蛋白质间相互作用研究

volume 33 no. 1 pp 15 - 16

Does mapping reveal correlation between gene expression and protein–protein interaction?

Ralf Mrowka1, Wolfram Liebermeister2 & Dirk Holste3

1. Johannes Müller Institute for Physiology, Humboldt University, Tucholskystr. 2, D-10117 Berlin, Germany. e-mail: ralf.mrowka@charite.de
2. Max Planck Institute for Molecular Genetics, Berlin, Germany.
3. Department of Biology, Massachusetts Institute for Technology, Cambridge, Massachusetts, USA.

Genome-wide approaches make systematic inferences about function, regulation and interaction of genes and their corresponding protein products. The challenge is to integrate different sources of information1-3, such as mRNA abundance4 and protein–protein interaction data5, to derive new, biologically relevant and testable hypotheses.

Ge et al.6 carried out a large-scale mapping analysis of gene expression and protein–protein interaction data in the yeast Saccharomyces cerevisiae. The authors contrasted patterns of pair-wise combinations of genes within the same expression cluster (intracluster) and between different expression clusters (intercluster), and focused on an important biological problem: the relationship between coordinately expressed genes and interaction of their protein products.

Ge et al.6 reported a significantly higher fraction of protein interaction densities (PIDs), that is, the number of observed protein interaction pairs over the total number of possible pair-wise combinations, in intracluster protein pairs as compared with intercluster pairs. They interpreted their findings as evidence that genes with similar expression profiles are more likely to encode interacting proteins. Analyzing two different protein–protein interaction databases, one derived from literature searches7 (Munich Information Center for Protein Sequences database) and the other from genome-wide yeast two-hybrid (Y2H) experiments8, 9, the authors found that both data collections gave similar results. This contrasts with other observations of substantial differences between the literature survey data and Y2H assays10, 11. Furthermore, the extent of the correlation between the transcription and protein interaction reported in Ge et al.6 is markedly higher than that in a similar, previously reported analysis11.

Here, we wish to point out that these discrepancies can be resolved. Though it does not concern the potential usefulness of the algorithm applied by Ge et al.6, we find that their analysis favors an alternative explanation.

Ge et al.6 attributed their results of generally higher PID values in intracluster pairs versus intercluster pairs to the global pattern of correlation between expression-profiling and protein-interaction data in yeast. Using the protein–protein interaction data from Y2H assays8, 9 and mapping the data corresponding to the clusters introduced in Tavazoie et al.12, we were able to reproduce the findings. But we found that approximately 67% of the intracluster pairs constituted protein self-interactions. Although self-interacting proteins are valid in principle, they should have been excluded from the study under discussion6 because self-interacting pairs have identical expression patterns by definition. As the authors did not exclude protein self-interactions, we studied the extent to which self-interactions might explain the unusually high intracluster PID values. We assessed the change in global patterns of correlation by computing R, the ratio of average intracluster PIDs over average intercluster PIDs (see Figure). When self-interactions were excluded, the number of intracluster protein pairs did not differ significantly from the random expectation (P = 0.093 at 5% significance level, binomial distribution), and R 1.1 was close to R 1 expected for random pairs. It is, therefore, implausible that interactions between distinct proteins would give rise to R > 5 as observed by Ge et al.6

Transcriptome–interactome correlation map. a, Protein interaction density (PID) matrix for gene-expression and yeast two-hybrid protein–protein interactions. The rows and columns correspond to clusters of genes with similar mRNA abundance levels during the cell cycle, and the color of each matrix element encodes the PID value (scaled by a factor of 105). Following Ge et al.6, PIDs were computed as the scaled ratio between the observed number of interaction pairs and the number of all pair-wise combinations of proteins, but with the exclusion of protein self-interactions. b, PID values for intercluster and intracluster protein pairs (scaled by a factor of 105). Protein self-interactions were excluded, PIDs for intra- and intercluster elements of the PID matrix (a) were averaged separately, and the arithmetic mean is shown. The average PID inter- to intracluster ratio of about 1.1 was close to the value expected for randomly interacting pairs, and the number of intracluster pairs did not differ significantly from the random expectation (P = 0.093, binomial distribution). a,b correspond to Fig. 2c,d in Ge et al.6.

We finally wish to point out that the relationship between coordinately expressed yeast genes and Y2H protein interactions can be identified in an alternative analysis. A histogram of correlation coefficients (r) between mRNA abundance levels for protein pairs can be used to test for positively or negatively regulated pairs compared with random controls. For instance, using gene-expression data of the yeast's cell cycle13 and Y2H8, 9 data, we found a significant shift toward positive r values for interacting non-self protein pairs (P < 10-7, Kolmogorov–Smirnov test) when compared with random controls.

In conclusion, we found that the mapping approach may fail to identify a significant correlation between coordinated gene expression and protein interaction for non-self interactions, whereas a correlation effect was observed using alternative methods11. The high proportion of self-interactions may be of biological interest in its own right, for example, in the formation of regulatory homodimers14.

• BY-D-I经皮黄疸仪	• 卫生部：先看病后付费不宜推行
• 控费失当危及医保制度根在公立医院改革	• 公立医院改革将如何影响医疗器械产业？
• 医改持续深化影响渐显药品流通业利润增幅下滑	• 三明医改三回归：砍掉虚高抹掉灰色
• 黄洁夫：改善医疗生态环境是医改成功的关键	• 陈仲强委员：发展社会办医先要卸包袱
• 从常用药品短缺现象谈医药价格改革	• 回眸2012年新医改—迈向病有所医新时代

互邦电动轮椅HBLD1-A(	海波 811B-D双人电动
海波 811B 经济型电动	互邦电动轮椅HBLD2-A

VIP

推广服务

蛋白质间相互作用研究

平台客服