Exclusion of bird pollinators impacts mating system and reduces offspring fitness in a pollination-generalist tree

Abstract

Background and Aims

In comparison to pollinating insects and non-flying mammals (NFMs), nectarivorous birds might display behaviours leading to greater pollen carryover. Therefore, plants pollinated by birds might display higher levels of paternal diversity and outcrossing than those pollinated by insects and NFMs, with associated fitness benefits for seeds and seedlings. Here, we test these predictions using a plant where birds, insects and NFMs are all frequent visitors to flowers.

Methods

An experiment manipulating access to flowers of Banksia menziesii (Proteaceae) was conducted. Treatments applied to whole plants were: (1) open to all pollinators; (2) insect access, with birds and NFMs excluded; (3) NFM access, with insects and birds excluded; and (4) complete pollinator exclusion. Reproductive output was quantified in terms of fruit and seed production. The genetic consequences for offspring were tested using microsatellite markers to genotype individuals and quantify the mating system, and through field trials to quantify seedling vigour.

Key Results

When birds were excluded from flowers, maternal fitness was reduced. In comparison to open pollinated flowers, fruit set was reduced by 76 % when only NFMs could access flowers. When only insects (primarily introduced honeybees) could access flowers, the number of viable seeds per fruit was reduced by 20 % because the proportion of aborted seeds doubled, in comparison to pollination that included birds. For seedlings, heterozygosity was reduced by 22 %, outcrossing rates by 30 % and paternal diversity by 15 %, when birds were excluded. Seedling mortality was strongly associated with inbreeding, and selfing largely occurred only when birds were excluded. All parameters were lowest when only NFMs had access to flowers.

Conclusions

Although honeybees were effective pollinators of B. menziesii owing to their abundance, birds were inferred to be the most effective, with their exclusion resulting in a reduction of fecundity and offspring vigour. These negative effects were largely a consequence of selection against the products of self-pollination, which was associated with pollination by insects or NFMs. Our findings highlight how a genetic component of pollination that increases offspring fitness could favour the evolution of bird pollination.

INTRODUCTION

Pollinators differ widely in their foraging behaviour, morphology and pollen transfer mechanics (Faegri and van der Pijl, 1980). From a plant perspective, these differences mean that pollinators differ in their effectiveness in terms of the quantity and quality of the pollen they transfer among the plants (Sahli and Connor, 2007; Ne’eman et al., 2010; King et al., 2013; Valverde et al., 2019). Therefore, differences among pollinators can have a profound impact on plant reproduction and plant mating system properties such as outcrossing rates and paternal diversity (Devaux et al., 2014; Wessinger, 2021). These differences in pollinator effectiveness might have been important for driving evolutionary shifts in floral traits and the pollinators they attract, which can potentially lead to speciation (Stebbins, 1970; Fenster et al., 2004; Cronk and Ojeda, 2008; Pauw, 2019; Wessinger, 2021). Differences among pollinators in effectiveness can also have conservation consequences, especially if the more effective pollinator species undergo decline in anthropogenically modified environments (Potts et al., 2010; Regan et al., 2015; Powney et al., 2019). Furthermore, invasive pollinator species with a broad diet, such as the European honeybee Apis mellifera (Apidae), can be less effective than the native pollinators to which the plant is adapted (Whelan et al., 2009; Ayre et al., 2020; Diller et al., 2022).

Although the majority (∼82 %) of plant species are pollinated exclusively by insects, pollination by vertebrates has evolved repeatedly in a diversity of plant families across the world (Ollerton et al., 2011; Ollerton, 2017). It is estimated that ≥1000 bird species globally pollinate >10 000 plant species across ∼500 genera (Pauw, 2019; Johnson, 2022; Leimberger et al., 2022), with a bias towards the tropics and Southern Hemisphere temperate zone (Krauss et al., 2017). Among mammals, bats are the major pollinators globally (Fleming et al., 2009), although many species of non-flying mammals (NFMs) including primates, rodents and marsupials, are also involved in pollination (Janson et al., 1981; Carthew and Goldingay, 1997; Johnson et al., 2001; Zoeller et al., 2016; Kobayashi et al., 2021). Although some plant species have general pollination systems and are visited by multiple functional groups of pollinators (e.g. Dellinger et al., 2019), a meta-analysis of 126 experiments on vertebrate-pollinated plant species found that exclusion of birds from flowers that were also visited by insects reduced fruit and/or seed production by an average of 46 % (Ratto et al., 2018). There is no comparable meta-analysis for primarily mammal-pollinated plants, but experimental studies have demonstrated that they are critical for pollination in a range of plant species, with a centre of diversity in southern Africa (e.g. Johnson et al., 2001, 2011; Kleizen et al., 2008).

In addition to their role in plant fruit set, there are features of birds that can result in high pollen carryover and pollen loads comprising multiple donors deposited onto stigmas, from which fitness benefits are predicted for the offspring of the plants they pollinate (Krauss et al., 2009; Krauss et al., 2017; Ratto et al., 2018; Pauw, 2019). For example, birds are larger bodied and might move more frequently between plants compared with most insect pollinators. In addition, nectar-feeding birds often exhibit intra- and interspecies aggression (e.g. MacNally and Timewell, 2005; Temeles and Kress, 2010; Phillips et al., 2014), leading to disruptions to optimal foraging predictions, where energetic costs are minimized during foraging by preferentially moving between nearby flowers (Pyke, 1981; 1984; Heystek et al., 2014). Furthermore, birds do not groom themselves to harvest or remove pollen in the same way as many bees and mammals do (Westerkamp, 1991; Johnson and Pauw, 2014). Alternatively, pollination by NFMs is expected to lead to lower levels of pollen carryover and less diverse pollen loads for trees and shrubs, because the need to climb from one flowering plant to the next might favour pollen transfer predominantly within plants or to nearest neighbours (Garavanta, 1997; Garavanta et al., 2000). In a review of the genetic consequences of bird pollination, Krauss et al. (2017) found that plants pollinated by birds tend to show higher paternal diversity than plants pollinated by insects. High paternal diversity is theoretically advantageous because offspring diversity enables bet-hedging in the face of environmental variation, in addition to the more immediate benefits of heterozygosity (heterosis) (Breed et al., 2014; Barrett and Harder, 2017). However, the advantage of bet-hedging might be reduced if pollinators that are less effective at facilitating outcrossing predominate (Delmas et al., 2014).

A crucial yet largely neglected component of pollinator effectiveness is the consequence of pollen source for offspring quality, which can be measured genetically through estimation of mating system parameters following manipulation of access to flowers by pollinators (Steenhuisen et al., 2012; Diller et al., 2022). To our knowledge, no previous study has manipulated access by three pollinator groups (birds, NFMs and insects) and explicitly addressed the mating consequences that underlie their relative effectiveness at multiple life-history stages on the same plant species, nor extended that assessment to corresponding measures of seedling vigour and establishment (Breed et al., 2012; Nora et al., 2016). Our approach recognizes that detrimental effects associated with inbreeding can be expressed at multiple stages in the development of an individual (Charlesworth and Willis, 2009; Barrett and Harder, 2017). For example, inbreeding depression can impact pollen tube growth, ovule fertilization, embryo abortion, seed viability and germination, seedling emergence and growth, and competition among individuals at multiple life stages (Stephenson, 1981). When reduced fitness is associated with inbreeding, pollinators that promote outcrossing and paternal diversity are expected to have fitness advantages for the plant and drive the evolution of floral traits that favour them as pollinators (Hopper, 2009; Pauw, 2019).

In many regions of Australia, bird pollination is particularly prominent, with honeyeaters (Meliphagidae) the primary bird family acting as pollinators (Paton and Ford, 1977; Armstrong, 1979; Keighery, 1980). This high frequency of bird pollination has been hypothesized to have arisen from a large allocation of expendable energy towards the profuse production of nectar (Ford et al., 1979; Orians and Milewski, 2007) or, in some regions, through preferential outbreeding in species with naturally patchy populations (Hopper, 2009). Some plant genera, such as Banksia, an iconic genus of ∼170 species, are important food sources for multiple functional groups of animal pollinators, with most species studied thus far visited by birds, insects and mammals for nectar and/or pollen (Hopper, 1980; Collins and Rebelo, 1987; Ramsey, 1988; Collins et al., 2008; Phillips et al., 2010; Krauss et al., 2018). Such plant species provide a powerful opportunity to compare the effectiveness of multiple functional groups of pollinators. Although previous experimental studies have shown that vertebrates are important for fruit set in Banksia, it has been challenging to tease out the relative contribution of mammals and birds (Collins and Rebelo, 1987; Goldingay, 2000; Wooller and Wooller, 2001). Nonetheless, Wawrzyczek et al. (2024, 2025) have demonstrated by experimentally manipulating pollinator access that mammal visitation does lead to fruit set in banksias, and predominantly nocturnal floret opening suggests a role for nocturnal mammals for pollination in some banksias (Hopper, 1980).

Here, we test the relative effectiveness of birds, insects and NFMs as pollinators on reproduction, mating and offspring vigour in Banksia menziesii (Proteaceae). A dominant tree species of Banksia Woodlands of southwestern Australia (Ritchie et al., 2021a), B. menziesii has floral traits that are often associated with bird pollination (Fenster et al., 2004; Cronk and Ojeda, 2008; Pauw, 2019). These include red-coloured flowers arranged in large, robust inflorescences that provide a solid perch for foraging animals, large quantities of sucrose-rich nectar, and reproductive structures that are physically distant from nectaries found at the base of floral tubes (Fig. 1). Despite these traits and the frequent visitation by honeyeaters (Lewis and Bell, 1981; Ramsey, 1988), there is no empirical confirmation of adaptation to bird pollination, whereas NFMs can be frequent visitors (Krauss et al., 2018) along with a range of insect species (Lewis and Bell, 1981). Furthermore, introduced honeybees can be an order of magnitude more frequent visitors to the flowers of B. menziesii than birds (Ramsey, 1988). In Banksia (and many other Proteaceae), flowers are protandrous, and fertile pollen is presented around the stigma on the same structure (the pollen presenter), which deposits pollen on the heads of birds and mammals while they feed on basal nectaries (Collins and Rebelo, 1987; Ayre and Whelan, 1989). Pollen is also deposited on insects as they forage for nectar, but less effectively than seen in birds owing to their smaller size (Ramsey, 1988).

Fig. 1.

The main floral visitors to Banksia menziesii inflorescences at Ioppolo Nature Reserve, Western Australia. Clockwise from top left: western spinebill (Acanthorhynchus superciliosus; male pictured); brown honeyeater (Lichmera indistincta); native bee (Leioproctus sp.) and introduced European honeybee (Apis mellifera); and honey possum (Tarsipes rostratus).

By manipulating pollinator access to flowers, we tested the hypothesis that excluding birds will lead to reduction in plant fitness that can include: (1) decreased seed set; (2) reduced seed vigour; (3) lower genetic diversity and rates of outcrossing; and (4) reduced seedling vigour and survival. By combining a pollinator exclusion experiment with an ecological genetic assessment of the consequences for mating, we identify a genetic basis for observed fitness effects that is driven by variation in the effectiveness of different pollinator functional groups. In this way, we address a key determinant of mating system variation, and its fitness consequences, in a region that is not only a global biodiversity hotspot (Myers et al., 2000), but also a hotspot for vertebrate pollination (Hopper and Gioia, 2004).

MATERIALS AND METHODS

Study site

The study was conducted at Ioppolo Nature Reserve (NR), ∼65 km NE of Perth (31°28′54.20″S, 115°57′52.23″E; elevation ∼80 m). Ioppolo NR is a 1200 ha relatively pristine remnant of Banksia Woodland, a Federally listed threatened plant community growing over deep sand on the Swan Coastal Plain (Ritchie et al., 2021a). The study area (Fig. 2) has a canopy co-dominated by Banksia menziesii and Banksia attenuata, with a diverse understorey composed primarily of sclerophyllous shrub species (Keighery et al., 2021). This study builds on our previous research on bird and mammal pollination at the site (Krauss et al., 2018; van der Kroft et al., 2019; Kestel et al., 2021).

Fig. 2.

Pollinator exclosures over Banksia menziesii trees at Ioppolo Nature Reserve, Western Australia. Exclosure treatments were open pollination (OP, no exclosure), insect pollination (IP), honey possum pollination (HP) and complete exclusion of pollinators (CE). Note that there is a 5 cm gap at the base of HP to allow passage by honey possums; inset shows a camera trap photograph of a honey possum on an inflorescence taken from inside the exclosure.

Study species

Banksia menziesii is a long-lived small tree or shrub (typically 2–10 m tall), endemic to the Southwest Australian Floristic Region (George, 1996). It is a protandrous (sequential hermaphroditism) species reliant on pollinators for reproduction and was thought to be self-incompatible (Ramsey, 1988; Ramsey and Vaughton, 1991). Plants have a long flowering season (February–October, with a peak in June), with ∼600–1400 florets arranged within 4- to 12-cm-long inflorescences that open acropetally, at a rate of 30–70 flowers per 24 h and 95 % opening during the daytime (Ramsey, 1988). Like many other Proteaceae (Ayre and Whelan, 1989), fruit-to-flower ratios are very low, with Ramsey (1988) recording a mean of only 5.3 fruits per inflorescence in wild plants. Inflorescences are conspicuous, terminally positioned, pink–red or yellow in colour (George, 1996), and are sufficiently robust to support perching by honeyeaters (Fig. 1A, B) (Krauss et al., 2018). The flowers have a large stigma-to-nectary distance (∼4 cm) and transfer pollen when viable from anthers to a pollen presenter that is situated around the stigma. The pollen presenter and stigma are positioned at the tip of the style, which upon floret opening protrudes beyond the relaxed perianth. Flowers produce large amounts of nectar, with a much higher proportion of sucrose than glucose and fructose, which is important to the diet of nectarivorous birds (Wyk, 1998; Nicholson and Fleming, 2003) and honey possums (Wooller and Wooller, 2013).

The 4-cm-long gently curved styles of B. menziesii suggest that nectarivorous birds (predominantly honeyeaters in the family Meliphagidae) are better suited to pollination than mammals and insects (Ramsey, 1988, 1989; Saffer, 2004). Bird species known to visit the flowers of B. menziesii for nectar are brown honeyeater (Lichmera indistincta), western spinebill (Acanthorhynchus superciliosus), red wattlebird (Anthocaera carunculata), western wattlebird (Anthocaera lunulata), singing honeyeater (Lichenostomus virescens), white-cheeked honeyeater (Phylidonyris nigra), New-Holland honeyeater (Phylidonyris novaehollandiae) (all Melaphagidae), and silvereyes (Zosterops lateralis; Zosteropidae) (Lewis and Bell, 1981; Ramsey, 1988, 1989; Krauss et al., 2018; Fig. 1; Supplementary Data Fig. S2). Other floral visitors include the nocturnal honey possum (Tarsipes rostratus) and several insects, predominantly beetles (Coleoptera: Staphylinidae, Nitidulae and Alleculidae), ants (Hymenoptera: Formicidae), native bees (e.g. the banksia bee, Hylaeus alcyoneus) and the introduced European honeybee (Apis mellifera) (Lewis and Bell, 1981; Ramsey, 1988; Houston, 2000; Krauss et al., 2018; Supplementary Data Figs. S3, S4).

Introduced honeybees tend to be by far the most frequent visitor to B. menziesii flowers (Ramsey, 1988). The vast majority (90 %) of honeybees foraging on B. menziesii do so for nectar and spend on average >4 min per inflorescence visit (and 11 % spend >10 min at an inflorescence), whereas those foraging for pollen spend an average of 37 s at an inflorescence (Ramsey, 1988). In contrast, birds foraging for nectar tend to spend 20–40 s per visit on a B. menziesii inflorescence (Ramsey, 1988), efficiently probe an average of nine flowers per inflorescence visit, visit multiple trees per foraging bout, and are often interrupted by other birds while foraging (Ramsey, 1988, 1989; Krauss et al., 2018; Ritchie et al., 2021b). No comparable data exist for mammals visiting B. menziesii. However, in a study using camera traps, honey possums accounted for 15 % of all recorded visits by vertebrates to the flowers of B. menziesii, all of which were nocturnal, with a peak visitation rate during the hour after sunset (Krauss et al., 2018).

Selective pollinator exclusion experiment

At our study site, B. menziesii grow at low density (∼300 plants in an area 200 m × 200 m, i.e. less than one plant per 100 m², Supplementary Data Fig. S1), with most neighbouring plants being sufficiently distant that there is typically a gap of many metres between plants, and their canopies are not in contact. This enabled us to conduct a pollinator exclusion experiment where our whole-tree exclusion nets did not disturb pollinators foraging nearby or affect the ability of mammals to move between neighbouring plants. A selective pollinator exclusion experiment comprising four treatments was applied to plants prior to flowering in March 2019 and sustained into infructescence maturity in February 2020. Treatments were as follows (Fig. 2):

1. Open pollinated (OP): the plants were marked and left unmanipulated through the flowering season, allowing free access by all potential pollinators.

2. Complete exclusion of pollinators (CE): all pollinators were prevented from accessing flowers by high-density poly-ethylene monofilament insect screen (20 × 10 threads per square centimetre, Advanced Netting, Nedlands, Perth, WA, Australia) supported on a 3.5 m × 3.5 m × 3.5 m metal frame to cover the entire tree and pinned to the ground.

3. Insect pollinated (IP): vertebrate pollinators (birds and NFMs) were prevented from visiting the flowers by an ultraviolet-treated, fine black polyethylene netting (16 mm × 16 mm openings) stretched over the frame (as in CE above) and pinned to the ground. The opening in the mesh would permit only small insects, such as bees, which were observed to move through the mesh unimpeded.

4. Honey possum pollinated (HP): all flying pollinators excluded as in CE (above), but with a 5 cm gap between the ground and the base of the frame to allow scansorial honey possums to access the flowers. It is possible that some flower-visiting insects were able to enter via this 5 cm gap, but they were rarely observed inside the structures during flowering.

Photographs from motion-triggered cameras also confirmed that honey possums visited flowers on the plants in the HP and OP treatments (that we intended to allow them to be able to access; Fig. 1C). No other NFMs have been recorded visiting the flowers of this species at the study site (Krauss et al., 2018).

For each treatment, two to four plants of similar size were arbitrarily selected for this experiment based on their ability to fit within the structures, with similar maturity, abundance of inflorescences and proximity to a similar pool of potential pollen donors (∼100 m maximum distance). We acknowledge that with few maternal plants, our results are potentially subject to confounding effects. However, we were cognisant of the issues and implemented measures to address this issue. For example, maternal plants were randomly assigned to treatments and were widely distributed within the study plot to remove potential bias in genetic outcomes from spatial genetic structure and potential near-neighbour mating. In total, the number of inflorescences:plants per treatment were 31:4 (OP), 32:2 (CE), 48:3 (IP) and 74:3 (HP). Only two plants were included in the CE treatment because it was already known that B. menziesii does not set seed in the absence of pollinators (Ramsey and Vaughton, 1991). With an average of ∼1000 flowers per inflorescence, this study equates to an assessment of ∼185 000 potential pollination events over the course of a 6-month flowering season (April–September). The extended flowering period adds a temporal dimension to our experimental design, which reduces the risk of biased results attributable to pollination treatments being replicated over a small number of plants. Because the phenology of plants varies substantially within a population, the flowering of nearest neighbours to the experimental maternal plants varies through time. Likewise, differences in flowering phenology between plants will foster differences in pollinator foraging routes over the course of the flowering season. The treatments with densest artificial covering, (2) and (4), permitted ultraviolet radiation, minimizing the impact on plant photosynthesis and therefore fruit set.

Reproductive output

Infructescences (cones with or without fruit) were harvested in January 2020, when follicles (fruit) were mature. The number of cones per plant and the number of fruits per cone were recorded. We also noted whether a fruit was predated upon, typically evident in the form of a small round hole in the woody follicle. Predated fruits were included in the count of the total number of fruits but recorded as containing no seed. Cones were oven-dried at 55 °C for 2 weeks to open fruits, and seeds were removed and counted. The source of all seed (plant, cone, fruit) was recorded.

Seed and seedling vigour

To assess seed quality, seeds were weighed and viability was assessed by seed mass, then confirmed via X-ray (Faxitron MX-20 X-ray cabinet, Tucson, AZ, USA). Seeds containing a solid white endosperm filling the entire area within the seed coat were scored as viable, whereas those that lacked endosperm or where endosperm was not fully formed were scored as not viable. In May 2020, all viable seeds were sown into labelled forestry tubes filled with three parts native potting mix to one part coarse white sand, then kept moist in a glasshouse. At 12 weeks, the seedlings were transferred to an outdoor nursery under reticulation. Labelled tubes were distributed randomly in trays and rotated weekly to ensure even light distribution. Fifteen months after germination, individually tagged seedlings were randomly planted into an area of ∼2 ha adjacent to source maternal plants in Ioppolo NR in August (winter) 2021. Prior to planting, all seedlings were measured for stem diameter and height. Seedling survival was recorded in early summer (9 December 2021), which is a critical time for seedling establishment because soils dry out after the last of the winter–spring rains in this mediterranean-type climate. Only 14 mm of rain fell (over 2 days) in November of the year the seedlings were planted at the study site, and there was no rain at all in December–March (Western Australian Bureau of Meteorology). Seedlings were assessed again in mid-summer (8 February 2022), at which time all were deceased owing to the extreme summer drought.

Genetic diversity and mating system

Genomic DNA was extracted from fresh leaves of maternal plants and from a fresh cotyledon of all seedlings after emergence of the first true leaves. We used the method of Carlson et al. (1991) with modifications as outlined by Anthony et al. (2016). All individuals sampled were genotyped at eight microsatellite loci previously developed for B. menziesii and B. attenuata (He et al., 2007; Ritchie et al., 2019; Table 1; Supplementary Data Tables S6, S7). Alleles were scored using Geneious v.7.1.4 (https://www.geneious.com). Software-assigned scores were checked manually and adjusted with binning as required, based on inheritance and family arrays. Multiple PCR repeats of the same individuals were routinely run to check and eliminate scoring error from possible PCR effects. Null alleles were identified during scoring of individual family arrays, and maternal alleles present in offspring were identified. Where apparently homozygous maternal single-locus genotypes were identified as null heterozygotes based on offspring genotypes, offspring in those families that expressed a non-maternal allele as an apparent homozygote genotype were adjusted to null heterozygote, and the null allele was coded as a dummy value (500). Known null alleles occurred in two families (OP4 and IP3) at the same locus (Bm A1).

Table 1.

Genetic diversity and mating system parameters for eight maternal families of Banksia menziesii across pollinator-access treatments.

	Birds permitted				Birds excluded				Group means (s.e.)		U-test
Maternal	OP1	OP2	OP2	OP4	IP1	IP2	IP3	HP3	BP	BE	P-value
n offspring	14	12	46	45	16	46	46	45	117	153
N a	4.13	3.75	5.63	5.5	3.88	5.63	4.38	5.63	4.75 (0.48)	4.88 (0.44)	0.28
N e	1.94	2.61	2.32	2.22	2.04	2.24	1.77	1.97	2.27 (0.14)	2.00 (0.10)	0.12
H E	0.39	0.51	0.53	0.48	0.42	0.45	0.32	0.41	0.48 (0.03)	0.40 (0.03)	0.07
H O	0.48	0.54	0.6	0.54	0.46	0.51	0.3	0.43	0.54 (0.02)	0.42 (0.04)	0.02
I	0.79	0.95	1.02	0.95	0.78	0.91	0.64	0.81	0.92 (0.05)	0.79 (0.06)	0.04
t m	0.99	0.92	1.00	0.94	0.60	0.88	0.45	0.65	0.95 (0.02)	0.66 (0.12)	0.01
t m –t s	0.09	0.01	0.09	0.05	0.00	0.02	0.07	0.02	0.06 (0.02)	0.03 (0.02)	0.12
r p	0.10	0.10	0.11	0.08	0.10	0.03	0.05	0.08	0.10 (0.01)	0.07 (0.02)	0.05
h (pollen pool)	0.50	0.55	0.64	0.59	0.49	0.54	0.42	0.51	0.57 (0.03)	0.49 (0.03)	0.04
dt m	1.00	0.92	1.00	0.93	0.56	0.87	0.45	0.69	0.96 (0.02)	0.64 (0.11)	0.01

Abbreviations: HP1, honey possum pollinated; IP1–IP3, insect pollinated; OP1–OP4, open pollinated (access by birds, insects and honey possums). Shown are the number of offspring per family (n), mean number of alleles (N_a), effective number of alleles (N_e), expected heterozygosity (H_E), observed heterozygosity (H_O), Shannon's information index (I), multi-locus outcrossing rate (t_m), bi-parental inbreeding (t_m–t_s), correlated paternity (r_p), haplotype diversity of the pollen pool (h) and the proportion of seedlings that were detectable outcrosses (i.e. they expressed at least one non-maternal allele) (dt_m). Maternal trees within treatments are categorized by birds being permitted (BP) or being excluded (BE), with means (±s.e.) reported for these two groups. Note that HP1 and HP2 were not included owing to the low number of viable seeds (1 and 4, respectively), and IP2 and IP3 sample sizes were capped at 46 to minimize sample size variation for these analyses. Significant differences between group means are indicated by Mann–Whitney U-test P-values in bold.

Parameters of genetic variation were estimated with Genalex v.6.51b2 (Peakall and Smouse, 2012) for each family. The mating system parameters multi-locus outcrossing rate (t_m), bi-parental inbreeding (t_m–t_s) and correlation of outcrossed paternity (r_p) were estimated with MLTR (Ritland, 2002) for each family. The correlated paternity measure (r_p) is the probability that two siblings are outcrossed full sibs (Ritland, 2002). Values range from zero to one, where one indicates that all outcrossed offspring share the same sire and are full sibs and zero indicates no shared sires. MLTR parameters were set at: standard error based on 1000 bootstraps; individuals within families resampled; expectation-maximization method; pollen = ovule gene frequency; initial value of outcrossing (t) all set at one and all other parameters at 0.1 (Ritland, 2002). The genetic diversity of pollen haplotypes (h) was estimated for each family, as outlined by Torres-Vanegas et al. (2021). This measure corresponds to the probability that the paternal alleles of two randomly chosen offspring from one maternal are different and is consequently an estimate of the effective diversity of sires for all offspring (selfed and outcrossed). The proportion of detectable outcrosses per maternal family was also recorded based on the presence of at least one non-maternal allele in each offspring. The number of non-maternal alleles expressed by each offspring across the eight loci reflects the genetic dissimilarity among its parents. Given sufficient power in the marker polymorphism, zero non-maternal alleles reflects self-pollination, and six non-maternals (the maximum recorded here) reflects outcrossing among the least-related parents. To infer the role of birds in pollination, we then tested for significant differences in mating system parameters between plants that had birds excluded (BE = IP and HP treatments combined; n = 4 maternal plants) and those for which birds had access (OP; n = 4 maternal plants).

Statistical hypothesis testing

Statistical tests of reproductive output were performed in R v.4.2 using generalized linear mixed models (GLMMs) as implemented in the R package ‘glmmTMB’ (Brooks et al., 2017), specifying maternal plant identity as a random effect in all models to account for nestedness of the data (sampling multiple inflorescences from each maternal plant). We fitted models appropriate for the type of data: a negative binomial distribution for the number of fruits per cone (nbinom2 with ‘log’ link function), a beta binomial distribution for the proportion of viable seeds per cone and mean number of viable seeds per fruit (with ‘logit’ link function adjusting the values to fit within the zero–one interval by multiplying by [(n − 1) + 0.5/n] and, in the latter case, dividing by two), a Gaussian distribution for seed mass, a Poisson distribution for the number of non-maternal alleles (with ‘log’ link function) and a binomial distribution for seedling survival (Smithson and Verkuilen, 2006; Zuur et al., 2009). For all models, we used the likelihood ratio test to determine the significance of the fixed effect using the anova function of glmmTMB. For significant models, we used the R package ‘emmeans’ (Lenth et al., 2023) to test for pairwise differences among treatments. We used a Mann–Whitney U-test to assess the statistical significance of differences between genetic diversity and mating system parameter estimates for bird-permitted and bird-excluded treatments. We used linear regression to test for an association between the number of non-maternal alleles and seedling size at the time of planting, and logistic regression to test for association between the number of non-maternal alleles and seedling survival at the beginning of summer. We used a χ² test of association between seedling number and number of non-maternal markers expressed per seedling for each pollinator treatment, and we present residuals to show where departures from expectations were found. We used the measures of reproductive output and seed quality to quantify the cumulative effect of different pollinators on reproduction by the conversion rate of inflorescences to a surviving seedling for an ultimate measure of pollinator effectiveness for early seedling establishment.

RESULTS

Reproductive output

The two plants for which all floral visitors were excluded produced no fruit from 13 inflorescences and were excluded from further analyses. For the remaining treatments, the number of cones bearing fruit per plant varied from 4 to 22, depending on the number of inflorescences per plant and exclusion treatment. The OP treatment plants had fruit on all cones (100 %), while the IP and HP treatments had fruit on 93.5 and 38.8 % of cones, respectively.

The percentage of fruit that were predated was 19 % for OP (25 of 131 fruits predated), 15 % for IP (46 of 306 fruits predated), and 6 % for HP (5 of 88 fruits predated). Predated fruit contained no seeds or did not open on burning or crumbled when pliers were used to open it. Although absolute numbers of seeds set were impacted by predation, it was not a confounding factor in assessing the impact of open- (OP) versus insect-pollinator (IP) treatments on subsequent measures of reproduction, because both treatments were affected similarly. Predation was lower for fruit in the HP treatment, inflating relative estimates of seeds set per cone compared with OP and IP cones.

There was very strong evidence that fruit set (number of predated and unpredated follicles per cone combined) varied among pollination treatments (likelihood ratio test, χ² = 14.3, d.f. = 2, P < 0.001). Although plants pollinated by insects tended to produce more follicles per cone than OP plants, the difference was not significant [mean IP = 7.60 ± 0.56, n = 48(4) cones (plants), compared with OP = 4.61 ± 0.52, n = 31(3); P = 0.38]. However, both IP and OP treatments were greater than HP plants [mean HP = 1.15 ± 0.25, n = 74(3); P < 0.001; Fig. 3; Supplementary Data Table S8, Fig. S9.

Fig. 3.

Reproductive output for ten Banksia menziesii maternal plants from three pollinator-access treatments: open pollinated (OP, yellow); insect pollinated (IP, i.e. birds and mammals excluded, green); and honey possum pollinated (HP, i.e. birds and insects excluded, blue). Reproductive output is described by the number of fruits per cone (A); percentage of all seeds that were viable per cone (B), number of viable seeds per fruit per treatment (C); and viable seed mass (D).

There was very strong evidence that the proportion of viable seeds per cone varied among pollination treatments (likelihood ratio test, χ² = 16.9, d.f. = 2, P < 0.001). Viable seeds per cone was highest in the OP treatment, lower in the IP treatment, and lowest in the HP treatment (OP, 75 ± 4.0 % > IP, 51 ± 1.8 % > HP, 36 ± 4.7 %; Fig. 3; Supplementary Data Table S8, Fig. S9.

There was strong evidence that the mean number (at the inflorescence level) of viable seeds per (non-predated) fruit differed among the three treatments (likelihood ratio test, χ² = 12.0, d.f. = 2, P = 0.002) and was highest for OP inflorescences and lowest for the HP inflorescences (OP, 1.15 ± 0.088, n = 31 > IP, 0.80 ± 0.053, n = 44 = HP, 0.55 ± 0.092, n = 32; Fig. 3; Supplementary Data Table S8, Fig. S9.

Seed mass also varied significantly among pollination treatments (likelihood ratio test, χ² = 16.9, d.f. = 2, P < 0.001). Although there was no evidence that the seed mass differed between OP and IP flowers (P = 0.73), there was strong evidence that it was lower for the HP flowers (OP, 89.3 ± 0.9, n = 165 = IP, 90.2 ± 0.9, n = 296 > HP, 67.6 ± 1.5, n = 53; P < 0.01; Fig. 3; Supplementary Data Table S8, Fig. S9.

Genetic diversity and mating system

Genetic diversity and mating system parameters varied among families, and most parameters were higher when birds were allowed access to flowers than when they were excluded (Table 1). When birds were excluded from the flowers, heterozygosity (H_o = 0.42) and Shannon's information index (I = 0.79) in offspring were significantly (P < 0.05) lower than that of open-pollinated flowers, where visits by birds were permitted (H_O = 0.54, I = 0.92). When birds were excluded from pollinating flowers, outcrossing rate (t_m = 0.66), pollen pool diversity (h = 0.49) and the proportion of offspring with detectable outcrosses (dt_m = 0.64) were all significantly (P < 0.05) lower than that of open-pollinated flowers that permitted birds (t_m = 0.95, h = 0.57, dt_m = 0.96). Bi-parental inbreeding and correlated paternity were largely equivalent between pollinator access treatments where birds were either permitted or excluded (Table 1).

The frequency of non-maternal alleles expressed in offspring from eight microsatellite loci showed an approximately normal distribution around a median of three, and a maximum of six for OP offspring. GLMM indicated that the differences in the number of non-maternal alleles were marginally non-significant (χ² = 5.26, d.f. = 2, P = 0.072), with moderate evidence for differences between the seedlings resulting from OP and HP flowers (P = 0.024). However, for treatments where birds were excluded, there were many more offspring with zero non-maternals than expected based on the frequency distribution for open pollination (Fig. 4). The range of values for the number of non-maternal alleles reflects the relative genetic similarity among the parents of each offspring. The distribution suggests near zero probability of zero non-maternal alleles for outcrossed offspring, meaning that the high number of offspring with zero non-maternal alleles are almost certainly the product of animal-mediated self-pollination, because all outcrossed offspring are expected to express at least one non-maternal allele (see Supplementary Data S10). In support of this interpretation, the proportion of these offspring with zero non-maternal alleles (4 % for OP, 37 % for IP and 33 % for HP) is equivalent to estimates of the selfing rate from MLTR (Table 1).

Fig. 4.

Mean (±s.e.) of the relative proportion of the number of non-maternal alleles expressed by seedlings of Banksia menziesii from three pollinator-access treatments: open pollinated (OP, yellow); insect pollinated (IP, i.e. birds and mammals excluded, green); and honey possum pollinated (HP, i.e. birds and insects excluded, blue). Note the normal distribution around a peak of three non-maternal alleles, and that when birds were excluded, the mean proportion of seedlings with zero non-maternal alleles (0.36) was higher than for open pollination that allowed birds (0.04). See also Table 3.

The distribution of the number of non-maternal alleles expressed by offspring across pollinator access treatments showed a significant departure from assumptions of no effect (χ² = 45.5, d.f. = 12, P < 0.001; Table 2; Fig. 4). In particular, there were far fewer than expected offspring with zero non-maternals (selfs) among the offspring from the OP treatment [4 of 96 (4 %) seedlings] and far greater than expected offspring from the HP and IP treatments [HP, 14 of 38 (37 %); IP, 43 of 149 (29 %) seedlings]. Thus, selfing was almost exclusively associated with pollination for the treatments where birds were excluded. There was then largely a significant deficit of individuals with one or more non-maternal alleles for the HP treatment, reflecting relatively less outcrossing in comparison to OP and IP (Table 2).

Table 2.

Residuals from χ² analysis of seedling number for each pollinator treatment and number of non-maternal markers expressed per seedling.

		Pollinator treatment
Number of non-maternals	OP	IP	HP
0	13.5*	3.7	4.1
1	10.8	5.4*	0.4*
2	0.2	0.0	0.5*
3	1.0	1.5*	0.6
4	0.0	0.3	1.1*
5	0.0	0.2	0.9*
6	0.5	0.0	0.8*

Abbreviations: HP, honey possum pollinated; IP, insect pollinated; OP, open pollinated. Values indicate the extent of departure from expectation; *Indicates fewer than expected under an assumption of no effect, all other values indicate more than expected. Largest values are highlighted (in bold) to emphasize the relative importance for overall χ² test statistic (χ² = 45.5, d.f. = 12, P < 0.001). Note far greater observed than expected offspring with zero non-maternals (selfs) for HP (N_observed = 14:N_expected = 8) and IP (43:32), and far fewer observed than expected offspring with zero non-maternals (selfs) for OP (4:21), and greater observed than expected offspring with one non-maternal for OP (17:8), and fewer than expected for IP (4:12).

Seedling vigour

There was strong evidence that seedling growth was positively associated with outbreeding, as quantified by the number of non-maternal alleles for individual seedlings (height R² = 0.122, P < 0.001; stem diameter R² = 0.084, P < 0.001, n = 283 seedlings). There was also strong evidence that seedling survival was associated with the number of non-maternal alleles for an individual seedling (likelihood ratio test, χ² = 13.0, d.f. = 1, P < 0.001, n = 283 seedlings; Fig. 5). Higher than expected survival of seedlings with one non-maternal [10 of 23 (43 %) seedlings; observed = 43 %, expected = 10 %] was a notable outlier of otherwise strong linear associations. Seedling survival to early summer ranged from 6.5 % for individuals with zero non-maternal alleles (selfs) to 50 % for individuals with six non-maternal alleles. Overall, seedling survival for detectable outcrosses (one or more non-maternal alleles) was 29.7 %. The very high mortality for seedlings that were the product of self-pollination (94 %) was almost exclusively [57 of 61 (93 %) seedlings] associated with IP and HP pollination treatments (i.e. when birds are excluded; Fig. 5). Overall seedling survival rates were equivalent (GLMM, P = 0.42) for the seedlings resulting from OP (30.2 %; n = 96) and IP (27.5 %; n = 149) pollination treatments, both of which were higher (GLMM, P = 0.012) than the survival rate of seedlings resulting from HP (7.9 %; n = 38).

Fig. 5.

Significant positive association (R² = 0.463, P < 0.001) between the proportion of Banksia menziesii seedlings that survived in a summer field trial (y-axis) and the number of non-maternal alleles expressed per seedling (x-axis), for each pollinator treatment (coloured bars). Here, zero non-maternal alleles is inferred to indicate a selfed seedling, and one to six non-maternal alleles reflects increasing heterozygosity of outcrossed seedlings from increasing degree of genetic dissimilarity of parents. Note that survival of selfed seedlings (0.06) was lower than the mean survival of outcrossed seedlings (0.3). Abbreviations: HP, honey possum pollinated; IP, insect pollinated; OP, open pollinated.

Overall conversion rates of inflorescences to a surviving seedling in early summer were ultimately very low, but similar for open-pollinated flowers [1.65 seedlings per inflorescence (comprising ∼1000 flowers)] and insect-pollinated flowers (1.67 seedlings per inflorescence). The conversion rate for honey possum-pollinated flowers was negligible (0.06 seedlings per inflorescence) and an order of magnitude less than that of open-pollinated flowers (Table 3).

Table 3.

Cumulative consequence of pollinator treatments on seed and seedling vigour of Banksia menziesii.

		Pollinator treatment
Variable	OP	IP	HP
N (inflorescences to cones)	31	48	74
Mean number of fruits per cone	4.6	7.6	1.1
Percentage of all seeds viable	74	51	35
Mean number of viable seeds per fruit	1.2	0.8	0.6
Number of viable seeds per cone	5.5	6.1	0.7
Percentage seedling survival in field trial	30.2	27.5	7.9
Number of viable seeds that survived as seedlings per cone	1.65	1.67	0.06
Conversion rate: flowers to seedling1	0.0016	0.0017	0.00001

Note that each cone bears fruit called follicles, which contain one or two seeds. Abbreviations: HP, honey possum pollinated; IP, insect pollinated; OP, open pollinated.

¹Conversion rate based on an estimated mean of 1000 flowers per inflorescence.

DISCUSSION

All pollinator functional groups contributed to pollination of Banksia menziesii, but with varying levels of effectiveness and efficiency. Given that B. menziesii possesses several traits suggestive of bird pollination, our results highlight a pollination system that is more generalized than might be predicted from floral traits. We provide the first direct evidence of the role of honey possums as effective pollinators of B. menziesii, albeit less effective than birds or insects (in this case, primarily introduced A. mellifera). However, when birds were excluded from flowers that were visited by either insects or honey possums, post-fertilization seed abortion more than doubled, most measures of genetic diversity in offspring decreased, the percentage of seedlings inferred to be the product of self-pollination increased from 4 to 29 %, paternal diversity in offspring decreased, and seedling survival was much lower for selfed compared to outcrossed offspring. This is a crucial new finding that establishes higher offspring fitness resulting from outcrossing as a mechanism underpinning variation in pollinator effectiveness. We were also able to demonstrate, for the first time, that B. menziesii is self-compatible, in contrast to a previous study suggesting otherwise based on hand-pollination with self-pollen (Ramsey and Vaughton, 1991).

Although self-compatible, we have shown that B. menziesii is preferentially outcrossing. Crucially, from our bird-exclusion treatment, we infer that nectar-feeding birds are more effective than insects or mammals in delivering the preferred (outcross) pollen. Ideally, we would implement a bird-only treatment to confirm this conclusion, but such an experiment was not feasible in our system. Despite this limitation, our results allow us to suggest why these birds are more effective pollinators of B. menziesii than insects or mammals. Firstly, their foraging behaviour promotes outcross pollination over self-pollination. Secondly, selfing is harmful in comparison to outbreeding because it results in higher offspring mortality. Deposition of predominantly self-pollen is a consequence of pollinator foraging behaviour that results in more movement among the flowers of an inflorescence and/or plant and fewer movements among plants (Ramsey, 1988, 1989; Paton, 1993; Richardson et al., 2000; Whelan et al., 2009). Behaviours by honeyeaters that are likely to promote deposition of a diversity of cross pollen onto flowers include sourcing nectar from multiple plants within their territory and/or home range, high mobility, little grooming, and intra- and interspecies aggression disrupting movements between near-neighbour plants (Pyke, 1980; Ramsey, 1989; Krauss et al., 2017, 2018; Wessinger, 2021). Strikingly, the declines in outcross pollination we found to be associated with insect and mammal pollination were relative to an open pollination treatment that included birds, insects and mammals. Given that in B. menziesii honeybee visitation can be an order of magnitude more frequent than that of birds (Ramsey, 1988), the extent of decline is thus likely to be grossly underestimated compared with a bird-pollination-only treatment. Treatments that somehow enable access by birds but not by insects and other pollinators are, however, notoriously difficult to implement.

From an evolutionary perspective, the associated fitness benefits arising from near complete outcrossing and higher paternal diversity provide a mechanism that might favour the evolution of pollination by birds (Stiles, 1978; Cronk and Ojeda, 2008; Hopper, 2009; Pauw, 2019). This mechanism might also apply to lineages with bird-pollinated plants on other continents, given that many of the main pollinating birds in the Americas and Africa/Asia exhibit some behaviours similar to Australian honeyeater species (Pyke, 1980). For example, many sunbirds and hummingbirds also engage in interspecific conflict over food resources (Murray, 1981). However, there is also evidence for behavioural differences among bird species, which would be likely to lead to different consequences for plant mating (Krauss et al., 2017). For example, some hummingbirds (particularly in the Phaethornithinae) display traplining foraging rather than holding territories (Stiles, 1975; Leimberger et al., 2022), which might lead to high outcrossing and high pollen dispersal distances. Alternatively, lorikeets feed intensively on particular trees, which might lead to low outcrossing rates, but also move between distant flowering trees, potentially leading to some long-distance pollen dispersal (Southerton et al., 2004). These examples highlight the importance of further testing of how genetic consequences vary across different bird-pollinated systems. Nonetheless, fitness benefits for plant offspring associated with pollination by nectar-feeding birds is likely to have played a crucial role in the evolution of diverse communities of bird-pollinated plants, including some habitats in the temperate Southern Hemisphere (Armstrong, 1979; Johnson, 2010) and tropics (Stiles, 1978).

Our results also highlight conservation concerns associated with introduced honeybees reducing the fitness of plants adapted to bird pollination (Diller et al., 2022). These concerns stem from increased self-pollination in plant species that have evolved highly heterozygous genotypes through a predominantly outcrossing mating system. Such species have a high genetic load and are therefore susceptible to inbreeding depression with increased selfing (James, 2000). For B. menziesii, introduced honeybees, but also native honey possums, are the key drivers of reduced seed quality that is associated with increased selfing. This finding is likely to apply to many other Banksia, given that in self-compatible Banksia inbreeding depression is pronounced (Vaughton and Carthew, 1993; Vaughton and Ramsey, 1997, 2006; Heliyanto et al., 2005, 2006).

From an overall plant fitness perspective, an extended flowering period and a very high abundance of honeybees can dilute the individual pollinator contributions that stem from differences in pollinator behaviours and traits. Previous studies have yielded contrasting results when testing whether insects lead to reduced seed quality in comparison to birds for plants seemingly adapted for bird pollination (England et al., 2001; Steenhuisen et al., 2012; Gilpin et al., 2016; Hung et al., 2018; Duffy et al., 2020; Page et al., 2021; Diller et al., 2022). These contrasting results highlight that impacts of honeybees will vary among plant species, even among those that are bird pollinated. Our results suggest that when highly abundant, introduced honeybees can be effective pollinators for B. menziesii, despite reduced fitness associated with increased self-pollination. This generalized pollination and mating system provides reproductive assurance, especially when more effective bird pollinators might not be available (e.g. when populations might be too small to sustain bird pollinators). A capacity to be pollinated by a range of vectors might also contribute resilience in a rapidly changing world (Suijkerbuijk et al., 2025).

Although we confirmed experimentally that honey possums contribute to pollination of B. menziesii, they were surprisingly ineffective pollinators. Dramatically lower seed set might have been attributable, in part, to lower rates of floral visitation compared with the inflorescences in OP and IP treatments (Krauss et al., 2018). Low visitation might arise from a low abundance of honey possums at our study site or because our exclosures deterred visitation and/or natural foraging behaviour. Alternatively, the exclosures could also encourage foraging by providing protective coverage from predators. Nocturnal activity could also reduce their contribution to pollination of B. menziesii, because almost all flowers open in the morning (Clifford and Sedgley, 1993), meaning that birds and insects have the earliest access to pollen and, consequently, pollinate the majority of flowers. This does not apply to our study, because birds and insects were excluded by our HP exclosures.

The negative consequences of pollination by honey possums for seed quality were pronounced, resulting from a high incidence of self-pollen deposition onto stigmata. Movement distances of honey possums vary depending on the availability of nectar sources, with shorter distances moved in nectar-rich heathlands (Garavanta et al., 2000; Bradshaw and Bradshaw, 2002), although males have been recorded to travel as much as 500 m between feeding and refuge sites (Bradshaw et al., 2007). However, as a scansorial mammal, once in the feeding area, movement between flowers on a single inflorescence, or inflorescences on the same plant, appears to predominate (Garavanta et al., 2000; Wooller and Wooller, 2013). If the same applies in our study site, then our estimates of paternal diversity (h = 0.51) for honey possum-pollinated seedlings suggests either a diversity of outcross pollen via visits by multiple individuals and/or changes in individual foraging behaviour during the many months of flowering of a B. menziesii plant, or that different individuals have different foraging paths. Although honey possums were less effective pollinators of B. menziesii, it will be of interest to test whether they are comparatively more effective pollinators in geoflorous species of Banksia. Several of these species have rigid, dull-coloured flowers with strong odours, which are traits suggestive of adaptation to mammal pollination (Wawrzyczek et al., 2024, 2025).

Surprisingly, in B. menziesii, the estimates per tree of correlated paternity (r_p), an inverse measure of paternal diversity in outcrossed offspring, were similar between open and bird-excluded treatments. This contrasts with the pollen pool diversity estimate (h), where a significant difference was found between bird-included and bird-excluded treatments, reflecting different levels of selfing. Equivalent correlated paternity (r_p) suggests that there is largely an equivalent level of paternal diversity for outcrossed seed regardless of whether birds did or did not have access to flowers. This is also reflected in the seedling survival data, where values were largely equivalent between OP and IP treatments for all but selfed seedlings. Although the insect-only treatment experienced more frequent self-pollination, equivalent paternal diversity of outcrossed offspring between pollinator treatments suggests that bees and potentially other insects ultimately transfer pollen among plants in a way that is comparable to birds in terms of resulting paternal diversity. This is likely to be, at least in part, a consequence of the typically long period of flowering for B. menziesii, which for an individual tree can extend to ≥6 months and tens of thousands of flowers. However, equivalent paternal diversity values are also likely to reflect the very high abundance of honeybees on the OP treatment that might overwhelm the expected contribution of more diverse pollen loads from birds alone. Despite this, our estimates of the mean effective number of sires (N_ep = 1/r_p), a measure of paternal diversity, ranged from 9 to 33 across all our maternal plants, which are amongst the highest recorded and are equivalent to other bird-pollinated banksias (Krauss et al., 2017).

In conclusion, our results suggest that for a preferentially outcrossing plant species where birds, insects and mammals all contribute to pollination, birds best deliver the outcross pollen that these plants prefer. Our findings emphasize the importance of interpreting pollination results in an ecological genetic context, which includes an understanding of how multiple paternity and the genetic composition of pollen loads deposited on stigmas affect offspring quality (Karron et al., 2009; Barrett and Harder, 2017; Rhodes et al., 2017; Valverde et al., 2019; Hung et al., 2023; Travis and Kohn, 2023). By comparing fitness for the offspring arising from different pollinator functional groups, we have demonstrated that variation in pollinator effectiveness extends beyond pollen deposition and fruit set mechanisms to include also the genetic quality of pollen deposited (Ne’eman et al., 2010; Valverde et al., 2019; Wessinger, 2021).

SUPPLEMENTARY DATA

Supplementary data are available at Annals of Botany online and consist of the following.

S1: map showing location of Banksia menziesii plants in the study site. S2: microsatellite information for Banksia menziesii. S3: microsatellite data for Banksia menziesii families. S4: justification of zero non-maternal alleles as evidence of selfing in offspring of Banksia menziesii. S5: GLMM statistics summary. S6: summary boxplots for pollination treatments on Banksia menziesii. S7: video of bees on Banksia menziesii inflorescence. S8: video of honey possum on Banksia menziesii inflorescence. S9: video of red wattlebird on Banksia menziesii inflorescence. S10: photo montage for journal cover. S11: text for cover image.

FUNDING

This work was supported by the Australia & Pacific Science Foundation (APSF20049), the Wettenhall Environmental Trust, Friends of Kings Park, Edith Cowan University and The University of Notre Dame Australia.

AUTHOR CONTRIBUTIONS

L.M.A., D.K., R.D.P., E.v.E. and S.L.K. designed the study; L.M.A. and S.L.K. executed the fieldwork; L.M.A. executed the laboratory work; L.M.A., S.W. and S.L.K. analysed the data; L.M.A. and S.L.K. led the writing, with contributions from all authors.