The compounding effects of high pollen limitation, selfing rates and inbreeding depression leave a New Zealand tree with few viable offspring

Abstract

Background and Aims Interactions between species are especially sensitive to environmental changes. The interaction between plants and pollinators is of particular interest given the potential current global decline in pollinators. Reduced pollinator services can be compensated for in some plant species by self-pollination. However, if inbreeding depression is high, selfed progeny could die prior to reaching adulthood, leading to cryptic recruitment failure.

Methods To examine this scenario, pollinator abundance, pollen limitation, selfing rates and inbreeding depression were examined in 12 populations of varying disturbance levels in Sophora microphylla (Fabaceae), an endemic New Zealand tree species.

Key Results High pollen limitation was found in all populations (average of 58 % reduction in seed production, nine populations), together with high selfing rates (61 % of offspring selfed, six populations) and high inbreeding depression (selfed offspring 86 % less fit, six populations). Pollen limitation was associated with lower visitation rates by the two endemic bird pollinators.

Conclusions The results suggest that for these populations, over half of the seeds produced are genetically doomed. This reduction in the fitness of progeny due to reduced pollinator service is probably important to population dynamics of other New Zealand species. More broadly, the results suggest that measures of seed production or seedling densities may be a gross overestimate of the effective offspring production. This could lead to cryptic recruitment failure, i.e. a decline in successful reproduction despite high progeny production. Given the global extent of pollinator declines, cryptic recruitment failure may be widespread.

INTRODUCTION

Pollinators provide a key ecosystem service in both natural and agricultural ecosystems. Over 87 % of flowering plant species rely on biotic pollination to reproduce (Ollerton et al., 2011). Similarly, although few crop species rely solely on biotic pollinators, one estimate suggests that only 7 % of crop species receive no benefit from pollinators (Klein et al., 2007). Interactions between species, such as those between plants and their pollinators, are especially sensitive to environmental changes (Tylianakis et al., 2008; Dunn et al., 2009; Berg et al., 2010; Kiers et al., 2010; Yang and Rudolf, 2010; Northfield and Ives, 2013). Current environmental changes are resulting in negative effects on pollinator abundance, fuelling concerns of a global pollination crisis and its effect on agricultural and ecological services (Biesmeijer et al., 2006; Hegland et al., 2009; Potts et al., 2010; Thomann et al., 2013; but see Ghazoul, 2005). These pollinator declines have the potential to affect both the quantity (through pollen limitation) and quality (through selfing and inbreeding depression) of offspring negatively. While negative effects of pollinator abundance on the quantity of progeny are well documented (Linhart and Feinsinger, 1980; Spears, 1987; Steffan-Dewenter and Tscharntke, 1999; Chittka and Schurkens, 2001; Quesada et al., 2003), the effect on the quality is virtually unknown (but see Eckert et al., 2010; Delmas et al., 2015).

Reduced pollinator service can negatively affect the quality of offspring if plants supplement their pollen receipt with self-pollen, leading to an increase in selfing rates. Many plant species can self-pollinate in the absence of pollinators (autonomous self-pollination), termed reproductive assurance (Baker, 1955; Jain, 1976; Herlihy and Eckert, 2002; Busch and Delph, 2012). Despite the fact that selfing usually results in less fit offspring due to inbreeding depression, reproductive assurance is generally considered an adaptation to variable pollinator abundance in species where all selfed offspring do not die (inbreeding depression <1; Lloyd, 1992). However, consistently higher selfing rates due to low pollinator abundance could negatively affect both short-term (decreased germination, growth, competitive ability) and long-term processes (loss of genetic diversity, reduced gene flow; Stebbins, 1957; Takebayashi and Morrell, 2001). Thus, understanding the impact of pollinator declines on plant population processes requires information on the type of selfing that occurs, the amount of selfing and the severity of inbreeding depression.

New Zealand is unfortunately an excellent location to examine the effect of human disturbance on plant–pollinator interactions. Prior to human arrival, bird species were the primary vertebrates with roles as pollinators, fruit dispersers, herbivores and predators (Holdaway, 1989). Human colonization of the islands from approx. 1280 AD (Wilmshurst et al., 2008), and the resulting deforestation and introduction of mammal predators, led to the extinction of at least 41 % of endemic bird species (Holdaway et al., 2001). Range reductions of the remaining birds have continued; from 1979 to 2004, 44 % of forest bird species have experienced range restrictions (Innes et al., 2010). Because bird species were integral to ecosystem functioning, the loss of so many species led to the disruption of many interactions, including pollination (Kelly et al., 2010). Understanding the impact of these changed interactions is the first step in conserving the remaining species and their interactions.

Here we address the issue of pollinator abundance, selfing rates and inbreeding depression using a widespread, self-compatible, canopy tree species, Sophora microphylla. Plants flower spectacularly in the spring over a period of about 1 month, during which time a tree can produce many thousands of yellow to gold flowers. Flowers are visited by a variety of animals that are thought to vary in their effectiveness. The native honeyeaters are considered to be the primary pollinators, particularly tui (Prosthemadera novaeseelandiae) and bellbirds (Anthornis melanura) (Castro and Robertson, 1997; Anderson, 2003), both of which are territorial (Craig et al., 1981; Bergquist and Craig, 1988). Other floral visitors include introduced insects such as bumble-bees and honey-bees (Anderson, 2003), the recently arrived silvereye (Zosterops lateralis), the house sparrow (Passer domesticus) and the chaffinch (Fringilla coelebs) (McCann, 1952), all of which are most probably less effective pollinators due to their size relative to the floral tube and robbing behaviour (Anderson, 2003; A. W. Robertson, pers. comm.). Although S. microphylla is distributed throughout New Zealand, it prefers open habitats including steep slopes along rivers, lake margins, dunes and even pastures (Heenan et al., 2001). Due to this restricted range of appropriate habitats, few large populations occur in areas with high bird abundance. Previous work has suggested that selfing can occur in the absence of pollinators (Rattenbury, 1979) and that inbreeding depression is strong; selfed offspring were 96–99 % less fit than outcrossed offspring after 11 years of growth in a common garden (Robertson et al., 2011). The combined effect of populations occurring in marginal habitats, the ability to self-pollinate and strong inbreeding depression could lead to substantial negative effects of pollinator declines on offspring quality.

Using a combination of field observations, mating system analyses and population genetics, we asked several questions. (1) Are populations pollinator and/or pollen-limited? (2) Does reproductive assurance lead to an increase in selfing rate as pollinator abundance decreases? (3) Is inbreeding depression strong enough to negate the benefits of selfing? Because S. microphylla is pollinated by a range of bird species and is widespread across New Zealand, it serves as a useful indicator species for the health of plant–pollinator interactions. Any changes in the quality of offspring for this species would probably be an underestimate of changes experienced in rarer or more specialized taxa.

MATERIALS AND METHODS

Sophora microphylla populations were chosen throughout New Zealand with varying amounts of human disturbance – from isolated trees in grazed pasture, to a conserved ‘mainland island’ whose purpose is to preserve the unique collection of plant species (Fig. 1; Supplementary Data Table S1). All populations have a disrupted avifauna compared with pre-human estimates as none of these populations occurs in predator-free areas (Diamond and Veitch, 1981). According to the New Zealand Bird Atlas (Robertson, 2007), both tui and bellbirds currently occur in the areas around most of the populations, with the exceptions that bellbirds are not found near Lake Waikare and tui are not found near Waimakariri, Ahuriri or Rakaia Gorge.

Fig. 1.

Map of locations of each population, with the symbol type indicating pollinator abundance measures sampled (circle, both observations and pollinations; square, observations; triangle, pollinations; diamond, none), the symbol fill indicating if the mating system was measured (black, measured; white, not measured), and symbol size indicating the approximate population size. See Supplementary Data Table S1 for sample sizes for each type of measurement and additional population information.

Several closely related Sophora species co-occur near some of the populations chosen (Supplementary Data Table S1) and are known to hybridize with S. microphylla (Heenan et al., 2001). Thus, some populations may include a mixture of pure parental species and hybrids. This was a particular problem at Wenderholm, which was the only site with a mixture of S. microphylla and S. chathamica, as we discovered after the 1999 pollination season (S. chathamica was not formally described until 2001). In the 2000 season at Wenderholm on ten S. microphylla trees, we hand outcrossed approximately ten flowers per plant with S. microphylla pollen from a different tree, and approximately ten flowers with S. chathamica pollen. Fruit set for the microphylla × chathamica crosses (13·2 %) was only half that for microphylla × microphylla crosses (26·8 %), and natural fruit set (11·6 %) was comparable with that of the microphylla × chathamica crosses, suggesting that low fruit set from hybrid crosses may be an issue at Wenderholm. Species differ only slightly in floral morphology (Heenan et al., 2001); however, it is unknown if self-compatibility or selfing rates differ amongst them.

Pollinator observations

To quantify bird visitation rates, ten populations were chosen in which to observe the types of birds visiting and the length of these visits (Fig. 1; Supplementary Data Table S1). Within a population, several observation stations were chosen from which multiple trees could be observed simultaneously. During observation periods, ranging from 5 to 20 min, the type of floral visitor and the length of their foraging bout were recorded. To standardize across floral abundance and observation lengths, for each observation period we calculated the visitation rate as seconds of visitation per 100 flowers per hour. Observations were conducted in 1999, 2000 or 2013 depending on the population (Table S1).

Although observations differed in many ways (year, observers and length), we were interested in statistically determining how populations differed in their pollination rates. To this end, we used population means in a Kruskal–Wallis non-parametric test to determine if population type (pasture or conserved, Supplementary Data Table S1) differed in the visitation rates of legitimate pollinators (tui or bellbirds, hereafter ‘good’) or potentially illegitimate (all others, hereafter ‘bad’) visitors or the total visitation rates. Additionally, we used Pearson’s correlation to determine if population size affected any visitation rates [the results were qualitatively the same if using Spearman’s rank correlation or population size categories in an analysis of variance (ANOVA)].

To determine how much nectar is generally produced, we measured nectar amounts and sugar concentration on pollinator-excluded flowers in the Waimakariri population in 2003. Flowers that were bagged prior to flower opening produced on average 13·9 μL ± s.e. 2·19 of nectar (n = 59) with an average Brix reading of 20·17 ± 1·74 % (n = 13). This gives a sucrose equivalence of 3·01 mg per flower (Bolten et al., 1979), which is comparable with hummingbird-pollinated flowers (13 species average = 11·32 μL or 2·39 mg of sucrose; Cruden, 1976).

Manipulative crosses

To determine the potential for autonomous self-pollination, pollen limitation and early inbreeding depression, we performed several types of crosses in nine populations (Fig. 1; Supplementary Data Table S1) in 1999 and/or 2000. On each of up to 11 plants in a population, flowers were chosen for: hand pollination with a mixture of outcrossed pollen from at least three donor trees (mean = 10 flowers per tree); self-pollination (mean = 10); pollinator exclusion (mean = 17); or natural pollination (mean = 37). Flowers for the self-pollination and pollinator exclusion treatments were placed into a mesh bag prior to opening and after treatment for approx. 1 month to ensure pollinators did not visit receptive flowers. Fruit set and seeds per fruit were determined after 3 months. In Lake Rotoroa, Waimakariri and Ahuriri, the same individuals were scored in two consecutive years. In 2000 no trees set fruit at Lake Rotoroa, probably due to flooding, and these were excluded from the analysis.

From these results, a variety of measures were calculated. Capacity for autonomous self-pollination was determined by calculating the mean seeds per flower in the pollinator exclusion treatment for each population. Pollen limitation was analysed by comparing the natural treatments with hand-outcrossed treatments for fruits per flower and seeds per flower. We chose not to include seeds per fruit because fruits per flower is easier to compare with other studies, seeds per flower is a more complete measure of reproductive success and seeds per fruit was positively correlated with fruit set (Pearson’s r = 0·192, P = 0·001). We used a generalized linear model with specific probability distributions (fruit set = binomial, seeds per flower = negative binomial) in proc GENMOD in SAS v9.3. The population:year combination, pollen source (natural or outcrossed) and their interaction were used as predictors. A significant interaction effect would indicate differences among populations in the severity of pollen limitation. We also calculated the population pollen limitation index (PLI) as 1 − (natural/outcrossed) from the least-square means (LS means) derived from the above models (Larson and Barrett, 2000). To determine if pollen limitation was correlated with pollinator visitation rates, for the populations that had both measures we used a Spearman’s rank correlation for both total pollinator visitation rates (all birds including silvereyes) and ‘good’ visitation rates using proc CORR in SAS v9.3. To determine if population type (pasture or conserved) affected the PLI, we used an ANOVA with the population means as the dependent variable and population type as the predictor.

To compare pollen limitation with previous studies, a variety of other calculations were made. Limitation indices have been tabulated in reviews using several different methods, which we attempted to calculate using the most comparable data. For comparison with Ashman et al. (2004), we averaged the standardized effect size [(outcrossed − natural)/s.d.] of seeds per flower for each population. To compare with results from Knight et al. (2005), we averaged the log response ratio [ln (outcrossed/natural)] of fruit set for each population. For comparison with Newstrom and Robertson (2005), we averaged their measure of PLI (natural/outcrossed) of fruit set or seeds per flower for each population.

Pre-dispersal inbreeding depression was similarly analysed by comparing the self-pollination and outcross pollination treatments using the same predictors and probability distributions as above. Population pre-dispersal inbreeding depression (i.e. from pollination to developed seed) was calculated as δ_pre = 1 − (self/outcrossed) using the LS means.

Mating system and inbreeding depression

For the mating system analysis, leaves and up to ten mature fruits from up to 20 trees from six populations (Fig. 1; Supplementary Data Table S1) were collected in the summers of 2012 and 2013. Tissue collection for DNA extraction from these seeds differed among populations due to necessary changes in the germination method. Initially seeds from two populations (Ahuriri and Kowhai Point) were scarified, their seed coats removed and then planted in soil. However, this resulted in low germination (average of 55 %), which could severely downwardly bias selfing rate estimations if caused by early acting inbreeding depression. To determine if estimated selfing rates differed between soil-germinated and Petri dish-germinated seeds, we chose one of these populations (Ahuriri) to compare selfing rates estimated from soil-germinated and Petri dish-germinated seeds. To reduce the effect of low germination on the selfing rates, we chose to germinate seeds in Petri dishes for the remaining four populations. For soil-grown plants we collected a single leaf, while for Petri dish-germinated seeds we collected the shoot apex after the root had begun growing. Sample sizes for each population and collection type can be found in Table 2.

Table 2.

Population genetic parameters (s.e.) of adult plants (without null allele loci), MLTR mating system parameters (s.e.) (without null allele loci) for seeds (Petri dish-germinated if not otherwise specified), and post-dispersal inbreeding depression for each population calculated from COLONY results

Population	Population genetics			MLTR
	Ho	He	FIS	Nfam	Nprog	sm	b = ss − sm	rpm	rs	δpost
Paengaroa	0·800 (0·065)	0·830 (0·05)	0·038 (0·038)	8	73	0·673 (0·099)	0·057 (0·024)	–0·022 (0·301)	0·897 (0·064)	0·967
Vinegar Hill	0·715 (0·045)	0·824 (0·043)	0·132 (0·032)	21	202	0·766 (0·060)	0·050 (0·018)	0·164 (0·108)	0·836 (0·079)	0·867
Kowhai Point (soil-germinated)	0·727 (0·067)	0·819 (0·037)	0·108 (0·084)	23	144	0·495 (0·063)	0·04 (0·025)	0·106 (0·140)	0·867 (0·080)	0·646
Waimakariri	0·738 (0·029)	0·820 (0·002)	0·100 (0·036)	20	200	0·628 (0·066)	0·088 (0·027)	0·286 (0·148)	0·708 (0·120)	0·868
Ahuriri	0·785 (0·059)	0·832 (0·032)	0·052 (0·077)
Petri-dish germinated				18	59	0·431 (0·172)	0·168 (0·055)	0·188 (0·226)	0·525 (0·213)	0·792
Soil-germinated				26	169	0·346 (0·064)	0·107 (0·018)	0·054 (0·035)	0·654 (0·108)	0·539
Rakaia Gorge	0·828 (0·026)	0·861 (0·019)	0·037 (0·04)	18	176	0·335 (0·062)	−0·007 (0·019)	0·082 (0·022)	0·925 (0·055)	0·815
Average	0·766 (0·020)	0·831 (0·013)	0·078 (0·021)

H_o, observed heterozygosity; H_e, expected heterozygosity; F_IS, Wright’s inbreeding coefficient; N_fam, number of families; N_prog, number of progeny; s_m, multilocus selfing rate; s_s, single locus selfing rate; b, biparental inbreeding; r_pm, multilocus correlation of paternity; r_s, correlation of selfing among loci; δ_post, inbreeding depression from Petri dish or soil-germinated seeds to adult stage calculated per Ritland (1990).

For the inbreeding depression analysis, we collected leaf tissue from up to 50 adults and 50 juveniles (when present), along with an estimation of their height under the assumption that this correlates with age. Leaf tissue was dried and stored in silica gel.

DNA was extracted using a modified cetyltrimethylammonium bromide (CTAB) protocol (Doyle and Doyle, 1987). For Ahuriri and Kowhai Point, six loci from Van Etten et al. (2014) were amplified as described in the paper (Sop-248, Sop-802, Sop-806, Sop-807, Sop-808 and Sop-825) with the exception of 40 cycles of PCR rather than 35. Samples were genotyped at Massey Genome Services (Massey University, New Zealand) on an ABI3730 DNA Analyzer (Applied Biosystems, Carlsbad, CA, USA). Genotypes were scored using GeneMapper v 3.7 (Applied Biosystems). Comparisons of progeny and maternal genotypes revealed four loci with null alleles (Sop-802, Sop-806, Sop-807 and Sop-808), especially in the Kowhai Point population. Decreasing the annealing temperature from 53 to 48 °C reduced the frequency of null alleles in loci Sop-806 and Sop-808. For Sop-802 and Sop-807, we redesigned the primers from the original sequence (GenBank accession nos KF672187 and KF672189, respectively), which decreased the frequency of nulls. The new primers were Sop-802 forward 5' to 3', CACGACGTTGTAAAACGACAAGCTCTCAAGAGATCCTC; reverse, GTTTCTTTCAGGTCGTGGTATGAGTC; Sop-807 forward, CACGACGTTGTAAAACGACAATAGGTTGCTCTTGACCC; reverse, GTTTCTTTCTAAGTTGCATGCAGTGG. The PCR program was as described in Van Etten et al. (2014) with a 53 °C annealing temperature. For all other populations, we replaced the original loci with the newly designed primers and PCR programs.

From the progeny genotypes, we estimated selfing rates using both COLONY (Jones and Wang, 2010) and MLTR (Ritland, 2002). COLONY uses maximum-likelihood to identify putative parents and thereby the selfing rate (s) and inbreeding coefficient (F). We chose this program in addition to the more common MLTR because COLONY can use loci with null alleles and has been shown to be less biased than other similar programs (Wang et al., 2012). We used the following settings: monoecious species, inbreeding present, diploid, polygamy for males and females, sibship complexity prior, full-likelihood method, medium length run, medium precision and no updating allele frequencies. Locus-specific error rates per population were estimated from an initial run. Seeds from a mother were coded as a known maternal sibship with their known maternal individuals. The estimated probability of the father being in the potential genotyped pool was dependent on the population (Paengaroa 0·3, Vinegar Hill 0·3, Kowhai Point 0·3, Waimakariri 0·7, Ahuriri 0·5 and Rakaia Gorge 0·3). Average selfing rates and F-values were estimated for each population separately. Additionally, because Ahuriri had genotypes for a reasonable number of progeny from both Petri dish- and soil-germinated seeds, we estimated selfing rates for each group separately. We compared selfing rates using the 95 % confidence intervals. To test if visitation rates correlated with selfing rates, we used Pearson’s correlation between population mean visitation rates (‘good’, ‘bad’ and total) and the populations’ selfing rate. Differences in selfing rates by population type were not tested because only one pasture population had selfing rate data. To compare with previous research and to validate the COLONY results, we also ran the traditional MLTR analysis without the two loci with the most frequent null alleles (Sop-806 and Sop-807). Default options were used, except that we increased the bootstrap replicates to 1000, which were used to compute standard errors.

To account for inbreeding depression before the genotyping stage (from fertilization through seed development), we also calculated the zygotic selfing rate (s_z) using the equation from Maki (1993). Because the inbreeding depression analysis above indicated a range of inbreeding depression in fruit and seed set, we calculated s_z based on low (25th quartile δ_pre = 0·361) and high (75th quartile δ_pre = 0·804) estimates from that analysis. For the selfing rates from post-germination seedlings, we used the estimates of inbreeding depression in germination from Robertson et al. (2011) of low (δ_germ = 0) and high (δ_germ = 0·571) to calculate the combined inbreeding depression as: 1 − [(1 − δ_pre) × (1 − δ_germ)].

To estimate the inbreeding depression between seed and adult stages and to investigate its timing, we calculated inbreeding coefficients for adults and juveniles. A decrease between life stages suggests that inbreeding depression has decreased the proportion of selfed progeny. Because the data were known to have null alleles and few programs are available to account for this, we used COLONY to calculate the F-values for the juveniles and adults. From the adult F-values and selfing rates of progeny, we calculated the post-dispersal inbreeding depression using Ritland’s (1990) equation: δ_post = 1 − [2(1 − s)F/s(1 − F)]. This estimate differs from the previous inbreeding depression study (Robertson et al., 2011) in that it is a cumulative estimate from germination to adulthood under field conditions; the previous estimate was only from seed to 11 years in a common-garden experiment, which may lead to an underestimate of inbreeding depression in natural populations (Dudash, 1990; Husband and Schemske, 1996). To compare with other studies we also calculated observed and expected heterozygosity (H_o and H_e, respectively) and F_IS of the adults without the loci with frequent null alleles (806 and 807) using GenAlEx v.6.501 (Peakall and Smouse, 2006, 2012).

RESULTS

Pollinator observations

A range of birds were observed visiting flowers, with tui, silvereyes, house sparrows (Passer domesticus at the University of Canterbury) or chaffinches (Taihape) being the most common depending on the population (Fig. 2A). Other visitors included (with number of sightings): blackbirds (Turdus merula; 4), starlings (Sturnus vulgaris; 3), kaka (Nestor meridionalis; 1), yellowheads (Mohoua ochrocephala; 1) and New Zealand pigeons (Hemiphaga novaeseelandiae; 1).

Fig. 2.

(A) Visitation rate by bird type, (B) seed set by pollination treatment type, (C) pollen limitation for fruit set (black bars) and seed set (white bars), and (D) early inbreeding depression for fruit set (black bars) and seed set (white bars) for each population:year.

Visitor types and visitation rates varied among populations and years. Many of the populations received visits from both ‘good’ (tui or bellbirds) and ‘bad’ pollinators (others). However, the populations furthest south had very low ‘good’ pollination rates. In fact, only three out of 130 observation periods had any ‘good’ pollinators in the Waimakariri, University of Canterbury or Ahuriri populations. Visitation rates tended to differ with population type; populations in pastures had slightly lower ‘good’ visitation rates (χ²₁ = 3·62, P = 0·06) but higher ‘bad’ visitation rates (χ²₁ = 6·51, P = 0·01). However, within each population type there was considerable variation. For example, the highest ‘good’ visitation rates were found at Wenderholm (former pasture) in 1999 and Lake Rotoroa (conserved forest) in 2000. Population size was not correlated with ‘good’, ‘bad’ or total visitation rates (P > 0·4). Additionally, in two out of the three populations for which multiple years of data were collected, visitation rates differed significantly between years (Lake Rotoroa, χ²₁ = 5·93, P = 0·01; Waimakariri, χ²₁ = 1·83, P = 0·2; Ahuriri, χ²₁ = 11·34, P = 0·0008).

Pollen limitation index

Overall, bagged flowers generally showed the lowest seed production, followed by naturally pollinated and hand-pollinated selfed flowers, and hand-pollinated outcrossed flowers producing the most seeds (Fig. 2B). Bagged flowers generally produced very few seeds, with the highest average of 0·4 seeds per flower in the University of Canterbury population. Across populations, naturally pollinated flowers produced 60 % fewer seeds per flower than hand-pollinated outcrossed flowers (natural_lsmean = 0·738 ± 0·088 s.e., outcross_lsmean = 1·854 ±0·149; P < 0·0001; Table 1; Fig. 2B) partly due to decreased fruit set (natural_lsmean = 0·199 ± 0·007, outcross_lsmean =0·420 ± 0·017; P < 0·0001).

Table 1.

ANOVA results (χ² values) from pollination treatments from 11 population:year combinations

	Distribution	Population:year (d.f. = 10)	Treatment (1)	Interaction (10)
Pollen limitation (natural vs. outcross hand-pollinated)
Fruits/flower	Binomial	330·83***	164·28***	39·87***
Seeds/flower	Negative binomial	28·51**	42·21***	8·77
Early inbreeding depression (self- vs. outcross hand-pollinated)
Fruits/flower	Binomial	265·83***	90·89***	47·41***
Seeds/flower	Negative binomial	65·26***	48·30***	14·25

*P < 0·05; **P < 0·01; ***P < 0·0001.

Pollen limitation varied widely among populations (PLI fruits/flower range = 0·242–0·719, mean = 0·478; seeds/flower range = 0·303–0·807, mean = 0·575; Fig. 2C) and differed significantly among sites for fruits per flower (population × pollen source P < 0·0001; Table 1) but not seeds per flower (P = 0·55). Higher rates of ‘good’ pollinators had a negative relationship with pollen limitation after removing an outlier (Wenderholm 1999; see the Materials and Methods and Discussion) for both seeds/flower (F_1,8 = 7·91, P = 0·026) and fruits/flower (F_1,8 = 10·66, P = 0·014). There was no relationship between total visitation rate and the PLI (F_1,8 = 0·10, P = 0·76 for seeds per flower) even with Wenderholm excluded. Population type did not significantly affect pollen limitation, although pasture populations had slightly higher pollen limitation scores (F_1,9 = 2·44, P = 0·15; PLI seeds/flower means: pasture = 0·61 ± 0·05 s.e., conserved = 0·43 ± 0·06).

Pre-dispersal inbreeding depression

Selfed flowers produced 68 % fewer seeds per pollinated flower than outcrossed flowers (self_lsmean = 0·602 ± 0·092 s.e., outcross_lsmean = 1·854 ± 0·149; P < 0·0001; Table 1). In part this was due to selfed flowers having 51 % lower fruit set than outcrossed flowers (self_lsmean = 0·206 ± 0·014 s.e., outcross_lsmean = 0·420 ± 0·017; P < 0·0001; Table 1). Populations differed in the severity of inbreeding for fruit set (population × cross type P < 0·0001; Table 1; Fig. 2D) with Wenderholm, Waimakariri 2000 and University of Canterbury being low and Lake Rotoroa and Ahuriri 2000 being higher.

Selfing rates

On average, 61 % of genotyped seeds were estimated to result from selfing (Fig. 3A). Selfing rates from seeds ranged from 40 to 76 %, with high rates at Vinegar Hill and Paengaroa and low rates at Rakaia Gorge. The selfing rates from soil-germinated seeds in the Ahuriri population were the lowest (34 %) and lower than the rates estimated from the Petri dish-germinated seeds in the same population (54 %). Estimated zygotic selfing rates ranged from 45 to 94 % depending on the strength of inbreeding depression used in the calculation (Fig. 3A). Results from MLTR (Table 2) were qualitatively similar to results from COLONY, with a slightly lower average selfing rate for seeds of 57 %. Biparental inbreeding rates were low (mean = 0·06), as was the correlation of paternity (mean =0·134), suggesting that within a tree seeds rarely shared a father. Visitation rates did not have a significant relationship with selfing rates (‘good’ F_1,3 = 0·50, P = 0·53; ‘bad’ F_1,3 = 0·01, P = 0·93; total F_1,3 = 0·12, P = 0·75).

Fig. 3.

(A) Selfing rates from Petri dish-germinated (circles) and soil-germinated (triangles) seeds, and back-calculated zygotic selfing rates (grey area; see text for calculation). (B) Inbreeding coefficient with 95 % confidence intervals by stage.

Post-dispersal inbreeding coefficient

The average inbreeding coefficient changed dramatically between life stages (Fig. 3B). Adult inbreeding coefficients were very small in all populations, ranging from 0·04 to 0·17. However, seeds and seedlings had high inbreeding coefficients, ranging from 0·21 to 0·61 (Fig. 3B). This large difference between the inbreeding coefficients of adults and seeds resulted in estimates of inbreeding depression ranging from 0·79 to 0·97 (Table 2). The low inbreeding coefficient for juveniles (average = 0·19, Fig. 3B) suggests that most of this reduction occurs prior to the juvenile stage.

DISCUSSION

We found that the abundance and type of pollinators differed by population, with some indication that pasture populations had fewer ‘good’ pollinators. Higher visitation rates of these ‘good’ pollinators were negatively correlated with pollen limitation, suggesting that both the abundance and effectiveness of visitors limited seed production in many of the populations. We found high rates of selfing, especially after correcting for early inbreeding depression. Little of this seems to be accounted for through autonomous self-pollination, suggesting that reproductive assurance does little in this species to compensate for pollinator limitation. Despite high selfing rates, adult trees were not highly inbred, suggesting very strong inbreeding depression. Due to the high selfing rates and high inbreeding depression, almost half of the seeds produced are genetically doomed. This study highlights the importance of assessing both the quantity and quality of offspring when determining population health.

Pollen limitation

We found variation among populations in the types of birds visiting and the abundance of birds. Populations in pastures tended to have higher rates of ‘bad’ pollinators, which ranged from exotic silvereyes to native parrots, and lower rates of ‘good’ pollinators. Not surprisingly, types of visitors differed geographically. In particular, populations near Christchurch (Waimakariri, University of Canterbury and Ahuriri) had effectively no ‘good’ pollinators. While seed production does occur in these populations, suggesting that ‘bad’ visitors are doing some pollination, our results suggest they are not as effective; visitation rates by the presumed legitimate pollinators (the two honeyeaters, bellbirds and tui) were negatively related to the degree of pollen limitation after removing Wenderholm (Wenderholm appeared to have fruit set depressed by hybridization with S. chathamica), while it was not related to PLI for ‘bad’ visitation rates. This suggests that low pollination is due to local shortages of the two large, long-tongued endemic birds that are best able to probe S. microphylla flowers, and that other native and exotic birds are not adequate substitute pollinators.

In the populations studied, S. microphylla plants were more pollen limited than most species globally (Ashman et al., 2004; Knight et al., 2005). We found a range of pollen limitation, with Lake Waikare being the lowest (approx. 0·30 for seeds/flower) and Wenderholm being the highest (approx. 0·81, but note caution above about possible hybridization). The average pollen limitation found in this study was much higher than for most plant species based on comparison of our results with those in several reviews. The average standardized effect size (1·10, range: 0·38–2·62) was higher than in 90 % of the studies reviewed by Ashman et al. (2004). Even the least pollen-limited population (Lake Waikare) had greater pollen limitation than in 58 % of the studies. Similarly, compared with the results from Knight et al. (2005; range: 0·58–0·87), our average log response ratio (0·67, range: 0·27–1·26) was higher than that of 75 % of the species. However, comparisons with New Zealand species show S. microphylla to have about average pollen limitation. Compared with results from Newstrom and Robertson (2005), our values of 0·53 (fruit set) and 0·43 (seeds per flower) were slightly lower than for other trees, but similar to the average over all life forms (0·42). Similarly, comparisons in Kelly et al. (2010) indicated that S. microphylla was in the middle of the range of PLIs for New Zealand, bird-pollinated species. These results show that S. microphylla is more pollen limited than most plants worldwide, but about average for plant species measured in New Zealand.

Selfing

We found high selfing rates, ranging from 34 to 76 %. The rate increases to a range of 45–94 % when accounting for early selfed seed and fruit abortion prior to genotyping. Our average of 61 % selfing in seeds is higher than approx. 60 % of animal-pollinated species (Vogler and Kalisz, 2001). If using the estimated zygotic selfing rate, it is higher than 65–90 % of the species. Whether our rates are higher than most species in the New Zealand flora is more difficult to determine as few studies have measured the selfing rates of species in New Zealand. A review of New Zealand plant species suggests that approx. 21 % of species are self-incompatible and approx. 21 % are apomictic or autonomously selfing, leaving 58 % of the species as possibly mixed-mating (Newstrom and Robertson, 2005). Despite this high percentage, general conclusions are difficult to draw given that only three species have been examined (Barrell et al., 1997; Schmidt-Adam et al., 2000, 2009; Jesson et al., 2006; Howell and Jesson, 2013). However, considering that pollen limitation is higher in New Zealand species and many of the plants can have mixed-mating systems, higher selfing rates in New Zealand compared with elsewhere might be expected.

There are several types of selfing, including autonomous within-flower self-pollination, facilitated within-flower self-pollination and between-flower self-pollination (geitonogamy; Lloyd and Schoen, 1992). Our results from the pollinator exclusion treatment suggest that autonomous self-pollination is effectively zero, indicating that selfing does little to ensure reproduction in this species. Facilitated self-pollination within flowers should also be low due to the presence of both temporal and spatial separation between male and female function. Between-flower selfing, on the other hand, may be quite frequent due to trees having thousands of flowers open at one time and pollinators visiting multiple flowers before leaving a tree. The limited data we have on the number of flowers visited within trees from Wenderholm in 1996 (n = 51 tui visits; average of 32·7 ± 8·4 s.e. flowers per tree) suggest that considerable geitonogamy could be occurring. Comparing the selfing rates of emasculated (only between-flower selfing) and non-emasculated flowers (within- and between-flower selfing) would measure how much selfing is due to geitonogamy vs. autogamy. Regardless of the type of selfing, it appears that selfing is not a guaranteed way to reproduce in the absence of pollinators in this species.

The absence of a negative correlation between selfing rates and visitation rates also suggests that reproductive assurance is not occurring in this species. If plants supplement pollen receipt through self-pollination when pollinators are rare, there should be a negative correlation between the selfing rate and visitation rates, which our data do not support. Other population characteristics may be more important in determining the selfing rate in this species, particularly mate availability (Delmas et al., 2015). Mate availability can potentially affect the selfing rate via two mechanisms. First, pollinator behaviour may differ when resources are abundant. Optimal foraging theory suggests that when resources are abundant pollinators should visit fewer flowers per plant (Charnov, 1976), which would decrease geitonogamous pollen transfer. Secondly, the presence of more pollen-producing individuals decreases the relative frequency of self-pollen in the population, which should decrease self-pollination rates. Unfortunately, most populations of S. microphylla are small (10–100 individuals), limiting the possibility of experimentally testing these hypotheses. However, Rakaia Gorge, the largest of the populations we sampled, had the lowest selfing rate, suggesting that population size may be important in determining selfing rates.

Inbreeding depression

Selfing could increase fitness if plants are pollen limited and inbreeding depression is not too high (Lloyd, 1992). Unfortunately, in S. microphylla, inbreeding depression is very strong, leading to most selfed offspring not surviving to adulthood. Our hand pollinations show strong inbreeding depression prior to seed dispersal and our comparison of seeds with adults shows additional strong inbreeding depression after seed dispersal. Calculating a cumulative inbreeding depression from the population averages yields δ = 1 − (1 − δ_{pre-dispersal}) ×(1 − δ_{post-dispersal}) = 0·95, i.e. selfed progeny have only 5 % of the fitness of outcrossed progeny. This estimate has two caveats. First, the decrease in seed production between outcrossed flowers and selfed flowers could be due to late-acting self-incompatibility (LASI) rather than pre-dispersal inbreeding depression. LASI could manifest at a variety of stages including pollen tube growth and successful fertilization, differential provisioning of seeds or selective fruit abortion (Seavey and Bawa, 1986). While the mechanisms and evolutionary pressures may be very different between LASI and early acting inbreeding depression, distinguishing which of the two is causing the difference is not easy (Seavey and Bawa, 1986). Regardless of which factor is causing our results, it does not change the result that selfed pollen does little to help with seed production.

The second caveat is that the method we used to estimate inbreeding depression in natural populations relies on several assumptions that may be violated in our populations (Ritland, 1990). In particular, the method assumes a constant inbreeding coefficient of adults, i.e. that the seedlings sampled will eventually reach the same F-value as the adults sampled. This assumption could be violated if inbreeding depression or selfing rates varied among generations. Given our disturbed habitats it is possible that both of these violations could have occurred. For example, the adults may have been produced when pollinators were more abundant, leading to lower selfing rates, which would explain their low F-values rather than high inbreeding depression. In this case, our inbreeding depression estimate would be upwardly biased. However, two results suggest that while the estimate calculated here might not be precisely correct, inbreeding depression is high in this species. First, a common-garden experiment measuring inbreeding depression for the first 11 years resulted in similarly high values (Robertson et al., 2011). Secondly, the juveniles in our study, which should be from a similar pollination and selective environment to the seeds, show similar F-values to adults, suggesting that the reduction in F-values is due to inbreeding depression and not changes in the selfing rates. Thus, we feel it is safe to say that inbreeding depression is very strong in this species.

This strong inbreeding depression starts early and continues through the juvenile stage. As found in a previous study (Robertson et al., 2011), our hand crosses showed that inbreeding depression started prior to seed dispersal. The dramatic selfing rate differences between the Ahuriri seeds and seedlings show that inbreeding depression is strong even over this short growth period. Strong, early-acting inbreeding depression is common in long-lived species (Charlesworth and Charlesworth, 1987; Husband and Schemske, 1996; Scofield and Schultz, 2006). Various reasons for this have been proposed, including: small fitness differences between selfed and outcrossed individuals accumulate over multiple reproductive bouts to higher lifetime fitness; a longer pre-reproductive growth period increases cumulative inbreeding depression; and selfing is more costly due to between-season seed discounting (reviewed in Petit and Hampe, 2006). This early inbreeding depression means that by the time juveniles are established, most of the selfed individuals have died.

Conservation implications

Declines in bird pollinators caused by human impacts appear to have reduced seed quantity in S. microphylla and, due to the high levels of selfing and the strong inbreeding depression, most of the seeds produced are effectively useless. If on average over half of the seeds produced are selfed and approx. 86 % of the selfed seeds die before adulthood, nearly half of the seeds produced in a population are doomed. We can incorporate these values into a measure of the effective pollen limitation, i.e. the reduction of fitness due to too little pollen from non-self sources, as [1 − ((1 − PLI) − (1 − PLI) × s × δ)]. Using the average PLI (0·57), average selfing rate (0·61) and average inbreeding depression (0·86), the average effective PLI increases to 0·80. Described another way, seed fitness could increase on average by 80 % if flowers were entirely outcrossed.

Because S. microphylla is a relatively common species with abundant nectar production and large floral displays that receive much pollinator attention, we suspect that bird declines will have had an even greater effect on other New Zealand species. Other less rewarding or less apparent species may have even higher selfing rates, especially if they can autonomously self-pollinate or occur at low adult densities. In addition, inbreeding depression in New Zealand trees is probably high, given that it is a trait common to long-lived species (Charlesworth and Charlesworth, 1987; Husband and Schemske, 1996; Scofield and Schultz, 2006). Therefore, we suggest that high selfing rates and high inbreeding depression could lead to many offspring being low quality in New Zealand.

More broadly, our results suggest that measures of seed production or seedling densities may be a gross overestimate of the effective offspring production. This could lead to cryptic recruitment failure, i.e. a decline in successful reproduction despite high progeny production. Given the global extent of pollinator declines, cryptic recruitment failure may be widespread. Future studies measuring the impact of pollinator declines should consider both offspring quantity and quality when determining population health and making conservation recommendations.

SUPPLEMENTARY DATA

Supplementary data are available online at www.aob.oxfordjournals.org and consist of Table S1: details of the populations studied.