Resilience was historically ignored in commercial swine breeding programs. One reason was that breeding companies need to supply high health animals to the commercial sector of the swine industry. Nucleus herds needed to be naïve or at least negative for major diseases such as porcine reproductive and respiratory syndrome virus (PRRSv), Actinobacillus pleuropneumoniae (APP), Mycoplasma hyopneumonaie, and other major diseases. Traits such as mortality have not been added into commercial breeding program selection indices until more recently. Hermesch et al. (2014) estimated that post-weaning mortality should account for roughly 44.5% of the terminal sire index. This represents a major economic value for a trait that has been largely ignored. Post-weaning mortality may be the single largest economic cost due to the cost of the piglet and resources like feed, but many other infectious and non-infectious diseases cause loss of productivity in the commercial sector with causing significant mortality (e.g. heat stress; see Martínez-Miró et al., 2016).
Many authors have attempted to define and redefine the definition for resilience over the years. Other terms such as robustness sometimes appear with resilience or as a separate definition altogether. Defining resilience and robustness has continued to be a hot topic for authors over the years. Clunies-Ross (1932) first recognized the difference between “resistance to infestation” and “resistance to the effects of infestation”. This was the beginning of the separation between ‘resistance’ and ‘resilience’, the realization that infestations of parasites or infection by bacteria or virus was not completely correlated to the production of that animal under these diseases.
While studying parasites in sheep, Albers et al. (1987) was very careful to define traits that represented resistance, production (healthy and under challenge), and resilience. Resilience traits were identified by adding ‘DEP’ to the production trait of interest that stood for ‘depression’. This represented the difference (depression) between uninfected and infected environments (i.e. how much production loss occurs for that individual). The loss of production from the uninfected environment is what quantifies resilience in this case.
Confusion has arisen from alternative definitions over the years. Bishop and Woolliams (2014) defined resilience as the productivity of an animal under infection. However, animals differ in their ability to produce under non-limiting conditions (i.e. healthy). Two animals will still perform differently due to alleles making them more suited for the production trait of interest. For instance, melanocortin-4 receptor gene (MC4R) influences feeding behavior and body weight in pigs (Kim et al., 2000). This gene likely influences feed intake regardless of infection status. Therefore, both immune defense and normal productivity are components of production under stressed environments.
Colditz and Hine (2016) chose to separate resilience and robustness altogether. The authors defined resilience as ‘the capacity of animals to cope with short-term perturbations in their environment and return rapidly to their prechallenge status’. Authors contrast this definition to robustness defined as ‘the capacity to maintain productivity in a wide range of environments without compromising reproduction, health and wellbeing’. The main contrast being between the macro-level (robustness across environments) to the micro-level (resilience within a certain environment).
To explain resilience, a simple equation can be adapted from van der Waaij et al. (2000).
where 𝑃𝑐 is the commercial production of the animal, 𝑃𝑝 is the production potential of the animal, and f(R) is a function of resilience. This simple equation makes it clear that production under challenge (𝑃𝑐) and resilience (R) are not the same trait, however it shows there is a clear relationship among the three variables. For example, if an animal is 100% resilient, its commercial productivity is equal to the performance potential of the animal. There is clearly additive genetic variation in resilience (Elgersma et al., 2018; Putz et al., 2019).
To encompass a broad definition of resilience, both infectious disease and non-infectious stressors must be included. Martínez-Miró et al. (2016) reviewed stressors and categorized them into social interactions among pigs, environmental, metabolic, immunological, and human interactions. Heat stress as an environmental stressor has been estimated to cost the US swine industry $299 million per year (St-Pierre et al. 2003). PRRS is estimated to cost US swine producers $664 million annually (Holtkamp et al., 2013). Out of feed events can cause metabolic stress to animals while growing. Air quality and dust is another common environmental stressor in pigs not commonly studied (Donham, 1990; Senthilselvan et al., 1997). There is no known current estimate of the economic impact of all the above stressors combined, but it is likely in the billions of dollars. Holck et al. (1998) estimated a 30% loss in productivity between a non-limiting environment and stressed commercial conditions.
Although definitions have changed over the years and continue today, too much time should not be spent on defining resilience and robustness. More time needs spent researching and developing genetic evaluation strategies. Commercial data collection schemes should also be evaluated to breed for more resilience pigs in all phases of pork production with a focus on post-weaning survivability.
Selection for uniformity linked to residual variance
Uniformity models have been developed in recent years. This is accomplished by looking into the residual variance of the individual records relative to their contemporary group and additive genetic value. Recently, Iung et al. (2019) published a review summarizing the methods and results of the studies on residual variance (uniformity) to date. Theory for these models has developed in recent years from extensions of the classical quantitative genetics model. Obtaining genetic values for uniformity in a swine breeding program can be accomplished relatively easily because there only needs to be a single record observed such as off-test weight in pigs, making it slightly more practical than other methods (see novel phenotyping below). For instance, some programs are already collecting hot carcass weight in commercial pigs that could be used for this type of evaluation (Dufrasne et al., 2013).
There are several different methods for uniformity models including Bayesian methods (Sorensen and Waagepetersen, 2003), the two-step REML approach (Mulder et al., 2009), and double hierarchical generalized linear models (DHGLM; Rönnegård et al., 2010). Each of these contains two models, one for the mean (typical animal breeding model) and one for the residual variance (with an additive genetic effect embedded within).
To get an understanding of these models, the two-step REML model can be described as follows (Mulder et al., 2009; Iung et al., 2019),
where y and ln(ê2) are the response variables for each model, respectively, 𝒃(𝒃𝒗) and 𝒂𝒂(𝒂𝒂𝒗𝒗) are the vectors of fixed effects and additive genetic (random) effects for each model, respectively, X(𝑿𝒗) and Z(𝒁𝒗) are the coefficient matrices for fixed and additive genetic (random) effects for each model, respectively, and e(𝒆𝒗) is the random residual effects for each model, respectively. Taking the natural logarithm of the estimated squared residuals makes the distribution closer to a normal distribution. This model has been implemented in the popular mixed model software ASReml (Gilmour et al., 2015).
Variance components have been estimated for these models. The heritability of the residual variance (ℎ𝑣2) has a median of 0.012 across 32 studies, most estimates falling within the range of 0.0 to 0.05 (Iung et al., 2019). However, the genetic coefficient of variation of residual variance (GCVE) has a median estimate across studies of 0.27 (Iung et al., 2019). The GCVE indicates the potential response to selection for residual variance. Current estimates suggest genetic progress is possible, but likely very slow.
Implementation of selection against residual variation into a breeding program is possible existing implementation into ASReml and MiXBLUP (Mulder et al., 2018). However, with the extremely low heritability estimates, very large family groups would be required. Estimates in aquaculture species tends to be higher and it is possible to obtain large family groups for increased accuracy in these species. Therefore, although it is possible, diverting part of the economic selection index to selection against residual variation using these models in livestock species would be risky for a commercial breeding program, excluding aquaculture species. Furthermore, trials comparing competition in the commercial genetics industry is typically based on the mean, with little to no value typically placed on variation or uniformity even when these traits have economic value (Hubbs et al., 2008).
Precision livestock farming and novel resilience phenotyping
Using technology in agriculture is not new, but one that is a hot topic right now in academia and across different industries (Berckmans, 2017; Koltes et al., 2019). For many years, academia and industry partners have been working on implementing technology into the livestock sector. Examples include individual feed intake stations and ultrasound machines to get carcass measures on live animals. Dairy appears to be leading the way in many respects with other industries not far behind. What seems to separate precision livestock farming (PLF) from normal technology in farms today is the collection of real-time data to make decisions and collect information on farms. Also, most of the technology applied in the past, such as individual feed intake stations, were only implemented on nucleus farms for breeding companies. Today, PLF has the aim at being implemented at the commercial level as well to help farmers make decisions and track performance of their animals over time. Labor is an increasing issue and finding quality people to be stockmen is a difficult task. PLF has the ability to improve economic viability and increase animal welfare.
There are two main ways to utilize PLF data and technology. Researchers and producers can use the data to 1) collect and monitor real-time data to improve management decisions and catch issues quickly and 2) use the high throughput data to retrospectively analyze the data to quantify resilience for each individual animal. Both methods are currently under rapid developments in research, although the use of this type of data is not new. Madsen et al. (2005) monitored drinking patterns in a group of pigs in order to detect issues early for management (from a group basis). Today, researchers are finding ways to use video to predict weights in pigs (Fernandes et al., 2018), monitor body temperature (Garrido-Izard et al., 2019), and even detect social interactions between pigs (Chen et al., 2020). Camera vision technology is also being used to quantify activity, eating, and drinking behavior in individual pigs (Mittek et al., 2016). More uses will continue to come up over time, however it will require animal scientists working with other fields, especially computer scientists (Koltes et al., 2019).
The second use of PLF data, retrospectively analyzing the data to quantify each individual animal’s level of resilience to obtain novel resilience phenotypes, has just begun. This is a much newer research area and more applicable to breeding programs due to the individual nature of the phenotyping. Novel phenotypes extracted from PLF data could be integrated into a breeding program as indicator traits or breeding objective traits. Elgersma et al. (2018) first quantified dairy cows’ resilience by taking the standard deviation in daily milk yield over a lactation as well as quantifying the number of drops in milk yield from several days prior. Daily milk yield was collected using data from robotic milking machines. These resilience traits were genetically correlated to existing health traits in dairy. Poppe et al. (2020) expanded on this research by adding a regression curve of milk yield on day in milk, which should always be done if there are any systematic mean changes over time. Authors found that quantile regression was the best method to extract the residual variation for each individual cow (log of the variance around the regression of daily milk yield on day in milk per cow).
In swine, Putz et al. (2019) quantified resilience in growing pigs using daily feed intake data from individual feed intake stations. Two traits were developed including the root mean square error (RMSE) and quantile regression (QR) phenotypes. For the RMSE phenotype, a simple linear regression was fit for feed intake on age and extracting the RMSE from the regression (see Figure 1 for a simulated example). The QR phenotype was quantified by counting the number of days within animal that were off-feed or below the 5% quantile regression line of all daily feed intake records on age. Both of these phenotypes showed a higher heritability than mortality and were genetically correlated to mortality and treatment rate. These traits are easy to implement into current evaluations as each animal receives one phenotype per time period (e.g. lactation) or lifetime (e.g. juvenile growth). They do not require complex new statistical frameworks like DHGLM models for residual variance that may not be well understood (see above).
Figure 1. Simulated example of a feed intake curve in a pig with two periods of depressed feed intake due to a stressor such as disease. The regression line can be used to extract the residual mean square error (RMSE) to quantify resilience for each individual animal.
It is important to note that both uniformity models and novel phenotypes from PLF technology quantify general resilience. Although infectious diseases likely cause the most significant economic losses, there are other major losses from heat stress for example. By looking at variation in milk yield or feed intake, it cannot be determined what caused these issues and should therefore be thought of as general resilience to all stressors that impact production traits.
There are many challenges with PLF technology, see Koltes et al. (2019) for a full review on this subject. For commercial producers looking to use PLF technology to help them manage farms, two main concerns are reliability and return on investment. The cost of some of these technologies can be quite high, especially as many technology companies are just entering this market. For breeding companies (e.g. swine) and national evaluations systems (e.g. Holstein dairy cattle), cost is still a limiting factor, but given the large economic value attached to resilience, the return on investment should be much more attractive, especially if competitors ignore this value.
Inclusion of commercial crossbred data
Improved crossbred performance should be the breeding goal of any breeding company (Hermesch et al., 2015). In the past, companies have relied on or assumed a high, positive additive genetic correlation between purebreds and crossbreds to justify not collecting commercial data. There are also difficulties tracing pedigreed commercial animals back to nucleus animals’ reliably and collecting quality data at that level. Estimates of the purebred-crossbred genetic correlation tend to be significantly lower than 1, with 50% of estimates between 0.45 and 0.87 (Wientjes and Calus, 2017). This has lead companies to reconsider collection of crossbred data over the years along with the large economic value attached to resilience (e.g. Hermesch et al., 2014).
Commercial post-weaning mortality data may be the easiest trait to implement into a breeding program. The minimum requirements are 1) single-sire mated F1 (Landrace x Large White) dams, 2) tagging piglets at birth or weaning to link their pedigree to the nucleus sire (from the central nucleus AI station), and 3) recording individual mortality dates post-weaning. Some companies may want to also have a pedigreed F1 sow population in order to eliminate confounding between the litter and additive genetic effects of the dam (unpublished results), otherwise a sire model can be fit to the data.
Literature containing commercial post-weaning mortality data are limited as this was not a traditional trait in selection indexes and has not be heavily researched in academic circles. Dufrasne et al. (2015) analyzed commercial mortality data and found the post-weaning survival had a heritability of 0.06, with nursery and finishing mortality showing slightly higher heritability estimates of 0.14 and 0.10, respectively. Authors also found that pre- and postweaning mortality have a low genetic correlation and therefore suggest the control by different sets of genes.
Combined crossbred and purebred selection (CCPS) has been studied over the years (Bijma et al., 2001; Dekkers, 2007). Dekkers (2007) found that adding marker assisted selection (MAS) with crossbred data increased response in crossbred performance 34% over purebred selection. More research is warranted on the advantage of adding crossbred data in the age of computer simulation where more questions can be answered. The issue seems to be the difficulty in programming and computational load to simulate such a large purebred-crossbred breeding scheme and make it flexible enough for any breeding program.
Stressors in swine cause tremendous losses, with infectious disease leading the way. Management has improved over the years to reduce these losses, however despite these advancements, large losses still persist in the commercial sector. Genetic improvement is one way to make slow but cumulative gain over time to combat losses from stressors. Many commercial genetics suppliers are now collecting commercial data through CCPS systems. Due to the large economic cost to the swine industry, it is important that breeding companies start to focus on commercial productivity by integrating resilience into their breeding goals.
Published in the proceedings of the International Pig Veterinary Society Congress – IPVS2020. For information on the event, past and future editions, check out https://ipvs2022.com/en.