This is a curious paper: if its conclusions had been correct, the whole foundation of modern genome biology would have come into question. Fortunately, its conclusions are wrong, and the key principle of comparative genomics is safe for now.

The reason why I believe this paper's conclusions are wrong is that the authors relied on an increasingly popular approach where, instead of studying the problem itself, researchers study what others say about that problem. Such 'meta-analysis' has done much damage to social and medical sciences and now appears to be creeping into bioinformatics.

Nehrt and colleagues are correct in noting that the current genome annotation implicitly relies on what they call the 'ortholog conjecture', the assumption that orthologs (genes with a common origin that were vertically inherited from the same gene in the last common ancestor of the host organisms) typically retain the same function or have closely related ones. When checked on model organisms, such as Escherichia coli, Bacillus subtilis, yeast, worm, and mouse, this assumption mostly held true, although there were certain cases of function loss, non-orthologous gene displacement, promiscuous enzyme activities, moonlighting, and so on (please see refs {1-3} and ref {4}, a book, where I am a co-author). Accordingly, this assumption underlies gene annotations in non-model organisms, whose functions we’ll never be able to check experimentally and, therefore, have to deduce from sequence comparisons. A thorough, genome-wide check of the 'ortholog conjecture' is a potentially useful task; it would be interesting to learn what fraction of genome annotations actually proves to be incorrect. Unfortunately, Nehrt and colleagues used the meta-analysis approach to this problem: instead of painstakingly comparing functions of orthologs in two reasonably well characterized genomes, they merely compared the Gene Ontology (GO) annotations {5} for orthologous and paralogous genes in human and mouse. GO is a valuable resource that uses structured, controlled vocabularies for functional annotation of genes and gene products {6}. However, GO annotations for human and mouse are not independent, as experimentally characterized functions are being propagated to homologous (orthologous and paralogous) genes in other organisms. The results presented by Nehrt and colleagues are stunning indeed: orthologous genes had fewer overlapping GO terms (e.g. in the Biological Process and Molecular Function categories) than paralogous genes. However, instead of concluding that there might be systematic bias in the GO assignments (which would be interesting but not terribly exciting), they questioned the whole 'ortholog conjecture' (which resulted in the above paper). I am not sure this was an appropriate use of the GO system. Anyway, the key question, as I see it, is whether these results reflect some kind of biological reality. I don’t think so: I was unable to find a single example where orthologous genes in human and mouse have been both experimentally characterized and shown to have distinct functions. I could see that the examples that Nehrt et al. cite in their paper all suffer from the same drawback: when the GO assignments for orthologous genes are indeed different, the actual function of both genes either has never been tested or, if tested, proved to be identical. This is not to say that no orthologous genes in human and mouse would have different functions, the authors have just not used the right approach to identify such genes.

In conclusion, I see no reasons to believe that the 'ortholog conjecture' is seriously threatened by the article of Nehrt and colleagues; we can continue annotating new genomes the same way we did before. However, this key proposition of genome biology certainly deserves further scrutiny, and this could be a positive effect of the publication by Nehrt and colleagues, even if its conclusions are not convincing.

References:
{1} Sonnhammer and Koonin, Trends Genet 2002, 18:619-20 [PMID:12446146].
{2} Koonin EV, Annu Rev Genet 2005, 39:309-38 [PMID:16285863].
{3} Jeffery CJ, Mol Biosyst 2009, 5:345-50 [PMID:19396370].
{4} Koonin and Galperin, "Sequence-Evolution-Function: Computational Approaches in Comparative Genomics" Boston: Kluwer Academic, 2003 ISBN:978-1402072741 [PMID:21089240].
{5} Hill et al. BMC Bioinformatics 2008, 9 Suppl 5:S2 [PMID:18460184].
{6} Gene Ontology Consortium, Nucleic Acids Res 2010, 38:D331-5 [PMID:19920128].