease, climate change, negative social interactions, etc. [5, 6, 10]. Thus, we explored the genetic background of LNVAR, AUTO-R, and SKEW of the deviations of weekly egg production of an individual hen from batch average to assess their potential as selection criteria for resilience in layers. Our results indicate that these resilience indicator traits are lowly heritable and are genetically different in purebred and crossbred laying hens. Moreover, LNVAR showed favorable genetic correlations with egg production.
Egg production
Our estimates of heritability for EP in purebreds were low and consistent with those from previous studies that were performed on EP data for similar time frames and environments (ranging from 0.01 to 0.20 across the laying period [32, 33]). In general, reported estimates of heritability for egg production traits are higher for purebreds than for crossbreds [34, 35], which appears unexpected because part of the dominance variance being captured in the additive genetic variance, hence heritability is usually higher in crossbreds than in purebreds [36]. However, estimates of residual variances vary greatly between studies and are mostly higher in crossbreds than in purebreds, which may be due to environmental conditions of crossbreds varying more than those of purebreds and makes comparisons complicated. Because purebred EP is recorded in individual cages and crossbred EP is recorded in group cages, social interactions may increase phenotypic variation for crossbred EP. Differences in heritability estimates between individual and group records have also been reported in the literature [28, 29, 37]. These differences can be explained by differences in the estimation of variance components, either because of incomplete information on the relationships between individuals due to the use of group information instead of individual information (in purebreds the complete pedigree is known while in purebred only the sire branch is known), or modelling choices (e.g. taking social interactions into account or not). Yet, when using individual data to simulate group records, estimates of the variance components based on individual and group records are the same [30].
Using LNVAR and AUTO-R to improve resilience to disturbances
LNVAR is expected to be an indicator of both the severity and of the duration of a perturbation that the animal is experiencing, i.e. animals with a small variance are less affected by a perturbation than those with a large variance [6]. In our study, LNVAR was the most promising resilience selection criterion, because it had the highest heritability and also had a favorable genetic correlation with EP.
Our estimates of heritability for LNVAR of purebred EP (estimates of 0.11 in WA and 0.12 in BD) were similar to the estimate of heritability for LNVAR of body weight (repeated records) of layers related to the WA line (estimate of 0.10) [18]. Few studies have reported genetic parameters for LNVAR for multiple observations on one individual. LNVAR is by definition the natural logarithm of the variance of the residuals from an animal model of a trait [6]. In layers, heritabilities of 0.01 have been reported for residual variation of eggshell color at the individual egg level and of 0.15 at the level of 15 to 20 eggs measured per hen [20], and of 0.10 for residual variation of egg weight (also repeated records) [19]. In broilers, low heritabilities (lower than 0.05) have been reported for residual variation of body weight (one record per individual) [38,39,40]. Other studies in dairy cattle, pigs, and aquatic species, with either one record per individual or repeated records have also reported mostly low heritabilities for residual variance (ranging from 0.01 to 0.14; [16, 41,42,43,44,45]).
In our study, the genetic correlation of LNVAR with EP was favorable and negative (ranging from − 0.83 to − 0.54). In other species, LNVAR has also been shown to have favorable genetic correlations with other traits. In pigs, lower day-to-day variance in feed intake (root mean square error of prediction of feed intake) was genetically correlated with lower mortality and fewer health treatments [17]. In dairy cattle, LNVAR of milk yield deviations was found to be favorably genetically correlated with health (udder and hoof), longevity (productive longevity, calf and maternal survival at birth), fertility (calving interval, first to last service interval), metabolic status (ketosis resistance), and production traits (milk yield, persistency, body condition score, and feed intake) [15, 16].
To our knowledge, our study is the first to report genetic correlations of LNVAR with EP in layers. These correlations imply that selecting for a reduced LNVAR would increase EP, which is desirable. EP has been a selection criterion for many generations and, therefore, has probably contributed to indirect genetic improvement of LNVAR. Given the low heritability of EP, adding LNVAR in the breeding goal is an opportunity to improve not only resilience but also production, in spite of the low heritability of LNVAR. The next step towards the inclusion of LNVAR in a breeding program is to derive economic values or to set the targeted genetic gain. A simplified simulation study has already shown that selecting for resilience indicators has a beneficial economic impact, and even more if health traits are not included in the breeding program, because of reduced labor and treatments [6].
In summary, although selecting layers on EP already improves LNVAR, including LNVAR in breeding programs may further enhance the improvement of resilience. The absence of trade-offs makes it relatively easy to increase resilience without a loss in selection response in EP. In addition, literature on other species shows that LNVAR is favorably associated with general immunity and health [6, 45].
The resilience measure AUTO-R is expected to be informative regarding the duration of a perturbation and thus to be an indicator of an animal’s recovery capacity: close to 0 (subsequent deviations of EP are uncorrelated) for animals with high recovery rates, close to + 1 (subsequent deviations of EP are correlated) for animals with a slow recovery rate; and close to − 1 (subsequent deviations of EP are opposite) for animals showing a compensatory response to the perturbation [6]. In our study, AUTO-R had a low heritability estimate in both purebreds and in crossbreds. These heritability estimates are similar to the heritability for AUTO-R of body weight deviations in layers (value of 0.09) [18]. Similarly, in dairy cattle, the heritability for AUTO-R of milk yield deviations was estimated at 0.09 [16]. In our study, AUTO-R was estimated to be unfavorably genetically correlated with EP in purebreds, but this relationship was less clear in crossbreds. Other studies on layers have not shown a clear genetic relationship of AUTO-R with health: a non-significant genetic correlation between AUTO-R of body weight deviations and titers of natural antibody (NAbs) isotypes in serum was reported [18]. However, in dairy cattle, AUTO-R of milk yield deviations was found to be favorably genetically correlated with health, longevity, fertility, metabolic, and production traits, but at a lower level than LNVAR [16].
Before using AUTO-R as a selection criterion, its definition as a trait needs to be further discussed to reach a consensus, and its biological impact on production traits and its relationship with health needs to be understood. Knowing both the LNVAR and AUTO-R of an animal can lead to complicated interpretations: phenotypically, animals with a low LNVAR could display an AUTO-R close to 1 because they are hardly affected by perturbations, and their subsequent deviations from EP are more alike. In the present study, we found that LNVAR and AUTO-R were favorably genetically correlated in all populations but their genetic correlations were less than 1. However, if layers have genetically a high AUTO-R when faced with a severe perturbation, they will show a slow recovery. Recovery from a perturbation may be an interesting trait in the case of severe perturbations (such as a virus outbreak or a heatwave), thus it may be more appropriate to calculate AUTO-R during periods with large deviations instead of over the whole period.
In summary, selecting purebred layers on EP is expected to increase AUTO-R, which could decrease recovery rates from perturbation. Including AUTO-R in breeding programs may further enhance the improvement of resilience because it would put more emphasis on recovery rate. The trade-off between AUTO-R and EP also highlights the need to quantify economic values when considering the inclusion of AUTO-R in breeding programs.
SKEW is expected to be an indicator of both the direction (positive or negative) and severity of a perturbation that an animal is experiencing, i.e. animals with a negative SKEW experience more negative deviations than others, which is interpreted as poor resilience [6]. However, heritability estimates for SKEW were very low in the purebreds and crossbreds from both lines, similar to previous studies in layers [18] and in cattle [16], which reported that SKEW of body weight deviations and of milk yield deviations are not heritable. SKEW is known to be highly sensitive to outliers and thus to contain more noise [6]. We conclude that SKEW does not seem to be a useful resilience indicator trait when breeding for resilient laying hens. Overall, estimates of the genetic correlation of SKEW with LNVAR and AUTO-R were favorable, which means that genetic selection on LNVAR or AUTO-R is not expected to result in degraded SKEW.
Resilience indicators in purebreds and crossbreds are genetically different
In our study, all \(r_{pc}\) estimates were statistically significantly lower than 1, ranging from 0.16 to 0.63. This suggests that EP and resilience indicators traits are genetically different between purebreds and crossbreds. Our estimates of \(r_{pc}\) were rather low compared to those in the literature, which range from 0.45 to 0.87 in pigs [21] and poultry [46]. Three main factors can contribute to \(r_{pc}\) being lower than 1: (i) genotype-by-genotype interactions (G × G), (ii) genotype-by-environment (G × E) interactions, and (iii) differences in measurement or definition [20, 21].
First, crossbreds only share half of the genetic characteristics of each purebred parental line. Allele frequencies likely differ between purebreds and crossbreds and such differences may have direct effects on the trait (causal variants) or indirect effects (also known as G × G: dominance or epistasis), depending on the loci involved.
Second, G × E interactions can contribute to the same trait having a different genetic architecture in different environments. Purebreds and crossbreds were raised in very different environments in our study in terms of climate, management, feed, barns, disease pressure, etc. The farms where purebreds were kept have a higher level of biosecurity than regular farms because the breeding companies cannot take any risk with the population under selection. RT-farms generally belong to farmers who collaborate with breeding companies and, although their health and care practices are usually among the best ones, the layers will experience more challenges on an RT-farm than on a nucleus farm. Commercial farm environments can still be very different from RT-farm environments (cage-free systems, winter garden access, outdoor access, etc.). This diversity of environments and their susceptibility to changes (heat waves, disease outbreak, etc.) may influence the resilience performance of hens due to differences in types, severity, and duration of the perturbations. In addition, purebreds were housed individually, while crossbreds were in group cages, in which case the presence of cage mates creates positive (e.g. grooming) as well as negative (e.g. pecking) social interactions that can affect the individuals’ wellbeing and performance. As a consequence of G × E interactions, re-ranking of sires may occur. In both lines, estimates of \(r_{pc}\) were lower for LNVAR than for EP or AUTO-R, which may be due to G × E interactions having a greater role in traits based on variances than in traits based on means. In trout, G × E interactions were reported to be larger for the variance of body weight than for body weight itself [47].
Third, differences in the definition or measurement of the traits may contribute to \(r_{pc}\) being lower than 1. Although the definition of the variable may be the same, values may differ due to differences in the recording machines and protocols used, (e.g. housing conditions such as individual vs. group), the commitment of the collaborators, units, errors, etc., that affect the quality of the raw data. These differences can contribute to reducing the \(r_{pc}\). Although the recording of the performance was the same between purebreds and crossbreds, dealing with records from individually-housed hens or group-housed hens may lead to different traits (e.g. implication of social interactions) and thus result in reduced \(r_{pc}\) (theoretical derivation [20]).
All these elements can contribute to the value of recording the traits in crossbreds in a commercial environment. Breeding programs would benefit from considering both pure- and crossbred data to improve response to selection for crossbred performance in commercial environments [23, 48, 49], because it can increase the accuracy of the estimated breeding values (EBV) of purebred animals for crossbred performance.
Longitudinal recording to genetically improve general resilience in farm animals
Our study shows that longitudinal recording offers opportunities to develop resilience indicators that are heritable. Here, resilience indicator traits are based on deviations of production performance from the undisturbed phenotype (here defined as the contemporary group average). An alternative is to estimate the undisturbed phenotype of an individual based on various modeling approaches [14, 16, 50]. Both approaches may underestimate the undisturbed phenotype, leading to underestimation of deviations and, thus, overestimation of resilience. On the one hand, if a perturbation affects the majority of the individuals in the contemporary group, the contemporary group average will be lower than the true undisturbed phenotype. On the other hand, if a perturbation affects the phenotype of an individual permanently (i.e. poor recovery), the individual’s trajectory based on observations will be lower than the undisturbed phenotype. Another approach is to estimate breeding values that allow to calculate an undisturbed trajectory (e.g. breeding values for curve parameters) [6]. Nevertheless, when modeled, a correction for the contemporary group effect is needed to account for the difference in disturbance level between contemporary groups. More research is needed to provide insight into the genetic correlations between resilience indicator traits estimated with all three approaches and which approach would best predict resilience [51].
General resilience is likely a combination of resilience of various biological functions, including food intake, digestion, growth, maintenance, reproduction, thermoregulation, immunity, behavior, etc. [5, 6, 10]. In our study, LNVAR and AUTO-R of deviations of weekly EP may capture the resilience of (re)production, i.e. the capacity of EP to be minimally affected by disturbances or to rapidly return to the undisturbed performance. This could lead to resilience of EP but not necessarily to general resilience. Thus, future studies should jointly study LNVAR of EP, LNVAR of bodyweight, and LNVAR of feed intake (e.g. cross-correlations of deviations). Longitudinal data on different traits (egg weight, body weight, and feed intake) could also be used to identify periods with a clear perturbation at the herd level [51], which would enable the study of the capacity to recover from a perturbation (i.e. AUTO-R or slope of reaction norm). Resilience indicators based on different traits could also be combined in a selection index for general resilience.
Tools are available now for phenotype recording in layers, such as electronic nests, feeders, weighing scales, cameras, microphones, RFID chips (radio-frequency identification), other embedded sensors, NIRS or MIRS (near- or mid-infrared spectroscopy), etc., which allow the concomitant high-throughput phenotyping of different traits. Based on our results and those reported in other livestock species that show the potential of resilience indicator traits for genetic improvement of resilience, especially LNVAR, further investigations in chickens are needed (i) to find the best (combination of) production traits to calculate resilience indicators, and (ii) to record available perturbation-related indicators and relate them with the resilience indicator traits.