Genomic positions of co-expressed genes: echoes of chromosome organisation in gene expression data

Abstract

Background: The relationships between gene expression and nuclear structure, chromosome territories in particular, are currently being elucidated experimentally. Each chromosome occupies an individual, spatially-limited space with a preferential position relative to the nuclear centre that may be specific to the cell and tissue type. We sought to discover whether patterns in gene expression databases might exist that would mirror prevailing or recurring nuclear structure patterns, chromosome territory interactions in particular.

Results: We used human gene expression datasets, both from a tissue expression atlas and from a large set including diverse types of perturbations. We identified groups of positional gene clusters over-represented in gene expression clusters. We show that some pairs of chromosomes and pairs of 10 Mbp long chromosome regions are significantly enriched in the expression clusters. The functions of genes involved in inter-chromosome co-expression relationships are non-random and predominantly related to cell-cell communication and reaction to external stimuli.

Conclusions: We suggest that inter-chromosomal gene co-expression can be interpreted in the context of nuclear structure, and that even expression datasets that include very diverse conditions and cell types show consistent relationships.

Figure 1

The numbers of chromosomes within co-expression clusters. For example, in Symatlas data, the most common number of chromosomes represented in an expression cluster is eight, and there are more than 40 such expression clusters. Dots - Symatlas (left) and NeMo (right) data. Bars - randomised data with standard deviation (SD) shown (see Methods).

Figure 2

Distances in sequence between genes from the same co-expression cluster. Dots - Symatlas (top) and NeMo (bottom) data. Bars - randomised data with SD shown. The number of inter-chromosomal gene pairs within one co-expression cluster is 163 (compared to an average 26.81 + - 5.26 in randomised data) and 214 (compared to an average 23.42 + - 5.28 in randomised data) for Symatlas – and NeMo data, respectively.

Figure 3

Standard deviations of gene densities [genes/Mbp] within groups of positional clusters. Groups of positional clusters are defined as positional clusters belonging to the same co-expression cluster. SD was calculated for gene densities from each group of positional clusters separately (dashed lines indicate the mean SD) and for gene densities from all positional clusters together (red lines). Symatlas (top) and NeMo (bottom) data.

Figure 4

Differences in gene densities between genes from the same co-expression cluster and from different chromosomes. Dots - Symatlas (top) and NeMo (bottom) data. Bars - randomised data with SD shown.

Figure 5

Chromosome pairs from which genes are found together in one co-expression cluster. Square colours correspond to the frequency of observation of genes from a pair of chromosomes within the same co-expression cluster. The upper triangle colours indicate the number of standard deviations between co-expression gene pair counts for real and randomised data, the lower triangle – p-value (0.05 threshold for each one-tailed test). Chromosome order is by chromosome size (top) or chromosome gene density (bottom). Symatlas data (left), NeMo data (right).

Figure 6

Ten Mbp-wide regions in the genome from which genes are found together in one co-expression cluster. Square colours correspond to the frequency of observation of genes from a pair of genomic regions within the same co-expression cluster The upper triangle colours indicate the number of standard deviations between co-expression gene pair counts for real and randomised data (cut-off 3, 5 or 10 standard deviations), the lower triangle – p-value (0.05 threshold for each one-tailed test). Symatlas data (top), NeMo data (bottom).