Routes
A homology search that returns nothing outside a narrow clade is compatible with at least eight different histories. Five of them are real ways a young gene can arise. Three are ways an old gene — or no gene at all — can produce the same signature. Telling them apart is the whole methodological problem, and nothing about the search output distinguishes them.
These five all produce a protein that did not exist in the ancestor. They differ in where the coding sequence came from and in what order the two required properties — transcription and an open reading frame — were acquired.
The reading frame is already there by chance in ancestral non-coding DNA. The novelty is the acquisition of expression — a new promoter, or an enhancer captured from a neighbour.
Transcription comes first. The locus spends time as a lncRNA while random mutation lengthens an ORF inside the existing transcript. This is the route the proto-gene continuum model puts most weight on.
Expression and ORF length drift up and down together over long periods. "Gene" and "not a gene" are the two ends of a gradient populated by thousands of intermediates, and asking when the gene was born is asking the wrong question.
A new ORF opens in a different reading frame, or on the antisense strand, of DNA that already codes for something else. The DNA is old; the protein is entirely new and shares no ancestry with any existing protein.
An intron, an untranslated region, or an inserted transposable element acquires splice sites and joins the mature mRNA as new coding exon. New protein-coding sequence — but its raw material was not truly random, which matters for how you interpret it.
Scenarios 1 and 2 differ only in ordering. That is exactly what the game makes you feel: reach for O/P before Q/W and you have played ORF-first; the reverse is expression-first; interleave them and you get the continuum. The end-of-run screen tells you which one you played.
These produce the same empty homology search and are, in practice, the more common explanation. Every one of them has been found in real phylostratigraphy output — including in this project's own runs.
An ancient gene evolving fast enough — or simply short enough — that its real orthologs fall below the search's detection threshold. Called homology detection failure, and it is the single largest source of false de novo calls.
A young duplicate freed from constraint can diverge past recognition within one genus. It looks lineage-specific and it is new — but it is built from an old part, which is a different evolutionary claim entirely.
Runs in both directions. A contaminated database entry gives an ancient-looking hit to a young gene; a mis-annotated ORF invents a gene that was never transcribed. This one is not hypothetical here — see the COR15A case.
De novo genes are expected to be short, fast-evolving, weakly expressed and lineage-restricted. So are the sequences that defeat homology search. The two categories are correlated with the same properties, so a method that scores only "did we find a hit" cannot separate them — it will systematically over-call de novo birth for exactly the sequences most likely to be misdated.
What each piece of evidence rules in or out. No single row is decisive; the argument is made by the combination.
| Observation | Supports | Rules against |
|---|---|---|
| Syntenic outgroup region, alignable, no intact ORF | de novo (1–3) | divergence, duplication |
| Syntenic outgroup region with an intact but diverged ORF | divergence beyond recognition | de novo |
| Profile HMM or structural search recovers an old family | divergence, duplication | de novo |
| Novel ORF lies inside an older gene, frame-shifted | overprinting | independent de novo locus |
| Novel exon matches a known TE family | exonization | de novo from random sequence |
| Segregating in the population, not fixed | proto-gene caught in transit | an established ancient gene |
| Ribosome profiling shows engagement | a real, translated product | spurious ORF annotation |
| Single distant hit, all intermediate ranks empty | contamination / HGT artefact | a genuinely ancient gene |
| Protein is very short (< 100 aa) with low complexity | — uninformative on its own; both real de novo genes and detection failures look like this | |