Ernst Wit

Markers of Adenocarcinoma Characteristic of the Site of Origin: Development of a Diagnostic Algorithm

Purpose: Patients with metastatic adenocarcinoma of unknown origin are a common clinical problem. Knowledge of the primary site is important for their management, but histologically, such tumors appear similar. Better diagnostic markers are needed to enable the assignment of metastases to likely sites of origin on pathologic samples. Experimental Design: Expression profiling of 27 candidate markers was done using tissue microarrays and immunohistochemistry. In the first (training) round, we studied 352 primary adenocarcinomas, from seven main sites (breast, colon, lung, ovary, pancreas, prostate and stomach) and their differential diagnoses. Data were analyzed in Microsoft Access and the Rosetta system, and used to develop a classification scheme. In the second (validation) round, we studied 100 primary adenocarcinomas and 30 paired metastases. Results: In the first round, we generated expression profiles for all 27 candidate markers in each of the seven main primary sites. Data analysis led to a simplified diagnostic panel and decision tree containing 10 markers only: CA125, CDX2, cytokeratins 7 and 20, estrogen receptor, gross cystic disease fluid protein 15, lysozyme, mesothelin, prostate-specific antigen, and thyroid transcription factor 1. Applying the panel and tree to the original data provided correct classification in 88%. The 10 markers and diagnostic algorithm were then tested in a second, independent, set of primary and metastatic tumors and again 88% were correctly classified. Conclusions: This classification scheme should enable better prediction on biopsy material of the primary site in patients with metastatic adenocarcinoma of unknown origin, leading to improved management and therapy.

Probabilistic relabelling strategies for the label switching problem in Bayesian mixture models

The label switching problem is caused by the likelihood of a Bayesian mixture model being invariant to permutations of the labels. The permutation can change multiple times between Markov Chain Monte Carlo (MCMC) iterations making it difficult to infer component-specific parameters of the model. Various so-called ‘relabelling’ strategies exist with the goal to ‘undo’ the label switches that have occurred to enable estimation of functions that depend on component-specific parameters. Existing deterministic relabelling algorithms rely upon specifying a loss function, and relabelling by minimising its posterior expected loss. In this paper we develop probabilistic approaches to relabelling that allow for estimation and incorporation of the uncertainty in the relabelling process. Variants of the probabilistic relabelling algorithm are introduced and compared to existing deterministic relabelling algorithms. We demonstrate that the idea of probabilistic relabelling can be expressed in a rigorous framework based on the EM algorithm.

Bayesian inference in hidden Markov random fields for binary data defined on large lattices.

Hidden Markov random fields represent a complex hierarchical model, where the hidden latent process is an undirected graphical structure. Performing inference for such models is difficult primarily because the likelihood of the hidden states is often unavailable. The main contribution of this article is to present approximate methods to calculate the likelihood for large lattices based on exact methods for smaller lattices. We introduce approximate likelihood methods by relaxing some of the dependencies in the latent model, and also by extending tractable approximations to the likelihood, the so-called pseudolikelihood approximations, for a large lattice partitioned into smaller sublattices. Results are presented based on simulated data as well as inference for the temporal-spatial structure of the interaction between up- and down-regulated states within the mitochondrial chromosome of the Plasmodium falciparum organism. Supplemental material for this article is available online.

Protein biogenesis machinery is a driver of replicative aging in yeast

An integrated account of the molecular changes occurring during the process of cellular aging is crucial towards understanding the underlying mechanisms. Here, using novel culturing and computational methods as well as latest analytical techniques, we mapped the proteome and transcriptome during the replicative lifespan of budding yeast. With age, we found primarily proteins involved in protein biogenesis to increase relative to their transcript levels. Exploiting the dynamic nature of our data, we reconstructed high-level directional networks, where we found the same protein biogenesis-related genes to have the strongest ability to predict the behavior of other genes in the system. We identified metabolic shifts and the loss of stoichiometry in protein complexes as being consequences of aging. We propose a model whereby the uncoupling of protein levels of biogenesis-related genes from their transcript levels is causal for the changes occurring in aging yeast. Our model explains why targeting protein synthesis, or repairing the downstream consequences, can serve as interventions in aging. DOI: http://dx.doi.org/10.7554/eLife.08527.001

Sparse time series chain graphical models for reconstructing genetic networks

We propose a sparse high-dimensional time series chain graphical model for reconstructing genetic networks from gene expression data parametrized by a precision matrix and autoregressive coefficient matrix. We consider the time steps as blocks or chains. The proposed approach explores patterns of contemporaneous and dynamic interactions by efficiently combining Gaussian graphical models and Bayesian dynamic networks. We use penalized likelihood inference with a smoothly clipped absolute deviation penalty to explore the relationships among the observed time course gene expressions. The method is illustrated on simulated data and on real data examples from Arabidopsis thaliana and mammary gland time course microarray gene expressions.

Bayesian Structure Learning in Sparse Gaussian Graphical Models

Decoding complex relationships among large numbers of variables with relatively few observations is one of the crucial issues in science. One approach to this problem is Gaussian graphical modeling, which describes conditional independence of variables through the presence or absence of edges in the underly- ing graph. In this paper, we introduce a novel and efficient Bayesian framework for Gaussian graphical model determination which is a trans-dimensional Markov Chain Monte Carlo (MCMC) approach based on a continuous-time birth-death process. We cover the theory and computational details of the method. It is easy to implement and computationally feasible for high-dimensional graphs. We show our method outperforms alternative Bayesian approaches in terms of convergence, mixing in the graph space and computing time. Unlike frequentist approaches, it gives a principled and, in practice, sensible approach for structure learning. We illustrate the efficiency of the method on a broad range of simulated data. We then apply the method on large-scale real applications from human and mammary gland gene expression studies to show its empirical usefulness. In addition, we implemented the method in the R package BDgraph which is freely available at http://CRAN.R-project.org/package=BDgraph

An information-rich alternative, chemicals testing strategy using a high definition toxicogenomics and zebrafish (Danio rerio) embryos

Large-scale toxicogenomic screening approaches offer great promise for generating a bias-free system-wide view of toxicological effects and modes-of-action of chemicals and ecotoxicants. However, early applications of microarray technology have identified relatively small groups of responding genes with which to define new targets for analysis by conventional means. We have trialled a more intensive approach to the design and interpretation of array experiments incorporating a balanced interwoven ANOVA design with higher levels of biological replication, a more thorough analysis of errors and false discovery rates, and an analysis of response patterns using gene network models. Zebrafish embryos were exposed from 1.5 h post-fertilization for 72 h to ecotoxicants representing different classes--2,4-dichlorophenol, 3,4-dichloroaniline, pentachlorophenol, and cadmium chloride--at low concentrations producing a developmental disturbance to 10% of embryos and half of this dose. Extracted whole embryo RNA was then analyzed on microarrays. Analysis revealed responses of 3000-5000 genes, which is 10-1000 times greater than previously reported, with significance at lower levels of fold change. Some gene responses were common to multiple toxicants, and others were restricted to just one or two toxicants. The gene expression profiles for the different toxicants were distinctive, and analysis using network-based models provided a high level of detail of affected processes, some of which were novel. This approach provides a more highly refined view of toxic effects, from which meaningful patterns of response can be discerned and related to functional deficits and from which more reliable indicators of toxicological effect can be predicted.

Reconstructing repressor protein levels from expression of gene targets in Escherichia coli

The basic underlying problem in reverse engineering of gene regulatory networks from gene expression data is that the expression of a gene encoding the regulator provides only limited information about its protein activity. The proteins, which result from translation, are subject to stringent posttranscriptional control and modification. Often, it is only the modified version of the protein that is capable of activating or repressing its regulatory targets. At present there exists no reliable high-throughput technology to measure the protein activity levels in real-time, and therefore they are, so-to-say, lost in translation. However, these activity levels can be recovered by studying the gene expression of their targets. Here, we describe a computational approach to predict temporal regulator activity levels from the gene expression of its transcriptional targets in a network motif with one regulator and many targets. We consider an example of an SOS repair system, and computationally infer the regulator activity of its master repressor, LexA. The reconstructed activity profile of LexA exhibits a behavior that is similar to the experimentally measured profile of this repressor: after UV irradiation, the amount of LexA substantially decreases within a few minutes, followed by a recovery to its normal level. Our approach can easily be applied to known single-input motifs in other organisms.

Inferring slowly-changing dynamic gene-regulatory networks

Dynamic gene-regulatory networks are complex since the interaction patterns between their components mean that it is impossible to study parts of the network in separation. This holistic character of gene-regulatory networks poses a real challenge to any type of modelling. Graphical models are a class of models that connect the network with a conditional independence relationships between random variables. By interpreting these random variables as gene activities and the conditional independence relationships as functional non-relatedness, graphical models have been used to describe gene-regulatory networks. Whereas the literature has been focused on static networks, most time-course experiments are designed in order to tease out temporal changes in the underlying network. It is typically reasonable to assume that changes in genomic networks are few, because biological systems tend to be stable.We introduce a new model for estimating slow changes in dynamic gene-regulatory networks, which is suitable for high-dimensional data, e.g. time-course microarray data. Our aim is to estimate a dynamically changing genomic network based on temporal activity measurements of the genes in the network. Our method is based on the penalized likelihood with ℓ1-norm, that penalizes conditional dependencies between genes as well as differences between conditional independence elements across time points. We also present a heuristic search strategy to find optimal tuning parameters. We re-write the penalized maximum likelihood problem into a standard convex optimization problem subject to linear equality constraints. We show that our method performs well in simulation studies. Finally, we apply the proposed model to a time-course T-cell dataset.

Calorie restriction does not elicit a robust extension of replicative lifespan in Saccharomyces cerevisiae

Significance Calorie restriction (CR) has been shown to extend the lifespans of various organisms. Consequently, a considerable amount of research has been performed to elucidate its mechanisms, especially in the yeast Saccharomyces cerevisiae. Here, we show that due to small sample sizes, large variation exists between measurements. In addition, the effect of CR on lifespan has been routinely overestimated in yeast due to the use of short-lived experimental controls, which together may explain why contradictory mechanisms were found to mediate CR-induced lifespan extension. Moreover, we did not observe any lifespan-enhancing effect of CR using an alternative measurement technique. The inability of CR to robustly extend lifespan suggests that calories alone do not modulate the lifespan of this important model organism. Calorie restriction (CR) is often described as the most robust manner to extend lifespan in a large variety of organisms. Hence, considerable research effort is directed toward understanding the mechanisms underlying CR, especially in the yeast Saccharomyces cerevisiae. However, the effect of CR on lifespan has never been systematically reviewed in this organism. Here, we performed a meta-analysis of replicative lifespan (RLS) data published in more than 40 different papers. Our analysis revealed that there is significant variation in the reported RLS data, which appears to be mainly due to the low number of cells analyzed per experiment. Furthermore, we found that the RLS measured at 2% (wt/vol) glucose in CR experiments is partly biased toward shorter lifespans compared with identical lifespan measurements from other studies. Excluding the 2% (wt/vol) glucose experiments from CR experiments, we determined that the average RLS of the yeast strains BY4741 and BY4742 is 25.9 buds at 2% (wt/vol) glucose and 30.2 buds under CR conditions. RLS measurements with a microfluidic dissection platform produced identical RLS data at 2% (wt/vol) glucose. However, CR conditions did not induce lifespan extension. As we excluded obvious methodological differences, such as temperature and medium, as causes, we conclude that subtle method-specific factors are crucial to induce lifespan extension under CR conditions in S. cerevisiae.