Improved annotation of protein-coding genes boundaries in metazoan mitochondrial genomes

Abstract

With the rapid increase of sequenced metazoan mitochondrial genomes, a detailed manual annotation is becoming more and more infeasible. While it is easy to identify the approximate location of protein-coding genes within mitogenomes, the peculiar processing of mitochondrial transcripts, however, makes the determination of precise gene boundaries a surprisingly difficult problem. We have analyzed the properties of annotated start and stop codon positions in detail, and use the inferred patterns to devise a new method for predicting gene boundaries in de novo annotations. Our method benefits from empirically observed prevalances of start/stop codons and gene lengths, and considers the dependence of these features on variations of genetic codes. Albeit not being perfect, our new approach yields a drastic improvement in the accuracy of gene boundaries and upgrades the mitochondrial genome annotation server MITOS to an even more sophisticated tool for fully automatic annotation of metazoan mitochondrial genomes.

Figure 1.

Overview of the annotated start codon (upper panel) and stop codon (lower panel) frequencies (f) ≥ 0.01 per gene and genetic code table inferred from the annotations in RefSeq 63; full data is given in Supplementary Figure S7.

Figure 2.

Usage of internal codons (inframe codons between annotated start and stop positions) in different coding tables. Shading of the boxes represents the frequency (f) of the internal codons according to PCGs annotated in RefSeq 63 (expressed as ⌊log₁₀f⌋). Frequencies of the corresponding tRNA anticodons (f-tRNA) and accepted amino acid are indicated by the circle sizes. Changes of encoded amino acid between code tables are indicated by changes of the line profile or their single-letter code (e.g. AGR codes for stop in code table 2, for arginine in code table 4, and—together with AGY—for serine in code table 5).

Figure 3.

Cumulative plot of the differences (in base pairs) of the start (Δ start) and stop positions (Δ stop) predicted with the method originally implemented in MITOS (left) and the new method presented here (right) with respect to the reference annotations for those entries that are present in RefSeq 89 but not in RefSeq 63. Positive (negative) values correspond to predictions outside (inside) of the annotation. Differences are shown on an inverse hyperbolic sine scale (). For a comparison to RefSeq 63, see Supplementary Figure S8.

Figure 4.

Cumulative plot of the differences (in base pairs) of the predicted start and stop codon positions and the positions annotated in MitoAnnotator for genomes present in RefSeq 89 but not in RefSeq 63.