Fig. 2: BPA predicts individual caste phenotypes independent of external morphological traits.

a, BPA predictions of early caste identity in M. pharaonis, showing that the first two PC projections from first instar larvae (n = 54) match reproductives (an unknown mix of gynes and males) and workers among second instar larvae (n = 66). Colours of first instar sample symbols reflect the predicted probability to be a reproductive individual (left panel) and identify individuals with morphologically validated caste and sex in second instar larvae while also visualizing body length (right panel). b, HCR-FISH staining for two of the best predictor genes of caste in first instar M. pharaonis larvae based on presence (+) or absence (−) of gonad tissue, that is LOC105839887 (red; expressed in fat bodies) and Smyd3 (purple; expressed in gonads, white arrow), indicating that transcriptome-wide BPA assignments as gynes were correct because these two genes can only be detected in individuals with a germline (left Gonad+ panel). Germline presence was independently checked by vasa expression (green), showing that these transcripts were always undetectable in individuals without a germline (right Gonad− panel). The housekeeping gene Tubulin (yellow) was stained as a positive control (right images in both panels). c, BPA predictions of early caste identity in A. echinatior larvae, showing that the first two PC projections from second instar larvae (n = 84) matched the third instar individuals (n = 67) with known caste morphology. Individual samples are coloured according to their predicted probability of being gyne or worker in the second instar (left panel) while visualizing individual body lengths via the symbol diameters. Known gynes and workers in the third instar are represented by triangles and squares (right panel) confirming BPA segregation. The inserted epifluorescence microscope hair images refer to the ventral thorax region of a typical predicted second instar gyne (left) or worker (right).