Mitigation of Pollen Limitation in the Lowbush Blueberry Agroecosystem: Effect of Augmenting Natural Pollinators

blueberry agroecosystem: effect of augmenting natural pollinators. Ecosphere 6(11):235. Abstract. Growers of small fruit crops often supplement the natural pollinator community by introducing pollinators into commercial orchards and fields, but there are relatively few studies that test the extent to which such interventions increase fruit yield. To test whether plants are limited by pollen availability, inflorescences in 78 commercial lowbush blueberry fields during three years were hand-pollinated either with supplemental outcross pollen, or marked and left as controls (open-pollination). Maximum fruit set with supplemental pollination was in the range 50–65%, whereas with open-pollination, it was slightly, but significantly lower, in the range of 47–57%, suggesting that pollen limitation can affect fruit set. In a two-year experiment, we augmented native pollinators with introduced honey bees, bumble bees, and leaf cutter bees and all combinations in 48 fields. Stigmatic pollen loads were influenced by the total numbers of managed bees in a field in 2010 but not 2011. The presence of leaf cutter bees had a small but positive effect on seed set in 2010, and honey bees had a small but negative influence on seed set in 2011. There was a strong correlation between fruit set reproductive output of supplementally pollinated and open-pollinated plants, suggesting that plant health or plant resources influence reproductive success. Temperature variation among regional groupings of fields was minimal and is unlikely to have accounted for differences among fields in fruit set. We propose several reasons why, despite pollen-limitation of fruit and seed set of blueberry plants, augmentation of the pollinator community of lowbush blueberry did not significantly increase reproduction. These include pollinator-mediated transfer of self-pollen followed by subsequent fruit abortion due to inbreeding depression and resource limitation for fruit maturation. Management practices that focus on increasing outcross pollen receipt, or plant resources for fruit set may have greater overall benefit than pollinator augmentation alone. Copyright: Ó 2015 Fulton et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.


INTRODUCTION
Fruit set per plant rarely attains levels near 100% (Stephenson 1981, Sutherland 1986, 1987, Charlesworth 1989).This may be an evolved feature of plants, selected to increase male fitness through pollinator attraction and pollen dispersal, or to increase female fitness through selective abortion of low quality fruit, or as a ''bethedging'' strategy in an environment where the availability of pollinators or resources fluctuate (Stephenson 1981, Sutherland and Delph 1984, Stanton et al. 1986, Sutherland 1986, 1987, Burd 1995, 1998, 2008, Ashman et al. 2004).Fruit production can be costly (Snow andWhigham 1989, Obeso 2002) and may be limited by resource availability (Bateman 1948, Stephenson 1981).If resources limit the production of fruit, the addition of extra pollen to stigmas would not be expected to influence fruit set (Ashman et al. 2004).Although this is true in some cases, there are many instances where the supplemental addition of pollen does increase fruit and seed set (Larson and Barrett 2000, Ashman et al. 2004, Knight et al. 2005).
Pollen limitation can arise because of inadequate deposition of pollen, deposition of low quality pollen (e.g., self-pollen), or both (Aizen and Harder 2007).Insufficient receipt of pollen is often associated with features of the available pollinator community.For example, Morandin and Winston (2005) found less pollen limitation in canola fields that contained higher abundances of wild bees.Other studies have also shown that pollination can depend on abundance or diversity of pollinators (e.g., Klein et al. 2003, Hoehn et al. 2008).
While pollen limitation is common (Burd 1994, Larson and Barrett 2000, Ashman et al. 2004, Knight et al. 2005), it can vary among regions and years for plants of the same species (Burd 1994).The environmental factors associated with pollen limitation (e.g., resource availability, pollinator community) are known, but differences in the magnitude of pollen limitation are difficult to predict (Cosacov et al. 2008).Furthermore, the relative influence of pollen quantity and pollen quality remains largely unknown, as these two distinct factors are confounded in most hand pollination experiments (Aizen and Harder 2007).The quality and quantity of pollen deposited on stigmas is largely dictated by the foraging behaviors of pollinators and by the degree of relatedness of nearby plants and clone size.
In many systems, effective pollination, leading to high seed or fruit production, is often correlated with pollinator abundance (Jennersten and Nilsson 1993, Klein et al. 2003, Morandin and Winston 2005, Hoehn et al. 2008).An abundance of pollinators alone, however, may not guarantee pollination if pollinators are not adapted to the particular floral design.Specialist bees that are able to manipulate floral morphology efficiently (Strickler 1979, Laverty and Plowright 1988, Utelli and Roy 2000) may deposit a large amount of pollen relative to the amount of pollen removed (see Harder and Thomson 1989).In principle, there should be selection for pollinators that deposit large numbers of outcross pollen grains on stigmas.But in outcrossing plant species, especially those with large floral displays, self-pollen is also deposited and this can reduce reproductive success through inbreeding depression (early fruit abortion), stigma clogging and pollen discounting (Aizen and Harder 2007).Patterns of pollinator movement within and between individual plants are likely to influence not only the amount of pollen deposited, but also the relative proportions of self or outcross pollen transferred.Many pollinators forage on multiple flowers within an inflorescence and travel short distances between plants (e.g., Levin and Kerster 1974, Price and Waser 1979, Waddington 1979, 1980, 1981, Waser 1982).These movement patterns could potentially increase rates of self-pollination through geitonogamy, resulting in fruit abortion and lower yield (Eckert 2000, Charpentier 2001, Barrett 2002, 2003).
Lowbush blueberry (Vaccinium angustifolium) is an important small fruit crop that has the capacity to set fruit on all flowers produced, but fruit set rarely surpasses 40-50% with native pollinators in agricultural fields (Wood 1968, Hepler andYarborough 1991).A key issue identified by the blueberry industry and researchers alike is the variability in fruit set seen in fields (Hepler andYarborough 1991, Bell et al. 2010).Lowbush blueberry yield has been shown to range from 300 to 17,000 kg/ha (Hepler and Yarborough 1991).Stigmatic pollen-loading has been associated with correlates of blueberry yield such as fruit set, seed number and fruit weight (Aalders andHall 1961, Dogterom andWinston 1999).
Determining whether lowbush blueberry yield can be increased through additional pollination (e.g., as could be achieved by the introduction of pollinators into blueberry fields) is critical to pollination management for producers and growers of this fruit crop.However, the large clone size of lowbush blueberry (commonly 2-10 m in diameter; Bell et al. 2009, Bobiwash et al. 2013) and the simultaneous display of thousands flowers on a clone increases the likelihood of selfpollen transfer by pollinators that forage on flowers within a clone (Eckert 2000, Charpentier 2001, Barrett 2002, 2003), and so it is unclear whether augmenting numbers of native pollinators by introducing non-native pollinators, as commonly practiced by growers, is sufficient to increase yields.
In lowbush blueberry, the ideal pollinator community would likely consist of species that can extract pollen from the poricidal anthers.Given the high level of inbreeding depression observed in this species (Bell et al. 2010, Bobiwash et al. 2013), the ideal pollinators would be those that bring about outcrossing.Lowbush blueberry has evolved alongside a variety of native bee genera (Andrena, Bombus, Halictus, Agapostomon, Augochlora, Augochlorella and Lasioglossum) that are adapted to inclement weather conditions typically present during the flowering season, as well as to poricidal anther dehiscence (Sampson 1993, Javorek et al. 2002).
This paper focuses on a number of questions that pertain to lowbush blueberry pollination.These are as follows: (1) How severe is pollinatorlimited fruit set and seed set in commercial fields of lowbush blueberry?For a total of 78 fields in one of three years we contrasted fruit set and seed set in flowers on open pollinated shoots with flowers where the all flowers on the shoot had been supplementally pollinated with outcross pollen.(2) Is there evidence that abundance of any one (or more) particular species of managed pollinators results in the highest fruit set or seed set values?We used a pollinator addition experiment in 48 fields over two years to examine the effect of the addition of managed pollinators either separately or in combination.
(3) Does pollen quantity or pollen quality (outcross versus self-pollen) limit fruit set in lowbush blueberry?We used a path-analysis approach to examine whether the effect of pollinator treatment increased seed set or fruit set directly as a result of an added pollinator, and/or indirectly through increases in observed numbers of pollen grains on stigmas.( 4) Is there evidence that any one (or more) particular species of pollinators results in larger rates of between-clone transfer of pollen?We followed foraging managed pollinators to examine whether differences in visitation patterns between and within clones might influence selfing rates.

Study area
The study was conducted in a total of 78 commercially harvested blueberry fields located near Neguac and St. Stephen, New Brunswick over three years.We studied ten commercial fields in St. Stephen and Neguac in 2009, a further 10 fields in St. Stephen in 2010, and 24 fields in Neguac in 2010 and 2011.The Neguac 2010 and 2011 fields were included in the pollinator-manipulation experiment.Fields ranged in size from 1-20 ha, were surrounded by forest or agricultural land, and were separated from one another by 1 to 5 km.Growers in St.
Stephen either added managed honeybees or no pollinators.Growers in Neguac added honey bees, leaf cutter bees or bumble bees (Table 1).We refer to these added pollinators as ''managed'' pollinators.In fields with added pollinators, honey bee hives were placed at the edge of the fields, while bumble bee boxes and leaf cutter bee domes were dispersed across the fields, following regional pollinator introduction practices.

Pollen limitation studies
Hand-pollination experiments were conducted in Neguac andSt. Stephen between 2009 and2011.Two shoots per clone were selected, with the shoots separated within the clone by 2-3 m (Table 1).Paired shoots were determined to originate from the same clone by following the shoot to the crown of the clone.Clones were also differentiated from one another based on morphology and growth pattern (Bell et al. 2009).For each clone two pollination treatments were compared: (1) an open-pollinated inflorescence (i.e., pollinated by managed and native pollinators present in the field), and (2) an inflorescence supplementally pollinated by hand with outcross pollen collected from a separate clone located at least 5 m away from the experimental transect (Table 1).Flower counts and supplemental pollinations were conducted a minimum of five times during the flowering season until flowering finished (approximately 30 days).We attempted to pollinate all open flowers on supplemental shoots.Once fruit set began, all berries on an inflorescence were collected and frozen.Fruit set (proportion of flowers setting fruit) was calculated directly by comparing fruit with flower counts.In 2010 and 2011, seeds were counted from at least one fruit per plant (unless that plant set no fruit) from all fields (24 fields in Neguac and 10 fields in St. Stephen).Seeds were only counted from two and eight fields in 2009 in St.
Stephen and Neguac respectively.Fruit were chosen haphazardly for seed counts (i.e., the ripest berries were not preferentially chosen).Seeds were counted under a dissecting scope and categorized as either ''viable seed,'' ''aborted seed,'' or ''ovule'' based on size and morphology; identification of viable versus non-viable seeds was modeled after Krebs and Hancock (1991) and Desjardins and de Oliveira (2006).

Pollination addition experiment
We coordinated with blueberry growers in the Neguac region to identify fields with different combinations and numbers of managed pollinators throughout the region and to measure openpollinated seed set and fruit set as well as seed set and fruit set on supplementally pollinated shoots (see previous section).The 24 fields in Neguac in 2010 and 2011 fell into eight pollinator treatments (replicated in each of three fields).
These consisted of control fields with no introduced pollinators, fields with one introduced pollinator species: honey bee (Apis mellifera), bumble bee (Bombus spp.), or leaf cutter bee (Megachile rotunda), fields with two introduced pollinators (honey bee and bumble bee, honey bee and leaf cutter bee, or bumble bee and leaf cutter bee), or fields with all three introduced pollinators (Table 2).As per general practice, the blueberry growers introduced pollinators to the fields a few days before flower bud opening.In fields with added pollinators, honey bee hives were placed at the edge of the fields, while bumble bee boxes and leaf cutter bee domes were dispersed across the fields.These treatments were dispersed across the Neguac region, so the effect of bees was not confounded by local regional effects.The quantity of pollinators added to these fields ranged as follows: 1-10 quads of four boxes of bumble bees, 4-80 hives of honey bees, and 4-53 liters of leaf cutter material (leaf cutter bee pupae; see Table 2).These values depended on total farm acreage and farmer investment in pollination strategies and reflect the quantities of pollinator units typically deployed by growers.In all fields, managed pollinators were observed foraging within a day of introduction to the fields.This is particularly important for fields with leaf cutter bees, as the development of pupae into adults must coincide with the predicted onset of flowering.Fruit set and seed set in open-and supplementally pollinated inflorescences in these fields was determined as described above in the pollen limitation experiment above.

Stigmatic pollen load studies
To determine whether the introduction of pollinators influenced pollen deposition on stigmas, stigmatic pollen loads were measured in the 24 Neguac fields during the peak of the 2010 and 2011 seasons (June 11 in 2010 and June 13 in 2011).In each field, stigmas were collected from 12 plants every 10 m along two transects (for a total of 24 plants).To minimize spatial effects, the first transect was oriented from the edge of the crop toward the center of the field, and the second transect was in the center of the field, oriented perpendicular to the first.Unopened flower buds were marked using a spot of acrylic paint, and collected after three days exposure to pollinators, and stigmas were stored in 95% ethanol.All pollen tetrads were counted whether on the stigma or style.The remaining ethanol was centrifuged and sampled for pollen tetrads.

Measurements of native pollinator abundance
To get an estimate of background levels of native pollinators, in 2011 we quantified the native pollinators in the pollinator addition study by conducting field surveys across all fields for three days.For all 24 fields five observers walked through the field for two hours (10 person hours per field) and recorded each pollinator seen visiting a flower.Observers were dispersed so the same pollinator was not counted twice on the same flower.Because the ability to identify pollinators varied with observer, we limited our observations to the most frequent visitors: andrenid bees, halictid bees, bumble bees (which could be native or managed) and the managed honeybees and leaf cutter bees.

Pollinator flight and visitation behavior
To investigate the potential for within or between clone pollen transfer, in 2010 we quantified flight distances and number of flowers per clone visited for each managed pollinator.Managed bees were followed in ''bee chases'' as they visited blueberry flowers.Bee chases were performed in 11 fields with managed pollinators in 2010.Bee chases were performed once a bee was observed visiting a flower.It was then followed for as many visits as possible.We were able to follow honey bees and bumble bees to the end of a foraging bout, but were for the most part unable to follow leaf cutter bees in this way, as leaf cutter bees tended to only visit one flower before changing behaviors (usually mating attempts).For all visits (including leaf cutter bees) we noted the number of flowers visited and distances between flowers.

Temperature measurements in fields
The fields in the pollinator augmentation study were clustered into regional groupings.To determine whether climatic differences might have influenced our results, during the 2010 season, nine of the fields distributed across the regions were equipped with HOBO pendant temperature data loggers (Onset Company, Cape Notes: For the ''Treatment'' column, ''B'' ¼ bumble bees added, ''H'' ¼ honey bees added, ''L'' ¼ leaf cutter bees added.Under columns headed ''Bumble bees,'' ''Honey bees,'' and ''Leafcutter bees'' are the estimated numbers of pollinators introduced into the fields by growers.The ''Area'' column denotes the size of the field.The ''Cost'' column denotes cost to the grower of purchasing pollinators, calculated as bumble bee unit ¼ 1 four colony box (C$150); honey bee unit ¼ 1 two super hive (C$120), leafcutter bee ¼ 1 gallon package of leaf cutter cocoons (C$100).
v www.esajournals.orgCod, Massachusetts, USA) set at 30-minute intervals.These fields were selected because they encompassed all regional groupings of fields.Data loggers were deposited in the field prior to bloom and collected once bloom finished.

Statistical analyses of data from pollen limitation and pollinator addition experiments
We examined pollen limitation of fruit set and seed set by estimating the mean difference in fruit or seed production after supplemental pollination compared to the control.We used general linear models with a Gaussian distribution to examine whether the proportion of flowers that became fruits, or ovules that became seeds vary with treatment, year and/or locations.We tested the effect of all-two way interactions on seed set and fruit set.Both fruit and seed set were untransformed.
To further explore the relationship between control fruit and seed set and supplemental seed set, we regressed mean field fruit set of supplementally pollinated shoots on the mean field fruit set of control shoots.We examined how these slopes varied among years and regions by using analysis of covariance to test the interaction of year and location on the slope of the association between yield (fruit set or seed set) of control and supplementally pollinated shoots.
We used path analysis to examine the influence of direct and indirect effects of pollinator number and pollinator type on fruit set and seed set.We built our final model in a sequential manner.We started with a simple model, one that hypothesized that measures of yield (seed set and fruit set) are correlated, that yield is influenced by plant health, temperature, nutrients and other external conditions, and that pollen number on stigmas directly increases both seed set and fruit set.Since our experimental plots varied widely in the numbers of pollinators added to fields by growers (Table 2), and we were interested in how this variation in pollinator number influenced our measures of yield, we added additional paths to our simple model to test whether: (1) pollen deposition, seed set or fruit set were influenced by the total number of managed pollinators in a field; and (2) if there was an effect of a particular species separate from the effect of total managed pollinator number in a field.We did not explicitly include interactions between pollinators in a field since our preliminary analyses revealed no strong interaction effects.
To disentangle the effects of pollen number versus of pollen quality we added paths from pollinator number to the simple path model to ask whether the effect of the addition of pollinators was due only to increases in pollen load on stigmas-i.e., do pollinators have an indirect effect on yield as a result of changes in pollen load?We compared the results from this analysis to those from models where we included a direct path between either the total number of pollinators, or the numbers of individual pollinator species and the variables: seed set and fruit set.A significant direct effect of pollinators (one not passing from pollinator to pollen load to seed or fruit set) would indicate pollinators influenced yield through other means than just pollen number, for example, if pollinators influence pollen quality (transfer of self-pollen) rather than quantity.For 2011 data, we included the number of native bees that we counted in the native pollinator survey as a means to examine the effect of background pollination levels on fruit set.
We used the mean seed set and fruit set from each field in each year as a measure of yield.We separated our dataset into separate years as the yield estimates varied dramatically between the years, and in 2011 we had counts of native visitors.We used estimates of median stigmatic pollen load as our measure of pollen load.Proportion seed set (numbers of seeds set as a proportion of total ovules) was arc-sin squareroot transformed, and stigmatic pollen load and proportion fruit set was untransformed.Because our pollen addition experiment indicated a relationship between control fruit set and supplemental fruit set (see results below), we included estimates of the mean proportion seed set and mean proportion fruit set of supplemental pollinated shoots in each field-i.e., as a measure of plant health and factors other than pollinators that contribute to yield.We calculated the number of each pollinator in a field as the number of hives, bumble bee boxes or leaf cutter trays in a field multiplied by the average number of bees in each container.This was then summed over all three pollinators to provide an index of total bee number.Javorek et al. (2002) estimated the pollen deposition and pollination efficiency of the most common pollinators of blueberry, by estimating pollen load and seed set after visits to virgin flowers.He found that bumble bee workers were 5.4 times more efficient than honeybees; leafcutter females were 3.4 times more efficient than honeybees.Thus we also incorporated a measure of pollinator efficiency by weighting our measure of pollinator abundance by Javorek's efficiency index to ask if the efficiency of pollinators influenced yield more than would be predicted by number alone.The number of each bee species, total bee number and weighted bee number were all log transformed.All variables were kept unscaled.
To check the appropriateness of our hypothesized paths and necessary transformations, we ran linear models of each path to assess the fit of residuals versus fitted values, and to check for homoscedascity of variances.We used the sem function in the R package Lavaan (v 0.5-17) to fit the structural equation model using maximum likelihood estimation (R Core Team 2012).We assessed model fit by checking if the P-value of the maximum likelihood chi-square test statistic was greater than 0.05.We examined additional measures according to the criteria of Kline (2011): models with the lower 90% confidence interval of the root mean square error of approximation (RMSEA) less than 0.05 are preferred, the comparative fit index (CFI) should be greater than 0.9 and the standardized root mean square residual (SRMR) should be ,0.1.For all models we compared the effect of the modeled paths on the r 2 of pollen load, seed set and fruit set.To test hypotheses of the importance of particular paths we compared the AIC, BIC and log-likelihood estimates between nested models.All models included the direct effect of seed set and fruit set of supplementally pollinated shoots on the same yield measure of control shoots, as well as correlations between seed set and fruit set.We used nested model comparisons by constructing a full model, including all paths of interest and then removing non-significant paths sequentially until the model was reduced to our most simple model.
To test whether foraging behavior differed among the managed bee species we used linear or generalized linear mixed models with ML estimation.Nested random effects included field and individual bee, while fixed effects were the different bee species (bumble bee, honey bee, leafcutter bee).The response variables were: number of clones visited and distance traveled between shoots.Distance of each individual flight was log transformed and assumed to follow a normal error distribution.The number of clones visited was treated as count data and analyzed using a generalized linear mixed (GLM) model with a Poisson distribution (the R package lme4; R Core Team 2012).Model selection was performed via backward elimination, using likelihood ration test comparing GLS and LME (nlme package) or GLM (lme4 package) models.We used AIC values to compare an unreduced model (with all nested random effects) to a reduced model (removing field or individual bee).Tests of differences between pollinators were made via post hoc contrasts using the glht package in R (R Core Team 2012).

Pollen limitation studies
Fruit set in open pollinated inflorescences showed little overall variation from year to year and between Neguac and St. Stephen, with an average 46-57% of flowers producing berries, depending on year and location (Fig. 1).In each of the three years and in each of the two production regions, inflorescences on the same plants that received supplemental pollination set on average more fruit than inflorescences left exposed to natural pollinators but not more so compared with those that were supplementally pollinated (Fig. 1).General linear models with pollination treatment as a fixed effect showed a significant effect of supplemental pollination compared to open pollination (F ¼ 5.68, df ¼ 1, P ¼ 0.018).While consistent, the increase in fruit set when inflorescences were supplementally pollinated was modest, ranging from 3% to 10% of additional flowers setting fruit (Fig. 1).These results were obtained by averaging across fields within regions, though it is notable that the differences in fruit set between inflorescences that received supplemental versus open pollination treatments was about the same in Neguac and St. Stephen, despite the fact that in Neguac, many fields received inputs of pollinator species not introduced into the fields in St. Stephen (Table 1, Fig. 1)-that is, the pollination treatment 3 location interaction was not significant (F ¼ 0.025, df ¼ 1, P ¼ 0.875).
Seed set also varied significantly between supplemental and open pollinated shoots.General linear model analysis found a significant effect of treatment (control versus supplemental pollination) on seed set (F ¼ 12.82, df ¼ 1, P , 0.001).This effect was strongly driven by the Neguac fields in 2010 and 2011 (Fig. 1), with fruits from supplementally pollinated shoots setting 7-8% more seeds.However the average effect per field and region was less consistent, with supplementally pollinated shoots in St.
Stephen setting 2-5% fewer seeds.This interaction of pollination treatment with location was not significant, likely because of the small number of fields for which we had seed counts in St. Stephen in 2009 (treatment 3 location interaction: F ¼ 3.18, df ¼ 2, P ¼ 0.08).
While fruit production in fields was pollen limited, there was marked variation among fields in fruit set of both open pollinated and supplementally pollinated shoots (Fig. 2).At the field level there was a significant relationship between open-pollinated fruit set and supplemental fruit set (b ¼ 0.88, SE ¼ 0.1, r 2 ¼ 0.46, P , 0.001; Table 3).This result was also found within each field (data not shown).It is notable that this correlation between fruit set in supplementally and open-pollinated plants is based on fields whose inputs of pollinators were not the same-e.g., some fields received no additional pollinators, while others received additions of honey bees, bumble bees, leaf cutter bees, or various combinations of these.Instead, there was some indication that the slopes of this relationship varied between years (Table 3), suggesting that year-to-year variation in plant health or climate influences open pollinated fruit set.
As with fruit set, seed set on supplementally pollinated shoots was also correlated with open pollinated seed set, although the slope of this relationship was smaller (b ¼ 0.4, SE ¼ 0.09, r 2 ¼ 0.2; Table 3, Fig. 2).The effect of supplemental pollination varied between years, but there was no evidence that the slope of the regression of seed set on supplemental shoots differed between locations or years (Table 3).

Pollinator addition experiment
We used path analysis to examine how numbers and types of managed pollinators influence pollen load, seed set and fruit set.For all models there were strong influences of supplemental pollination on yield (either seed set or fruit set) and there was a weak negative correlation between seed set and fruit set (Fig. 3, Table 4).
The best supported path models for 2010 and 2011 (Fig. 3) suggest that, while the total number of pollinators may influence pollen loads, some species of pollinators affect seed set in ways that cannot be explained by pollen number alone.In 2010 stigmatic pollen load was better described by the sum of total managed pollinator number, rather than by fitting separate paths individually for each pollinator (Table 5).In 2011, a weighted index of bee number resulted in a better fit to the variance-covariance of the data than total bee number (Table 5), however this path was not significant.In both years, total pollinator number Table 3.The effect of supplemental pollination, year (2009,2010,2011), location of field (St.Stephen or Neguac), supplemental pollination treatment and all two-way interactions on mean fruit set of open pollinated shoots in 78 blueberry fields.Supplemental fruit set or seed set were treated as continuous covariates.''Suppseed set'' and ''Suppfruit'' are field mean seed set and fruit set of supplementally pollinated shoots, respectively.For clarity of presentation variances and covariances are not shown; a correlation matrix is presented in Table 4.  4).Removing total pollinator number from the model resulted in a significant effect of honey bees on pollen load in 2010, but not 2011 (results not shown), suggesting that in 2010 the effect of managed pollinator number on pollen loads is overwhelmingly driven by honey bee additions.Average pollen loads of stigmas in a field did not strongly predict seed set (Fig. 3, Table 6).Instead there was strong support that weighted pollinator abundance directly influenced seed set in 2011 (Fig. 3).In addition, pollinators significantly influenced seed set in ways that could not be attributed to pollinator number alone (Fig. 3).For example, in 2010, there was a significant direct positive effect of leafcutter bee number on seed set (Fig. 3), however the size of this effect was small (Fig. 4).And in contrast, in 2011, while weighted total bee number was positively associated with seed set (Figs. 3 and 4), the estimate of the partial coefficient of the direct effect of honeybees on seed set was negative (Fig. 3, Table 6).This apparently contradictory result is due to the high variance in seed set when pollinator numbers were low (Fig. 4).The proportion of seeds set was also greatly influenced by one field with a very high honey bee number and poor seed set (Fig. 4), and removal of this data point reduced significance (DAIC ¼ 7.54, v 2 difference ¼ 3.24, df ¼ 1, P ¼ 0.072).
There was no significant support for the inclusion of any paths predicting fruit set from numbers of a pollinator (Table 6).Our null model suggested here that the greatest influence on fruit set was the fruit set of the supplementally pollinated shoots in the field (Fig. 3), suggesting that plant health, resource or unmeasured biotic or abiotic conditions strongly influence the number of flowers that set fruits.
There was no support that adding estimates of native pollinator abundance into the path analysis model improved its explanatory powers for either pollen load, fruit set or seed set (Table 7).Comparisons of a model with all native pollinators included to one with all paths removed resulted in only a small change of AIC of 4-5 for pollen load, seed set and fruit set.

Pollinator flight and visitation behavior
In total 54 leaf cutter bees, 109 honey bees, and 230 bumble bees were followed in bee chases.When multiple shoots were visited, all of the bee species traveled short distances with mean flight distances less than 1 m (Fig. 5).On average, bumble bees flew longer distances between flower visits (mean ¼ 77.0 cm 6 8.98 SE) than leaf cutter bees (mean ¼ 40.8 cm 6 7.76 SE), and honey bees (mean ¼ 54.5 cm 6 7.19 SE), though these differences were not significant (linear mixed effect model, effect of bee species: likelihood ratio ¼ 3.08, df ¼ 2, P ¼ 0.2, DAIC ¼ 0.918).

Temperature measurements in fields
There was no indication that the regional grouping of fields around Neguac resulted in marked differences in temperature.In general, over the flowering period, average, minimum, and maximum temperatures trended upwards by about 108C, but there was virtually no variation among fields in Neguac (Fig. 6).The temperatures for 2010 were characteristic of the range recorded over the past 10 years in the Neguac region (daily means: 3.4-18.88C,daily minimum: 7-13.28C,daily maximum 19.6-23.88C).Converting these data into days above the minimal threshold required for managed pollinator activity (bumble bees 88C; honey bees 138C; and leafcutter bees 168C) also showed very little difference across the regions.Between May 26th and June 22nd 2010, local regions ranged from 2-6 days that had a maximum temperature lower than 168 which is required for leaf cutters to be active, and 1-2 days with temperatures lower than a maximum of 138 when honey bees were not likely to be active.

DISCUSSION
The production of the blueberry crop is akin to such natural resource harvesting activities as fisheries and forestry.In the case of such enterprises, community-level factors such as the diversity of mutualists together with physical factors in the environment are under relatively little control by the harvester.In the past, blueberry producers relied solely upon native pollinators to influence fruit production, but with perceived declines in these pollinator species, they have turned increasingly to introduced species of pollinators to conduct this important ecosystem service.Lowbush blueberry especially relies upon Table 6.Sequential nested model comparisons of the effect of removing each path compared to a path model with the effect retained.Paths were removed from lowest contribution to highest, stopping when a chi-square likelihood test between the two models was significant, or the simplest path model (determined in v www.esajournals.orgpollinators capable of removing pollen from the poricidal anthers in the flowers of this species.Some of the introduced species used in blueberry fields (e.g., honey bees and leaf cutter bees) may be less effective than others (e.g., bumble bees) at mediating such pollen removal.It is thus reasonable to ask whether the costs of introducing pollinators in fields of this crop justify the benefits (e.g., increased fruit yields).This question is likely to become increasingly relevant if wild populations of pollinators decline.
Pollen is only one of the factors limiting fruit and seed set Our results showed that there was pollen limitation of fruit set (but not seed set) across many of the 78 fields measured in one of two regions and three years, however the increase in yield associated with pollen supplementation is modest.While open-pollinated inflorescences set 47-57% of their flowers as berries, supplemental pollination by hand increased this figure only to 50-65% (Fig. 1).In addition, there is a negative association between seed number and fruit number, and supplemental pollination reduced seed set compared to control plants in some fields, suggesting that increases in berry number may be offset by lighter berries.
From our results we can surmise that there are a number of factors that may help to explain why Fig. 4. Predicted partial effects from the path analysis of the effect of supplemental seed set and bee number on the proportion of seeds set in blueberry fields.Predicted effects were estimated at the mean level of all other predictors in the regression model.Predicted values are back-transformed to the original scale.
v www.esajournals.orgthe increase in fruit set with hand-pollination was relatively small.One of these is the environmental condition of the individual fields themselves.Our pollen limitation study shows that there is a significant positive relationship between the level of average fruit set in openpollinated and supplementally pollinated inflorescences (Fig. 2, Table 3).This suggests that factors independent of pollination, possibly soil fertility or disease organisms may vary among fields.Resource limitation, including that arising as a result of disease, is also consistent with the negative association found between fruit and seed set.That said, one of the physical environmental features that appears to be less important in the field-to-field variation in fruit set is temperature variation.There was relatively little variation between local regions around Neguac (Fig. 6).In addition, patterns of fruit set across two very distinct regions of New Brunswick only varied by 11%, despite markedly different weather patterns, suggesting that weather and climatic effects on fruit set are not dramatic.Weather conditions clearly vary from year to year, and in some years they likely have a significant role in pollinator availability and fruit set (e.g., see Retamales and Hancock 2012).But within the two years of our augmentation experiment, and the three years of our pollen limitation study, there was no evidence that unusual variation in weather patterns confounded our results.

Supplementing native pollinators has minimal effect on fruit and seed set
While previous studies have shown that the introduced pollinators used in this study differ with respect to their ability to deposit pollen tetrads on stigmas (Javorek et al. 2002) and to pick up blueberry pollen (Moisan-DeSerres et al. 2014), we note that the addition of honey bees, leaf cutter bees, and bumble bees, either as single species introductions, or in pairwise or the threeway combination did not in general lead to higher levels of pollen deposition, seed set and fruit set than that seen in control fields, where no   3).This may be because the total amount of bees added by growers tends to be confounded with species.Estimates by Javorek et al. (2002) suggest that honey bees are the least efficient species at pollination, and growers tend to add these in greatest numbers (Table 1).However, weighting pollinator abundance by some measure of efficiency did not increase the explanatory power of our models on fruit or seed set.In general, the variation within individual pollination treatments was large, and even differences among the separate pollination treatments themselves were difficult to detect.
There were some small and detectable effects of pollinators that could not be attributed to changes in numbers of pollinators, for example, fields with added leaf cutter bees had slightly higher inflorescence seed set in 2010, and fields with honey bees had lower seed set than expected in 2011, suggesting that the addition of pollinators may have unintended influence on pollen quality.However the effect of honey bee number on seed set can be attributed to high variation in seed set in fields with low numbers of added pollinators (Fig. 4), rather than to a strong linear decline.In the path analysis, the lack of a significant path between pollen loads and yield, as well as support for a direct path between pollinator number and yield in 2011 is further evidence that pollen quality may be an important determinant of yield.As our flight distance data show, most of the pollinator flight distances between flowers fall below the mean v www.esajournals.orgdiameter of a single clone (Fig. 5).Lowbush blueberry seed and fruit set are well known to suffer from inbreeding depression (Bell et al. 2010, Bobiwash et al. 2013), and it may be the case that much of the pollen transferred by all introduced pollinators was self-pollen that ultimately led to fruit abortion (i.e., many pollinations were geitonogamous).In this regard, the positive influence of numbers of leaf cutter bees in 2010 is interesting as our bee chases for this species were in the most part unsuccessful since leaf cutters did not often visit more than one flower sequentially.We can speculate that this contrast in leaf cutter floral visitation behavior (with respect to what we observed with honey bees and bumble bees) may have resulted in lower selfing rates.As with our pollination studies, however, the overall positive effect of leaf cutters was not particularly high (Fig. 4).
Given the costs associated with maintaining managed pollinators (Table 1), it is important to ask if the benefits achieved from them merit the investment.We were unable to obtain yield estimates (e.g., kg/ha) for the entire fields serviced by the introduced pollinators, though we have no reason to question that our inflorescence-level measurements of fruit set cannot be treated as surrogate measures of yield at the whole field level, and other studies have found a correspondence between inflorescence-level percentage fruit set and whole field yields (e.g., Jordan and Eaton 1995).A province-wide economic assessment suggests that in 2010 the value of blueberry harvest per hectare is C$3524.Increases in fruit set at similar levels as seen by our hand pollen supplemental addition experiments would result in a value of added pollen ranging from $0-$948 (Table 8), assuming no reduction in fruit weight as a result of lower seed numbers.This is similar to or less than the total invested in adding pollinators (Table 2).It is possible that the years in which the effects of introduced pollinators were studied could have, by chance, been non-representative of the pollination conditions experienced by these plants in other years.For instance, perhaps in the years of our study, wild bee populations were sufficient to service most of the flowers, and the additional pollination attained with the introduction of managed pollinators was near or above the levels Note: The mean farmgate value in 2010 was C$20,000,000, resulting in an estimated value of blueberries of C$1.36 per kilo or C$3524.85/ha.at which pollen was not a limiting factor-as our results on open-versus hand-pollination have suggested (Fig. 1).It is known that wild bee populations can fluctuate from year to year (Williams et al. 2001).In that case, the practice of introducing managed pollinators as pollination ''insurance'' can be viewed as a reasonable investment, especially if it is unknown in advance of flowering whether native pollinators are in abundant supply or not.Alternatively, the amounts of pollinators introduced into the fields that we monitored, while generally characteristic of the production practices in the region, may not have been sufficient to boost fruit set levels sufficiently to detect a signal above the plant-toplant ''noise'' inherent in our samples.
We do not doubt that some species of pollinators are more effective at extracting pollen from the poricidal anthers of lowbush blueberry flowers, as shown by Javorek et al. (2002).What remains to be seen in longer-term studies is whether these pollinator behaviors matter with respect to elevating fruit production, and if so, whether they only become manifest once much larger levels of pollinator introductions are used, or alternatively, once conditions that permit selfpollination are minimized.We believe that improvements in pollination and fruit set might best be achieved by participation among producers in producing a database in which information on types and amounts of pollinators, fruit yield, use of herbicides and fungicides, whether fields are burned or mowed in off-harvest years.Such longer-term record keeping can only benefit the community of growers and producers as a whole.

Fig. 1 .
Fig. 1.Fruit set and seed set by location across three years.The effect of supplemental pollination on fruiting success was significant across both locations, but pollen limitation (difference between supplemental and open pollination) varied by year.Lines indicate 6 1 SE.

Fig. 3 .
Fig. 3. Significant paths in the final path model of the direct and indirect effects of pollinators on pollen grain deposition, proportion seed set and proportion fruit set.Values are the standardized path-coefficients, and the arrow weight is proportional to the size of the path-coefficient (red arrows depict a negative association; green arrows a positive association).Arrows with double heads are correlations between the dependent variables.''Suppseedset'' and ''Suppfruit'' are field mean seed set and fruit set of supplementally pollinated shoots, respectively.For clarity of presentation variances and covariances are not shown; a correlation matrix is presented in Table4.

Fig. 5 .
Fig. 5. Histogram of the distance travelled between flowers by bees during bee chases in 2010.Vertical line represents the mean diagonal length of lowbush blueberry clones in the region.

Fig. 6 .
Fig. 6.Temperature in 9 different field sites in the Neguac region each represented by colored lines between 9 May and 15 June 2010.

Table 1 .
Experimental design and methodology summary for field seasons in New Brunswick, Canada.

Table 2 .
Pollination additions to fields in Neguac, New Brunswick in 2010.

Table 4 .
Correlation matrix of transformed parameters used in path analysis.Correlation coefficients above the diagonal are from 2010; below from 2011.Abbreviations are: PL pollen load, FS fruit set, SFS supplemental fruit set, SS seedset, SSS supplemental seed set, TBN total bee number, WBN weighted bee number, HN honey bee number, LN leaf cutter bee number.

Table 5 .
Path analysis model fit parameters for the construction of an initial path model testing the effect of total bee number or weighted bee number on pollen load deposition, fruit set or seed set.CFI is Comparative Fit Index, LogL is Log-likelihood estimate, AIC is Akaike Information Criterion.

Table 5
Note: v 2 diff is the difference in chi-square estimate between the two models; df diff is the difference in degrees of freedom.

Table 7 .
Sequential nested models comparison comparing a full path analysis model that had all native pollinators added, to models sequentially removing the non-important effect of native pollinators.Significant paths from the best model (see Fig.3) were retained in this analysis.''Allbees'' is the sum of all bees observed (identified or unidentified) during an observation period.Notes: v 2 diff is difference in chi-square estimate between the two models; df diff is difference in degrees of freedom.

Table 8 .
Economic analysis of Neguac 2010 fields.Values are estimated using a crop average of 2590 kg/ha.