Modeling phenotypic plasticity as an indicator of adaptability in beef cattle

Master of ScienceGenetics Interdepartmental ProgramMegan RolfAdaptability is an ambiguous term that has taken on several definitions over the history of the field of genetics. At its most simple definition, adaptability refers to the ability of the organism in question to prosper or adjust to a new environment, which also implies the possibility of genetic-by-environment effects (i.e., alleles conferring different value depending on the environment; G×E). However, what this definition entails has grown ever more complex as our understanding of genetics has evolved. Initially, it was simply the consequence of selection, either natural or artificial, changing allele frequencies to match the species to the environment. In livestock production, however, local adaptation, whereby individuals or subpopulations are adjusted to a specific environment, and phenotypic plasticity, or the ability of an individual to perform across a wide variety of environments, are also relevant. Local adaptation can merely be thought of as an extension of selection for subpopulations in different environments. Phenotypic plasticity, however, involves consistency of performance across diverse environments. Phenotypic plasticity, which can also be thought of as a lack of environmental sensitivity to performance, may be important to livestock producers given the extensive use of seedstock across the United States and world via artificial insemination. However, it is unknown whether it is better to produce many sires suited for a particular environment, which would reduce selection intensity, or many sires selected for phenotypic plasticity, which would allow for an increase in selection intensity. However, selecting for increased phenotypic plasticity would likely have tradeoffs on other selection criteria. Furthermore, how to define, model, and select on phenotypic plasticity is still debatable. Local adaptation has traditionally been modeled via multi-trait animal models, where each trait is merely the phenotypic records for the trait of interest in different categories of the environment. As an example, yearling weight could be measured in many regions of the United States and the weaning weights in each region would be different traits. Each region then has a genetic correlation, which would correspond to the degree of G×E effects and environmental sensitivity. Phenotypic plasticity, however, can be thought of as an infinite number of “regions” or environments. In such a case, it may be better to characterize the environments, which can be classified by a continuous variable like temperature, by a function. Thus, random regression, which allows for random intercepts and slopes (or higher order polynomials) is the natural extension of a categorical variable into a continuous case for individuals, much like ANOVA and regression are represented by ANCOVA. Individuals with a random slope that counteracts the population expectation to move it towards zero would be considered more phenotypically plastic. The slope variance indicates the degree of G×E effects and variance. Furthermore, the random intercept (which can intuitively be centered around a baseline environment of interest) and random slope share a covariance structure, which define the relationship between baseline production and environmental sensitivity. Random regression has thus far been applied to evaluate G×E and life cycle or growth curves. The applications to G×E remain limited in the literature and focus mostly on dairy cattle with a few applications to beef cattle. With some notable exceptions, the prevailing trend appears to be an unfavorable relationship between baseline (intercept) performance and environmental sensitivity (slope) in most traits. Dry matter intake (DMI) and respiration rate were modeled as a function of water restriction (WR) and estimated breeding values (EBV) for the intercept and slope of WR were computed. Genetic correlations between the intercept and slope, permanent environment (PE) parameters, WR-specific genetic variances, WR-specific PE variances, WR-specific heritabilities, genetic correlations between different levels of WR were computed. Spearman rank correlations between EBVs from different levels of WR, and Beef Improvement Federation (BIF) accuracy at different levels of WR were also examined. Finally, a genome-wide association study was conducted on the intercept and slope traits to evaluate biological pathways and processes that might be contributing to each trait. The population slope for DMI under WR was -4.10 kg DMI per 100% WR. This indicated that DMI decreased on average as WR increased. Therefore, selection for positive slope EBVs is warranted to increase phenotypic plasticity. The slope variance was great enough at 3.16 〖((kg DMI)/(100% WR))〗^2 such that the phenotypic mean could easily be moved towards zero change in DMI as WR increased. However, the genetic correlation between the intercept, which represents DMI under normal, non-restricted conditions, was highly negative at -0.75. This indicates selection to increase phenotypic plasticity will decrease DMI under ad libitum conditions, which would result in a production loss under normal management. While it was log-transformed, respiration rate followed a similar trend with a much greater magnitude of genetic correlation between the intercept and slope at -0.98. This is interesting as respiration rate is predicted to decrease as WR increases. However, respiration rate may be a proxy for shedding metabolic heat generated from production (e.g., milk yield or muscle deposition and weight gain), which would explain this apparent conundrum. In general, estimated genetic variance decreased by nearly half from just greater than two kg DMI^2 at 0% WR to less than one kg DMI^2 at 50% WR; however, uncertainty was large. There was no evidence of a PE effect until about 40% or greater WR and, even then, the estimated variance was relatively small compared to the genetic variance. Log-transformed respiration rate variances were not shown for respiration rate given the difficulty in interpreting a transformed variable. Heritabilities of DMI and respiration rate varied by group, due to the inclusion of heterogeneous variances by group. Dry matter intake heritability followed a similar trend as the genetic variance, with estimates at 0% WR between 0.25 and 0.40. However, the declining trend was much smaller. It is worth noting the 95% credible intervals highly overlapped, indicating there may be little probability for a difference in heritability as WR increases. Interestingly, the trend for the DMI repeatability was even less steep, likely due to the evidence for PE effects as WR increased. Respiration rate WR-specific heritabilities and repeatabilities followed a similar pattern. However, there was little evidence for a heritable component past approximately 25% WR as the 95% credible intervals overlapped zero. Therefore, genetic correlations and accuracies of respiration rates were not considered past 25% WR for respiration rate. In general, the Spearman correlations between the EBVs at different levels of WR and genetic correlations between different levels of WR were in near agreeance. This is generally to be expected. Genetic correlations for DMI between different water restricted environments were generally high except at the most divergent WR values (e.g., 0% and 40% or 50%). Genetic and Spearman correlations between the most divergent environments for DMI were as low as 0.80, which is high and indicates small/few G×E effects and predictions in one environment should be accurate in others. The correlations for respiration rate at divergent WR values were even greater, only reaching as low as 0.975. However, this was only between 0% and 25% WR due to a lack of heritability beyond this point. While the correlations generally indicate selection based on non-restricted environments is an accurate indicator of performance in water restricted environments, it is worth noting the decreasing genetic variance is still potentially of concern for long term genetic improvement. Finally, the BIF accuracies generally increased as WR increased (with large 95% credible intervals) for both DMI and respiration rate. This is likely because of the greater amount of data present at higher WR levels or decrease in overall variance, which would likely have greater influence on the EBVs. Finally, GWAS was performed for both the intercept and slope of DMI and respiration rate. For DMI, metabolic signaling and exocytosis pathways and biological gene ontology (GO) terms were enriched for the intercept. In contrast, the slope for DMI was mostly related to central metabolism, including fat metabolism. A previous, traditional GWAS (i.e., no G×E included) also identified fat metabolism pathway genes for DMI, which may indicate models not accounting for G×E pick up signals from both the intercept and slope that are strong enough in both. This would corroborate with the high genetic correlation between the intercept and slope. Due to the lack of a heritable component at most values of WR, respiration rate GWAS was not conducted. Respiration rate and DMI were also modeled in a random regression model with a temperature humidity index (THI) as a covariate to model heat stress. Analyses mirror the WR covariate. However, there was less data available for respiration rate as two of the seven groups did not reach high enough THI levels. The population slope for DMI was -0.046 (kg DMI)/(1 unit THI), indicating daily DMI decreases, on average, as THI increases. This would be expected a priori and corresponds to a nearly one kg DMI decrease as THI approaches 90. Multiplied across an entire feedlot population, this represents a serious loss of production. Fortunately, the point estimate and bounds of the 95% credible interval for the slope variance indicate there is ample room for selection to improve phenotypic plasticity and move the population slope towards zero. Consistent with the G×E literature, there is an unfavorable genetic correlation of -0.78 between selection for increased production under thermoneutrality (the intercept) and environmental sensitivity (the slope). This indicates there is a tradeoff in selecting for phenotypic plasticity and productivity under optimal conditions. The population slope for the log-transformed respiration rates was positive, which would be expected as a response to heat stress a priori. While there was a great amount of uncertainty, the genetic correlation between the respiration rate intercept and slope under THI was negative at both bounds of the 95% credible interval. This is initially strange, as selection to reduce respiration rate under thermoneutral conditions would be expected to increase environmental sensitivity and decrease phenotypic plasticity whereas selection to increase respiration rate under thermoneutral conditions would apparently increase phenotypic plasticity. However, respiration rate under thermoneutral conditions has been associated with shedding metabolic heat generated from production. Thus, selection to increase respiration rate (and possibly production) in thermoneutral conditions would appear to decrease environmental sensitivity. In reality, selection to increase respiration rate in thermoneutral conditions would likely change the frequency of alleles associated with production. Therefore, the seemingly nonsensical relationship between the additive genetic intercept and slope is not favorable and selection to increase respiration rate/production in thermoneutral conditions would decrease production in heat-stressed conditions and increase environmental sensitivity. The THI-specific genetic variance for DMI rapidly decreased as THI increased, but appeared to stabilize past 80 THI at about 50% of thermoneutral conditions. At lower levels of THI, there was no evidence of a PE effect, but there was evidence for moderate PE effects at higher THI. The heritability and repeatability of DMI within each group reflected the trends of the variance components with heritability point estimates ranging from 0.3-0.4; but a great degree of overlap of the 95% credible intervals made it difficult to determine whether the decrease in heritability and repeatability was meaningful. As would be expected, the repeatability was similar to heritability at low THI, but declined less than heritability at high THI due to the additional PE variance. Respiration rate was very lowly heritable, with point estimates ranging from 0.02 to 0.04 at 70 THI for the different groups. However, uncertainty placed the point estimates anywhere from just above 0 to 0.08, indicating respiration rate is lowly heritable under varying THI. Genetic correlations between different levels of THI for DMI or respiration rate were both low. Genetic correlations for DMI and respiration rate between 70 and the upper bound of THI (85 for DMI and 80 for respiration rate) were 0.50 and 0.40, respectively. This is low and indicates performance at thermoneutrality is not a great indicator of performance in heavily heat stressed environments. Spearman correlations between the EBVs at the same THI levels were likewise low, with Spearman correlations between THI levels of 0.40 and 0.20 for DMI and respiration rate, respectively. Therefore, phenotypic plasticity likely plays a large role in environments with high THI. The BIF accuracies for EBVs at different levels of THI generally increased as THI increased. However, past 75 THI, there was no increase in accuracy observed. The GWAS for DMI intercept and slope under THI identified several variants associated with the traits. The intercept seemed to be mostly associated with energy balance signaling to generate adenosine triphosphate generation, whereas the slope was associated with a variety of pathways, including gastric acid secretion, cAMP signaling, Ras signaling, and fat metabolism. Most of these are growth pathways, but fat metabolism has been implicated in previous studies and provides a connection to pathways identified for the intercept, because energy balance signaling may be related. The intercept for respiration rate was the most interesting, as cardiomyopathy and heart function were the primary GO terms and pathways implicated. Cardiomyopathy and heart function are interesting, as high-altitude disease pulmonary arterial pressure and similar issues in the feedlot are thought to be related. A high-altitude disease GWAS identified similar gene candidates that were associated with heart function and cardiomyopathies. This may indicate respiration rate in thermoneutral environments could be related to either the ability of the animal to properly supply oxygen or a relationship between heart issues in the feedlot and high-altitude disease. The associated variants for the slope, however, were associated with GO terms and pathways related to metabolic signaling and metabolism. This strengthens the previous untested hypothesis relating respiration rate to increased metabolic heat