There are some things in biology that only make sense in the light of evolution, and the gradual and de novo gain of novel protein coding genes is one such thing. I mean just take a look at this section in this paper:
De Novo Genes Have Unique Sequence Characteristics as Compared with Conserved Genes
The de novo candidates share a number of structural properties that differentiate them from the genes conserved outside the two genera. They are significantly shorter, have a lower codon adaptation index and a higher aggregation propensity compared to conserved genes (supplementary fig. S5, Supplementary Material online). Their biosynthetic cost is also lower than those of noncoding sequences, in agreement with an intermediate stage from a noncoding to a coding state (supplementary fig. S5, Supplementary Material online). When recent, de novo genes are not enriched in intrinsically disordered regions compared to conserved genes. The low propensity of recent genes to disorder was previously reported in S. cerevisiae (Carvunis et al. 2012). When ancient, but in Lachancea only, de novo genes have a higher proportion of predicted disorder than conserved genes (fig. 3), suggesting contrasted evolutionary pressures (see Discussion section).
Okay so, this makes perfect sense in an evolutionary context. Imagine shooting blindly into protein sequence space to hit a functional gene. Regardless of how likely you think that is a priori, in so far as you manage to hit one - is it more or less likely that you’re going to hit the peak of some hill in the landscape, or just some place further down? Clearly, since the peak only occupies a tiny fraction of all sequences that make up the hill, it is more likely that randomly hitting a hill produces a sub-optimal sequence that doesn’t occupy the local, or global optimum.
De novo genes are further from the hill compared to older more conserved genes. A designer could have just made it optimal to begin with. Score one for evolution.
Aggregation propensity is usually selected against, since all proteins have some mutual affinity, it can interfere with protein function. Over time we expect evolution to reduce unwanted side effects on existing proteins that have been under selection for longer periods of time. What do we find? De novo genes are more aggregation prone. A designer could have just designed them without that. Score two for evolution.
Expression levels. Among all possible levels of expression, the most active is only a tiny fraction. A priori, again, we expect that a blind shot into sequence space does not produce the most active promoter. What do we find? De novo genes have overall lower expression levels. Score three for evolution.
Oh and they’re also generally shorter than older genes.
It gets better:
De Novo Genes Preferentially Emerge Next to Divergent Promoters in GC-Rich Intergenic Regions
We found that de novo genes are significantly enriched in opposing orientation with respect to their direct 5′ neighboring gene (fig. 4 A ). Similar enrichment was already observed for mouse-specific genes (Neme and Tautz 2013). This suggests that de novo genes would benefit from the divergent transcription initiated from bidirectional promoters. In contrast, tandemly duplicated genes are significantly enriched in co-orientation with respect to their 5′ neighbor (69% and 74% in Saccharomyces and Lachancea , respectively) (not shown). Therefore, the bias toward opposing orientations strongly suggests that the de novo gene candidates do not actually correspond to tandemly duplicated genes that would have diverged beyond recognition. In addition, the bias towards divergent orientation is the strongest for the reliable de novo genes which correspond to the most recently emerged genes (see above), suggesting that divergent transcription from bidirectional promoters, which are widespread in eukaryotes (Core et al. 2008; Neil et al. 2009), is critical in the early stages of origination.
Okay so, again. How does a promoter function? To transcribe a region of DNA, supercoiled double-stranded DNA must be unwound and the double-strand “unzipped” so a transcriptional initiator can bind and recruit an RNA polymerase to read the DNA sequence and produce a complementary RNA transcript. That means DNA regions that are already more often unwound and unzipped are more frequently accessible to DNA binding proteins, such as transcriptional initiators. That implies we expect mutations near frequently transcribed regions to be more likely to produce novel promoters as transcription initiators have more opportunity to interact with exposed single-stranded DNA, and we are most likely to find frequently transcribed regions near existing promoters, leaving the DNA strand in the opposite direction where the sequence is more free to mutate as the most likely candidate region for a de novo gene. That’s what we find for novel genes, they are most typically found near existing promoters on the anti-parallel DNA strand going in the opposite direction. Score four for evolution.
Further down in the paper they also show de novo genes are more often found at recombination hotspots. Again, makes perfect sense in the context of evolution, both because of the accessibility again, and as recombination greatly facilitates the blind exploration of sequence space, as recombination can basically function in a way equivalent to multiple simultaneous mutations.
De Novo Genes Are Significantly Enriched at Recombination Hotspots
In multiple eukaryotic taxa, including yeasts and humans, heteroduplexes formed during meiotic recombination are repaired by gene conversion biased toward GC-alleles, thus increasing the GC content of recombination hotspots (RHS) (Lamb 1984; Jeffreys and Neumann 2002; Mancera et al. 2008; Duret and Galtier 2009). Furthermore, it provides a nucleosome-free region (Berchowitz et al. 2009; Pan et al. 2011) that promotes transcriptional activity. It follows then that RHS could be favorable locations for the emergence of de novo genes in yeasts.
We’re supposed to think these novel genes refute common ancestry, and that they couldn’t possibly evolve because there’s a lack of a plausible mechanism. That makes no sense whatsoever when we actually look at the data. Their pattern of distribution is how we can see their incremental accumulation over time, and their characteristics only make sense in light of gradual evolution from non-coding DNA. They begin short, with weak fitness effects, at low expression, near already existing genes and in recombination hotspots. They then grow in length over geological time, get expressed more strongly, climb to higher on the fitness peak and come unders stronger purifying selection, and are passed on to subsequent descendant species.