What drives codon choices in human genes?

Synonymous codon usage is based and the bias seems to be different in different organisms. Factors with proposed roles in causing codon bias include degree and timing of gene expression, codon-anticodon interactions, transcription and translation rate and fidelity, codon context, and global and local G + C content. We offer a new perspective and new methods for elucidating codon choices applied especially to the human genome. We present data supporting the thesis that codon choices for human genes are largely a consequence of two factors: (1) amino acid constraints, (2) maintaining DNA structures dependent on base-step conformational tendencies consistent with the organism's genome signature that is determined by genome-wide processes of DNA modification, replication and repair. The related codon signature defined as the dinucleotide relative abundances at the distinct codon positions (1,2), (2,3), and (3,4) (4 = 1 of the next codon) accommodates both the global genome signature and amino acid constraints. In human genes, codon positions (2,3) and (3,4) containing the silent site have similar codon signatures reflecting DNA symmetry. Strong CG and TA dinucleotide underrepresentation is observed at all codon positions as well as in non-coding regions. Estimates of synonymous codon usage based on codon signatures are in excellent agreement with the actual codon usage in human and general vertebrate genes. These properties are largely independent of the isochore compartment (G + C content), gene size, and transcriptional and translational constraints. We hypothesize that major influences on codon usage in human genes result from residue preferences and diresidue associations in proteins coupled to biases on the DNA level, related to replication and repair processes and/or DNA structural requirements.