Phylogeny and floral hosts of a predominantly pollen generalist group of mason bees (Megachilidae: Osmiini)

Abstract

Within the genus Osmia, the three subgenera Osmia, Monosmia, and Orientosmia form a closely-related group of predominantly pollen generalist (‘polylectic’) mason bees. Despite the great scientific and economic interest in several species of this clade, which are promoted commercially for orchard pollination, their phylogenetic relationships remain poorly understood. We inferred the phylogeny of 21 Osmia species belonging to this clade by applying Bayesian and maximum likelihood methods based on five genes and morphology. Because our results revealed paraphyly of the largest subgenus (Osmia s.s.), we synonymized Monosmia and Orientosmia under Osmia s.s. Microscopical analysis of female pollen loads revealed that five species are specialized (‘oligolectic’) on Fabaceae or Boraginaceae, whereas the remaining species are polylectic, harvesting pollen from up to 19 plant families. Polylecty appears to be the ancestral state, with oligolectic lineages having evolved twice independently. Among the polylectic species, several intriguing patterns of host plant use were found, suggesting that host plant choice of these bees is constrained to different degrees and governed by flower morphology, pollen chemistry or nectar availability, thus supporting previous findings on predominantly oligolectic clades of bees.

Apiformes, bee, flower relationship, evolution, Osmia bicornis, Osmia cornifrons, Osmia cornuta, Osmia lignaria, pollination, polylecty

Introduction

Among the 19 900 bee species described so far (Ascher & Pickering, 2013), only a tiny fraction of these is used commercially for crop pollination (Torchio, 1987; Bosch & Kemp, 2002), including several closely-related pollen generalist species of the genus OsmiaPanzer, 1806 (Megachilidae, Osmiini), which are promoted mainly as orchard pollinators around the world. These commercially managed species belong to the subgenus Osmia, which consists of 25 species (Ascher & Pickering, 2013) that form a monophyletic group with the two species-poor subgenera MonosmiaTkalců, 1974 and OrientosmiaPeters, 1978 (Peters, 1978; Praz et al., 2008a). The phylogeny of this group of Osmia bees (‘Osmia s.s. group’) remains largely unresolved, despite the great scientific and economic interest in some of its representatives. Two recent studies including several species of the Osmia s.s. group did not satisfactorily uncover their phylogenetic relationships (Bosch, Maeta & Rust, 2001; Kwon, Lee & Suh, 2003).

Among the commercial Osmia pollinators, Osmia cornifrons has been used to pollinate apple trees in Japan since the 1940s (Kitamura & Maeta, 1969). Subsequently, Osmia cornuta and Osmia bicornis have been established for orchard pollination in Europe (Bosch, 1994; Biliński & Teper, 2004; Krunic & Stanisavljevic, 2006), and Osmia lignaria and O. cornifrons are now used as pollinators of rosaceous fruit trees in the USA (Torchio, 1976, 1981; Abel & Wilson, 1998; Bosch & Kemp, 2001; Bosch, Kemp & Trostle, 2006). Furthermore, Osmia ribifloris has been tested as a potential pollinator of blueberries in the USA (Torchio, 1990; Sampson & Cane, 2000), and Osmia pedicornis, Osmia excavata, and Osmia taurus have been considered for orchard pollination in Asia (Maeta, 1978; Wei et al., 2002).

The Osmia species promoted for commercial pollination are broadly polylectic (i.e. they harvest pollen on the flowers of two or more different plant families). Apart from these and the few additional Central European and North American species, for which the pollen hosts are reasonably well known (Free & Williams, 1970; Tasei, 1973; Raw, 1974; Torchio, 1976, 1981, 1990; Rust, 1986, 1990; Westrich, 1989; Márquez, Bosch & Vicens, 1994; Vicens, Bosch & Blas, 1994; Teppner, 1996; Bosch & Kemp, 2001; Kraemer & Favi, 2005; Sheffield et al., 2008; Teper & Biliński, 2009), the pollen host spectra of the remaining species of the Osmia s.s. group have never been analyzed in detail. Indeed, pollen host preferences, as well as the physiological ability to develop on certain pollen diets, were found to differ markedly between two polylectic Osmia species within the subgenus Osmia (Tasei, 1973; Sedivy, Müller & Dorn, 2011), suggesting that even pronounced pollen generalist bees exhibit patterns of pollen host use, which may differ from those of related generalist species.

Present knowledge of the host plant choice of bees of the Osmia s.s. group suggests that most species are polylectic, whereas a small proportion of species, including Osmia (Osmia) cerinthidis and Osmia (Monosmia) apicata, are considered to be oligolectic (Westrich, 1989; Teppner, 1996) (i.e. they restrict pollen collection to plants of a single plant genus or family). The prevalence of polylecty in combination with the occurrence of oligolectic species renders the Osmia s.s. group suitable to address questions about evolutionary transitions between polylecty and oligolecty.

In the present study, we explored the phylogenetic relationships of the Osmia species of the subgenera Osmia, Monosmia, and Orientosmia based on the analysis of five genes and a morphological data set. In addition, we investigated the pollen host spectra of these species by microscopically analyzing the pollen loads of collected females. We addressed the following questions: (1) What are the phylogenetic relationships between and within the three subgenera of the Osmia s.s. group? (2) What are the pollen hosts of the different species? (3) Are there species-specific patterns of host plant use in the broadly polylectic species? (4) Is oligolecty the ancestral or derived state within this mainly polylectic group of bees?

Material and Methods

Bee species

The subgenus Osmia contains 25 described species, 23 of which occur in the Palearctic and two in the Nearctic (Rust, 1974; Müller, 2013). It is closely related to the exclusively Palearctic subgenera Monosmia and Orientosmia, which contain one and three species, respectively (Peters, 1978; Praz et al., 2008a; Müller, 2012). For the present study, we included 17 species of the subgenus Osmia and all four species of the subgenera Monosmia and Orientosmia (Table 1). For O. bicornis and O. lignaria, two subspecies each were included that show considerable differences in the colour of the metasomal pilosity and in facial morphology, respectively. As an outgroup, we chose five Osmia species that are among the closest relatives of the ingroup based on the phylogenetic hypotheses of Praz et al. (2008a) and Rightmyer, Griswold & Brady (2013). Voucher specimens are deposited in the Entomological Collection of the ETH Zurich.

Table 1.

Pollen host spectrum of 21 Osmia species of the subgenera Osmia, Monosmia and Orientosmia. Definitions of bee host ranges after Müller & Kuhlmann (2008)

Bee species	n	N	Origin of pollen loads	Pollen grain volume (%)	Pure loads of preferred family (%)	Host range
Osmia (Monosmia) apicata Smith, 1853	52	34	AM, GR, IL, IR, JO, TR	BOR (Onosma) 82.9, BOR (Echium) 14.4, BOR (Podonosma) 2.5, BOR (indet) 0.3	100	Oligolectic on Boraginaceae with strong preference for Onosma
Osmia (Osmia) bicornis (Linnaeus, 1758)	50	48	AT, CH, DE, ES, FR, LI, GR, IT, IR, KS, MA, NL	RAN (Ranunculus) 22.1, FAG (Quercus) 16.7, CIS 14.1, PAP (Papaver) 10.6, ACE (Acer) 10.4, ROS 4.9, BRA 4.5, BOR (Echium) 3.6, LAM 2.8, CAP (Lonicera) 2.4, SAL (Salix) 1.9, FAB 1.2, JUG (Juglans) 0.7, PLA (Plantago) 0.7, MAL (Tilia) 0.6, MON 0.3, AST 0.2, CAR 0.1, CAM 0.1, unknown 2.2	10	Polylectic (19 plant families)
Osmia (Osmia) cerinthidis Morawitz, 1876	52	38	AT, CZ, DE, FR, HR, IT, MK, PL, TR, UA	BOR (Cerinthe) 90.5, BOR (Onosma) 2.2, BOR (Echium) 0.1, ROS 4.9, MON 1.4, unknown 0.9	90	Oligolectic on Boraginaceae with strong preference for Cerinthe
Osmia (Osmia) cornifrons (Radoszkowski, 1887)	28	21	JP, KOR, RU, US	FAB 43.6, ROS 26.3, BRA 13.6, LAM 4.2, SAL (Salix) 4.2, ERI 2.5, AQU (Ilex) 1.3, AMA (Allium) 0.7, RAN (Ranunculus) 0.2, unknown 3.4	18	Polylectic (9 plant families)
Osmia (Osmia) cornuta (Latreille, 1805)	50	41	CH, DE, DZ, ES, GR, HR, IR, IT, JO, KZ, LI, TUR	ROS 57.5, SAL (Salix) 13.8, PAP (Corydalis) 7.2, ACE (Acer) 7.0, MON 6.7, BRA 2.4, ERI 1.8, RAN (Anemone) 1.7, unknown 2.0	40	Polylectic (8 plant families)
Osmia (Osmia) emarginata Lepeletier 1841	24	17	ES, FR, MA, TN	BOR (Echium) 38.0, CIS 27.7, FAB 26.7, PAP (Papaver) 7.1, LAM 0.6	25	Polylectic (5 plant families)
Osmia (Osmia) excavata Alfken, 1903	19	15	JP	BRA 67.8, FAB 12.9, CAP (Lonicera) 6.3, MON 4.7, ROS 4.1, RAN (Ranunculus) 0.7, BOR (Echium) 0.1, unknown 3.4	37	Polylectic (7 plant families)
Osmia (Osmia) fedtschenkoi Morawitz, 1875	22	12	AF, PK, TJ, UZ	FAB 77.9, ROS 13.5, BRA 5.1, CAP (Lonicera) 1.5, SAL (Salix) 1.4, unknown 0.6	59	Polylectic (5 plant families) with strong preference for Fabaceae*
Osmia (Osmia) kohli Ducke, 1899	33	29	IT, MT	PAP (Papaver) 43.3, FAB 32.3, BOR (Echium) 8.9, RAN (Ranunculus) 4.4, BRA 4.1, CIS 3.3, FAG (Quercus) 1.1, LAM 0.9, JUG (Juglans) 0.5, API 0.4, ROS 0.1, unknown 0.8	19	Polylectic (11 plant families)
Osmia (Osmia) lignaria Say, 1836	50	50	CA, US	SAL (Salix) 27.2, BOR (Hydrophylloideae) 18.0, BOR (Echium) 0.7, FAB 17.1, ROS 16.9, ACE (Acer) 11.0, OLE 1.2, GRO (Ribes) 0.5, AST 0.3, FAG (Quercus) 0.1, unknown 6.9	16	Polylectic (9 plant families)
Osmia (Osmia) longicornis Morawitz, 1875	3	3	UZ	ROS 98.1, SAL (Salix) 1.9	67	Unknown
Osmia (Orientosmia) maxillaris Morawitz, 1877	23	15	KS, KZ, UZ, TM	FAB 100	100	Oligolectic on Fabaceae
Osmia (Orientosmia) maxschwarzi Müller, 2012	6	5	IR, TR	FAB 99.7, CAM 0.3	83	Oligolectic on Fabaceae
Osmia (Osmia) melanocephala Morawitz, 1875	12	12	KS, KZ, MN	FAB 73.0, ROS 10.4, MON 9.3, CAP (Lonicera) 2.4, BOR 1.5, unknown 3.5	25	Polylectic (5 plant families) with strong preference for Fabaceae
Osmia (Osmia) mustelina Gerstaecker, 1869	50	43	AT, CH, CZ, DE, FR, GR, IT, IL, JO, TR	FAB 58.4, PAP (Papaver) 16.7, CIS 16.4, BOR (Echium) 4.7, BOR (Anchusa) 0.3, PLA (Plantago) 1.5, ROS 1.1, AST 0.5, RAN 0.1, unknown 0.2	36	Polylectic (8 plant families)
Osmia (Osmia) nigrohirta Friese, 1899	27	23	GR, IR, TR	FAB 51.6, BOR (Echium) 16.3, LAM 12.8, CIS 7.5, AST 3.8, BRA 3.1, ROS 1.4, PLA (Antirrhineae) 1.2, API 0.3, PAP (Papaver) 0.2, unknown 1.8	26	Polylectic (10 plant families)
Osmia (Osmia) pedicornis Cockerell, 1919	32	23	CN, JP	FAB 34.1, FAG (Quercus) 23.3, RAN (Ranunculus) 19.1, BRA 12.7, JUG (Juglans) 5.3, ROS 3.4, AST (Cichorioideae) 1.0, unknown 1.1	25	Polylectic (7 plant families)
Osmia (Osmia) ribifloris Cockerell, 1900	50	42	US	ERI (Arbutus, Arctostaphylos, Vaccinium) 50.5, FAB (Cercis, Sophora) 18.1, BER (Berberis, Mahonia) 15.3, GRO (Ribes) 8.0, EBE (Diospyros) 3.3, ROS (Prunus) 2.1, LAM (Salvia) 1.4, ANA (Rhus) 1.4	48	Polylectic (8 plant families)
Osmia (Orientosmia) scheherezade Peters, 1978	18	10	IR, TR	FAB 100	100	Oligolectic on Fabaceae
Osmia (Osmia) taurus Smith, 1873	12	9	JP, RU	JUG (Juglans) 38.1, FAB 21.2, FAG (Quercus) 12.9, ALT (Liquidambar) 9.5, CAP (Lonicera) 5.7, BRA 5.0, ROS 4.8, AST 2.6, RAN (Ranunculus) 0.2	25	Polylectic (9 plant families)
Osmia (Osmia) tricornis Latreille, 1811	50	41	DZ, ES, IT, MA, TN	CIS 67.1, PAP (Papaver) 14.1, FAB 5.6, BOR (Echium) 3.3, BOR (Anchusa) 0.4, BRA 3.0, RAN (Ranunculus) 1.5, ROS 0.8, LAM 0.8, MON 0.5, PLA (Antirrhineae) 0.3, MIM 0.3, AST 0.1, unknown 2.2	24	Polylectic (12 plant families)

n, total number of pollen loads; N, number of pollen loads from different localities. States: AF, Afghanistan; AM, Armenia; AT, Austria; CA, Canada; CH, Switzerland; CN, China; CZ, Czech Republic; DE, Germany; DZ, Algeria; ES, Spain; FR, France; GR, Greece; HR, Hungary; IL, Israel and Palestine; IR, Iran; IT, Italy; JO, Jordan; JP, Japan; KOR, South Korea; KS, Kyrgyzstan; KZ, Kazakhstan; LI, Liechtenstein; MA, Morocco; MK, Macedonia; MN, Mongolia; MT, Malta; NL, Netherlands; PK, Pakistan; PL, Poland; RU, Russia; TJ, Tajikistan; TN, Tunesia; TR, Turkey; UA, Ukraine; US, United States of America; UZ, Uzbekistan.

Plant families: ACE, Aceraceae; AMA, Amaryllidaceae; ANA, Anacardiaceae; ALT, Altingiaceae; API, Apiaceae; AQU, Aquifoliaceae; AST, Asteraceae; BER, Berberidaceae; BOR, Boraginaceae; BRA, Brassicaceae; CAM, Campanulaceae; CAP, Caprifoliaceae; CAR, Caryophyllaceae; CIS, Cistaceae; EBE, Ebenaceae; ERI, Ericaceae; FAB, Fabaceae; FAG, Fagaceae; GRO, Grossulariaceae; JUG, Juglandaceae; LAM, Lamiaceae; MAL, Malvaceae; MIM, Mimosaceae; MON, Monocots; OLE, Oleaceae; PLA, Plantaginaceae; PAP, Papaveraceae; RAN, Ranunculaceae; ROS, Rosaceae; SAL, Salicaceae.

*

The observed pollen spectrum of O. fedtschenkoi might be biased as a result of the large number of specimens collected at the same locality and date (N = 6).

Molecular data

DNA was extracted from specimens preserved in 96% ethanol and from up to 7-year-old pinned specimens using DNeasy Blood and Tissue Kit (Qiagen). One mitochondrial gene and four nuclear genes were amplified by the polymerase chain reaction (PCR): cytochrome oxidase subunit I (COI; 1236 bp), CAD (948 bp), elongation factor1-α (F2-copy) (EF; 1493 bp), long-wavelength rhodopsin (ops; 874 bp), and wingless (wnt1-paralogue) (wnt; 663 bp). Details regarding the primers selected and the PCR conditions applied are provided in the Supporting information (Table S1). PCR products were purified using ExoSAP (Thermo Fisher Scientific) and sequenced on an automated 3130xl DNA analyzer (Applied Biosystems) using BigDye technology (Applied Biosystems). The sequences were trimmed and assembled in SEQUENCHER, version 4.10.1 (Gene Codes Corporation) and visually aligned with MACCLADE, version 4.08 (Maddison & Maddison, 2005). Reading frame and intron/exon boundaries were determined by comparison with published sequences of Osmia cornuta (CAD, EF, ops) and O. lignaria (wnt). To determine the absence of stop codons, the coding sequences were converted to amino acid sequences before analysis. All intron sections that could not unambiguously be aligned were excluded. The sequences of all five genes were concatenated to one single matrix comprising 5214 characters. GenBank accession numbers and specimen data are given in the Supporting information (Table S2).

Morphological data

No fresh material for DNA extraction was available for Osmia (Osmia) kohli and Osmia (Osmia) longicornis. To add these two species to the phylogeny, we studied and recorded morphological characters for all 26 species included in the present study by external examination of both females and males under a dissecting microscope. Additionally, we dissected the male metasoma to analyze the otherwise hidden sterna and genitalia. The search for morphological characters was facilitated by the publications of Peters (1978) and Rust (1974). The morphological analysis resulted in a data matrix containing 41 adult morphological characters (see Supporting information, Tables S3, S4).

Phylogenetic analysis

We performed phylogenetic analyses applying Bayesian and maximum likelihood methods based on the molecular data matrix alone. In addition, Bayesian analyses were performed based on the morphological data matrix alone and on the ‘total evidence’ data matrix that contained both the molecular and morphological characters. The latter analysis was performed for the 24 species with molecular data, as well as for all 26 species included in the present study, treating the molecular partitions of O. kohli and O. longicornis as missing data.

To establish a suitable partitioning regime, a preliminary Bayesian analysis was conducted. Each of the five genes was partitioned into first, second, and third codon position (e.g. CAD1, CAD2, and CAD3). The introns of CAD, EF, ops, and wnt were combined and treated as one additional partition, resulting in a dataset comprising 16 partitions. We ran an analysis in MRBAYES, version 3.2.1 (Huelsenbeck & Ronquist, 2001; Ronquist & Huelsenbeck, 2003) for five million generations using a general time-reversible model. The resulting parameter files were examined in TRACER, version 1.5 (Rambaut & Drummond, 2009b) and an appropriate burn-in was discarded. Based on the substitution rates and nucleotide compositions for the 16 partitions, similar partitions were grouped together, resulting in the following partitioning regime: partition 1 included COI1, COI3, CAD3, EF3, ops3, wnt3, and the introns (2624 bp); partition 2 included COI2, ops1, and ops2 (866 bp); partition 3 included CAD1, EF1, and wnt1 (862 bp); and partition 4 included CAD2, EF2, and wnt2 (862 bp).

Applying MRMODELTEST, version 2.3 (Nylander, 2008), 24 models of nucleotide substitution were tested for each partition and the following models, associated with the lowest Akaike information criterion, were selected: GTR+I+G (partitions 1 and 2), GTR (partition 3), F81 (partition 4). Morphological data were defined as an additional partition, for which the standard model for morphological data implemented in MRBAYES was applied.

Bayesian analyses were performed using MrBayes under the partitioning and model regime specified above. Partitions were unlinked to allow all parameter values and overall rate of substitution to differ. Markov chain Monte Carlo analyses were conducted with one cold and three heated chains. We ran four independent analyses for a total of 120 million generations, sampling trees every 2000 generations. From each run, a burn-in of 10% was discarded as determined in TRACER. The resulting 54 000 trees were sampled and combined to a 50% majority rule consensus tree using PAUP*4.0a125 (Swofford, 2002). Finally, a maximum clade credibility tree that combined the branch lengths of all the post burn-in trees was generated using TREEANNOTATOR, version 1.5.3 (Rambaut & Drummond, 2009a).

Maximum-likelihood analyses were performed using RAXML, version 7.0.4 (Stamatakis, Ludwig & Meier, 2005). The rapid bootstrapping algorithm with a GTR + CAT approximation was applied to perform 1000 bootstrap replicates. Bootstrap replicates were sampled and combined to produce a 50% majority rule consensus tree in PAUP.

Host plants

To assess the pollen host spectra of the 21 Osmia species, we microscopically analyzed the scopal pollen contents of 663 female specimens from museum and private collections. Before removing pollen from the abdominal scopae of the female specimens, the amount of pollen in the scopae was assigned to five classes, ranging from 5/5 (full load) to 1/5 (filled only to one-fifth). The pollen grains were stripped off the scopae with a fine needle and embedded in glycerine gelatine on a slide. When a pollen load was composed of different pollen types, we estimated their percentages by counting the grains along two lines chosen randomly across the cover slip at a magnification of × 400. Pollen types represented by less than 5% of the counted grains were excluded to prevent potential bias caused by contamination. For pollen loads consisting of two or more different pollen types, we corrected the percentages of the number of pollen grains by their volume. After assigning different weights to scopae according to their degree of filling (full loads were weighted five times more strongly than scopae filled to only one-fifth), we summed up the estimated percentages over all pollen samples for each species. We identified the pollen grains at a magnification of ×400 or ×1000 with the aid of Beug (2004) and our own extensive reference collection down to family or, if possible, to subfamily or genus level. Flower records on the collection labels often facilitated pollen identification to a taxonomic level lower than the plant family. A species was classified as oligolectic if 95% or more of the pollen grain volume belonged to one plant family or if 90% or more of the females carried pure loads of the preferred hosts (Müller & Kuhlmann, 2008).

Ancestral state reconstruction

To reconstruct the evolution of host plant choice, we applied parsimony mapping in MACCLADE using the topology of the majority rule consensus tree of the Bayesian analysis of the total evidence matrix. In addition, maximum likelihood inference of ancestral character states was conducted for three strongly supported nodes with BAYESTRAITS (Pagel, Meade & Barker, 2004) after the outgroup taxa and the non-nominotypical subspecies of O. bicornis and O. lignaria had been excluded with MESQUITE, version 2.75 (Maddison & Maddison, 2011). Transition rates between the two states ‘oligolectic’ and ‘polylectic’ were constrained to be equal in BayesTraits. We analyzed a subset of 1000 randomly chosen trees from the Bayesian analyses of the total evidence matrix each for the 19 ingroup species with molecular data and for all 21 ingroup species, including O. kohli and O. longicornis for which only morphological data were available. When the latter two species were included, their branch lengths could not satisfactorily be estimated as a result of the missing molecular data. However, because these species were well nested within the clades for which ancestral state reconstruction was performed, the biased branch lengths are not expected to substantially affect the results. To assess the robustness of the ancestral state reconstructions, we successively constrained the ancestral state of each node to one of the two states by using the ‘fossil’ command in BAYESTRAITS. We compared the mean ln-likelihood associated with each state and considered a difference of two log units as evidence of a ‘significantly’ more likely ancestral state (Pagel, 1999).

Results

Phylogeny

Bayesian analysis of the molecular data matrix yielded a well resolved phylogeny of the 19 Osmia species of the subgenera Osmia, Monosmia, and Orientosmia (Fig. 1). The two subspecies each of O. bicornis and O. lignaria clustered together. Addition of the morphological data set considerably increased the posterior probability values (PP) at three nodes at the base of and within the bicornis clade (Fig. 1). Maximum likelihood analysis yielded a less well resolved phylogeny, although there were no topological incongruences compared to the Bayesian analysis (Fig. 1). Bayesian analysis including those two species for which molecular data were missing (see Supporting information, Fig. S1) placed O. kohli as sister to Osmia tricornis with maximal support (PP = 1) and O. longicornis as sister to O. cerinthidis with only weak support (PP = 0.72). Bayesian analysis of the morphological data set alone (see Supporting information, Fig. S2) yielded a poorly resolved phylogeny. There were no topological incongruences compared to the Bayesian analysis of the molecular data set, except for the position of O. ribifloris; this species was placed within a clade comprising all species of the ingroup except the apicata clade (Fig. 1).

Figure 1.

Phylogeny of Osmia bees of the subgenera Osmia, Monosmia, and Orientosmia. Maximum clade credibility tree based on 54 000 post burn-in trees from four independent Bayesian analyses. Bayesian posterior probabilities are given for the analysis of the molecular data set and for the combined molecular plus morphological data set. Maximum likelihood bootstrap values are given for the analysis of the molecular data set.

The three subgenera form a maximally supported clade (PP = 1; Fig. 1), as do the two subgenera Monosmia and Orientosmia (PP = 1). However, the monophyly of the subgenus Osmia was not supported because the clade consisting of Monosmia + Orientosmia turned out to be sister to only part of the species of the subgenus Osmia (PP = 0.96–0.98), rendering the latter subgenus paraphyletic. In addition, O. (Osmia) ribifloris appears to be sister to all other ingroup species, although with only weak support (PP = 0.73–0.77). Based on our phylogeny, the species other than the basal O. ribifloris can be grouped into three morphologically well circumscribed clades (Fig. 1): (1) the apicata clade comprising all species that lack clypeal horns and possess extraordinarily elongated mouthparts, which are as long as the body length when fully extended and still reach beyond the mesosoma when folded together; (2) the emarginata clade comprising all species that have neither clypeal horns, nor elongated mouthparts; and (3) the bicornis clade comprising all species that have a pair of horns or horn-like structures laterally on the clypeus and possess mouthparts of normal length (Fig. 1).

Host plants

Based on the microscopical analysis of the scopal pollen loads, we classified five of the 21 species as oligolectic (Table 1). Three of these pollen specialist species restrict pollen harvesting to Fabaceae and two species are strictly specialized to Boraginaceae exhibiting a clear preference for Onosma and Cerinthe, respectively. Fifteen species were classified as polylectic collecting pollen from five to 19 plant families. The host plant spectrum of O. longicornis could not be determined conclusively as a result of the low number of pollen samples.

Among the polylectic species, several intriguing patterns of pollen host use emerge (Fig. 2, Table 1): (1) Fabaceae play a predominant role as hosts for the species of the emarginata clade and are also regularly exploited by Osmia ribifloris and several species of the bicornis clade; (2) Asteraceae remain almost unexploited; pollen of this family was found in low proportions of mostly below 20% and invariably mixed with other pollen types in only ten pollen loads of six species; (3) O. ribifloris exhibits a distinct preference for plant taxa that bear their anthers inside bell-shaped flowers such as Ericaceae (Arbutus, Arctostaphylos, Vaccinium), Berberidaceae (Berberis, Mahonia), Ebenaceae (Diospyros), and Grossulariaceae (Ribes); (4) Flowers containing no or very little nectar such as Quercus (Fagaceae), Ranunculus (Ranunculaceae), Juglans (Juglandaceae), Cistaceae, Papaver (Papaveraceae) or Liquidambar (Altingiaceae) play an important role as hosts of the clade composed of O. taurus (60.7% pollen from nectarless or -poor flowers), O. kohli (52.6%), O. tricornis (82.7%), O. pedicornis (47.7%), and O. bicornis (64.9%); (5) Two pairs of sister species show very similar host plant preferences: Fabaceae, Cistaceae, Echium (Boraginaceae), and Papaver are the almost exclusive pollen hosts of Osmia emarginata (99.4% pollen from these four plant taxa) and Osmia mustelina (96.2%); Ranunculus and Quercus are among the most important pollen hosts of O. bicornis (38.8% pollen from these two plant taxa) and O. pedicornis (42.4%); (6) Several species appear to incline towards the use of a main single pollen host such as the species of the emarginata clade (up to 77.9% pollen of Fabaceae) and several species of the bicornis clade: O. cornuta (57.5% pollen of Rosaceae), O. cornifrons (43.6% pollen of Fabaceae), O. excavata (67.8% pollen of Brassicaceae), O. kohli (43.3% pollen of Papaver), and O. tricornis (67.1% pollen of Cistaceae).

Figure 2.

Pollen hosts and evolution of host breadth in Osmia bees of the subgenera Osmia, Monosmia, and Orientosmia. Parsimony mapping of host breadth is based on the topology of the 50% majority rule consensus tree of 54 000 trees from four independent Bayesian analyses. Outgroup species as well as subspecies were omitted. Oligolectic branches are coloured grey, polylectic branches are black, and equivocal branches are shown dashed. The pie charts at three well resolved nodes (A–C) give the maximum likelihood probabilities of the two different states, with asterisks indicating that the analyses constraining the more likely state had significantly higher log-likelihood values than analyses with the alternative state constrained. The coloured pie charts beside the species names give the proportions of the different pollen hosts collected by each species except O. longicornis (for which only three pollen samples were available) based on microscopical analysis of pollen loads from collected females.

Ancestral state reconstruction

Based on the total evidence phylogeny, parsimony mapping of host plant breadth suggests polylecty to be the ancestral state and oligolectic lineages to have evolved twice from polylectic ancestors (Fig. 2). Maximum likelihood inference of host plant breadth at the three selected nodes A–C, including the two species for which only morphological data were available, confirmed this result. The ancestor at all three nodes was most probably polylectic with likelihood probabilities of 99.0% (node A), 79.0% (node B), and 98.6% (node C), and negative log likelihood differences of 3.3, 4.0, and 4.6. These values did only marginally differ for the analysis that excluded the two species with missing molecular data (likelihood probabilities of 98.5% (node A), 78.8% (node B), and 99.0% (node C), and negative log likelihood differences of 3.2, 3.9, and 4.5).

Discussion

The present study provides for the first time a well resolved phylogeny of the commercially promoted Osmia pollinators and their relatives. It additionally uncovers a number of intriguing patterns of host plant use in this predominantly polylectic group of bees and identifies two evolutionary shifts from polylecty to oligolecty.

Phylogeny

Our phylogeny is not consistent with the current morphology-based subgeneric classification (Michener, 2007) in that the subgenus Osmia appears to be paraphyletic as a result of the sister group relationship of (Monosmia + Orientosmia) with the emarginata clade. Although the basal position of O. ribifloris is only weakly supported, the strongly differing host plant spectrum of this species is in line with its distant position. Given the paraphyly of the subgenus Osmia in conjunction with the pronounced morphological resemblance among the species of all three subgenera (Peters, 1978; Müller, 2012), we propose to merge the three subgenera Osmia (25 species), Monosmia (one species), and Orientosmia (three species) into a single large subgenus Osmia. By uniting these three subgenera, a phylogenetically strongly supported clade emerges, which morphologically differs from most other Osmia taxa by: (1) the wide malar space in both sexes; (2) the medioventrally broadened middle femur of the male; (3) the usually simple and little exposed male tergum 7; and (4) the long penis valve that attains or even projects over the apex of the gonoforceps, which is slender and parallel-sided medially (Michener, 2007).

Patterns of host plant use

Among the most important pollen hosts of bees of the Osmia s.s. group are Fabaceae, which are the exclusive hosts of three species of the apicata clade, major hosts of all species of the emarginata clade and regular hosts of O. ribifloris, as well as of several species of the bicornis clade. The importance of Fabaceae for these bee species is in line with the fact that zygomorphic flowers are common hosts of many higher megachilid lineages such as the Osmia s.s. group, whereas the basal representatives of both the Megachilidae and the Osmiini predominantly exploit actinomorphic flowers with well accessible anthers (Litman et al., 2011; A. Müller and C. Sedivy, unpubl. data). Fabaceae are also among the most important pollen hosts of the closely related subgenera Pyrosmia, Helicosmia, and Melanosmia (Müller, 2013), suggesting that the exploitation of Fabaceae by species of the Osmia s.s. group is an ancestral trait. Indeed, O. bicornis and O. cornuta were experimentally found to be able to develop on pure pollen diets of two Fabaceae species (M. Haider, S. Dorn and A. Müller, unpubl. data), although both belong to those few species of the O. bicornis group that do not appear to commonly exploit Fabaceae (Table 1).

Osmia apicata and O. cerinthidis are specialized on Onosma and Cerinthe, respectively, both members of the Boraginaceae. The closest relatives of these two oligolectic bee species are either strictly specialized on Fabaceae or regularly exploit Fabaceae for pollen. This pattern corresponds to the ‘Boraginaceae–Fabaceae paradox’, which describes the counterintuitive affinity of bees of a clade of Hoplitis species towards both Boraginaceae and Fabaceae, which are neither closely related, nor share similar flower morphologies (Sedivy et al., 2013). The affinity of the Hoplitis bees towards these two plant families was hypothesized to be a result of similar chemical properties of their pollen, which might require similar physiological adaptations to digest it. In the apicata clade, the considerably elongated mouthparts, which are typical for all its species (Müller, 2012), might have contributed to promote switches between Fabaceae with long flower tubes (e.g. Astragalus) and Onosma, because such long mouthparts facilitate access to the deeply hidden nectar during pollen harvesting.

Asteraceae are ubiquitous in most terrestrial habitats and their inflorescences produce considerable amounts of pollen, which is easily harvestable for any flower visitor. However, among the 663 pollen loads of 21 Osmia species analyzed in the present study, only ten contained small quantities of Asteraceae pollen, which is remarkable given the high phylogenetic and morphological diversity of the pollen hosts exploited by these mostly polylectic bees. The almost complete absence of Asteraceae pollen in the pollen spectra of bees of the Osmia s.s. group is in line with the observation that females of O. lignaria rejected the collection of pollen of Heliantheae (Asteraceae) both in choice and nonchoice experiments (Williams, 2003). Recent findings indicate that the pollen of many Asteraceae taxa possesses chemical properties that interfere with its digestion by the larvae of unspecialized bees and again suggests that bees need physiological adaptations to overcome these unfavourable properties (Müller & Kuhlmann, 2008; Praz, Müller & Dorn, 2008b). Indeed, the larvae of O. bicornis and O. cornuta proved to be incapable of developing on a pollen diet composed of pure pollen of Tanacetum (Asteroideae) (Sedivy et al., 2011), pollen of Taraxacum (Cichorioideae) was experimentally found to exert lethal effects on developing larvae of O. lignaria if provided as sole food (Levin & Haydak, 1957; Rust, 1990), and larval survival of O. lignaria was considerably reduced on pollen of three Heliantheae species (Asteraceae) (Williams, 2003).

The pollen host spectrum of O. ribifloris differs strongly from that of all other members of the Osmia s.s. group. Species of Ericaceae, Berberidaceae, Ebenaceae or Grossulariaceae play a predominant role as pollen hosts and they all have their anthers more or less concealed inside bell-shaped flowers of rather small size. Because these host plant taxa are not closely related except for Ericaceae and Ebenaceae, which both belong to the Ericales (APG, 2009), flower architecture was probably pivotal in shaping the host plant preferences of O. ribifloris, supporting previous findings that flower morphology and mode of pollen presentation may play an important role for bees in acquiring new pollen hosts (Müller, 1996; Sipes & Tepedino, 2005; Sedivy et al., 2008). Indeed, females of O. ribifloris apply a specialized behaviour when collecting pollen on Vaccinium (Torchio, 1990). They insert their forelegs into the flowers and rapidly strike the staminal filaments with sufficient force to cause the anthers to vibrate, resulting in pollen release from the poricidal anthers. A similar pollen harvesting technique is applied on flowers of Berberis and Diospyros (J. Neff, pers. comm.). Interestingly, females of O. ribifloris possess a modified pilosity on the basitarsi of their forelegs composed of apically curved bristles, which probably help to remove pollen from the host flowers, as do similarly modified bristles on proboscides or forelegs of other bees that harvest pollen on narrow-tubed flowers (Müller, 1995; Sedivy et al., 2013). Another possible adaptation of O. ribifloris to its hosts is the particular pilosity of the ventral scopa. Compared to all other species of the Osmia s.s. group, this pilosity is relatively sparse and composed of thick hairs, which might facilitate the transport of the large pollen grains of Ericaceae, Berberidaceae, and Diospyros, similar to the extremely sparse scopal pilosity of some Hoplitis and Tetralonia species that harvest the giant pollen grains of Malvaceae (A. Müller, unpubl. data) and the stout, unbranched scopal hairs of Xenoglossa species that collect the very large pollen grains of Cucurbita (Roberts & Vallespir, 1978).

By contrast to all other species of the Osmia s.s. group, the pollen host spectra of O. bicornis, O. kohli, O. pedicornis, O. taurus, and O. tricornis are dominated by plants that provide no or very little nectar, which supports previous observations on the favoured pollen hosts of O. bicornis and O. tricornis (Free & Williams, 1970; Tasei, 1973; Vicens et al., 1994). The brood cell provisions of O. bicornis, O. taurus, and O. tricornis appear to be relatively dry, which has been attributed to their low nectar content (Maeta, 1978; Westrich, 1989; Vicens et al., 1994). However, sugar content in provisions of O. bicornis was not found to significantly differ from that in provisions of O. cornuta (A. Bühler and A. Müller, unpubl. data), a species that usually does not collect pollen on nectarless flowers (Table 1). This suggests that O. bicornis and its close relatives must satisfy their nectar needs on other plant taxa than their preferred pollen hosts, as has been assumed previously (Raw, 1974). Indeed, O. lignaria has been hypothesized to compensate for the moderate nectar content of single flowers of Salix, one of its preferred pollen hosts (Table 1), by regularly combining visits to both the pollen-rich flowers of Salix and the nectar-rich flowers of Hydrophyllum on the same foraging bouts (Williams & Tepedino, 2003). During nectar uptake on the Hydrophyllum flowers, the females also collect their pollen, probably because the simultaneous harvesting of pollen does not entail many additional costs. Correspondingly, the particularly high diversity of pollen hosts of O. bicornis encompassing 35% of pollen from plants that produce nectar (Table 1) might partly be a result of casual pollen harvesting during flower visits that are primarily intended to collect nectar. Even if the pollen of such nectar hosts does not support larval development when provided as the sole food, its unfavourable properties might be compensated or mitigated by the admixture of favourable pollen (Williams, 2003; Eckhardt et al., in press; M. Haider, S. Dorn & A. Müller, unpubl. data).

Two pairs of polylectic sister species of the Osmia s.s. group each show intriguing similarities in host plant choice irrespective of their geographical range. The western Mediterranean O. emarginata and the eastern Mediterranean O. mustelina almost exclusively exploit hosts from the very same four plant taxa. Similarly, Ranunculus and Quercus are among the most important pollen sources both of the western Palearctic O. bicornis and the eastern Palearctic O. pedicornis; these two species are also the only two species of the Osmia s.s. group that collect Ranunculus pollen in considerable amounts. The larvae of O. bicornis were experimentally found to be able to develop on a pure Ranunculus pollen diet in contrast to the larvae of O. cornuta, where only a small proportion of the tested individuals managed to develop into dwarfish adults (Sedivy et al., 2011; Haider, Dorn & Müller, 2013). This suggests that the physiological ability to digest Ranunculus pollen has been newly acquired by the ancestor of O. bicornis and O. pedicornis. The overlapping host plant spectra, as observed for these polylectic sister species, are consistent with previous studies on predominantly oligolectic clades, which found that closely-related oligoleges often exploit the same host plant taxa. Furthermore, emerging polyleges usually retain the host used by their ancestors at the same time as including new hosts into their diet that are already exploited by closely-related species (Larkin, Neff & Simpson, 2008; Michez et al., 2008; Sedivy et al., 2008, 2013). Thus, we hypothesize that the host plant spectra of broadly polylectic bees are usually conserved to some degree and governed by constraints regarding pollen digestion or flower recognition and handling as is assumed for oligolectic bees (Sedivy et al., 2008).

Evolution of host range

Polylecty appears to be the ancestral state in the Osmia s.s. group with oligolectic species having evolved twice independently from polylectic ancestors, once in the ancestor of the apicata clade and once in O. cerinthidis. The specialization of O. cerinthids does not appear to be complete because this species collects pollen to a small extent from plant families other than the Boraginaceae. Its strong preference for Cerinthe, however, stands in sharp contrast to the broadly polylectic habit of the other species of the bicornis clade. The case of O. cerinthidis thus represents a transition from polylecty to oligolecty that might become more pronounced in future. It is tempting to speculate that those polylectic species of the bicornis clade exhibiting an affinity towards one single host (i.e. O. excavata, O. cornuta, O. cornifrons, O. kohli, and O. tricornis) might represent a transitional stage between polylecty and oligolecty (or vice versa) as has been hypothesized for polylectic species of Colletes bees that show a pronounced preference for a single host plant taxon (Müller & Kuhlmann, 2008).

In the Osmia s.s. group, no switch from oligolecty to polylecty could be observed, which is in contrast to other studies on mostly oligolectic taxa, where shifts from oligolecty to polylecty predominated (Müller, 1996; Larkin et al., 2008; Michez et al., 2008; Sedivy et al., 2008, 2013). Growing evidence suggests that oligolecty is the ancestral state in bees because the most basal clades appear to be highly host specific (Danforth et al., 2013). Because the basal lineages of both the Megachilidae and the Osmiini are also characterized by a narrow host plant range (Sedivy et al., 2008; Litman et al., 2011), the polylectic ancestry of the Osmia s.s. group is consistent with its more recent origin.

Conclusions

The present study provides a well resolved phylogeny of a clade of mason bees of the genus Osmia, whose species have been established as commercial pollinators around the world. Being the first study on the evolution of host plant choice in a group of mainly pollen generalist bees, our results clearly demonstrate that even broadly polylectic bee species may exhibit distinct patterns of host plant use, suggesting that flower morphology, pollen chemistry, and nectar availability may have played crucial roles in shaping their pollen host preferences.

Taxonomy

Genus Osmia Panzer, 1806

Subgenus Osmia Panzer, 1806s.s.

OsmiaPanzer, 1806: 230. Type species: Apis bicornis Linnaeus, 1758 = Apis rufa Linnaeus, 1758, by designation of Latreille (1810: 439).

Amblys Klug, in Illiger (1807: 198); Klug (1807: 226). Type species: Apis bicornis Linnaeus, 1758 = Apis rufa Linnaeus, 1758, by designation of Latreille (1811: 577). Monobasic in Illiger (1807); Klug (1807) was published simultaneously and listed two species.

Osmia (Ceratosmia) Thomson, 1872: 232. Type species: Apis bicornis Linnaeus, 1758 = Apis rufa Linnaeus, 1758, by designation of Sandhouse (1939: 9).

Osmia (Aceratosmia) Schmiedeknecht, 1885: 19. Type species: Osmia emarginata Lepeletier, 1841, by designation of Sandhouse (1939: 9).

Osmia (Pachyosmia) Ducke, 1900: 18. Type species: Apis bicornis Linnaeus, 1758 = Apis rufa Linnaeus, 1758, by designation of Sandhouse (1939: 9).

Osmia (Monosmia) Tkalců, 1974: 337. Type species: Osmia apicata Smith, 1853, by original designation. Syn. nov.

Osmia (Orientosmia) Peters, 1978: 332. Type species: Osmia maxillaris Morawitz, 1875, by original designation. Syn. nov.

Acknowledgements

We particularly thank C. Praz for his help with the phylogenetic analyses. We are indebted to all people listed in the Supporting information (Table S1) for providing bee specimens for DNA extraction. We are especially grateful for the help of the following colleagues who kindly provided bee material for morphological analysis and the analysis of scopal pollen loads: G. Le Goff, S. Schmidt, M. Schwarz, S. Tomarchio, F. Turrisi, and Y. Astafurova (Russian Academy of Sciences); F. Bakker (Naturalis Leiden); H.-S. Lee (Plant Quarantine Technology Center, South Korea); T. Griswold (USDA Logan); B. Harris (Natural History Museum of Los Angeles County); R. Miyanaga (Shimane University); J. L. Neff (Central Texas Melittological Institute); D. Notton (Natural History Museum London); P. Peters (Senckenberg Naturmuseum Frankfurt); J. Schuberth (Zoologische Staatssammlung München); J. Thomas (University of Kansas Biodiversity Institute); O. Tadauchi (Kyushu University); E. Wyman (American Museum of Natural History); and R. Zuparko (California Academy of Sciences). We thank F. Schmid for help with pollen preparations, as well as T. Torrossi and A. Minder for advice with molecular work. Comments made by four anonymous reviewers substantially improved the manuscript. All molecular data analyzed in the present study were generated in the Genetic Diversity Centre of ETH Zurich.