Predicting breed by environment interaction using ecological modelling

In most of African countries, livestock production branches from an ancient tradition where nearly all rural and peri-urban families keep different indigenous breeds in scavenging systems. In sub-Saharan Africa, where these production systems are the most prominent, livestock mainly forages for resources that are highly dependent on the local environment and season. Even though these breeds are said to be adapted to the local conditions, their productivity is still low compared to commercial breeds. There have been several efforts from researchers, policy makers and livestock specialists to introduce commercial breeds to support the generation of food security and poverty alleviation. However, most of these attempts have failed because of the non-adaptability of introduced breeds to the local conditions. Typically there is no prior knowledge on the environmental sensitivity from these breeds to this new tropical environments. Throughout this thesis I use Geographic Information Systems (GIS) that describe the environment, and models used in ecology to investigate the match of animals with their environment. The aim of this thesis was to evaluate how the environment plays a role in shaping differences in breed performance across agro-ecological zones, and what implications this can have in genetic improvement of livestock.

Several animal breeding studies tested breeds in different environments to evaluate whether genotypes respond differently to changes in the environment (i.e. G x E). To estimate if there is a re-ranking in breed/genotype performance between environments, a genetic correlation is estimated. To accurately estimate this correlation, an optimal mating design should be established. Breeding programs use full-sibs or half-sibs to perform testing in different environments. Within families, common environmental effects can be present generating a covariance between siblings, and should therefore be taken into account when estimating genetic correlations. In chapter 2, I used stochastic simulation to find the optimal population structure to accurately estimate the genetic correlation between environments using a combination of full-sibs and half-sibs groups under different levels of common environmental effects. Simulation results showed that when there are no common environmental effects present in the population, the mating ratio that gives the lowest standard error of the genetic correlation is of one female per male with 10 offspring per sire per environment. Not accounting for common environmental effects when these are present in the population will lead to an upward bias of the genetic correlation. Increasing the number of females per male to a minimum of 5, with 10 offspring per sire per environment will alleviate the impact of common environmental effects lowering the standard error of the genetic correlation. I suggest for studies that aim to estimate the magnitude of the G x E, to acknowledge the presence of common environmental effects and to take this into account when deciding the mating ratio.

In chapter 3, using GIS and habitat distribution models a methodology to predict breed suitability for different agro-ecological zones was developed. The methodology was tested on the current distribution of two introduced chicken breeds in Ethiopia. Results show that this methodology is effective in predicting breed suitability for specific environmental conditions. For both chicken breeds the model predicts suitable areas beyond their current extent, hence suggesting areas for breed introduction. The most significant variables that explain the current breed distribution were similar to the environmental conditions from which the breeds originate.

In chapter 3, only information on the location of the breeds was taken into account. This was extended in chapter 4, leading to an approach that predicts the productivity of the breeds. I present a methodology to model breed performance by using growth data from five different introduced chicken breeds in Ethiopia part of the African Chicken Genetic Gains project (ACGG; https://africacgg.net/). The suitability of these breeds was tested by predicting the response of body weight as a function of the environment in Ethiopia. Across the Ethiopian landscape, predicted body weights varied for all of the breeds. The variation in body weight was explained by different environmental variables, highlighting the importance of understanding the role of the environment in predicting breed productivity.

In chapter 4, breed performance was predicted within a single country. In chapter 5 breed performance was predicted across countries. Growth data was used from two chicken breeds that were introduced in Ethiopia, Nigeria and Tanzania by the ACGG project. The aim was to assess if the data from one country could be used to predict the performance of the same breed in the other two countries. The variation found in breed performance could be attributed to each breeds’ environmental sensitivity. The environmental variables responsible for shaping the variation in performance were different for each breed in each country. The accuracy of the prediction models projected from one country to the other show they can be used to identify areas for successful breed introduction.

In chapter 6 I discussed how the tools developed in this thesis can be used in animal breeding for different approaches. I suggest for different disciplines such as landscape genomics and ecology to work together with animal breeding to understand the role that the environment plays in shaping the observed phenotypic differences. This knowledge has implications for the development of breeding programs for different agro-ecologies, taking into account the continuous environmental variation. Furthermore, I recommend the use of these tools to generate knowledge on the impact of climate change on livestock to help generate mitigation plans and policy frameworks that will help in enhancing food security and preserving the current biodiversity.