Abundant novel transcriptional units and unconventional gene pairs on human chromosome 22

Abstract

Novel transcriptional units (TUs) are EST-supported transcribed features not corresponding to known genes. Unconventional gene pairs (UGPs) are pairs of genes and/or TUs sharing exon-to-exon cis-antisense overlaps or putative bidirectional promoters. Computational TU and UGP discovery followed by manual curation was performed in the entire published 34.9-Mb human chromosome 22 euchromatic sequence. Novel TUs (n = 517) were as abundant as known genes (n = 492) and typically did not have nonprimate DNA and protein homologies. One hundred seventy-one (33%) of TUs, but only 13 (3%) of genes, both lacked nonprimate conservation and localized to gaps in the human-mouse BLASTZ alignment. Novel TUs were richer in exonic primate-specific interspersed repetitive elements (P = 0.001) and were more likely to rely on splice junctions provided by them, than were known genes: 19% of spliced TUs, versus 5% of spliced genes, had a splice site within a primate-specific repeat. Hence, novel TUs and known genes may represent different portions of the transcriptome. Two hundred nine (21%) of chromosome 22 transcripts participated in 77 cis-antisense and 42 promoter-sharing UGPs. Transcripts involved simultaneously in both UGP types were more common than was expected (P = 0.01). UGPs were nonrandomly distributed along the sequence: 89 (75%) clustered in distinct regions, the sum of which equaled 4.4 Mb (<13% of the chromosome). Eighty (67%) of the UGPs possessed significant locus structure differences between primates and rodents. Since some TUs may be functional noncoding transcripts and since the cis-regulatory potential of UGPs is well recognized, TUs and UGPs specific to the primate lineage may contribute to the genomic basis for primate-specific phenotypes.

Figure 1.

Schematic illustrations of principal UGP genomic structures. Each arrow represents one transcribed feature. Only the orientation of the transcribed features is shown. Individual exons of the features are not shown. (A) Three types of exon-to-exon cis-antisense pairs. (Left) A tail-to-tail cis-antisense pair of transcribed features (genes, TUs, or one gene and one TU) on opposite strands in the same locus, with overlap of the last exons (“tails”) of the two features. Only the last exons are involved in the overlap. (Center) A head-to-head cis-antisense pair of transcribed features on opposite strands in the same locus, with overlap of the first exons (“heads”) of the two features. Only the first exons are involved in the overlap. (Right) cis-antisense pairs that are neither tail-to-tail nor head-to-head are collectively referred to as “other” or “complex” in this article. Only one of the many possible configurations of cis-antisense pairs is shown (B) A pair of transcribed features with a putative bidirectional promoter. (C) A gene (gray) that is antisense to multiple other genes (black). (D) A gene (gray) that shares a bidirectional promoter with another gene (solid black) and harbors an antisense transcript (hatched black).

Figure 2.

A contiguous chain of five genes and one TU linked by five UGPs on chr22 (BC004346-SLC7A4). Arrows indicate the direction of transcription of the labeled genes and TU. Cis-antisense overlaps are represented by hatched black vertical rectangles. Putative bidirectional promoters are represented by solid gray vertical rectangles. For clarity, individual exons are not shown. Drawing is not to scale.

Figure 3.

Clustering of UGPs along 35 Mb of chr22q. UGPs cluster near one another more frequently than expected by chance on the 35 Mb of human chr22q. Cis-antisense gene pairs (A) are defined as forming an island (B) if two pairs lie within 250 kb of one another. Similarly, pairs of genes that may share a putative bidirectional promoter (C) are defined as forming an island (D) if they lie within 250 kb of one another. Sixty-three of 77 (82%) cis-antisense gene pairs and 26 of 42 (62%) of gene pairs sharing a putative bidirectional promoter lie in islands. Ten thousand simulations of distributions of these features into 35 Mb of genomic space did not yield any distribution with clustering as great as that observed (see text for details). Five regions of 22q harbor both cis-antisense islands and putative-bidirectional promoter islands (E).