T.H.E. Meuwissen

Natural Resources Institute Finland, Helsinginkylä, Province of Western Finland, Finland

Are you T.H.E. Meuwissen?

Claim your profile

Publications (253)472.25 Total impact

  • J. Ødegård · T. Meuwissen

    No preview · Article · Feb 2016 · Journal of Animal Breeding and Genetics
  • Theo Meuwissen · Ben Hayes · Mike Goddard

    No preview · Article · Jan 2016
  • Source
    S van den Berg · M P L Calus · T H E Meuwissen · Y C J Wientjes
    [Show abstract] [Hide abstract]
    ABSTRACT: Background: The use of information across populations is an attractive approach to increase the accuracy of genomic prediction for numerically small populations. However, accuracies of across population genomic prediction, in which reference and selection individuals are from different populations, are currently disappointing. It has been shown for within population genomic prediction that Bayesian variable selection models outperform GBLUP models when the number of QTL underlying the trait is low. Therefore, our objective was to identify across population genomic prediction scenarios in which Bayesian variable selection models outperform GBLUP in terms of prediction accuracy. In this study, high density genotype information of 1033 Holstein Friesian, 105 Groningen White Headed, and 147 Meuse-Rhine-Yssel cows were used. Phenotypes were simulated using two changing variables: (1) the number of QTL underlying the trait (3000, 300, 30, 3), and (2) the correlation between allele substitution effects of QTL across populations, i.e. the genetic correlation of the simulated trait between the populations (1.0, 0.8, 0.4). Results: The accuracy obtained by the Bayesian variable selection model was depending on the number of QTL underlying the trait, with a higher accuracy when the number of QTL was lower. This trend was more pronounced for across population genomic prediction than for within population genomic prediction. It was shown that Bayesian variable selection models have an advantage over GBLUP when the number of QTL underlying the simulated trait was small. This advantage disappeared when the number of QTL underlying the simulated trait was large. The point where the accuracy of Bayesian variable selection and GBLUP became similar was approximately the point where the number of QTL was equal to the number of independent chromosome segments (M e ) across the populations. Conclusion: Bayesian variable selection models outperform GBLUP when the number of QTL underlying the trait is smaller than M e . Across populations, M e is considerably larger than within populations. So, it is more likely to find a number of QTL underlying a trait smaller than M e across populations than within population. Therefore Bayesian variable selection models can help to improve the accuracy of across population genomic prediction.
    Full-text · Article · Dec 2015 · BMC Genetics
  • M Lillehammer · A K Sonesson · T H E Meuwissen
    [Show abstract] [Hide abstract]
    ABSTRACT: The aim of this study was to test how genetic gain for a trait not measured on the nucleus animals could be obtained within a genomic selection pig breeding scheme. Stochastic simulation of a pig breeding program including a breeding nucleus, a multiplier to produce and disseminate semen and a production tier where phenotypes were recorded was performed to test (1) the effect of obtaining phenotypic records from offspring of nucleus animals, (2) the effect of genotyping production animals with records for the purpose of including them in a genomic selection reference population or (3) to combine the two approaches. None of the tested strategies affected genetic gain if the trait under investigation had a low economic value of only 10% of the total breeding goal. When the relative economic weight was increased to 30%, a combination of the methods was most effective. Obtaining records from offspring of already genotyped nucleus animals had more impact on genetic gain than to genotype more distant relatives with phenotypes to update the reference population. When records cannot be obtained from offspring of nucleus animals, genotyping of production animals could be considered for traits with high economic importance.
    No preview · Article · Dec 2015 · animal
  • Source
    Theo H. E. Meuwissen · Morten Svendsen · Trygve Solberg · Jørgen Ødegård
    [Show abstract] [Hide abstract]
    ABSTRACT: In dairy cattle, current genomic predictions are largely based on sire models that analyze daughter yield deviations of bulls, which are derived from pedigree-based animal model evaluations (in a two-step approach). Extension to animal model genomic predictions (AMGP) is not straightforward, because most of the animals that are involved in the genetic evaluation are not genotyped. In single-step genomic best linear unbiased prediction (SSGBLUP), the pedigree-based relationship matrix A and the genomic relationship matrix G are combined in a matrix H, which allows for AMGP. However, as the number of genotyped animals increases, imputation of the genotypes for all animals in the pedigree may be considered. Our aim was to impute genotypes for all animals in the pedigree, construct alternative relationship matrices based on the imputation results, and evaluate the accuracy of the resulting AMGP by cross-validation in the national Norwegian Red dairy cattle population. A large-scale national dataset was effectively handled by splitting it into two sets: (1) genotyped animals and their ancestors (i.e. GA set with 20,918 animals) and (2) the descendants of the genotyped animals (i.e. D set with 4,022,179 animals). This allowed restricting genomic computations to a relatively small set of animals (GA set), whereas the majority of the animals (D set) were added to the animal model equations using Henderson’s rules, in order to make optimal use of the D set information. Genotypes were imputed by segregation analysis of a large pedigree with relatively few genotyped animals (3285 out of 20,918). Among the AMGP models, the linkage and linkage disequilibrium based G matrix (G LDLA0 ) yielded the highest accuracy, which on average was 0.06 higher than with SSGBLUP and 0.07 higher than with two-step sire genomic evaluations. AMGP methods based on genotype imputation on a national scale were developed, and the most accurate method, G LDLA0 BLUP, combined linkage and linkage disequilibrium information. The advantage of AMGP over a sire model based on two-step genomic predictions is expected to increase as the number of genotyped cows increases and for species, with smaller sire families and more dam relationships.
    Full-text · Article · Dec 2015 · Genetics Selection Evolution
  • K. H. Martinsen · J. Ødegård · D. Olsen · T. H. E. Meuwissen
    [Show abstract] [Hide abstract]
    ABSTRACT: Feed costs amount to approximately 70% of the total costs in pork production, and feed efficiency is, therefore, an important trait for improving pork production efficiency. Production efficiency is generally improved by selection for high lean growth rate, reduced backfat, and low feed intake. These traits have given an effective slaughter pig but may cause problems in piglet production due to sows with limited body reserves. The aim of the present study was to develop a measure for feed efficiency that expressed the feed requirements per 1 kg deposited lean meat and fat, which is not improved by depositing less fat. Norwegian Landrace ( = 8,161) and Duroc ( = 7,202) boars from Topigs Norsvin's testing station were computed tomography scanned to determine their deposition of lean meat and fat. The trait was analyzed in a univariate animal model, where total feed intake in the test period was the dependent variable and fat and lean meat were included as random regression cofactors. These cofactors were measures for fat and lean meat efficiencies of individual boars. Estimation of fraction of total genetic variance due to lean meat or fat efficiency was calculated by the ratio between the genetic variance of the random regression cofactor and the total genetic variance in total feed intake during the test period. Genetic variance components suggested there was significant genetic variance among Norwegian Landrace and Duroc boars in efficiency for deposition of lean meat (0.23 ± 0.04 and 0.38 ± 0.06) and fat (0.26 ± 0.03 and 0.17 ± 0.03) during the test period. The fraction of the total genetic variance in feed intake explained by lean meat deposition was 12% for Norwegian Landrace and 15% for Duroc. Genetic fractions explained by fat deposition were 20% for Norwegian Landrace and 10% for Duroc. The results suggested a significant part of the total genetic variance in feed intake in the test period was explained by fat and lean meat efficiency. These new efficiency measures may give the breeders opportunities to select for animals with a genetic potential to deposit lean meat efficiently and at low feed costs in slaughter pigs rather than selecting for reduced the feed intake and backfat.
    No preview · Article · Oct 2015 · Journal of Animal Science
  • Source

    Full-text · Dataset · Sep 2015
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: The pig is a well-known animal model used to investigate genetic and mechanistic aspects of human disease biology. They are particularly useful in the context of obesity and metabolic diseases because other widely used models (e.g. mice) do not completely recapitulate key pathophysiological features associated with these diseases in humans. Therefore, we established a F2 pig resource population (n = 564) designed to elucidate the genetics underlying obesity and metabolic phenotypes. Segregation of obesity traits was ensured by using breeds highly divergent with respect to obesity traits in the parental generation. Several obesity and metabolic phenotypes were recorded (n = 35) from birth to slaughter (242 ± 48 days), including body composition determined at about two months of age (63 ± 10 days) via dual-energy x-ray absorptiometry (DXA) scanning. All pigs were genotyped using Illumina Porcine 60k SNP Beadchip and a combined linkage disequilibrium-linkage analysis was used to identify genome-wide significant associations for collected phenotypes. We identified 229 QTLs which associated with adiposity- and metabolic phenotypes at genome-wide significant levels. Subsequently comparative analyses were performed to identify the extent of overlap between previously identified QTLs in both humans and pigs. The combined analysis of a large number of obesity phenotypes has provided insight in the genetic architecture of the molecular mechanisms underlying these traits indicating that QTLs underlying similar phenotypes are clustered in the genome. Our analyses have further confirmed that genetic heterogeneity is an inherent characteristic of obesity traits most likely caused by segregation or fixation of different variants of the individual components belonging to cellular pathways in different populations. Several important genes previously associated to obesity in human studies, along with novel genes were identified. Altogether, this study provides novel insight that may further the current understanding of the molecular mechanisms underlying human obesity.
    Full-text · Article · Sep 2015 · PLoS ONE
  • Source
    Jørgen Ødegård · Theo H E Meuwissen
    [Show abstract] [Hide abstract]
    ABSTRACT: Background: Genomic selection (GS) allows estimation of the breeding value of individuals, even for non-phenotyped animals. The aim of the study was to examine the potential of identity-by-descent genomic selection (IBD-GS) in genomic selection for a binary, sib-evaluated trait, using different strategies of selective genotyping. This low-cost GS approach is based on linkage analysis of sparse genome-wide marker loci. Findings: Lowly to highly heritable (h(2) = 0.15, 0.30 or 0.60) binary traits with varying incidences (10 to 90%) were simulated for an aquaculture-like population. Genotyping was restricted to the 30% best families according to phenotype, using three genotyping strategies for training sibs. IBD-GS increased genetic gain compared to classical pedigree-based selection; the differences were largest at incidences of 10 to 50% of the desired category (i.e. a relative increase in genetic gain greater than 20%). Furthermore, the relative advantage of IBD-GS increased as the heritability of the trait increased. Differences were small between genotyping strategies, and most of the improvement was achieved by restricting genotyping to sibs with the least common binary phenotype. Genetic gains of IBD-GS relative to pedigree-based models were highest at low to moderate (10 to 50%) incidences of the category selected for, but decreased substantially at higher incidences (80 to 90%). Conclusions: The IBD-GS approach, combined with sparse and selective genotyping, is well suited for genetic evaluation of binary traits. Genetic gain increased considerably compared with classical pedigree-based selection. Most of the improvement was achieved by selective genotyping of the sibs with the least common (minor) binary category phenotype. Furthermore, IBD-GS had greater advantage over classical pedigree-based models at low to moderate incidences of the category selected for.
    Full-text · Article · Apr 2015 · Genetics Selection Evolution
  • Source
    M L van Pelt · T H E Meuwissen · G de Jong · R F Veerkamp
    [Show abstract] [Hide abstract]
    ABSTRACT: Longevity, productive life, or lifespan of dairy cattle is an important trait for dairy farmers, and it is defined as the time from first calving to the last test date for milk production. Methods for genetic evaluations need to account for censored data; that is, records from cows that are still alive. The aim of this study was to investigate whether these methods also need to take account of survival being genetically a different trait across the entire lifespan of a cow. The data set comprised 112,000 cows with a total of 3,964,449 observations for survival per month from first calving until 72 mo in productive life. A random regression model with second-order Legendre polynomials was fitted for the additive genetic effect. Alternative parameterizations were (1) different trait definitions for the length of time interval for survival after first calving (1, 3, 6, and 12 mo); (2) linear or threshold model; and (3) differing the order of the Legendre polynomial. The partial derivatives of a profit function were used to transform variance components on the survival scale to those for lifespan. Survival rates were higher in early life than later in life (99 vs. 95%). When survival was defined over 12-mo intervals survival curves were smooth compared with curves when 1-, 3-, or 6-mo intervals were used. Heritabilities in each interval were very low and ranged from 0.002 to 0.031, but the heritability for lifespan over the entire period of 72 mo after first calving ranged from 0.115 to 0.149. Genetic correlations between time intervals ranged from 0.25 to 1.00. Genetic parameters and breeding values for the genetic effect were more sensitive to the trait definition than to whether a linear or threshold model was used or to the order of Legendre polynomial used. Cumulative survival up to the first 6 mo predicted lifespan with an accuracy of only 0.79 to 0.85; that is, reliability of breeding value with many daughters in the first 6 mo can be, at most, 0.62 to 0.72, and changes of breeding values are still expected when daughters are getting older. Therefore, an improved model for genetic evaluation should treat survival as different traits during the lifespan by splitting lifespan in time intervals of 6 mo or less to avoid overestimated reliabilities and changes in breeding values when daughters are getting older. Copyright © 2015 American Dairy Science Association. Published by Elsevier Inc. All rights reserved.
    Full-text · Article · Apr 2015 · Journal of Dairy Science
  • J.A. Woolliams · P. Berg · B.S. Dagnachew · T.H.E. Meuwissen
    [Show abstract] [Hide abstract]
    ABSTRACT: Genetic contributions were first formalized in 1958 by James and McBride (Journal of Genetics, 56, 55-62) and have since been shown to provide a unifying framework for theories of gain and inbreeding. As such they have underpinned the development of methods that provide the most effective combination of maximizing gain whilst managing inbreeding and loss of genetic variation. It is shown how this optimum contribution technology can be developed from theory and adapted to provide practical selection protocols for a wide variety of situations including overlapping generations and multistage selection. The natural development of the theory to incorporate genomic selection and genomic control of inbreeding is also shown. © 2015 Blackwell Verlag GmbH.
    No preview · Article · Apr 2015 · Journal of Animal Breeding and Genetics
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Livestock production is the most important component of northern European agriculture and contributes to and will be affected by climate change. Nevertheless, the role of farm animal genetic resources in the adaptation to new agro-ecological conditions and mitigation of animal production's effects on climate change has been inadequately discussed despite there being several important associations between animal genetic resources and climate change issues. The sustainability of animal production systems and future food security require access to a wide diversity of animal genetic resources. There are several genetic questions that should be considered in strategies promoting adaptation to climate change and mitigation of environmental effects of livestock production. For example, it may become important to choose among breeds and even among farm animal species according to their suitability to a future with altered production systems. Some animals with useful phenotypes and genotypes may be more useful than others in the changing environment. Robust animal breeds with the potential to adapt to new agro-ecological conditions and tolerate new diseases will be needed. The key issue in mitigation of harmful greenhouse gas effects induced by livestock production is the reduction of methane (CH 4) emissions from ruminants. There are differences in CH 4 emissions among breeds and among individual animals within breeds that suggest a potential for improvement in the trait through genetic selection. Characterization of breeds and individuals with modern genomic tools should be applied to identify breeds that have genetically adapted to marginal conditions and to get critical information for breeding and conservation programs for farm animal genetic resources. We conclude that phenotyping and genomic technologies and adoption of new breeding approaches, such as genomic selection introgression, will promote breeding for useful characters in livestock species.
    Full-text · Article · Mar 2015 · Frontiers in Genetics
  • [Show abstract] [Hide abstract]
    ABSTRACT: The term functionality in animal breeding is used for traits that increase the efficiency of production by lowering the input cost, such as animal health and leg weakness related to longevity. The main objective of the study was to investigate the impact of genomic information, in a multivariate variance component analysis, on some of these traits. In addition, the effect of the inclusion was studied by testing the model's prediction ability based on best linear unbiased estimates for fixed and random effects. The material in this study consists of phenotypes from 76 683 animals, of which 4933 animals are genotyped. The heritabilities for front leg conformation, stayability, osteochondrosis and arched back, estimated using the traditional pedigree, were found to be between 0.12 and 0.29. When using the combined genomic and pedigree relationship matrix, the heritabilities were between 0.14 and 0.36. The results show that the combined relationship matrix can be used for the estimation of (co)variance components, and that the predictive ability of the model in this study marginally increases with the inclusion of genomic information. © 2015 Blackwell Verlag GmbH.
    No preview · Article · Mar 2015 · Journal of Animal Breeding and Genetics
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Reliability of genomic selection (GS) models was tested in an admixed population of Atlantic salmon, originating from crossing of several wild subpopulations. The models included ordinary genomic BLUP models (GBLUP), using genome-wide SNP markers of varying densities (1-220 k), a genomic identity-by-descent model (IBD-GS), using linkage analysis of sparse genome-wide markers, as well as a classical pedigree-based model. Reliabilities of the models were compared through 5-fold cross-validation. The traits studied were salmon lice (Lepeophtheirus salmonis) resistance (LR), measured as (log) density on the skin and fillet color (FC), with respective estimated heritabilities of 0.14 and 0.43. All genomic models outperformed the classical pedigree-based model, for both traits and at all marker densities. However, the relative improvement differed considerably between traits, models and marker densities. For the highly heritable FC, the IBD-GS had similar reliability as GBLUP at high marker densities (>22 k). In contrast, for the lowly heritable LR, IBD-GS was clearly inferior to GBLUP, irrespective of marker density. Hence, GBLUP was robust to marker density for the lowly heritable LR, but sensitive to marker density for the highly heritable FC. We hypothesize that this phenomenon may be explained by historical admixture of different founder populations, expected to reduce short-range lice density (LD) and induce long-range LD. The relative importance of LD/relationship information is expected to decrease/increase with increasing heritability of the trait. Still, using the ordinary GBLUP, the typical long-range LD of an admixed population may be effectively captured by sparse markers, while efficient utilization of relationship information may require denser markers (e.g., 22 k or more).
    Full-text · Article · Nov 2014 · Frontiers in Genetics
  • Theo H. E. Meuwissen · Anna K Sonesson · Jrgen degrd
    [Show abstract] [Hide abstract]
    ABSTRACT: Abstract Text: In the literature an abundance of genomic relationship matrices have been described which mainly differ in the age of the relationships that they trace. Marker based relationship matrices (G) generally trace very old relationships, since the marker mutations occurred. Pedigree (A) and linkage analysis relationship matrices (GLA) trace relationships since pedigree recording started, i.e. since the founder population. Genomic selection (GS) is based on three sources of information: a) pedigree relationships (A); b) linkage analysis (LA) information and c)linkage disequilibrium (LD) information, where LD is defined between alleles in the founder population. LD due to cosegregation of alleles in the known pedigree is denoted LA information. The described relationship matrices follow the same pattern, i.e. A, GLA and G, respectively. Keywords:genomic selection genomic relationships genetic modelling whole genome sequence data
    No preview · Conference Paper · Aug 2014
  • [Show abstract] [Hide abstract]
    ABSTRACT: Abstract Text: Purebred Duroc and Yorkshire sows were crossed with Göttingen minipig boars to obtain two separate F2 intercross resource populations (n=287 and 279 respectively). Several obesity, metabolic and slaughter measurements were recorded from birth to slaughter (220±45 days). In addition, body composition was determined at about two months of age (64±11 days) via dual-energy x-ray absorptiometry (DXA) scanning. All pigs were genotyped using Illumina Porcine 60k SNP Beadchip and a combined LDLA approach was used to perform genome-wide linkage and association analysis for body fat traits. Subsequently bioinformatic analysis was performed to identify genes in close proximity of chromosomal positions where statistically significant QTLs were identified. Several important genes previously linked to obesity (e.g. BBS4, CHRNA2, DLK1), along with other novel genes were identified, that together provide novel insights that may further the current understanding of the molecular mechanisms underlying human obesity. Keywords: LDLA, Pig model, QTL mapping
    No preview · Conference Paper · Aug 2014
  • Xijiang Yu · Theo H. E. Meuwissen · Matthew Baranski · Anna K Sonesson
    [Show abstract] [Hide abstract]
    ABSTRACT: Abstract Text: Atlantic cod families from year 2009 of the Norwegian national cod breeding program were challenged for viral nervous necrosis and vibriosis. Mortality was recorded. Around 1600 offspring and their parents were genotyped at 10,913 SNP loci, covering 2,285 scaffolds/contigs in the reference genome, which accounts for ~71.3% of total sequence length. Genomic enabled breeding values (GEBV) were estimated. A 10-folds cross-validation shows that the correlations of the survival states and corresponding GEBV were 0.085 for vibriosis and 0.55 for VNN. Whole genome resequencing of 111 parents was performed to an approximately 12x coverage per individual. Variant calling in the sequence of a subset of parents showed that all 12K SNP array SNPs were called and had matching genotypes. Imputation with Beagle and LDMIP software will enable inference of sequence data for all the challenge tested fish and the resulting improvement in accuracy will be investigated. Keywords: Atlantic cod Disease Sequence
    No preview · Conference Paper · Aug 2014
  • Source
    Kahsay G Nirea · Anna K Sonesson · Marie Lillehammer · Theo H. E. Meuwissen
    [Show abstract] [Hide abstract]
    ABSTRACT: Abstract Text: A higher accuracy of prediction was obtained for within family genomic selection as compared to the conventional selection method. The accuracy of selection was higher when the family structure was 10x10 followed by 1x10 and 2x2 specifically for within family genomic selection. This was in accordance with the number of full sibs and half sibs produced which increased relationships within or across test and candidate sibs. In all scenarios, accuracy of selection increased as family size increases but the increase was moderate when family size was beyond 40-50 individuals per family. In addition, the benefit would be higher if more full sibs and half sibs are available in the test and candidate group. Keywords: Aquaculture Within-Family Genomic selection Breeding Genetics
    Full-text · Conference Paper · Aug 2014
  • John A. Woolliams · Kahsay G Nirea · Theo H. E. Meuwissen
    [Show abstract] [Hide abstract]
    ABSTRACT: Abstract Text: Predicting gain for optimum contribution selection is associated with two issues, the first concerned with inter-generational dependence of the contributions, and the second concerned with dynamic desirability. By ignoring the latter, which is valid when the accuracy of candidates approaches 1, a formula for ΔG(T, ΔF, α) can be obtained, where ΔG(T, ΔF, α) is the maximum gain possible with T candidates per generation, rate of inbreeding ΔF, and degree of coancestry α. Simulation showed predictions were reasonable, although further validation is required. The developed theory made testable predictions that the importance of mating designs depended only on their impact on α as accuracy approaches 1, and simulations also validated this prediction. Mating designs that affect α retain impact because they affect both the variance of the Mendelian sampling term and the relationship between squared contributions and ΔF. Keywords: Rate of Gain, Rate of Inbreeding, Optimum Contributions, Genomic Selection, Mating Design.
    No preview · Conference Paper · Aug 2014
  • Binyam S. Dagnachew · Theo H. E. Meuwissen
    [Show abstract] [Hide abstract]
    ABSTRACT: Abstract Text: A novel iterative algorithm, Gencont2, for calculating optimum genetic contributions was developed. It was validated by comparing it with a previous program, Gencont, on three datasets obtained from practical breeding programs of three species (cattle, pig and sheep). The numbers of selection candidates were 2,929, 3,907 and 6,875 for the pig, cattle and sheep datasets respectively. In most cases, both algorithms select the same candidates and gave very similar results in genetic gain. In cases, when there were differences in number of animals to select, the extra selected candidates had contributions within the range of 0.006–0.08%. The correlations between assigned contributions were very close to 1; however, Gencont2 considerably decreased the computation time by 90% to 95% (13 to 22 times faster) compared to Gencont. This fast iterative algorithm makes the practical implementation of OC selection feasible in large scale breeding programs. Keywords: Inbreeding optimum genetic contributions genetic gain
    No preview · Conference Paper · Aug 2014

Publication Stats

8k Citations
472.25 Total Impact Points

Institutions

  • 2015
    • Natural Resources Institute Finland
      Helsinginkylä, Province of Western Finland, Finland
  • 2005-2015
    • Norwegian University of Life Sciences (NMBU)
      • Department of Animal and Aquacultural Sciences (IHA)
      Aas, Akershus, Norway
  • 2005-2014
    • Life University
      Marietta, Georgia, United States
  • 2005-2010
    • University of Melbourne
      • Department of Agriculture and Food Systems
      Melbourne, Victoria, Australia
  • 2004-2008
    • Wageningen University
      • Animal Breeding and Genomics Centre
      Wageningen, Provincie Gelderland, Netherlands
  • 2001
    • The University of Edinburgh
      • Institute of Cell Biology
      Edinburgh, Scotland, United Kingdom
  • 1995-2001
    • Merck Animal Health Netherlands
      Boksmeer, North Brabant, Netherlands
  • 1998
    • University of Nebraska at Lincoln
      Lincoln, Nebraska, United States
  • 1994
    • University of Guelph
      • Department of Animal and Poultry Science
      Guelph, Ontario, Canada
  • 1993
    • Sociaal en Cultureel Planbureau
      's-Gravenhage, South Holland, Netherlands