The aphid relatives that helped AlphaFold recover a mysterious protein shape

A small aphid can persuade a witch-hazel leaf to grow a cone around it. Hormaphis cornu induces these structures, called galls, in plant tissue. One clue to the insect’s influence is colour: a gene in the aphid is associated with whether its gall is red or green.[2]

The September PNAS study connects this biological puzzle to protein prediction. Sequencing the right relatives supplied information that AlphaFold2’s standard databases had missed.[1]

The earlier colour study, published in 2021, linked variation in the aphid gene determinant of gall color to reduced expression of that gene in the insect’s salivary glands. In the corresponding plant galls, seven genes involved in making anthocyanin pigments became more active, and red pigments accumulated. The researchers proposed that the aphid’s secreted protein changes the plant’s gene activity.[2]

That gene belonged to a large family named BICYCLE, after two recurring cysteine–tyrosine–cysteine sequences. The family’s members were especially active in the salivary glands of the generation that makes galls. This offered a collection of possible molecular tools for changing plant growth, while leaving their workings largely unexplained.[2]

Family resemblance can survive in unexpected places

Protein sequences change during evolution, sometimes so rapidly that a search for similar strings misses relatives. David Stern and Clair Han addressed this in a 2022 study by examining the organisation of genes. BICYCLE genes often contain many unusually short exons, the pieces retained when a gene’s RNA is assembled. The sizes of those pieces and the positions of the intervening introns can retain clues to ancestry after the encoded protein sequences have diverged.[3]

A logistic-regression classifier using gene lengths, exon lengths and related structural features found candidate family members that sequence searches missed. This included proteins lacking the motifs that had originally given BICYCLE its name. Recognising a relative therefore became a question about the arrangement of a gene as well as its letters.[3]

AlphaFold2 can use that kind of expanded family knowledge through a multiple sequence alignment: related proteins are lined up so corresponding positions can be compared. Evolution supplies many versions of a structure that must continue to work. Patterns of conservation and coordinated changes provide clues about how the chain can fit together.[4]

The alignment is evidence supplied when making a prediction. It need not mean retraining the model on newly solved structures. AlphaFold2 combines sequence information with learned relationships between residues and three-dimensional geometry, repeatedly refining its representations. Having a capable model and having useful relatives to compare are separate contributions to the prediction.[4]

Two measured structures provided the check

From 18 selected proteins, the researchers obtained two crystal structures: G3873 and G2703. Standard-database predictions failed. Three newly sequenced aphid species, combined with existing genomes, supplied 197 related sequences for G3873 and 11 for G2703. With these alignments, AlphaFold2 recovered both experimental folds.[1]

The team then attempted 4,924 proteins across seven species. Of these, 2,400 passed its confidence filter: at least 80% of residues scoring 60 or higher, over at least 70 residues. All contained a saposin-like domain, a bundle of helices, sometimes linked or interwoven with another. These remain predictions; the two crystal structures do not validate every later model.[1]

A confidence score is the model’s estimate of how reliably it has placed parts of a protein. It is useful for choosing which structural regions to interpret, but it does not identify the molecule the protein binds or establish what happens inside a plant cell.[4]

The biological family also extends beyond gall-making insects: the 2022 study found strongly expressed BICYCLE genes in the salivary glands of a non-galling aphid. Their roles may therefore encompass several ways of altering plant physiology.[3]

The shared scaffold brings those possibilities into structural view. The molecular targets in plants remain unknown; finding them is now a more sharply defined question.[1]

Report an error or suggest a correction