A conserved achaete-scute cluster in three tardigrade genomes, invisible to all three annotations

The achaete-scute complex is one of the best-studied gene clusters in animal development and nobody has looked for it in a tardigrade: PubMed returns no records for achaete-scute combined with tardigrad*, Hypsibius or panarthropod*. We searched the sequences rather than the annotations. An HMM search with the curated HLH model, at the family's own gathering threshold, finds 51 bHLH proteins in Hypsibius exemplaris, 44 in Ramazzottius varieornatus and 65 in Paramacrobiotus metropolitanus; a maximum-likelihood tree of all 184 domains with 23 references recovers an achaete-scute clade at UFBoot 100 containing all four Drosophila AS-C proteins, all five human ASCL proteins, and exactly two genes from each tardigrade genome — one grouping with ASCL1, one with ASCL3. In every genome the pair lies in a single short block: 17.7 kb in Hypsibius, 8.4 kb in Ramazzottius, 10.2 kb in Paramacrobiotus. In two of the three a third bHLH sits between them, and those two are orthologues of each other (UFBoot 98). None of this is in the annotations: nHd_3.1 calls every gene in the cluster "hypothetical protein", Rvar_4.0 does the same for 22,788 of its features, and RefSeq names one of the Paramacrobiotus pair and leaves the other "uncharacterized". Search the annotations and you find one gene in one genome; search the sequences and you find six genes in three genomes in conserved position. We then read the cluster's expression off a public developmental series and find the two paralogues running in opposite directions: the ASCL3-type rises across the five egg days (rho +0.70) while the ASCL1-type falls from an early, twenty-fold higher peak (rho −0.60) and is almost absent after hatching. We report, in its place, the wrong answer we got first: asking the annotations gave "not conserved", which was a measurement of annotation quality.