Robert J. Tempelman

Reassessing design and analysis of two-colour microarray experiments using mixed effects models

Gene expression microarray studies have led to interesting experimental design and statistical analysis challenges. The comparison of expression profiles across populations is one of the most common objectives of microarray experiments. In this manuscript we review some issues regarding design and statistical analysis for two-colour microarray platforms using mixed linear models, with special attention directed towards the different hierarchical levels of replication and the consequent effect on the use of appropriate error terms for comparing experimental groups. We examine the traditional analysis of variance (ANOVA) models proposed for microarray data and their extensions to hierarchically replicated experiments. In addition, we discuss a mixed model methodology for power and efficiency calculations of different microarray experimental designs.

A Bayesian Antedependence Model for Whole Genome Prediction

Hierarchical mixed effects models have been demonstrated to be powerful for predicting genomic merit of livestock and plants, on the basis of high-density single-nucleotide polymorphism (SNP) marker panels, and their use is being increasingly advocated for genomic predictions in human health. Two particularly popular approaches, labeled BayesA and BayesB, are based on specifying all SNP-associated effects to be independent of each other. BayesB extends BayesA by allowing a large proportion of SNP markers to be associated with null effects. We further extend these two models to specify SNP effects as being spatially correlated due to the chromosomally proximal effects of causal variants. These two models, that we respectively dub as ante-BayesA and ante-BayesB, are based on a first-order nonstationary antedependence specification between SNP effects. In a simulation study involving 20 replicate data sets, each analyzed at six different SNP marker densities with average LD levels ranging from r2 = 0.15 to 0.31, the antedependence methods had significantly (P < 0.01) higher accuracies than their corresponding classical counterparts at higher LD levels (r2 > 0. 24) with differences exceeding 3%. A cross-validation study was also conducted on the heterogeneous stock mice data resource (http://mus.well.ox.ac.uk/mouse/HS/) using 6-week body weights as the phenotype. The antedependence methods increased cross-validation prediction accuracies by up to 3.6% compared to their classical counterparts (P < 0.001). Finally, we applied our method to other benchmark data sets and demonstrated that the antedependence methods were more accurate than their classical counterparts for genomic predictions, even for individuals several generations beyond the training data.

Linear reaction norm models for genetic merit prediction of Angus cattle under genotype by environment interaction

The objectives of this work were to assess alternative linear reaction norm (RN) models for genetic evaluation of Angus cattle in Brazil. That is, we investigated the interaction between genotypes and continuous descriptors of the environmental variation to examine evidence of genotype by environment interaction (G×E) in post-weaning BW gain (PWG) and to compare the environmental sensitivity of national and imported Angus sires. Data were collected by the Brazilian Angus Improvement Program from 1974 to 2005 and consisted of 63,098 records and a pedigree file with 95,896 animals. Six models were implemented using Bayesian inference and compared using the Deviance Information Criterion (DIC). The simplest model was M(1), a traditional animal model, which showed the largest DIC and hence the poorest fit when compared with the 4 alternative RN specifications accounting for G×E. In M(2), a 2-step procedure was implemented using the contemporary group posterior means of M(1) as the environmental gradient, ranging from -92.6 to +265.5 kg. Moreover, the benefits of jointly estimating all parameters in a 1-step approach were demonstrated by M(3). Additionally, we extended M(3) to allow for residual heteroskedasticity using an exponential function (M(4)) and the best fitting (smallest DIC) environmental classification model (M(5)) specification. Finally, M(6) added just heteroskedastic residual variance to M(1). Heritabilities were less at harsh environments and increased with the improvement of production conditions for all RN models. Rank correlations among genetic merit predictions obtained by M(1) and by the best fitting RN models M(3) (homoskedastic) and M(5) (heteroskedastic) at different environmental levels ranged from 0.79 and 0.81, suggesting biological importance of G×E in Brazilian Angus PWG. These results suggest that selection progress could be optimized by adopting environment-specific genetic merit predictions. The PWG environmental sensitivity of imported North American origin bulls (0.046 ± 0.009) was significantly larger (P < 0.05) than that of local sires (0.012 ± 0.013). Moreover, PWG of progeny of imported sires exceeded that of native sires in medium and superior production levels. On the other hand, Angus cattle locally selected in Brazil tended to be more robust to environmental changes and hence be more suitable when production environments for potential progeny is uncertain.

A large-scale study of differential gene expression in monocyte-derived macrophages infected with several strains of Mycobacterium avium subspecies paratuberculosis

Mycobacterium avium subspecies paratuberculosis (MAP) is a significant concern to the American and European dairy industries and possibly to human health. MAP possesses the rare ability to survive and replicate in infected macrophages, cells that are typically able to destroy pathogens. Little is known about what changes occur in MAP-infected macrophages that prevent phagosome maturation and lead to intracellular survival of the bacteria. In this study, a bovine immunologically specific cDNA microarray was used to study genes whose expression was altered in monocyte-derived macrophages (MDM) when these cells were infected with 10 different strains of MAP bacteria. Although we used MAP strains isolated from four different host species, cluster analysis of each strains influence in infected MDMs showed no species of origin specific MAP alterations in the host transcriptome. However, MAP strain K10 was observed as a clear outlier in the cluster analysis. Additionally, we observed two SuperShedder MAP strains clustering very closely together compared to the other strains in this study. Overall, microarray analysis yielded 78 annotated genes whose expression was altered by MAP infection, regardless of strain. Few of these genes have been previously studied in the context of Johnes disease or other mycobacterium-caused diseases. Large groups of apoptosis genes, transcription factors and cytokines were found to be differentially expressed in infected monocyte-derived macrophages as well as several genes not previously linked to MAP-host interactions. Identifying novel host genes affected by MAP infection of macrophages may lead to a more complete picture of this complex host-pathogen interaction.

Invited review: Milk production and reproductive performance: Modern interdisciplinary insights into an enduring axiom

A general belief across the dairy community, both scientific and commercial, is that of an antagonistic association between milk production and reproductive performance of dairy cows. In this article, we critically review the evidence supporting this belief and discuss some of its limitations. Based on the fundamental principles of experimental design and inference, we consider relevant issues that, although critical to the very foundation of the perceived production-reproduction antagonism, seem to have been previously misrepresented or overlooked. In particular, we focus on issues of confounding, randomization, nature of inference, single- versus multiple-trait modeling, cow- versus herd-level modeling, and scope of inference, all within the context of dairy production systems. Taken together, these issues indicate that the production-reproduction antagonism may not be as pervasive as previously believed, suggesting the need for more rigorous methods of scientific investigation on this matter. We revisit the association between milk production and reproductive performance using a novel interdisciplinary approach based on cutting-edge statistical methods that accommodate some of the unique and previously ignored features of this problem. In fact, recent work supports a highly heterogeneous association between milk production and reproductive performance, whereby heterogeneity is partitioned across several scales and driven by many contributing factors, both physiological and managerial. We conclude that the relationship between milk production and reproductive performance is not necessarily that of a universal homogeneous antagonism and suggest better ways to study and even manage this association. A more comprehensive assessment that draws expertise from multiple scientific disciplines will be required to elicit management recommendations targeted to effectively optimize overall performance of dairy cows and commercial herds.

Genome-wide linkage analysis of global gene expression in loin muscle tissue identifies candidate genes in pigs.

Background Nearly 6,000 QTL have been reported for 588 different traits in pigs, more than in any other livestock species. However, this effort has translated into only a few confirmed causative variants. A powerful strategy for revealing candidate genes involves expression QTL (eQTL) mapping, where the mRNA abundance of a set of transcripts is used as the response variable for a QTL scan. Methodology/Principal Findings We utilized a whole genome expression microarray and an F2 pig resource population to conduct a global eQTL analysis in loin muscle tissue, and compared results to previously inferred phenotypic QTL (pQTL) from the same experimental cross. We found 62 unique eQTL (FDR <10%) and identified 3 gene networks enriched with genes subject to genetic control involved in lipid metabolism, DNA replication, and cell cycle regulation. We observed strong evidence of local regulation (40 out of 59 eQTL with known genomic position) and compared these eQTL to pQTL to help identify potential candidate genes. Among the interesting associations, we found aldo-keto reductase 7A2 (AKR7A2) and thioredoxin domain containing 12 (TXNDC12) eQTL that are part of a network associated with lipid metabolism and in turn overlap with pQTL regions for marbling, % intramuscular fat (% fat) and loin muscle area on Sus scrofa (SSC) chromosome 6. Additionally, we report 13 genomic regions with overlapping eQTL and pQTL involving 14 local eQTL. Conclusions/Significance Results of this analysis provide novel candidate genes for important complex pig phenotypes.

Heterogeneity in genetic and nongenetic variation and energy sink relationships for residual feed intake across research stations and countries

Our long-term objective is to develop breeding strategies for improving feed efficiency in dairy cattle. In this study, phenotypic data were pooled across multiple research stations to facilitate investigation of the genetic and nongenetic components of feed efficiency in Holstein cattle. Specifically, the heritability of residual feed intake (RFI) was estimated and heterogeneous relationships between RFI and traits relating to energy utilization were characterized across research stations. Milk, fat, protein, and lactose production converted to megacalories (milk energy; MilkE), dry matter intakes (DMI), and body weights (BW) were collected on 6,824 lactations from 4,893 Holstein cows from research stations in Scotland, the Netherlands, and the United States. Weekly DMI, recorded between 50 to 200 d in milk, was fitted as a linear function of MilkE, BW0.75, and change in BW (ΔBW), along with parity, a fifth-order polynomial on days in milk (DIM), and the interaction between this polynomial and parity in a first-stage model. The residuals from this analysis were considered to be a phenotypic measure of RFI. Estimated partial regression coefficients of DMI on MilkE and on BW0.75 ranged from 0.29 to 0.47 kg/Mcal for MilkE across research stations, whereas estimated partial regression coefficients on BW0.75 ranged from 0.06 to 0.16 kg/kg0.75. Estimated partial regression coefficients on ΔBW ranged from 0.06 to 0.39 across stations. Heritabilities for country-specific RFI were based on fitting second-stage random regression models and ranged from 0.06 to 0.24 depending on DIM. The overall heritability estimate across all research stations and all DIM was 0.15±0.02, whereas an alternative analysis based on combining the first- and second-stage model as 1 model led to an overall heritability estimate of 0.18±0.02. Hence future genomic selection programs on feed efficiency appear to be promising; nevertheless, care should be taken to allow for potentially heterogeneous variance components and partial relationships between DMI and other energy sink traits across environments when determining RFI.

Cumulative t-link threshold models for the genetic analysis of calving ease scores

In this study, a hierarchical threshold mixed model based on a cumulative t-link specification for the analysis of ordinal data or more, specifically, calving ease scores, was developed. The validation of this model and the Markov chain Monte Carlo (MCMC) algorithm was carried out on simulated data from normally and t4 (i.e. a t-distribution with four degrees of freedom) distributed populations using the deviance information criterion (DIC) and a pseudo Bayes factor (PBF) measure to validate recently proposed model choice criteria. The simulation study indicated that although inference on the degrees of freedom parameter is possible, MCMC mixing was problematic. Nevertheless, the DIC and PBF were validated to be satisfactory measures of model fit to data. A sire and maternal grandsire cumulative t-link model was applied to a calving ease dataset from 8847 Italian Piemontese first parity dams. The cumulative t-link model was shown to lead to posterior means of direct and maternal heritabilities (0.40 ± 0.06, 0.11 ± 0.04) and a direct maternal genetic correlation (-0.58 ± 0.15) that were not different from the corresponding posterior means of the heritabilities (0.42 ± 0.07, 0.14 ± 0.04) and the genetic correlation (-0.55 ± 0.14) inferred under the conventional cumulative probit link threshold model. Furthermore, the correlation (> 0.99) between posterior means of sire progeny merit from the two models suggested no meaningful rerankings. Nevertheless, the cumulative t-link model was decisively chosen as the better fitting model for this calving ease data using DIC and PBF.

Additive and Nonadditive Genetic Variation for Conformation Traits in Canadian Holsteins

Abstract Additive and dominance genetic variances for nine conformation traits in the Canadian Holstein population were estimated by REML using the derivative-free algorithm. All traits were analyzed under two sire and dam models differing in assumptions on the data structure. Within-herd relationships due to dam were defined first to be nested within sire in order to estimate both genetic variances, assuming unimportant biases on the estimates. The potential impact of some of these biases was estimated as the difference between the two parental variance components under a cross-classified model. Heritabilities were similar to those previously computed for Canadian Holsteins and ranged from .10 to .30. Results from the hierarchical dam-within-sire model suggest dominance genetic variation, as a proportion of the total variation, to be important for final score (.15), capacity (.16), and mammary system (.13). However, estimates of dam components, as a proportion of the total variance, were significantly greater than sire components for final score (.019), general appearance (.020), and feet and legs (.018) under the cross-classified model. Therefore, the significant dominance genetic parameter estimated for final score under the hierarchical model may have been biased upward.

An adjuvant-free mouse model to evaluate the allergenicity of milk whey protein

Milk allergy is the most common type of food allergy in humans with the potential for fatality. An adjuvant-free mouse model would be highly desirable as a preclinical research tool to develop novel hypoallergenic or nonallergenic milk products. Here we describe an adjuvant-free mouse model of milk allergy that uses transdermal sensitization followed by oral challenge with milk protein. Groups of BALB/c mice were exposed to milk whey protein via a transdermal route, without adjuvant. Systemic IgG1 and IgE antibody responses to transdermal exposure as well as systemic anaphylaxis and hypothermia response to oral protein challenge were studied. Transdermal exposure resulted in a time- and dose-dependent induction of significant IgE and IgG1 antibody responses. Furthermore, oral challenge of sensitized mice resulted in significant clinical symptoms of systemic anaphylaxis within 1 h and significant hypothermia at 30 min postchallenge. To study the underlying mechanism, we examined allergen-driven spleen cell T-helper 2 cytokine (IL-4) responses. There was a robust dose- and time-dependent activation of memory IL-4 responses in allergic mice but not in healthy control mice. These data demonstrate for the first time a novel transdermal sensitization followed by oral challenge mouse model of milk allergy that does not use adjuvant. It is expected that this model may be used not only to study mechanisms of milk allergy, but also to evaluate novel milk products for allergenic potential and aid in the production of hypo- or nonallergenic milk products.