Mining SAGE data allows large-scale, sensitive screening of antisense transcript expression

Abstract

As a growing number of complementary transcripts, susceptible to exert various regulatory functions, are being found in eukaryotes, high throughput analytical methods are needed to investigate their expression in multiple biological samples. Serial Analysis of Gene Expression (SAGE), based on the enumeration of directionally reliable short cDNA sequences (tags), is capable of revealing antisense transcripts. We initially detected them by observing tags that mapped on to the reverse complement of known mRNAs. The presence of such tags in individual SAGE libraries suggested that SAGE datasets contain latent information on antisense transcripts. We raised a collection of virtual tags for mining these data. Tag pairs were assembled by searching for complementarities between 24-nt long sequences centered on the potential SAGE-anchoring sites of well-annotated human expressed sequences. An analysis of their presence in a large collection of published SAGE libraries revealed transcripts expressed at high levels from both strands of two adjacent, oppositely oriented, transcription units. In other cases, the respective transcripts of such cis-oriented genes displayed a mutually exclusive expression pattern or were co-expressed in a small number of libraries. Other tag pairs revealed overlapping transcripts of trans-encoded unique genes. Finally, we isolated a group of tags shared by multiple transcripts. Most of them mapped on to retroelements, essentially represented in humans by Alu sequences inserted in opposite orientations in the 3'UTR of otherwise different mRNAs. Registering these tags in separate files makes possible computational searches focused on unique sense-antisense pairs. The method developed in the present work shows that SAGE datasets constitute a major resource of rapidly investigating with high sensitivity the expression of antisense transcripts, so that a single tag may be detected in one library when screening a large number of biological samples.

Figure 1

Analysis of SAGE tags mapping to the reverse complement of known transcripts. (A) Conceptual scheme and tag distribution in a U937-cell library (15). (B) Scatter plot showing sense versus antisense frequencies; the regression coefficient was calculated for 146 duplexes (gray dots). (C) Strand-specific RT–PCR of selected transcripts using RNAs from untreated (NT) and differentiated (Diff) U937 cells prepared as for SAGE libraries (15). Tag frequencies are indicated below. First-strand cDNA synthesis initiated with antisense (lanes 1 and 5) and sense-specific primers (lanes 2 and 6). Controls without reverse transcriptase for DNA contamination (lanes 3, 4, 7 and 8). (D) Strand-specific RT–PCR of beta-globin (HBB) mRNA antisense (lane a), sense (lane b) and controls without reverse transcriptase (lanes c and d) using reticulocytes RNA. Scatter plot of globin tag levels in the reticulocyte library, compared to levels in 260 SAGE libraries.

Figure 2

Search for complementary mRNAs pairs in the UniGene collection of cluster-representative sequences. (A) Illustration of the ‘Janus’ tags concept. (B) Flow chart and results of the selection. (C) Numbers of well-oriented transcripts (from UniGene) sharing an Alu-matching J-tag in sense (x-axis) and antisense (y-axis) orientation.

Figure 3

Comparative expression levels of SAGE tags mapping on to transcripts sharing complementary J-tags. Each dot corresponds to one SAGE library. (A) Expression levels of selected pairs in 260 SAGE libraries for cis-encoded, widely expressed genes. (B) Tissue-specific cis-encoded genes. (C) Tissue-specific trans-encoded genes including pseudogenes expressed in inverted orientation. (D) Co-expression of sense and antisense TNNI3 transcripts. (E) Alu elements inserted in the 3′UTR of multiple mRNAs.

Figure 4

Comparative expression levels of SAGE tags mapping on to syncoilin and RBBP4 cis-encoded, oppositely oriented transcripts. White dots, normal tissue samples; black dots, cancer tissue samples. Comparative expression levels in 260 SAGE libraries (A) as in Figure 3, and (B) in libraries prepared from breast normal and cancer samples.