The group 5 grass pollen allergens are a Pooideae-restricted lineage of a pan-Poaceae anther protein family

This preprint has not been peer reviewed. It is a computational study on public sequence data; its central prediction is untested. Grass pollen allergen groups are unevenly distributed across Poaceae. Group 1 beta-expansins are described across the family, whereas group 5 is reported only from Pooideae grasses. Because allergens are recorded only where they have been assayed, and assay effort has concentrated on temperate Pooideae species, it has been unclear whether this distribution is biological or an artefact of discovery. Combining the WHO/IUIS allergen registry with UniProtKB, Pfam, InterPro, NCBI Taxonomy, IEDB and the AlphaFold Protein Structure Database, this work shows the restriction is biological. The group 5/6 family (Pfam PF01620) occurs in 13 Poaceae genera, all Pooideae, and has no member outside Poaceae in UniProtKB. Fifteen non-Pooideae genera carry annotated expansins — the positive control — and no PF01620 protein, and no protein among 143,966 non-Pooideae sequences reaches the PF01620 gathering cutoff when curated Pfam profile HMMs are searched directly against four whole reference proteomes, one per subfamily. The family is not, however, without non-Pooideae relatives. PF01620 belongs to the homologous superfamily IPR035506, whose non-Pooideae members form a distinct, recently catalogued Pfam family, PF28983, previously unconnected to allergy. The work identifies 77 such proteins in 10 non-Pooideae genera spanning Panicoideae, Chloridoideae and Oryzoideae; 54 carry a predicted signal peptide and only 2 are curated in Swiss-Prot. Reciprocal cross-family HMM search confirms the homology independently of the InterPro assignment, and the seven sister-family proteins clearing PF01620's own cutoff are all Pooideae, leaving the restriction intact. PF01620 occurs in Brachypodieae as well as Triticeae and Poeae, placing the family's origin at or before the base of core Pooideae and addressing a question previously described as unresolved. Group 5 is therefore best understood not as a Pooideae invention but as a Pooideae-restricted radiation of an ancestral pan-Poaceae anther protein family whose other lineage persists across the grasses. The 77 sister-family proteins are candidate unrecognised pollen allergens, named at accession level and ranked into four evidence tiers. Sister-family proteins share only 22.7% residue identity with a group 5 reference against 61.8% within the allergen family, which argues they are candidate allergens in their own right rather than a reservoir of group 5 cross-reactivity. A separate methodological result: 21 of 45 officially recognised grass pollen allergens (47%) cannot be mapped to sequences reliably by name, through three specific and correctable failure modes — legacy synonyms that cross protein families, three coexisting numbering conventions (Arabic, Roman, and superseded group numbers), and numeric prefix collisions. The recommendation is to map allergens by domain architecture rather than by name. Limitations are stated in full in the manuscript: there is no experimental validation; 68 of 77 candidates have no evidence of existence beyond prediction; pollen presence is established for none; no evidence-backed candidate exists in Chloridoideae, the subfamily of greatest clinical interest; the epitope analysis yields a null result; and the fold comparison is inconclusive because most group 5/6 AlphaFold models are low confidence. The deposit is self-contained: pipeline code, all derived tables, and the complete cached API responses including the four searched proteomes and the AlphaFold models, so every number can be re-derived offline. Each pipeline stage carries an executable self-check. This is a comparative genomics and curation study. It is not clinical research, does not diagnose, and makes no treatment claim.