Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Mammalian Genes Preferentially Co-Retained in Radiation Hybrid Panels Tend to Avoid Coexpression

  • Ben-Yang Liao ,

    Affiliation Division of Biostatistics & Bioinformatics, Institute of Population Health Sciences, National Health Research Institutes, Zhunan, Miaoli County, Taiwan, Republic of China

  • Andrew Ying-Fei Chang

Mammalian Genes Preferentially Co-Retained in Radiation Hybrid Panels Tend to Avoid Coexpression

  • Ben-Yang Liao, 
  • Andrew Ying-Fei Chang


Coexpression has been frequently used to explore modules of functionally related genes in eukaryotic genomes. However, we found that genetically interacting mammalian genes identified through radiation hybrid (RH) genotypes tend not to be coexpressed across tissues. This pattern remained unchanged after controlling for potential confounding factors, including chromosomal linkage, chromosomal distance, and gene duplication. Because >99.9% of the genetically interacting genes were identified according to the higher co-retention frequencies, our observation implies that coexpression is not necessarily an indication of the need for the co-presence of two genes in the genome, which is a prerequisite for cofunctionality of their coding proteins in the cell. Therefore, coexpression information must be applied cautiously to the exploration of the functional relatedness of genes in a genome.


Coexpression refers to the coherent transcription of genes in spatial, temporal, or environmental dimensions [1][3]. Presumably, proteins functioning together need to be co-present in a cell or tissue; production of an individual protein without its partners may lead to cell energy and material waste. Therefore, coexpression information has been frequently used to detect the functional modules of genes in the genome [4][6].

Proteins that are produced together undoubtedly require the co-presence of their coding genes in the genome. However, regulation of protein abundance does not necessarily occur at the transcriptional level [7], [8], and gene expression does not determine the fate of tissue differentiation [9]. In addition, analyses in mammals [10], nematodes [11], and flies [12] indicate that many coexpressed gene clusters are unlikely to have originated to optimize gene regulation. Consequently, it remains elusive whether the requirement for the co-presence of two genes in a genome is reflected by an increased level of coexpression and, therefore, whether coexpression predicts the cofunctionality of genes.

To understand the biological implications of gene coexpression, we examined whether elevated coexpression predicts the need for the co-presence of genes in the genome, which is a prerequisite for the cofunctionality of their protein products. Exploiting genotypes of human, mouse, rat, and dog radiation hybrid (RH) panels, researchers recently calculated the co-retention frequencies of all mammalian gene pairs with an intergenic distance (D, see Methods) of ≥10 megabases (Mb) in the human genome, which resulted in the identification of >7×106 “genetic interactions” among >18,000 genes [13]. Because >99.9% of these interactions were identified through higher co-retention frequencies than by chance, such interactions can be considered as an index for the tendency of two genes to be co-present in the genome. In addition, because the topology of the resulted interaction network suggests the comprehensiveness of the interactions identified, the catalog of interactions is ideal for us to perform systematic analyses without inspection biases [14][16]. To our surprise, genes that were preferentially co-retained in the genome consistently showed lower coexpression compared to other gene pairs. This finding suggests that coexpression information must be used cautiously in the exploration of the functional relatedness of genes in a genome.

Results and Discussion

We measured coexpression between two genes from expression profile similarities across 63 human or 58 mouse tissues, using the equation ln[(1+CoExp)/(1−CoExp)] (see Materials and Methods). Larger values of ln[(1+CoExp)/(1−CoExp)] indicate higher coexpression. If coexpression predicts preference for the co-presence for two genes in the mammalian genome, then higher coexpression (and, hence, larger ln[(1+CoExp)/(1−CoExp)]) is expected to be found in pairs of “genetically interacting genes,” as defined in Lin et al. (2010) (hereafter referred to as “co-retained gene pairs” [CRGPs]), than in other “non-co-retained” gene pairs (nCRGPs) (Fig. S1).

Our initial analysis revealed that ln[(1+CoExp)/(1−CoExp)] values between CRGPs were significantly lower than those between nCRGPs (P<10−300, Mann-Whitney U test; Fig. 1A). However, chromosomal linkage promotes coexpression: D between linked genes is negatively correlated with their coexpression, even when D is on the order of tens of Mb in length [10]. Our result showed that the proportion of gene pairs located on the same human chromosome (linked) for CRGPs (61,986/2,615,153 = 2.43%) was only approximately half of that for nCRGPs (2,234,672/49,999,275 = 4.47%) (P<10−300, χ2 test). Compared to linked nCRGPs, the linked CRGPs had significantly larger D values (P<10−300, U test; Fig. 1B) in the human genome.

Figure 1. Coexpression in and characteristics of chromosomal linkage of CRGPs vs. nCRGPs.

Box plots of ln[(1+CoExp)/(1−CoExp)] of CRGPs vs. nCRGPs in (A) all gene pairs, (C) unlinked gene pairs, and (D) linked gene pairs with specified ranges of D. CoExp is measured by Spearman's ρ of expression levels between genes across human tissues (see Fig. S2 for CoExp measured by Pearson's r). (B) Box plots of logD of linked CRGP vs. linked nCRGPs. Upper quartile, median, and lower quartile values are indicated in each box. Bars outside the box indicate semi-quartile ranges. P-values are from a Mann-Whitney U test.

We can potentially explain the tendencies of CRGPs to be unlinked, or to have a larger D when linked, in two ways. First, CRGPs are more subject to transcriptional interference [10]. The human genome has evolutionarily shaped its architecture to avoid the deleterious effects of transcriptional interference [10]. Second, the “genetic interaction” data obtained by Lin et al. (2010) poses intrinsic biases in chromosomal linkage. Regardless of the cause, the bias in chromosomal linkage is an important factor that needs to be controlled in our analyses.

To determine whether a lower frequency of chromosomal linkage or larger D of linked genes sufficiently explains the lower coexpression of CRGPs (Fig. 1A), we classified all gene pairs into linked and unlinked groups (those located on different chromosomes), on the basis of their coordinates on the human genome. We further categorized linked gene pairs into 5 groups with similar D values, to control for D (Fig. 1D). For unlinked genes, the CRGPs still showed significantly lower ln[(1+CoExp)/(1−CoExp)] values than did the nCRGPs (P<10−300, U test; Fig. 1C); for linked gene pairs, CRGPs also consistently showed significantly lower ln[(1+CoExp)/(1−CoExp)] values than nCRGPs in nearly all groups (P≤10−7, U test; Fig. 1D), except for the group of 10–25 Mb (P = 0.37, U test; Fig. 1D). Hence, lower coexpression of CRGPs cannot be explained by chromosomal linkage or D. Duplicate genes share similarity in expression by ancestry and, thus, may confound our result [2], [17]. However, the removal of paralogous gene pairs from the analysis produced a virtually identical result (Fig. S3, S4), suggesting that lower coexpression between CRGPs is unrelated to gene duplication.

To determine whether our observations are specific to human, a parallel analysis was conducted on mouse data (see Materials and Methods). We used the mouse genes that are one-to-one orthologs to human genes mapped in Lin et al. (2010). When linkage was defined by mouse genome coordinates and coexpression was measured based on expression levels across 58 mouse tissues, the result remained consistent with Fig. 1, although the statistical significance of some of the comparisons was reduced (Fig. S5).

In several previous studies [4][6], [18], researchers have claimed that coexpression must be sufficiently high to be considered “biologically relevant” and to be used in exploring the functional relatedness of genes [19]. Hence, we examined coexpression from the aspect of frequencies of genes with high coexpression in the group. We used different thresholds of CoExp to define high coexpression (CoExp≥0.6, 0.65, 0.7, or 0.75) [3]. Linked and unlinked genes were separated to control for bias in the chromosomal linkage. As shown in Fig. 2, regardless of the threshold used, CRGPs consistently showed a lower proportion of highly coexpressed gene pairs than nCRGPs. Parallel analysis with mouse genome coordinates and gene expression data generated a result consistent with Fig. 2 (Fig. S7).

Figure 2. nCRGPs comprise a higher percentage of highly coexpressed genes.

Compared with nCRGPs, CRGPs have a lower percentage of gene pairs with high coexpression, as defined by the threshold of CoExp shown in the bottom, after controlling for chromosomal linkage. CoExp is measured by Spearman's ρ of expression levels between genes (see Fig. S6 when CoExp was measured by Pearson's r). Error bars show one standard error of the proportion. P-values are from a χ2 test.

In summary, our analysis showed no indication that coexpression between genes indicates a need for co-presence in the genome. In contrast, we consistently observed patterns indicating that co-retained genes tend to avoid coexpression in both human and mouse genomes. The result presented in this study implies that factors unrelated to functionality (e.g. transcriptional interference [10]), may cause the coexpression of mammalian genes. Because high coexpression between most of the coexpressed gene pairs is not necessarily evolutionarily conserved, our study implies that it is perhaps evolutionary conservation of coexpression [5], and not coexpression itself, that predicts cofunctionality.

Materials and Methods

CRGPs, which were defined as human genes with “genetic interactions” based on their having RH genotypes with an FDR threshold of ≤0.05, were obtained from the supplementary materials of [13]. The use of a more stringent FDR threshold (≤0.001) to define CRGPs and nCRGPs did not change the results of the analysis (Fig. S8). Chromosomal coordinates, one-to-one orthologs, and annotations of paralogous relationships of human and mouse genes based on Ensembl v62 were retrieved through BioMart ( The intergenic distance D was calculated as the distance in nucleotides between the transcriptional start sites of two genes.

Expression levels in 63 normal human tissues or 58 normal mouse tissues were obtained from Gene Atlas v2 [20] following a previous study [10]. Only 10,313 human genes with genetic interaction data, Ensembl annotations, and microarray data were used (Fig. S1). CoExp was defined by the Spearman's correlation coefficient (ρ) or Pearson's correlation coefficient (r) of expression levels across human tissues. Because CoExp measured by Spearman's ρ yielded statistically more significant results, those results are presented in the main text. Consistent results derived from using Pearson's r to calculate CoExp are shown as Figs. S2 and S6.

In addition to CoExp, the expression profile “dissimilarity” between genes was calculated from the Euclidean distance [21], where n is the number of tissues, and or is the relative transcriptional abundance of gene A or gene B, respectively, in tissue i. The relative transcriptional abundance was calculated from the expression level of a gene in the tissue examined divided by the summation of expression levels of that gene in all of the tissues in the dataset [21]. A lower d indicates a higher level of coexpression. Use of d yielded a result (Fig. S9) that was consistent with the result based on CoExp (Fig. 1), suggesting the robustness of the conclusion reached.

Supporting Information

Figure S1.

Flow chart illustrating the processes used to generate CRGPs and nCRGPs for comparisons in co-expression.


Figure S2.

Regenerated Figure 1 when CoExp is calculated by Pearson's r of expression levels between genes. See legend of Fig. 1 for detailed description.


Figure S3.

Regenerated (A) Fig. 1C and (B) Fig. 1D by excluding gene pairs that are paralogous from the analysis.


Figure S4.

Regenerated (A) Fig. S2C and (B) Fig. S2D by excluding gene pairs that are paralogous from the analysis.


Figure S5.

Regenerated Fig. 1 when CoExp is measured using mouse gene expression data and linkage and D are defined using mouse genome coordinates. See legend of Fig. 1 for detailed description.


Figure S6.

Regenerated Fig. 2 when CoExp is calculated by Pearson's r of expression levels between genes. See legend of Fig. 2 for detailed description.


Figure S7.

Regenerated Fig. 2 when CoExp is measured using mouse microarray data and linkage and D are defined using mouse genome coordinates. See legend of Fig. 2 for detailed description.


Figure S8.

Regenerated Fig. 1 using a more stringent FDR threshold (≤0.001) to define CRGPs and nCRGPs. See legend of Fig. 1 for detailed description.


Figure S9.

Regenerated Figure 1 when expression dissimilarity is calculated by d, the Euclidean distance of the relative transcriptional abundance between genes. A lower d indicates a higher level of coexpression. See legend of Fig. 1 for detailed description.


Author Contributions

Conceived and designed the experiments: BYL. Performed the experiments: BYL AYFC. Analyzed the data: AYFC. Wrote the paper: BYL.


  1. 1. Eisen MB, Spellman PT, Brown PO, Botstein D (1998) Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci U S A 95: 14863–14868.
  2. 2. Hurst LD, Pal C, Lercher MJ (2004) The evolutionary dynamics of eukaryotic gene order. Nat Rev Genet 5: 299–310.
  3. 3. Cohen BA, Mitra RD, Hughes JD, Church GM (2000) A computational analysis of whole-genome expression data reveals chromosomal domains of gene expression. Nat Genet 26: 183–186.
  4. 4. Lee HK, Hsu AK, Sajdak J, Qin J, Pavlidis P (2004) Coexpression analysis of human genes across many microarray data sets. Genome Res 14: 1085–1094.
  5. 5. Stuart JM, Segal E, Koller D, Kim SK (2003) A gene-coexpression network for global discovery of conserved genetic modules. Science 302: 249–255.
  6. 6. Shi Z, Derow CK, Zhang B (2010) Co-expression module analysis reveals biological processes, genomic gain, and regulatory mechanisms associated with breast cancer progression. BMC Syst Biol 4: 74.
  7. 7. Schrimpf SP, Weiss M, Reiter L, Ahrens CH, Jovanovic M, et al. (2009) Comparative functional analysis of the Caenorhabditis elegans and Drosophila melanogaster proteomes. PLoS Biol 7: e48.
  8. 8. Garcia-Sanz JA, Mikulits W, Livingstone A, Lefkovits I, Mullner EW (1998) Translational control: a general mechanism for gene regulation during T cell activation. The FASEB journal : official publication of the Federation of American Societies for Experimental Biology 12: 299–306.
  9. 9. Yanai I, Korbel JO, Boue S, McWeeney SK, Bork P, et al. (2006) Similar gene expression profiles do not imply similar tissue functions. Trends Genet 22: 132–138.
  10. 10. Liao B-Y, Zhang J (2008) Coexpression of linked genes in Mammalian genomes is generally disadvantageous. Mol Biol Evol 25: 1555–1565.
  11. 11. Yanai I, Hunter CP (2009) Comparison of diverse developmental transcriptomes reveals that coexpression of gene neighbors is not evolutionarily conserved. Genome Res 19: 2214–2220.
  12. 12. Weber CC, Hurst LD (2011) Support for multiple classes of local expression clusters in Drosophila melanogaster, but no evidence for gene order conservation. Genome Biol 12: R23.
  13. 13. Lin A, Wang RT, Ahn S, Park CC, Smith DJ (2010) A genome-wide map of human genetic interactions inferred from radiation hybrid genotypes. Genome Res 20: 1122–1132.
  14. 14. Bjorklund AK, Light S, Hedin L, Elofsson A (2008) Quantitative assessment of the structural bias in protein-protein interaction assays. Proteomics 8: 4657–4667.
  15. 15. Navratil V, de Chassey B, Combe CR, Lotteau V (2011) When the human viral infectome and diseasome networks collide: towards a systems biology platform for the aetiology of human diseases. BMC Syst Biol 5: 13.
  16. 16. Rual JF, Venkatesan K, Hao T, Hirozane-Kishikawa T, Dricot A, et al. (2005) Towards a proteome-scale map of the human protein-protein interaction network. Nature 437: 1173–1178.
  17. 17. Lercher MJ, Blumenthal T, Hurst LD (2003) Coexpression of neighboring genes in Caenorhabditis elegans is mostly due to operons and duplicate genes. Genome Res 13: 238–243.
  18. 18. van Noort V, Snel B, Huynen MA (2003) Predicting gene function by conserved co-expression. Trends Genet 19: 238–242.
  19. 19. Batada NN, Urrutia AO, Hurst LD (2007) Chromatin remodelling is a major source of coexpression of linked genes in yeast. Trends Genet 23: 480–484.
  20. 20. Su AI, Wiltshire T, Batalov S, Lapp H, Ching KA, et al. (2004) A gene atlas of the mouse and human protein-encoding transcriptomes. Proc Natl Acad Sci U S A 101: 6062–6067.
  21. 21. Liao B-Y, Zhang J (2006) Evolutionary conservation of expression profiles between human and mouse orthologous genes. Mol Biol Evol 23: 530–540.