Introduction

Body size influences many aspects of organismal biology, such as patterns of resource use1,2,3, habitat colonization potential4,5, and ecological strategy6,7. Among species with strong ecological interactions, body size can play an even stronger role, and this is especially important for parasites and their hosts8. The importance of size in parasitism is highlighted in Harrison’s rule (HR), which states that host and parasite body sizes generally positively covary9.

Given that parasite body size depends on that of their hosts to some degree, we must explore the evolutionary trends of host-parasite association to test for morphological similarity and divergence between parasites and hosts. Although HR is a common pattern across the animal kingdom, it remains poorly understood from a macroevolutionary viewpoint; the rule has been mostly demonstrated using phylogenetically-independent comparisons to date (i.e. simple allometry between hosts and parasites10,11,12,13,14,15,). Exploring HR under a molecular phylogenetic framework may offer new insights into the underlying adaptive basis and historical context16, while also providing invaluable quantitative metrics of relatedness. For example, are body size correlations a result of co-speciation or does it largely result from shifts to a host that has a similar body size? For such a study, it is ideal to focus on a monophyletic lineage of obligatory parasites with both high body size variation and good host-parasite association data.

Also, HR has been documented in many parasitic organisms: parasitic nematodes and their hosts12, parasitic barnacles and decapods17, avian lice and birds18, and fleas and rodents19. HR has been tested for several insect groups previously, in terms of both traditional parasitism and herbivory, but most insect groups remain unexplored11. One parasitic organism for which HR has never been extensively and quantitatively tested is the cleptoparasitic bees. About 13% of all bees and 20% of bees in the family Apidae exhibit the cleptoparasitic lifestyle20. Rozen21 indirectly proposed that HR applies to these bees, but quantitative analyses have yet to be conducted.

Cleptoparasitic bees secretly enter the nests of other bees and lay their eggs near or in the pollen balls that the hosts provisioned for their offspring, and either the cleptoparasitic female or the cleptoparasitic brood removes the host larva or egg22,23. The larvae consume the pollen that the host female prepared, pupate, and emerge from the host bee nest24. Body size may strongly affect the interaction between cleptoparasitic bees and their hosts in that i) nest entrance size physically constrains the width of parasitic females; ii) cell size physically limits the development of offspring; iii) the amount of the food source depends on the hosts and is generally correlated to the size of the host25; and iv) in the event of conflict, physical size may influence who is the victor. Consequently, cleptoparasitic bees are generally expected to only parasitize similar or smaller host bees on the individual level (physical constraint: i, ii) but, at the same time, it is advantageous to evolve to a similar size of the hosts to maximally utilize the food source and defend themselves (physiological constraint: iii, conflict constraint: iv). These considerations suggest that HR should be corroborated in cleptoparasitic bees, although they may also be slightly smaller, warranting formal testing.

We selected the cuckoo bee subfamily Nomadinae sensu Bossert et al. 201926 (Nomadinae hereafter) as a model group for this study. Nomadinae, with nearly 1,300 described species globally, is the largest lineage where all members are obligatory brood parasites27. They have repeatedly been recovered as a monophyletic group using several different datasets including Sanger multi-locus data28,29,30, transcriptomes and UCEs26, and UCEs alone31. Nomadinae represent an ancient origin of cleptoparasitism (c.a. 100 mya 28,31,) and they attack a wide variety of bees regardless of phylogenetic affinity (Andrenidae, Apidae, Colletidae, Halictidae, and Melittidae). A great deal of work has been done on determining hosts in the Nomadinae32,33,34,35. Alongside this, the nesting behaviors and immature biology of multiple genera have been meticulously investigated21,36,37. In Nomadinae, genera in the same tribe might infiltrate hosts in different families, while genera in different tribes might utilize exclusively the same host genera24. Remarkably, even small genera may use entirely different families (four species on two families, Townsendiella38).

Species that belong to the tribe Nomadini have the largest host range, using 10 genera of five families as hosts20; at the other end of the spectrum, the large genus Epeolus (tribe Epeolini) attacks only the genus Colletes39. Even though the tribe Nomadini has the highest species diversity and the vast host information, its phylogenetic relationships have not been entirely resolved. Following prior works on the group, Alexander conducted species-group classification in the genus Nomada, which is the sole genus in the tribe Nomadini, based on cladistic analyses40. There, he defined 16 species-groups, and this classification has been commonly used since35,41,42,43. However, he mentioned that one of the species-groups, the ruficornis species-group may be a paraphyletic due to the lack of distinct apomorphic characteristics. Recently, the incongruencies between the morphological study and molecular analysis was proposed by Won, and Odanaka et al.44,45. Yet, given its high species diversity in this tribe, it is crucial to reveal the hidden patterns of the phylogenetic relationships.

The majority of cuckoo bees are specialists in that they parasitize only one or a few relative hosts in the same genus25, although there are exceptions such as the more generalist genus Sphecodes in the family Halictidae. There are numerous ways in which cleptoparasitic bees have adapted to track their hosts, including phenological synchrony, and they can also exhibit similar spatial richness patterns46. Nomadine bees also vary greatly in size, making them ideal for studies on HR: the body length of one of the largest species (Acanthopus excellens Schrottky, 1902; ~ 23 mm) is more than ten times larger than one of the smallests (Oreopasites barbarae Rozen, 1992; 2.2 mm ~)36,47. Altogether, these traits provide a unique opportunity to explore HR in a molecular phylogenetic framework.

This study empirically tests HR in the cleptoparasitic bee lineage Nomadinae to infer the role of body size in their macroevolutionary dynamics for the first time. At the same time, we also briefly explore how closely related parasites and hosts are in this group, as Emery’s rule predicts that parasites should be closely related to their hosts48. We measured the body size of specimens, retrieved data from the literature (see Supplementary Data 1), and conducted allometric analyses of Nomadinae and their hosts. Then we revisited the phylogeny of Nomadinae with an increased sampling of the genus Nomada, the largest genus of cleptoparasitic bees35. Finally, we conducted ancestral state reconstruction of host associations and body size, using the resulting phylogenetic tree to infer their evolutionary history and correlation between the two traits (with focus on species whose hosts are known).

The main questions explored by this study are as follows: i) Does body size covary between hosts and parasites?; ii) How does body size change across the phylogeny, and do major shifts correlate with phylogeny or with host switches?; iii) How does host-specificity evolve across the tree and do nomadine bees follow Emery’s rule?

Results

Phylogenetic analysis

Regarding taxon sampling, we targeted species that have both host and body size information without bias. In total, 106 species including 2 species for the outgroup and 104 species for the ingroup were used. We extracted novel sequences from 35 species, and 71 species were obtained from NCBI (Table S1). About 30% of ingroup taxa belong to Nomada, which has an relatively unstable phylogenetic position and possesses the most host information in the subfamily. The dataset used for phylogenetic reconstruction contained a total of 4590 bp (657 bp of COI, 709 bp of Ef1α, 1463 bp of Nak, 459 bp of Opsin, 843 bp of PolII, 459 bp of Wng). Phylogenies obtained through Bayesian inference (BI) from MrBayes and maximum likelihood (ML) from IQtree indicate strong support for the monophyly of Nomadinae (Bayesian posterior probability, PP = 100, Maximum likelihood bootstrap values, BS = 99) (Fig. 1).

Figure 1
figure 1

Combined Phylogenetic tree of Nomadinae. Produced with MrBayes. Colors of circles on the node indicate bootstrap supporting values, and the one topological difference between BI and ML trees is presented as a red circle. The tribal classification followed31. Habitus images of cuckoo bees was conducted by the first author, Kayun Lim.

Full size image

Melectini was recovered sister to the remaining cleptoparasitic lineages with high support (PP = 100, BS = 98) (Fig. 1). Each of the tribes Caenoprosopidini, Ammobatini, Ammobatoidini, Hexepeolini represented independent clades. We found support for the relationship among the tribes Isepeolini + Epeoloidini (sensu Sless et al., 2022) + Protepeolini (PP = 100, BS = 100,) + Rhathymini + Ericrocidini (PP = 100, BS = 100), and Brachynomadini + Epeolini (PP = 100, BS = 94). The BI and ML trees showed largely congruent topologies except for a few nodes (Fig. 1). Topological differences between BI and ML analyses were found in the tribes Caenoprosopidini, Epeolini, and Nomadini. The tribe Caenoprosopidini was placed sister to Isepeolini + Epeoloidini + Protepeolini in BI, while it was sister to Ammobatini in ML.

The tribe Epeolini formed a distinct monophyletic clade (PP = 100, BS = 100), but some differences between analyses existed within the group. For example, in BI, Epeolus bifasciatus formed a subclade with E. brumleyi, E. scutellaris, E. basili, E. pusillus, E. autumnalis, E. ainsliei, E. lectoides, and E. lectus. However, E. bifasciatus grouped with E. interruptus in ML. Epeolus transitorius was sister to a subclade including E. asperatus in BI, but the species was recovered closer to E. flavociliatus in ML. Epeolus chamaesarachae was recovered as sister to remaining subclades in the tribe with high support in BL (PP = 100), but the species was adjacent to the E. autumnalis subclade in ML (BS = 88).

The tribe Nomadini was also strongly supported as monophyletic (PP = 100, BS = 100), and it was recovered sister to Hexepeolini (PP = 78, BS = 94) (Fig. 1). In comparison to the tribal relationships, the infra-tribal relationships of Nomadini were more complex. Some simple cases where single species represented species-groups could be valid, for example the roberjeotiana species-group (N. roberjeotiana) and basalis species-group (N. japonica) (sensu Alexander & Schwarz, 199441). Similarly, three members of the furva group, N. distinguenda, N. discedens, N. okubira, were supported as monophyletic with strong support (PP = 100, BS = 100). However, the ruficornis species-group was more problematic. For example, N. lathburiana, which was previously designated as the ruficornis species-group41, was part of the bifasciata species-group. Also, N. aswensis which is supposed to be part of the ruficornis species-group was part of a clade with N. nipponica (tripsona species-group), N. armata, and N. ginran (armata species-group). A difference between BI and ML was found in the placement of N. alboguttata. In BI, the species formed subclade with N. fulvicornis jezoensis, N. fervens, and N. marshamella. On the other hand, it grouped only with N. fulvicornis jezoensis according to ML.

Ancestral state reconstruction of host use

The ancestral states of host use in Nomadinae were analyzed at the family level to enable easier interpretation of the results, avoid an overestimation of shifts, and accommodate program analytical limits. Except for a few nodes, parsimony and RJ-MCMC analyses were largely congruent. According to the Bayestraits analysis, the common ancestral host of subfamily Nomadinae was the family Apidae, with a probability of ~ 63% (Fig. 2). This host family is also inferred as the ancestral host for multiple lineages in Nomadinae. Additionally, we found that the host use of Apidae by Melectini, Ericrocidini, Ammobatoidini was conserved. On the other hand, frequent reversal host shifts between Apidae and Andrenidae were detected. After the first reversal host transition from Andrenidae to Apidae happened in the tribal combination of Isepeolini + Parepeolini + Protepeolini + Rathymini + Ericrocidini, host switching also occurred from Apidae to Colletidae and Melittidae (Fig. 2). The common ancestral host of Epeolini was reconstructed as Family Apidae with a probability of ~ 62%. However, a host switch to the family Colletidae in the genus Epeolus was observed with a probability of > 99%.

Figure 2
figure 2

Ancestral state reconstruction of nomadine host associations at the family-level. Produced using the Bayesian analysis. The pie charts represent mean posterior probabilities assessed under RJ-MCMC analysis using Bayestraits. Branch colors indicate the result of parsimony ancestral state reconstruction performed by Mesquite (Black: Melittidae; Navy: Apidae; Sky blue: Colletidae; Ivory: Andrenidae; White: Halictidae; Mixed black/white lines: outgroups or unknown hosts).

Full size image

In the clade of Ammobatoidini + Neolarrini + Hexepeolini + Nomadini, their common ancestral host use was yielded as Family Andrenidae with a probability of > 96% (Fig. 2). Thereafter, a host shift occurred from Andrenidae to Halictidae at least three times in this group. To be specific, the first transition happened in the tribe Biasitini, and the remaining two transitions in different clades of the tribe Nomadini. In the tribe Nomadini, there was one switch from Andrenidae to Apidae within our sampling.

The most frequent host switches were observed in the tribe Ammobatini (Fig. 2), and the tribe showed a tendency of reversal host shifts from Andrenidae to Apidae. However, the ancestral host family reconstruction for the tribe was ambiguous and recovered associations with multiple families.

Allometry and ancestral state reconstruction of body size

The body length of the Nomadinae was on average 9.08 ± 0.36 (mean ± S.E.), widely ranging from 2.5 to 23 mm, and the intertegular distance (ITD) was 2.06 ± 0.11, ranging from 0.48 to 6.82 mm. The mean body length of the hosts was 10.66 ± 0.38, ranging from 3.6 to 23.28 mm and host ITD was 2.84 ± 0.17, ranging from 0.73 to 8.15 mm (see Supplementary Data 1). Our linear regression analysis strongly suggests that the size between Nomadinae and its hosts is highly related in terms of both body length (R2 = 0.6879, P < 0.05) and ITD (R2 = 0.7620, P < 0.05).

The ancestral state of ITD and body length was of moderate length (mean of full body length of Melectini: 11.06 mm, median ITD of Melectini: 3 mm; mean full body length of hosts: 13.43 mm; mean ITD of Melectini: 3.59 mm), and the length and width of body forms have evolved from medium to extremes (Fig. 3). It was observed in the tribe Ericrocidini that the body evolved to become longer and wider, but this phenomenon was found only in the tribe Ericrocidini. On the other hand, becoming shorter and narrower was observed in multiple lineages. This distinctly recognizable pattern of shrinking was detected in the tribes Ammobatini, Brachynomadini, Ammobatoidini, Neolarrini, and Nomadini.

Figure 3
figure 3

Size correlation between cuckoo bees and hosts. (A) Linear regression analysis of ITD (R2 = 0.7620, P < 0.05). (B) Ancestral character estimation of body width (ITD) of cuckoo bees (left) in accordance with size of the hosts (right). (C) Linear regression analysis of entire body length (R2 = 0.6879, P < 0.05). (D) Ancestral character estimation of body length of cuckoo bees (left) in accordance with size of the hosts (right). Illustrations in (A,C) were converted from photographs obtained by the first author using PRISMA mobile application.

Full size image

Intriguingly, we found potential indicators that changing size is related to host shifting across the phylogeny. To be specific, a considerable body size increase was detected in the tribe Ericrocidini as the host changes from Andrenidae to Apidae. In contrast, both body length and width remained relatively similar within the genus Epeolus, which is parasitic on just the genus Colletes. This was not always the case, however, as seen among the species-groups of Nomada. For example, N. flavoguttata, which belongs to the Nomada ruficornis species-group and parasitizes the genus Andrena, showed a similar body size with species of the Nomada furva species-group even though that species-group specializes on a different genus, Lasioglossum.

Discussion

Most of the relationships recovered in this study correspond with a prior investigation that used phylogenomic sampling31. Major differences from prior studies include the positioning of the tribes Caenoprosopidini and Ammobatoidini, and the sister group of Nomadini. Firstly, the tribe Caenoprosopidini is located sister to the tribes Isepeolini + Epeoloidini + Protepeolini in BI in our study, while it was placed in the ‘Nomadinae line’ in ML31, which includes the Caenoprosopidini, Ammobatini, Ammobatoidini, Brachynomadini, Epeolini, Hexepeolini, Neolarrini, and Nomadini. Second, the tribe Ammobatoidini belonged to a subclade with the tribes Neolarrini, Hexepeolini, and Nomadini in this study. Conversely, it is located closest to the tribes Brachnomadini and Epeolini in previous investigations31. Third, the tribe Hexepeolini was inferred as the sister group of the tribe Nomadini. The placement of this tribe has varied even among molecular studies as sometimes it is thought to be sister to the Ammobatoidini, Ammobatini, or Neolarrini26,28,29,30,31. However, our results are relatively similar to those of the most recent study31, where Nomadini was sister to Hexepeolini + Neolarrini. The discrepancy between our results and previous Sanger studies is likely due to differences in taxon sampling, since most prior studies have used relatively few samples. In general, however, we find the most recent Nomadinae phylogenomic study is likely representative31, given the depth of data used and their good coverage across taxonomic groups.

Odanaka and colleagues recently conducted the first densely-sampled molecular phylogenetic investigation of Nomada using UCE phylogenomic data with 16 species-groups, and their investigation suggested that the odontophora species-group is sister to the rest45, contrasting with our results that suggested the roberjeotiana species-group was sister to the rest. Notably, however, we did not include a sample of the odontophora species-group and consequently, here and elsewhere, it seems likely that the differences are due to taxon sampling. Further, difficulties in assigning species-group membership correctly are well known for this genus40, such that this could explain some disagreements between our two studies. For example, they recovered the superba and ruficornis species-groups as paraphyletic, while our study found that the armata species-group is paraphyletic, and ruficornis species-group is polyphyletic 45. Our multi-gene phylogeny supports the designation of N. ginran as a member of the armata species-group as suggested by42. On the other hand, in disagreement with Alexander and Schwarz41, N. aswensis and N. kaguya (previously treated in the ruficornis species-group) and N. nipponica (previously tripsona species-group) were also recovered in the armata group. Additionally, N. lathburiana (of the ruficornis species-group) was recovered in the same clade with N. succincta, N. comparata, N. goodeniana, which belong to the members of the bifasciata species-group. Therefore, further investigation with increased taxon sampling is necessary to resolve the true composition and relationships of the Nomada species-groups. Again, differences in results between our study and the most recent investigation may be due to differences in taxon sampling, since we primarily focused on species with valid host information. In addition, Odanaka et al. examined Nomada phylogeny at a deeper level by using the UCE phylogenomic data with a much denser sampling pool45. Given that the phylogenomic approach is possible to provide dept and resolution to the data, the incongruencies between the latest study and the present work are presumably due to the different research methodologies.

Our results largely agree with prior studies, with the family Apidae as an original host and multiple switches to other groups and sometimes back to Apidae. In this way, nomadines appear to initially follow Emery’s rule, with parasites only attacking close relatives as seen with Melectini attacking Anthophorinae, but later with many groups attacking entirely different families very successfully31.

In comparision to the most recent reconstruction of Nomadinae31, our study determined that host shifting from Apidae to Andrenidae in Nomadinae occurs much earlier. This is because the tribe Caenoprosopidini is shown as sister to the Isepeolini + Epeoloidini + Protepeolini + Rhathymini + Ericrocidini subclade which belongs to the Ericrocidini line (sensu Sless et al., 2022) in our investigation. Conversely, in the previous study, Caenoprosopidini was placed within the Nomadinae line31. Denser sampling of these groups will help resolve these and other disagreements.

Host specificity to certain groups has been frequently used to define taxonomic units, including in cuckoo wasps, fig wasps, and some mites24,49,50,51. Some similar examples exist in Nomadinae. For example, the host genus of Epeolus is well known to be limited to the genus Colletes52, and Ericrocidini is limited to Centridini39. The common feature of these two examples is that their hosts have specialized nesting strategies, and this may explain some instances of conservatism. Most colletid females apply a cellophane-like layer with their short, bilobed glossa in brood cells39,53. Conversely, many hosts of the tribe Ericrocidini collect floral oil for their brood-cell construction20. Applying floral oil and other secretions may make it difficult for parasites to access these nests54. Both of these specialized nesting behaviors, especially oils which may be hard to digest, could physically inhibit the offspring of parasites55. It may be that the adaptations for using these nests, in turn, make it harder for them to switch to other host groups with less specialized behaviors, further enforcing patterns of conservatism56.

Of course, exceptions to conservatism of host use exist, such as the use of many different hosts by Nomada. Coincidently, this group also represents another inconsistency with the most recent study31. We found three host shifts in the tribe, while they found one. There are over 800 species of Nomada35 and many of them have unknown hosts, so this result depends heavily on taxon sampling. With greater sampling, we would expect additional host shifts. For instance, the host of the Nomada emarginata Morawitz (ruficornis species-group) is recorded as Melitta haemorrhoidalis Fabricus57, which belongs to the family Melittidae, but it was not included in our analysis. Similarly, the unused Nomada articulata Smith (erigeronis species-group) takes Agapostemon sericeus Foster and A. virescens Fabricius as their hosts58, and this could add another shift. As there are numerous species with unknown or unconfirmed hosts, many of which might invoke additional host shifts when added to the phylogeny, it is difficult to estimate the expected total number of host shifts for this group at this time.

Our size allometry analysis strongly supports the positive correlation between parasites and hosts, commonly referred to as Harrison’s rule (HR 9;). Further, we found multiple instances where host shifts were linked with body size changes. It is likely that denser taxon sampling will reveal more such linkages between shifts in host use and body size, and it may be that this is a broader co-evolutionary pattern in Nomadinae.

Similar to prior studies on other taxa59,60,61,62, the body size of cleptoparasitic bees was typically smaller than their hosts in this study. We postulate that there is a “lock-and-key” relationship between the Nomadinae and their hosts (referred to in lice prior, based on the need to fit well on hair 13;). Most of the hosts of Nomadinae are soil-nesting, solitary bees. Females make their nests underground by digging, and their nest tunnels are typically just large enough to allow them passage. Because cleptoparasitic bees must enter nests to lay their eggs, they are constrained by the size of these tunnels, nest cells, and even the amount of food provisioned for offspring25,63,64. Our investigation supports this idea, with host shifts or stability seemingly influencing body size. For example, as shown in Fig. 3, there is relatively little size fluctuation in the genus Epeolus, all of which attack Colletes52. On the other hand, Nomada, which parasitizes a vast array of groups, shows scattered size-shifting corresponding with host switches. The ability of cleptoparasitic bees to modulate their size in response to hosts may contribute to the ability of these bees to switch hosts more readily, which could in turn enable adaptive radiations generated from switching to many new hosts (a possible component of the success of the huge group Nomada). Of course, plasticity in host recognition via visual and chemical pathways, as well as adaptability to different nectar and pollen resources, would also be important components.

Another example that highlights the relationship between host switching and body size modulation is found in the tribe Ammobatini. Even though the genus Melanempis was excluded from our study because its host is unknown, it is one of the largest nomadines and warrants discussion. According to the most recent study31, Ammobatini includes the small-sized Oreopasites (as small as 2.2 mm) and Melanempis (up to 22 mm), demonstrating huge size variation in the group36,65. Alongside this wide range of body sizes, they also parasitize many families: Apidae, Andrenidae, Colletidae, and Halictidae (See supplementary data in31). With further data from this group on host associations and body size, and greater taxon sampling, it could be an ideal finer-scale study for looking at the influence of host switching on body size.

Body size is thought to be one of the factors that could contribute to cophylogenetic patterns between hosts and parasites66. Food resource requirements being similar among close relatives can help explain such patterns, as they should be relatively similar13. Indeed, the ancestral host of Nomadinae was revealed as Apidae in this study (specifically Anthophorinae 31,), so co-evolution may have taken place more closely early in the evolution of the nomadines. However, behavioral flexibility can empower parasites to invade novel hosts56,67. Such potentials are enhanced when alternative hosts have similar-sized nests25, or when their increased abundance makes it more likely for them to be encountered, especially if in the same microhabitat68. Host chemical cues also likely play an important role but are relatively poorly understood69. Such factors may help explain how single species can exploit many varied hosts70. For example, Nomada flavoguttata parasites various Andrena species such as Andrena falsifica, A. minutula, A. minutuloides, A. semilaevis, and A. subopaca35. Notably, all these host species belong to the same subgenus, Micrandrena, which is well-known to have diminutive body size, and may also have similar chemical cues and metabolic requirements71. However, occasionally, a single Nomada species exploits multiple genera or even families as hosts, such as in N. imbricata, parasitizing Andrena and Halictus70. With greater emphasis on natural history studies to make more host-parasite associations, even more disparate host uses might be discovered.

With the combination of our multi-locus phylogeny and allometric data, our understanding of the evolutionary relationship of size fluctuation between cleptoparasitic bees and their hosts has greatly improved. Based on the scenarios and examples discussed above, host switches appear closely related to changes in body size. The ability to adapt to new hosts more quickly in this way could increase speciation rate in the presence of potential new hosts, explaining highly diverse groups like Nomada. Ultimately, we confirm Harrison’s rule, the positive correlation of body size between cleptoparasitic bees and hosts, and with further sampling we expect to find an even stronger impact of body size on the evolution of this group.

Methods

Taxon sampling

In total, we used 104 species for the ingroup. For the outgroup, Apis cerana and Colletes compactus were selected to represent a range of relatedness and performed well in pilot analyses, compared to the relationships recovered in a recent phylogenomic analysis31. Sequences for 71 species from published papers were taken from NCBI, and we newly added 35 species that have not been analyzed in previous phylogenetic investigations (Table S1). To minimize missing data and ensure better resolution in the phylogenetic tree, we used sequence data only if the number of available genes was greater than or equal to three. The aim of the sampling was to use species that have both host and body size information without bias. When species did not fulfill this condition, they were excluded from this study. However, we made exceptions where necessary to ensure that the resultant phylogenetic reconstructions were reliable, with reference to prior works informing our choices31,72.

We placed special focus on sampling the genus Nomada as it has highly-diverse host records31. Specifically, we expanded the sampling of the furva species-group because they are distinctly small-bodied and the parasite of genus Lasioglossum (Hymenoptera: Halictidae). Two species, N. aswensis, and N. kaguya, which belong to the ruficornis species-group but show similar biology with the furva species-group, were also included in this study. We included two type species, N. ruficornis from ruficornis species-group and Nomada armata from armata species-group, to fully investigate the composition of these species-group concepts by Alexander40, although greater sampling is necessary to fully define the species-groups45. The study used DNA samples of Finnish Nomada species from the Genomic Resources collection (https://laji.fi/en/theme/luomusgrc/instructions) of the Finnish Museum of Natural History Luomus (https://www.luomus.fi/en), and the HTTP-URIs of the DNA samples are listed in Table S1. In total, our sampling of the genus Nomada represents seven of the 16 species-groups, including a range of host information in three families representing Apidae, Andrenidae, and Halictidae. We could not represent all family-level hosts for this group because there are few species attacking some groups, such as Melittidae, and they were more difficult to sample.

DNA extraction, PCR amplification, and sequencing

Total genomic DNA was extracted by grinding up detached midlegs or heads of either alcohol vouchers or dried specimens; forceps used for extraction were sanitized before by flame and rinsed in 99% EtOH between each specimen. DNeasy Blood and Tissue kit (QIAGEN, Inc.) was used for the DNA extraction according to the manufacturer’s protocols and stored in − 20 °C. DNA vouchers were deposited in the Insect Biosystematics Laboratory, Seoul National University.

Genes commonly used in Anthophila phylogenetics were selected. One mitochondrial protein coding gene, cytochrome oxidase subunit I gene (COI)52,73, and five nuclear protein-coding genes (EF-1α, long-wavelength rhodopsin (opsin), NaK, pol II, and wingless) that were used in previously published literature28,29,52,55,74,75,76,77,78 were chosen for this study. In some cases, COI from NCBI were used.

PCR products were amplified using Accupowder PCR Premix (Bioneer, Daejeon, Korea) in a 20 µl reaction mixture. In case DNA extraction was needed from bad quality specimens, we redesigned the partial primers in threefold. The primers used in this study and specific PCR conditions are listed in Table S2. After the amplification process, PCR products were purified and sequenced at Bionics CO. (Seoul, Korea).

Using the SeqMan Pro version 7.1.0., raw sequence data were assembled, checked, and trimmed. Sequence alignment of six genes were performed in MAFFT version 7 (https://mafft.cbrc.jp/alignment/server/)79. All the sequences were adjusted in Mega 7 with the amino acid translation option. The aligned sequences were combined using SequenceMatrix Windows ver. 1.880.

Phylogenetic analyses

Phylogenetic analyses were performed using both Bayesian inference (BI) and Maximum likelihood (ML). The Bayesian inference analysis was conducted with MrBayes 3.2.7a81. The best substitution model for BI was selected for each partition under the Bayesian Information Criterion (BIC) using ModelFinder in IQTREE and the same protocol was used in searching for the best substitution model for ML82,83. GTR + F + I + G4 for COI, TIMe + I + G4 for ef1α, TIM3 + F + I + G4 for Nak, TIM3e + I + G4 for Opsin, TIM2 + F + I + G4 for PolII, and TIM + F + I + G4 for Wng were selected as the best fitting models for BI. For ML, GTR + F + I + G4 for COI, TIM + F + I + G4 for ef1α, TIM3 + F + I + G4 for Nak, TIM3e + I + G4 for Opsin, TIM2 + F + I + G4 for PolII, TIM + F + I + G4 for Wng were selected84. However, since TIM models cannot be used in MrBayes, GTR + I + G was used to run the BI analysis except COI. For the MrBayes analysis, we ran 20 million Markov chain Monte Carlo (MCMC) generations and trees were sampled every 100 generations. We ran one cold chain and three heated chains for each MCMC analysis. The first two million sampled trees were discarded as burn-in. Branch support for the maximum likelihood analysis was computed using the UltraFast bootstrap approximation (UBS 85;) with 1,000 replicates.

Size allometry

We measured two traits of the Nomadinae and their hosts. One is the intertegular distance (minimum distance between the tegulae), which is a useful estimator for the size of bees86. The other is total body length, which was measured as the maximum length distance from the head to end of the final tergite exclusive of exserted stingers or genitalia. (if curled, this was accounted for using a multi-stage measurement). In total, we obtained about 1,300 body size data from Nomadinae and their hosts (Body length: 73; ITD: 65 out of 106 species of Nomadinae and their corresponding hosts). Among them, we used specimens of 24 species (host: 7, Nomadinae: 17) and prepared photographs using the Microscope (DM 4000B, Leica Microsystem, Wetzlar, Germany) with a USB digital camera (Infinity3, Lumenera Corporation, Ottawa, Ontario) and measured the value of the traits using the measurement option. Since there can be a size difference between male and female bees, measurements were only taken from females. We also collected data from literature such as taxonomic papers, online-accessible specimen photographs in museums, or from taxonomists (see Supplementary Data 1). The number of measured individuals was different per species because we gathered the data from various sources. Therefore, we used the average value of measurements for the data analysis. Linear regressions were conducted using SPSS Statistics 25 (IBM, Armonk, NY, USA) to gauge the dependence of nomadine body length on host body length, as well as nomadine ITD on host ITD.

Ancestral character state reconstruction

The host information of each nomadine species was extracted from the literature (See supplementary data 1). Where multiple species were reported as hosts, the data was prioritized when the parasitic larvae were found in their nests or when direct intrusion of the parasite to the host nests was observed. However, when such confirmation was absent, data arose from phenological synchrony between parasites and hosts and information provided by taxonomists, including hypotheses raised in the literature.

The hosts of Nomadinae were coded as five discrete states by families: (A) Andrenidae, (B) Apidae, (C) Colletidae, (D) Halticidae, and (E) Melittidae. This may underestimate the number of shifts because different lineages are used in some families, but as it is difficult to exactly know when switches occurred this is in some cases conservative, and avoids the artificial inflation of the number of shifts. The trace character history function in in Mesquite 3.31 was used to map the evolutionary history of host use on a single tree with parsimony approach87 was adopted. The probability of the ancestral state of each node was calculated by Bayestraits v3.088 using reversible jump Markov Chain Monte Carlo (RJ-MCMC). An exponential distribution was implemented, seeding from a uniform prior in an interval of 0–100. We ran 50 million iterations, sampling every 1,000th iteration. The first million iterations were discarded as burn-in. Acceptance rates were automatically adjusted and achieved in the preferred range of near 35%.

To trace the evolutionary history of intertegular distance and body size, we used ancestral state reconstruction using the maximum likelihood (ML) method in the R package ‘Phytools’89. The resulting Bayesian tree was converted into a dendrogram, the missing data vector was estimated using ML, and the ContMap function was used to visualize their evolutionary history.