Based on the number of pollen hosts utilised, bees have been categorised as generalists (polylectic) or specialists (oligolectic). Faced with a changing habitat, polylectic bees can diversify their pollen ‘portfolio’, while oligolectic bees cannot and therefore may go locally extinct. Research into the evolution and maintenance of broad polylecty is scant. Instead, research has mainly focussed on the factors that constrain oligolectic species to a narrow diet. Here, we developed a molecular phylogeny of a native Australian subgenus Lasioglossum (Chilalictus), (Halictidae), to study the evolution of pollen host breadth within the group. We find that broad polylecty has evolved independently at least four times in L. (Chilalictus) and did not result in subsequent speciation. Oligolecty has evolved once and is found in at least three related species. In addition, broadly polylectic species have significantly larger areas of occurrence than oligolectic and polylectic species. Taken together, these results suggest that there is less opportunity for speciation in broadly polylectic than in polylectic and oligolectic species. As broad polylecty is uncommon in bees, we hypothesise the existence of genetic constraints to its evolution. Future studies on the evolution of broad polylecty should examine both the existence of constraints and selective advantages for host broadening in polylectic species.

Data S1. A morphological datamatrix comprising 80 characters was compiled by Ken Walker (Museums Victoria), an expert in Australian Lasioglossum, who kindly provided it to us for constructing a morphological phylogeny of L. (Chilalictus).

Data S2 and S3. Posterior probabilities of the ancestral states were produced by adding diet width characters into a data matrix of molecular sequence characters of L. (Chilalictus) and then conducting phylogenetic analysis while simulteneously reconstructing the ancestral diet width states of 12 well supported nodes of interest (with posterior probabilities ≥ 87). We used the full hierarchical Bayesian approach in MrBayes (v3.2) with a symmetric evolutionary model for the diet width character and without ordering the diet width character states (Data S2). The topology prior was constrained to fix the 12 well-supported nodes in every sampled tree during the analysis, so that their reconstructed ancestral states could be reported. The analysis was conducted 4 times and parameter and log files were analysed for convergence following guidelines from the Mr.Bayes manual. We repeated the analysis with ordered character states of the diet width, where we restricted the rate parameters for the transition between the two extreme diet width states (oligolecty and broad polylecty) to zero (Data S3).